Cognitive Vulnerability to Major Depression: View from the Intrinsic Network and Cross-network Interactions

Abstract

Although it is generally accepted that cognitive factors contribute to the pathogenesis of major depressive disorder (MDD), there are missing links between behavioral and biological models of depression. Nevertheless, research employing neuroimaging technologies has provided some elucidation of neurobiological mechanisms related to cognitive vulnerability factors, especially from a whole-brain dynamic perspective. In this review, we integrated well-established MDD cognitive vulnerability factors and corresponding neural mechanism in intrinsic networks under a dual-process framework. We propose that the dynamic alteration and imbalance among the intrinsic networks, both in the resting-state and the rest-task transition stages, contribute to the development of cognitive vulnerability and MDD. Specifically, we propose that abnormally increased resting-state default mode network (DMN) activity and connectivity (mainly in anterior DMN regions) contributes to the foundation of cognitive vulnerability. Furthermore, in the period of rest-to-task transition when subjects confront negative stimuli, the following three kinds of aberrant network interactions have been identified as facilitators of vulnerability and dysphoric mood through different cognitive mechanisms: DMN dominance over the central executive network (CEN), an impaired salience network (SN)-mediated switching between DMN and CEN, and ineffective CEN modulation of the DMN. Focus on interrelated networks and brain activity changes between rest and task states provides a neural-system perspective for future research on depressive vulnerability and resilience, and potentially may guide the development of new intervention strategies for MDD.

Introduction

Major Depressive Disorder (MDD) is a common, recurrent, and severe psychiatric disorder and a leading source of disease burden, ¹ with a lifetime prevalence of 16.2% in adults and a prevalence of 11.7% among 13-18 year old adolescents.^{2, 3} MDD is characterized not only by persistent negative mood, lack of motivation, but also by the maladaptive thinking styles and specific impairments in information integration. Additionally, those cognitive disturbances present both in people suffering from depressive disorders, and in people with elevated negative mood. Therefore, over the years, researchers have explored risk factors and potential treatment approaches of MDD from both psychological and biological perspectives.

Cognitive factors, which have been highlighted in all psychological models, suggest that the interaction of stress and these premorbid vulnerabilities contributes to depressive episodes throughout the life span.⁴ Recently, researchers suggested that cognitive vulnerability may represent an endophenotype for depression.⁵ At the same time, cognitive vulnerability models include a broad array of risk factors, such as Beck's cognitive model (e.g., information-processing biases, negative schemas),⁶ hopelessness theory (e.g., depressogenic attribution styles to event causes, consequences, and the self) ,⁷ and response styles theory (e.g., rumination).⁸ In light of the array of vulnerability factors, researchers have sought to explore a new model that integrates the core factors of prominent theories in order to provide a more holistic understanding of cognitive vulnerability.

The dual-process theory of cognitive vulnerability, which has been supported by many behavior experiments ^9-12, integrates the main vulnerability factors into one model from an information processing perspective¹³. According to this model, there are two information processing modes correlated with the cognitive vulnerability to depression: one is called associative/implicit mode characterized by quick, effortless processing based on the well-learned associations, and the other is reflective/ explicit mode characterized by slow, effortful processing which required more intention and awareness.^9,14 Dual process theory emphasizes negatively biased self-referential associative processing as the foundation for cognitive vulnerability, which may be overcome by corrective reflective processing. Additionally, it raised three scenarios of interplay between associative and reflective processing leading to a downward spiral into more severe forms of dysphoria (see Fig. 1). Therefore, the dual-process theory provides a relatively parsimonious framework of cognitive vulnerability with emphasizing the dynamic imbalance between two cognitive processes in the generation and persistence of negative mood. At the same time, this model integrates the core factors of prominent cognitive vulnerability theories, such as the negative information processing biases and negative self-schemas Beck's cognitive model, the depressogenic attribution in hopelessness theory, and the rumination in the response styles theory.¹⁴ However, there remains missing link between the symptom-behavior level psychopathological theory and the associated biological underpinnings of depressive vulnerability.

Fig 1.

Aberrant intrinsic network interaction model that integrates the dual-process model of cognitive vulnerability, which adapted from Beevers (2005)⁷. The dual-process model includes two modes of information processing: (i) an associative mode involving quick, effortless processing; and (ii) a reflective mode involving slow, effortful processing. We propose that abnormally increased resting-state DMN activity and connectivity and the corresponding biased associative processing (depressive rumination) contribute to the foundation of cognitive vulnerability. In the period of rest-to-task transition, aberrant network interactions contribute to three scenarios in which an associative bias cannot be corrected, thus promoting cognitive vulnerability. These three scenarios are: (i) DMN dominance over the CEN (cognitive resource depletion); (ii) abnormal SN switching between the DMN and CEN (associative bias does not violate internal expectancies and trigger reflective processing); and (iii) failure to activate the CEN effectively (reflective processing being triggered but failing to accurately adjust associative bias). As a result, a feedback loop between negative bias and dysphoria occurs, leading to a downward spiral. The solid lines in the part of triple network model indicate the enhanced interaction, while the dotted lines indicate the attenuated interaction.

Advances in brain imaging, especially in the field of intrinsic neural network research, may provide a useful tool to identify the missing neural-behavioral links (see Glossary). Over past two decades, neuroimaging researches using affective or cognitive tasks to study depressive vulnerability have shown that altered cortico-limbic connectivity contributes to the generation of excessive and persistent negative affect.^15,16 Additionally, as a great number of evidences have showed the functional importance of spontaneous BOLD activity on task-related responses and highlighted its predictive nature for subsequent behavioral and mental states,^17-20 resting-state functional magnetic resonance imaging (fMRI) has been increasingly utilized to probe cognitive vulnerability factors implicated in depression.²¹ Specifically, aberrant resting-state activity and functional connectivity (FC) in cortical midline structures have been reliably observed in at-risk,^22,23 depressed,^24,25 and formerly depressed populations.^26,27 Moreover, abnormality in the default-mode network (DMN), a principal intrinsic brain network including cortical midline structures, is increasingly associated with cognitive vulnerability among at-risk and depressed individuals.^28,29 Other large-scale neural networks, such as the central executive network (CEN) and salience network (SN), also seem to play a role in biased attention and processing of affective information in depression.^30-32 Thus, understanding how there might be simultaneous dysfunction in multi-networks (see Box) may elucidate the neural nature of cognitive vulnerability.^33-35

The intrinsic network research and the triple network model.

The intrinsic network research originally raised from the resting-state brain activity or connectivity studies. After the identification of “default mode” network by Raichle and Greicius, ^50-51 which exhibits high levels of activity at rest and become deactivated at the task needing specific goal-directed behavior, Fox et al. raised a model in 2005 which included two tightly locked but anti-correlated networks namely the Task Negative network (TNN) and Task Positive network(TPN).⁵² In 2007, Seeley et al demonstrated that the TPN is actually composed of two dissociable networks: an executive network (EN) and a salience network (SN).⁵³ Based on the previous studies, Menon et al. propose a triple network model consisted with Default-mode network (DMN), Salience network (SN), and Central executive network (CEN) as the ‘core’ neurocognitive networks, especially for investigating the psychopathology in psychiatric and neurological disorders.³³ The triple network model is now one of the most popular utilized model to understanding the neural bases of neuropsychiatric disorder and therefore used in this review:

DMN comprises mainly the medial prefrontal cortex (mPFC) and adjoining ventral anterior cingulate cortex, the posterior cingulate cortex (PCC), bilateral inferior parietal cortex (IPC) and the middle temporal lobe (MTL).⁵⁰ The DMN is involved in self-referential/internally directed information processing and typically deactivated during external stimulus-driven cognitive processing.⁵⁴

CEN comprises primarily the dorsolateral prefrontal cortex (DLPFC) and posterior parietal cortex (PPC) and response for high-level cognitive functions. The CEN is involved in control processes during goal-directed/externally oriented tasks and regulation of emotional responses, particularly mediated via the DLPFC.^53,55

SN is anchored in the anterior insular cortex (AI) and dorsal anterior cingulate cortex(dACC), as well as includes two key subcortical structures-the amygdala and substantia nigra/ventral tegmental area (SuN/VTA), which are important for detecting emotional and reward saliency.³⁶ The SN is involved in detecting and orienting to both external and internal salient stimuli and events.⁵³ Particularly, the AI is critically involved in maintaining and updating representations of current and predictive salience1, and contribute to appropriate behavioral responses to salient stimuli via switching between DMN-related self-referential and CEN-related goal directed cognitive activity.^37,56

Although originally being identified from the resting-state fMRI data, the CEN, SN, and DMN can also be readily identified across an extremely wide range of cognitive tasks. The responses of those networks increase and decrease proportionately, and often antagonistically, with general external cognitive task demands. For instance, the CEN and SN typically show increased activation during external stimulus-driven cognitive and affective information processing, whereas the DMN shows decreased activation with tasks in which self-referential processing is not required.^50-51 However, DMN and CEN also can positively couple when organizing internal self-relevant thought (i.e. the autobiographical planning).^40-42 Dynamic engagement and disengagement of those core neurocognitive networks is prominent in many cognitive tasks.^37,57-58

According the studies in healthy individuals, the CEN typically show increased activation during external stimulus-driven cognitive and affective information processing, whereas the DMN shows decreased activation with tasks in which self-referential processing is not required.^48-49 In addition, emerging evidences showed that the SN, especially the right fronto-insular cortex (rFIC), is responsible for switching DMN-CEN interactions.^36-39 Briefly, during the externally-oriented processing there is typically anti-correction relationship between DMN and CEN, which is regulated by the SN (see Fig.2).^40-42 Since vulnerability-stress models of depression emphasize an interaction of cognitive diatheses with negative life events leading to the development and onset of depressive episodes,⁴ investigating abnormal brain responses to emotional or cognitive stimuli in the rest-task transition period (from an internal oriented processing state to an external stimulus/task induced processing state) are especially important to understanding the emergence of cognitive vulnerability and depression. In that rest-task transition section, we focus on the role of DMN suppression in support of externally-oriented cognition to negative life events, and the dynamic interaction among three intrinsic networks. Previous studies showed that the patients with MDD, the at-risk individuals, and the remitted depression subjects consistently showed the abnormally connectivity and interaction between the DMN and CEN coupling with the dysregulation of SN when processing the external cognitive or affective stimuli.^27,43-44 Therefore, the resting-state-related prominent abnormality of DMN, along with an imbalance among intrinsic networks in the rest-task transition period, has been proposed as the specific neural susceptibility feature of depression.

Figure 2.

Schematic figure of triple network model showing the SN-induced coordination between DMN and CEN. According to this model, SN (blue) mediates the ‘switching’ between activation of the DMN (yellow) and of the CEN (green) to guide appropriate responses to salient stimuli.³⁶ Salience signals are integrated in AI of SN, then causally influence signals in the DMN and CEN, which support internally directed and externally directed cognition, separately.^37,58 In light of recent work that suggests the existence of distinct functional subdivisions within the insular cortex^147-148, the right AI is now thought to be the specific brain region that assists in switching between networks. The dots showed the key nodes of each network in triple network model: mPFC: medial prefrontal cortex; PCC: posterior cingulated cortex; dACC: dorsal anterior cingulated cortex; AI: anterior insular cortex; DPLFC: dorsolateral prefrontal cortex; PPC: posterior parietal cortex. Modified from Uddin (2014); Uddin & Menon (2009). ^39,149

The aim of this review therefore is to synthesize current neuroimaging research to reinterpret the formation of cognitive vulnerability and subsequent depression from the intrinsic organization and cross-network interaction perspective. Based on the dual-process model, we will describe how abnormal activities and interactions among DMN, CEN, and SN can provide insights into the emergence of cognitive vulnerability and MDD. Specifically, as first proposed by Sheline et al ^45-46 and subsequently discussed by multiple studies, ^{24, 28, 47-49} the hyperactivity in the DMN leads to a negative association processing bias commonly associated with negative self-referential processes, which forms the neural foundation of cognitive vulnerability to depression. In addition, abnormal cross-network interactions during rest-task transitions, which includes three scenarios, lead to the failure to correct negative associative processing bias, thereby facilitating and stabilizing the cognitive vulnerability (see Fig.1). We will describe these mechanisms in detail in the next sections. Though the functional connectivity and network researches are not the direct proxy for anatomic/structural connectivity, this integrative framework extend beyond the phenomenological dimensions and local functional impairments, suggests a new research directions addressing both behavioral and neuronal features in cognitive vulnerability and potentially informs the development of targeted interventions for MDD.

Aberrant resting-state DMN and associative processing bias

The dual-process model suggests that negatively biased self-referential associative processing is the foundation for cognitive vulnerability to depression.¹⁴ Though many vulnerability factors are associated with biased self-referent information, maladaptive rumination plays a primary role in mediating the internalization of negative self- representations and onset of depression symptoms.⁵⁹ Research suggests that maladaptive rumination is characterized by spontaneous narrowly-focused, self-referential associative processing.⁶⁰ In addition, there is a strong overlap between the DMN and the cortical network that mediates associative processing on medial prefrontal cortex (mPFC), medial temporal lobe (MTL), and medial-parietal cortex (MPC).⁶¹ Recently, increasing evidence suggests that maladaptive rumination-related resting-state DMN activity associated with the generation of cognitive vulnerability.⁴⁸ Specifically, Bar states that the neural basis of rumination is resting mPFC hyperactivity coupled with increased FC between mPFC and anterior cingulate cortex (ACC). This increased FC, in turn, suppresses MTL activity and constrains the normally broad associative processing and the generation of a positive mood, leading to ruminative processes and negative mood states.⁶⁰

Neuroimaging studies have supported Bar's hypothesis that the persistent hyperactivity in mPFC and its neighboring ventral ACC regions, as well as their relationship with the MTL play a core role in the cognitive vulnerability of depression. Berman et al. examined the relationship between DMN FC and rumination both at rest and in a short-term memory task.²⁸ During the rest period, MDD participants exhibited increased DMN connectivity in the subgenual ACC (sgACC), which correlated with self-report rumination and hyper-connectivity between the mPFC and MTL. Similarly, a study on first-episode unmedicated MDD subjects found significantly greater FC between the rostral ACC (rACC) and the parahippocampal gyrus at rest.⁶² Recently, Sambataro et al. investigated the function of multiple DMN subsystems at rest in MDD and showed increased connectivity, increased low-frequency band spectral power in the mPFC, sgACC, and rACC, and aberrant interactions between the rACC and hippocampus.⁶³ Moreover, using an unsupervised machine learning approach with a network derived from resting-state fMRI data, including sgACC, mPFC, and super temporal gyri, Zeng et al. demonstrated a highly discriminative power for accurate identification of MDD.⁶⁴ Schilbach's study using a meta-analytically informed network analysis also showed that sgACC as a hub of resting-state hyper-connectivity in the introspective socio-affective network in depressive individuals.⁶⁵ Thus, converging evidence from a number of brain imaging studies suggests that the generation of maladaptive rumination are associated with the resting mPFC hyperactivity, and the hyper-connectivity between mPFC and ventral ACC, as well as mPFC and MTL.

Bar's hypothesis has been further supported by consistent evidence of enhanced FC among anterior DMN regions in MDD patients. Using independent component analysis, Greicius et al. first explored the DMN abnormalities in major depression and suggested that the DMN functional connectivity in depression, which associated with the ruminative nature of depressive subjects, was disproportionately driven by the increased activity in the sgACC.⁶⁶ Importantly, multiple other papers have replicated the basic findings of that study.^45,67-68 In addition, using multivariate Granger causality analysis of resting-state data, Hamilton et al. found that mPFC and sgACC activities were mutually reinforcing in MDD, and that the mPFC-to-sgACC connectivity correlated with levels of depressive rumination.⁶⁹ In 2012, we provided the first direct evidence of increased resting-state FC in anterior medial cortex regions (especially the mPFC and rACC) correlating with rumination scores from first-episode, treatment-naive MDD patients.²⁴ That finding has been supported by studies in adolescents experiencing first-episode MDD.²⁵ Using a unique 4-year follow-up sample of children with a history of preschool-onset MDD, Gaffrey et al. also demonstrated abnormal sgACC FC and found correlations between dysregulated emotional behavior and sgACC/mPFC connectivity.⁷⁰ The sgACC and rACC formed the affective division of ACC, which is activated by tasks with affective or emotional content and also deactivated by cognitively demanding tasks.⁷¹ Hence, the joint hyperactivity in those DMN nodes may have a critical role in formation of cognitive vulnerability by leading to ruminative processes.

Neuroimaging evidence from subjects at-risk for MDD and non-clinical populations provides additional support for the supposition that DMN is involved in ruminative processes and contributes to depressive vulnerability. Norbury et al. examined resting-state fMRI data from 15 healthy at-risk participants having a biological parent with MDD and 15 healthy controls. Relative to the controls, the subjects with positive family history showed significantly greater DMN connectivity in the left dorsomedial PFC, left MTL, OFC, and left precuneus. The authors suggest that this pattern may represent a ruminative vulnerability.²² Similarly, Kross reports that rumination induction in healthy subjects activated a network including sgACC and mPFC, and this network activity was greater during induced rumination than during “analyze” or “accept” conditions.⁷² Moreover, activity in this network correlated positively with increases in negative affect during induced rumination. Recently, Liu and Fielder provided new evidence independently for the altered resting-state anterior DMN connectivity in healthy siblings of MDD patients⁷³ and in women with subclinical depression but without a history of MDD.²³

In summary, emerging evidence indicates strongly that elevated resting-state activity in anterior DMN regions and FC among them is associated with rumination: a narrow associative thinking pattern and cognitive vulnerability factor in depression. However, there are two points worth mentioning. First, we propose that abnormal anterior DMN activity at rest may only reflect brooding, a subcomponent of rumination, which involves negative self-focus and is hypothesized to be more strongly associated with depressive symptoms both in concurrent and longitudinal analyses..⁷⁴ The other subcomponent of rumination-reflection, more likely relates to reflective processing in the dual-process model because that is a purposeful turning inward to alleviate one's depressive symptoms. ^14,74-75 Indeed, Hamilton et al. reported that greater DMN activity in MDD patients at rest was associated with higher levels of brooding and lower levels of reflection.⁶⁹ Berman and his colleagues also reported a similar result.¹⁷ Second, there are some discrepant studies in the literature, which might be due to the differences in sample characteristics or analysis methods, such as using the priori defined ROI correlation approach vs. whole-brain analyses,^76-77 using ICA at the group level vs. on individual data sets,⁷⁸ and applying a newly developed method.⁷⁹ Deployment of more uniform methodologies may help resolve these discrepancies.

Aberrant network interactions during rest-task transitions: Three scenarios DMN dominance: Cognitive resource depletion

Although a persistence of resting-state DMN hyperactivity and heightened spontaneous rumination are core features of cognitive vulnerability, this state may be transient and quickly shift toward task-based deactivation to support stimulus-driven goal-directed cognitive processes.

In other words, the resting-state ‘imbalance’ situation within intrinsic networks may stop and the development of cognitive vulnerability and dysphoric mood may be suspended.⁸⁰ However, patients with MDD consistently show both lack of DMN suppression and impaired performance during task-induced states, indicating that the DMN dominate at rest sustains, and the residual activity interfere the subsequent attentional and cognitive processing. Johnson's study provided direct evidence that patients with MDD showed less DMN deactivation during externally oriented cognitive processing and, importantly, that effect was positively correlated with rumination.⁸¹ Hence, the author suggested that depression pathology may involve a difficulty with disengaging from self-reflection. Grimm et al. investigated negative BOLD responses (NBRs) in DMN regions during the presentation of emotional pictures and reported reduced rest-to-task attenuation of DMN activity in MDD patients Importantly, that reduced attenuation was associated with strong feelings of hopelessness.⁸² Furthermore, Mitterschiffthaler et al. reported significant engagement of rACC and right precuneus in MDD subjects during an emotional Stroop task, and the rACC activation in MDD subjects was positively correlated with impaired performance on trials with negative words.⁸³ A hyperactive DMN has also been shown to result in impaired cognitive performance in working memory task, passive viewing or implicit recognition tasks for negative images.^44,84-85

Based on the above neuroimaging studies, not only the elevated resting-state activity of DMN, but also the increased DMN activity (less suppression) during task-induced states plays an important role in the pathophysiology of MDD. The latter may be associated with the first scenario in dual processing developing to cognitive vulnerability (Fig. 1): reflective processing may not override the negatively biased associative processing when some cognitive load disrupts, and then depressive vulnerability takes place. Using a series of behavior tasks, Wenzlaff and colleagues demonstrated that view in remitted and currently depressed individuals.⁸⁶ Specifically, although remitted patients did not have a negative bias at baseline, their biased performance under a cognitive load condition was predictive of future depressive symptoms. In addition, behavioral,⁸⁷ pupillary,⁸⁸ and electrophysiological⁸⁹ data support the notion that depressed individuals engage in increased intrinsic processing (e.g. focusing on negative automatic thoughts or rumination), which potentially taxes resources that would otherwise be allocated to a cognitive task. Recently, Nixon et al. reported the first direct evidence of persistent excessive DMN activity during task within remitted MDD, which supported that the task-based DMN hyperconnectivity may result in cognitive and attentional bias and increase the risk of depression.⁹⁰ Therefore, the DMN dominance may be the neural bases for the role of cognitive resource depletion in developing vulnerability and depression.

Recent work on DMN-CEN anti-correlation provides potential evidence for this mechanism. The optimal attunement between DMN and CEN is thought to reflect efficient intrinsic brain organization.^52,91 During processing the externally directed stimuli, the deactivation of DMN and the increased activation in CEN show concurrently and timely for ensuring the appropriate and successful response.^53,57 However, in the rest-task transition period, both depressive and at-risk individuals show abnormal DMN dominance, which induces decreased DMN-CEN anti-correlation and subsequently depletes cognitive resources.⁹² This phenomenon may arise in two ways. In the first possibility, exaggerated spontaneous DMN activity may persist following the rest-task transition, reduce recruitment of cognitive control resources, and prevent distraction from rumination.^25,93 This DMN persistence is evidenced by the rumination-related lack of DMN suppression in a non-self- referential task.⁸¹ Similarly, using a new quantitative method to compute the extent to which DMN activity levels exceed task-related network activity over the course of the resting state scan, Hamilton et al. showed that increasing levels of DMN dominance were associated with higher levels of brooding and lower levels of reflection in MDD.⁶⁹ In addition, Davey et al reported that the sgACC-ventromedial PFC (vmPFC) connectivity in MDD increased in resting state, and this change in FC could predict the relative activity of CEN regions, thus suggesting that dysfunctional connectivity with sgACC may influence executive/attentional processes.⁹⁴ Thus, the attenuated DMN-CEN anti-correlation stemmed from abnormal DMN dominance in the rest-task transition may induce cognitive resource depletion for developing vulnerability to depression.

Additionally, abnormal DMN dominance in MDD may arise from the automatically, effortlessly negatively biased information processing at a preconscious level, which activate the DMN and attenuate the influence of normally distracting cognitive processing. Many studies related to negative stimulus processing mentioned above support this mode of DMN dominance development.^{44, 83, 95} Such studies consistently reported heightened responses in amygdala, along with the higher activity in DMN, and lower activity in CEN. Moreover, spontaneous activity in amygdala has been shown to positively predict the activity of mPFC, and negatively predict the activity of middle frontal gyrus.⁹⁶ A longitudinal study further showed that amygdala-vmPFC connectivity was predicted by early life stress, and associated positively with depressive symptoms in female adolescents.⁹⁷ A recent study further reported the increased sustained amygdala reactivity involving the involuntary processing of salient negative information associated with all the dimensions of trait rumination.⁹⁸ Thus, aberrant DMN dominance during prior processing of negative emotional stimuli in MDD may be caused by excessive input from of the SN nodes, and associated with aberrant attention allocation and switching. We will interpret the related mechanisms in the next section.

In summary, the DMN-dominant state during rest-task transition may arise from the persistence of resting DMN activity and/or quick activation of the DMN during negative stimulus processing. This persistence depletes cognitive resources and leads to a sustained negative cognitive bias, which then facilitates the development of cognitive vulnerability to depression.

Impaired SN switching between DMN and CEN: Expectancies are not violated

In Menon's triple network model, the SN, which anchored in dorsal anterior cingulate cortex (dACC), FIC, amygdala, and ventral striatum, is involved in bottom-up detection of salient events and switching between other large-scale networks to facilitate appropriate allocation of attentional and cognitive executive resources when a salient event is detected.³³ Using the Granger causality analysis method in both of resting-state and task-state data, Sridharan et al. discovered that the rFIC-ACC network, especially the rFIC, plays a causal role in DMN-CEN switching.³⁷ Recently, Goulden et al independently validated the role of SN driving the switching between DMN and CEN by applying a novel technique in resting-state data.³⁸ Another study involving a combination of fMRI and diffusion tensor imaging data also supported a critical switching role of rFIC between brain networks for complex, flexible cognitive processing.⁵⁸ Furthermore, through transcranial magnetic stimulation (TMS) combined fMRI method, Chen and colleagues successfully demonstrated that single-pulse TMS of an SN site enhanced both within-SN and within-CEN connectivity, whereas TMS stimulation of a CEN site enhanced connectivity only within-CEN psycho-physiological interaction connectivity.⁹⁹ In summary, the SN switching role between DMN and CEN (see Fig. 2) has been strongly supported by converging evidences in healthy subjects and highlight the possibility that the atypical engagement of SN is a feature of neuropsychiatric disorder.³⁹

Up to date, a growing body of evidence revealed that SN switching function were impaired in MDD and at-risk individuals.^100-101 For instance, medication-naive adolescents with MDD showed reduced activation in the FIC and dorsolateral PFC (dlPFC) during an attention switching task.¹⁰² Strigo et al. further demonstrated that altered rFIC function, along with increased vmPFC/rACC and decreased dlPFC activity, in MDD is associated with an impaired ability to effectively prepare for environmental changes.¹⁰³ Similarly, Hamilton et al. examined the rFIC response during initiation of ascents in DMN and CEN activity and showed that rFIC plays a differential role in DMN/CEN dominance switching in MDD subjects versus controls.⁶⁹ This model is further supported by Manoliu et al.'s recent study.¹⁰⁴ In addition, through comparing task-based brain function of individuals with high or low-risk to develop MDD, Peterson et al. reported a neural system as the risk endophenotype for MDD, which including insular and other nodes in cortical attention circuits.¹⁰¹ Recently, Kaiser et al compared the neural activation during emotion-word and color-word Stroop tasks in participants with subclinical depression. They suggested that affective interference stems from the increased salience of negative emotional information, coupled with the difficulty in shifting resources toward the external environment. That study further supported the abnormal switching function of SN between DMN and CEN directly in vulnerable individuals.⁴⁴

Besides the direct measurement of activity or functional connectivity of SN nodes in MDD and vulnerable individuals, a great number of studies have showed that the abnormal switching function of SN are associated with mood-congruent negative response biases, a well-known vulnerability factor being emphasized in Beck's cognitive model of depression.¹⁰⁵ Specifically, the amygdala contributes to the biased processing of both explicitly and implicitly negative emotional stimuli in at-risk depressive individuals,^31,106-107 MDD patients,^109-111 and remitted MDD subjects.¹¹² The anterior insula and dACC also participate in the formation of these biases.^113-114 A recent fMRI study showed that MDD patients develop a pessimistic attitude towards the emotional meaning of external events, and this attitude was associated with abnormal mPFC, dlPFC, and insular cortex activations.¹¹⁵ Therefore the misinterpretation of external cues by the insula due to unpleasant previous experiences may be a central underlying mechanism of depression-related dysregulation. Similarly, in a voxel-wise, whole-brain meta-analysis study, Hamilton et al. showed that MDD patients had reliably elevated activity in the amygdala, insula, and dACC, but decreased dlPFC activity, during negative emotional stimulus processing.¹¹⁶ The authors suggested that in depression, negative information could induce a heightened neural response in the SN but fail to be propagated to the CEN for contextual processing and reappraisal. Indeed in 2008, a pathway-mapping analysis provided direct fMRI evidence for that: regulation of emotion through amygdala to ventrolateral PFC could predict reappraisal failure and more sustained negative emotion.¹¹⁷ To sum up, several core regions in SN are critical involved in the formation of mood-congruent negative response biases and subsequently changed the CEN activity associated with the cognitive control processing.

The aforementioned SN-related mood-congruent negative bias processing could play a role in the second scenario of failing to correct biased associative processing in the dual process model (see Fig. 1): when the results of associative processing are congruent with one's negatively internal expectancy (e.g. negative self-referential schemas or representations related to one's self-worth), expectancies will not be violated. However, when healthy individuals with optimism biases confront negative external stimuli, cognitive conflict would be recognized automatically, and then attention would be disengaged from the negative thoughts. Previous studies showed that conflict signaling was reduced in individuals with negative thoughts/mood due to their mood-congruent attention and memory biases.¹¹⁸ Accordingly, in individuals with risk of depression, attentional resources could not reallocate appropriately to disengage from the situation, but continue to focus on self-referring negative information.¹¹⁹ At the same time, observations in the healthy individuals with high brooding scores show that more attentional control is needed to successfully disengage from negative information.¹²⁰ Therefore, the ‘impaired disengagement’ of attention based on the mood-congruent negative bias in MDD would facilitate not only sustained maladaptive rumination, but also impaired cognitive performance. In summary, we propose that in the context of mood-congruent negative bias processing, impaired SN switching between DMN and CEN underlies, at least in part, the cognitive vulnerability to depression.

Failure to effectively activate CEN: Reflective processing does not adequately adjust biased associative processing

In the third scenario of the dual-process model, which may involves a failure of CEN activation (see Fig. 1,), it is posited that cognitive vulnerability to depression is facilitated when reflective processing does not adequately adjust output from associative processing. In other words, when an individual tries to regulate emotion voluntarily (i.e. reappraise) through reflective processing but failed, an associative bias is maintained or even amplified. There are two cognitive vulnerability factors that are likely involved in emotion regulation-related reflection processes: reflective rumination⁷¹ and depressogenic attributional styles⁷ (i.e., regarding the self, cause, and consequences of negative events) based on the hopelessness theory. When vulnerable individuals confront negative events, both of those factors will guide them to spontaneously use emotion regulation strategies, which are directed against the increase negative emotion stemmed from elaboration on the meaning of unpleasant events. However, failure to regulate will lead the individual to fall into more negative cognition, depressive rumination and dysphoria.

Emotion regulation deficits are a central characteristic of depressed individuals. Though previous studies showed that different types of emotion regulation strategies (e.g., automatic vs. voluntary, ditraction vs. reappraisal) are associated with different neural mechanisms, both strategies are subserved by common “top-down” control areas in CEN to effectively allocate attention to the external stimuli, especially the dlPFC node. ^121-122 Furthermore, numerous studies supported that aberrantly voluntary emotion regulation in MDD was related with hypoactivity in CEN nodes.^123-124 However, recent evidence has also pointed to hyperactivity in DMN components. Smoski et al. found that remitted MDD showed significantly greater rACC activation and simultaneously decreased midfrontal gyrus activation during the reappraisal of sad images contrasted with attending to sad images.¹²⁵ Similarly, Sheline et al. reported that when subjects with MDD reappraised negative pictures actively, widely distributed elements of their DMNs failed to deactivate.⁴⁶ Erk and Johnstone also found that MDD patients show significantly diminished responses in dlPFC, but increased activation in DMN nodes, when performing an effortful reappraisal task for negative emotion.^126-127 In a small study including only 12 MDD patients, Dillon et al. reported recently that dlPFC activity correlated inversely with depressive severity, though no group differences in reappraising negative picture were found.¹²⁸ Another follow-up study of remitted MDD participants employing sad mood provocation has added new evidence showing that expansive mPFC reactivity can predict subsequent depressive relapse over an 18-month follow-up period.¹⁶ The authors linked mPFC reactivity in remitted MDD participants to inefficient recruitment of PFC in attempts to cognitively regulate negative emotion. Recently, two studies in remitted MDD reported similarly reduced prefrontal cortex reactivity during negative feedback, which associated with the rumination, possibly impaired the adaptive reappraisal of negative experience.^129-130 Thus, emerging evidences suggested that during the cognitive regulation of negative events, the failure to regulate and the generation of depressive vulnerability are not only associated with the decreased activation in CEN nodes, but also with increased activation in DMN nodes.

Since CEN hypoactivity and DMN hyperactivity could be occurring simultaneously during impaired cognitive regulation in MDD, it raised a question that the attenuated CEN modulation is a primary effect of CEN hypoactivity or a secondary effect of interference from DMN activity. In a recent study combining TMS with fMRI, Chen et al. reported direct evidence of a causal regulatory relationship primarily from the CEN to the DMN.⁹⁹ Specifically, single-pulse excitatory stimulation of a CEN node (posterior middle frontal gyrus) induced negative DMN connectivity with the CEN whereas inhibitory repetitive TMS to the same stimulation site induced the disinhibition of DMN activity on mPFC. Two other TMS study presented further evidences that rTMS on DLPFC could modulate interactions between the DMN and CEN by normalizing the depression-related DMN hyperconnectivity.^131-132 As the result, the more anti-correlation between two networks were induced, the better clinical efficacy of the TMS was found. Interestingly, a recent double-blind 6-month trial examining changes in the neural circuitry involved in emotional regulation after antidepressant treatments showed that only changes in CEN nodes (dlPFC and Brodmann area 10) that are engaged while subjects perform a negative affect regulating task correlate with changes in depression severity over the following 6 months.¹³³

In summary, previous studies suggest that failure to effectively activate CEN and the corresponding reduced effect on DMN mainly contribute to the unsuccessfully voluntary regulation of negative emotion, thus allowing negative cognitive/affective responses to persist and facilitating the development of depressive vulnerability.

CONCLUSIONS AND FUTURE DIRECTIONS

In this review, we integrated well established MDD cognitive vulnerability factors and corresponding neural mechanisms based on the literature of intrinsic networks under the dual-process framework. We propose that abnormally increased resting-state DMN activity and connectivity (mainly in anterior DMN regions) and corresponding depressive rumination contribute to the foundation of cognitive vulnerability. Furthermore, in the period of rest-to-task transition, three kinds of aberrant network interactions may facilitate the occurrence of cognitive vulnerability. Specifically, when confronting negative life events or stimuli, DMN dominance (persisting due to increased resting activity and facilitated by the automatic biased information processing), abnormal SN-mediated switching between DMN and CEN (related to a negative schema and mood-congruent negative bias), and failure to effectively activate the CEN (related to reflective rumination, the negative attribution style, and the corresponding emotion regulation impairment). A focus on interrelated networks and brain activity changes between rest-task transitions provides an approach for future research into inter-individual differences in vulnerability and resilience. However, several outstanding questions remain and need to be explored in depth.

First, although there are an increased number of neuroimaging studies investigating cognitive vulnerability to depression, more systematic research is required to test and confirm the validity of our framework. For instance, longitudinal fMRI studies with comprehensive cognitive vulnerability factor assessment would allow us investigate the intrinsic network changes in predicting future depressive occurrence. To date, only a few studies have compared brain system activity among never depressed individuals with vulnerability factors, currently depressed individuals, and remitted depressed individual. Therefore, a more systematic research approach including both cross-sectional and longitudinal studies is needed to develop the models for mapping the neural patterns.

Second, the neurobiological underpinnings of network dysfunction and impaired interactions remain poorly understood. However, recently, some theoretical hypotheses have been raised to understand the complex dynamics between large-scale neural systems. For example, Anticevic et al. proposed a model for the synaptic mechanisms of altered DMN suppression and DMN-CEN interactions: the anti-correlation between CEN and DMN activities during task performance is derived from the reciprocal network interaction through net inhibitory long-range projections, which are related to disruption of NMDA conductance onto GABAergic interneurons.⁹² According this model, local disinhibition induces hyperactivity of DMN-type microcircuitry and hyposensitivity to long-range suppressive inputs from task-activated cognitive-related microcircuits, which precludes silencing the already high-firing DMN at task onset. The specific mechanisms of GABAergic/glutamatergic neural interaction, as well as their regulation on the brain network activity remains to be resolved in future research.

Third, although there is strong evidence for the clinical effectiveness of psychotherapy in the treatment of MDD, the neural underpinning of depression-specific psychotherapies remains unclear. Our framework may help to explore this issue. For instance, in cognitive behavioral therapy (CBT), patients are given explicit instructions on how to regulate their negative thoughts and emotions, which may increase CEN activity and decrease DMN activity, and thereby enhance their ability to complete reflective processing successfully.^134-136 Moreover, mindfulness therapy, which highlights present-moment awareness and acceptance, may alter activity in the SN, especially anterior insular activity, by decreasing the incongruence between the outcome of associative processing and the individual's expectations.^137-138 Based on the result of meta-analysis, Ma further proposed that there are different neuropsychological mechanisms for antidepressant medication and CBT: CBT targets prefrontal function by increasing inhibitory executive control, while the antidepressants may act more directly on the network associated with abnormal emotion generation/experience.¹²¹ Furthermore, a recent review proposed psychotherapy may facilitate recovery and plasticity at the brain network-level after antidepressant drugs reactivate a window of juvenile-like plasticity in the adult cortex.¹³⁹Ongoing composite research on cognitive vulnerability, treatment effects, and brain network-level plasticity is very promising.

Forth, there is evidence showing some overlap of neural dysfunctional processing among different disorders. For instance, similar DMN abnormalities reflecting internally-oriented attention and thought are found in both depression and schizophrenia.¹⁴⁰ Our diagnostic categories are heterogeneous and likely encompass multiple biologically distinct entities. Future work using the research domain criteria for brain system dysfunction research may provide valuable insights into which patients demonstrate these network-level abnormalities and how these relate across diagnoses.

Finally, it is worth noting that much work remains for relating network models explicitly to cognition and neural computation.^141-143 The intrinsic functional connectivity and network model undoubtly provides a unique and powerful tool to provide insight into the organization of distributed association brain networks, especially serves as a organizing framework for characterizing biological substrates of brain in mental disorders. However, there are some methodological limitations in formation or computation of intrinsic networks, such as the sensitivity of functional connectivity MRI to head motion and physiological artifacts^{144, 145}, and the effect of global signal regression on detecting anti-correlation between networks. ^57,146 Difficulties also exist with the definition of node/edge and the interpretation of network measures. As an indirect, relative measure of neural activity fluctuations, functional connectivity MRI measures have no proven biological interpretation. Therefore, more studies are needed to promote the translation from the network properties to the realities of behavior and neurobiology.

In summary, a growing body of neuroimaging researches revealed that cognitive vulnerability, the most generally accepted psychological risk factor to depression, is associated with neural functional abnormalities. Under a dual-processing framework, we integrated the MDD-related cognitive vulnerability factors in the context of an intrinsic network and cross-network interaction perspective. Although the hypothesis in this review has some speculative nature, it till provide a unique link between in vivo brain measurements and depression vulnerability, and then enhance our insight into the biological underpinnings in development, maintenance, and treatment of MDD. The integrative framework suggests a paradigmatic shift in cognitive vulnerability research and potentially informs the development of targeted interventions for MDD.

Glossary

Functional magnetic resonance imaging (fMRI)

A form of noninvasive neuroimaging based on blood-oxygen-level-dependent (BOLD) signals in the brain in vivo.

Blood-oxygen-level-dependent (BOLD) signal

The measurement of metabolic activity in the brain based on the magnetic resonance imaging contrast of blood deoxyhemoglobin levels arising from changes in local blood flow.

Brain network

Network originally refers a physical system that can be represented by a graph consisting of nodes and edges. In the field of neuroscience, brain networks can be defined by structural connectivity or functional interdependence. The former is based on the anatomical linkage of its neurons, while the latter refers to joint activity in different brain structures that is co-dependent under variation of a functional or behavioral parameter.

Independent component analysis (ICA)

A computational technique that separates a multivariate signal into additive components based on the assumption that the components arise from statistically independent non-Gaussian sources.

Network node

Node refers the component of network linked by edges. In the field of neuroscience, the nodes in structural networks are typically considered to be brain areas defined by cytoarchitectonics, local circuit connectivity, output projection target commonality, and input projection source commonality. The functional nodes are commonly identified by inferences concerning the effects of brain lesions on cognitive function (historically), or relating the joint activation or deactivation of brain areas to different cognitive functions.

Functional connectivity

The statistical interrelation of variables representing temporal changes in different network nodes. In other words, functional connectivity refers the temporal correlation of a neurophysiological index measured in different brain areas, which usually being used to identify coactivating brain regions as functional brain networks.

Large-scale network

A term refers to neural systems that are distributed across the entire extent of the brain. Based on the technological and methodological advances of structural and functional brain connectivity, the large-scale network studies focus on revealing how cognitive functions arise from interactions within and between distributed brain systems.

Intrinsic network

Intrinsic networks are also called Intrinsic Connectivity Networks or ICNs. Originally, the intrinsic network refers large-scale network of interdependent brain areas observed at rest, which typically been identified by ICA on the BOLD spontaneous fluctuations. Recently, emerging evidences showed that the intrinsic network architecture also present across a wide variety of task state, suggesting this is an “intrinsic” standard architecture of functional brain organization.