Women with Major Depressive Disorder, Irrespective of Comorbid Anxiety Disorders, Show Blunted Bilateral Frontal Responses during Win and Loss Anticipation

Abstract

Background:

Electroencephalography (EEG) studies suggest that major depressive disorder (MDD) is associated with lower left than right frontal brain activity (asymmetry), a pattern appearing stronger in women than men, and when elicited during emotionally-relevant paradigms versus an uncontrolled resting state. However, it is unclear whether this asymmetry pattern generalizes to the common presentation of MDD with co-occurring anxiety. Moreover, asymmetry may differ for anxiety subtypes, wherein anxious apprehension (AnxApp: worry characteristic of generalized anxiety disorder) appears left-lateralized, but anxious arousal (AnxAro: panic characteristic of social anxiety, posttraumatic stress, and panic disorders) may be right-lateralized.

Methods:

This analysis attempted to replicate frontal EEG asymmetry patterns using functional magnetic resonance imaging (fMRI). Participants completed clinical interviews and a monetary incentive delay (MID) task during fMRI recording. We compared five groups of right-handed women from the Tulsa 1000 study, MDD (n=40), MDD-AnxApp (n=26), MDD-AnxAro (n=34), MDD-Both (with AnxApp and AnxAro; n=26), and healthy controls (CTL; n=24), as a function of MID anticipation condition (no win/loss, win, loss) and hemisphere on frontal blood oxygen-level-dependent (BOLD) signal.

Results:

CTL exhibited higher bilateral superior, middle, and inferior middle frontal gyrus BOLD signal than the four MDD groups for high arousal (win and loss) conditions. However, frontal attenuations were unrelated to current depression/anxiety symptoms, suggestive of a trait as opposed to a state marker.

Limitations:

This was a cross-sectional analysis restricted to women.

Conclusions:

Reduced prefrontal cortex recruitment during processing of both positively and negatively valenced stimuli is consistent with the emotion context insensitivity theory of MDD.

Introduction

Major depressive disorder (MDD) is a debilitating condition responsible for substantial personal and societal burden. Identification of markers conveying MDD vulnerability may ease this burden when used to screen individuals at risk, assess treatment outcome success, and/or identify treatment targets. Electroencephalography (EEG) studies demonstrate that right-handed individuals with lifetime MDD show higher right than left frontal cortical activity (corresponding to greater left than right alpha band power, a metric inversely associated with cortical arousal) than healthy controls (CTL) during rest and cognitive-emotional tasks (e.g., Cantisani et al., 2015; Gheza et al., 2019; Gotlib, 1998; Henriques & Davidson, 1990, 1991; Kemp et al., 2010; Nusslock et al., 2018; Shankman et al., 2013; Stewart et al., 2010; Stewart et al., 2011). This asymmetry pattern is thought to reflect a motivational and/or emotional imbalance characterized by reduced approach behavior/positive emotions paired with heightened withdrawal behavior/negative emotions (see Allen & Reznick, 2015 and Allen et al., 2018). Greater right than left frontal EEG activity within MDD samples is linked to heightened anhedonia (Smith et al., 2018), psychomotor retardation (Cantisani et al., 2015), dysphoria and apathy (Nelson et al., 2018), and melancholia (Liu et al., 2016). However, frontal EEG asymmetry findings for MDD are far from consistent (e.g., Arns et al., 2016; Cantisani et al., 2016; Nelson et al., 2018; Quinn et al., 2014), partly due to methodological differences regarding EEG reference montage/electrode sites, recording context (e.g., rest versus active tasks), and sample characteristics (e.g., group size, comorbid clinical symptoms/diagnoses; see Coan et al., 2006, Davidson, 1998, and Hagemann, 2004). Moreover, right-lateralized frontal EEG asymmetry in pure MDD may be more robust for women than men in some contexts but not others (Jaworska et al., 2012; Stewart et al., 2010; Stewart et al., 2011), and is associated with future MDD onset in never-depressed women, but not men (Stewart & Allen, 2018).

Further complicating the clinical picture, MDD is highly comorbid with anxiety disorders (e.g., Flory & Yehuda, 2015; Judd et al., 1998), and two types of anxiety symptoms appear to show differential brain lateralization (e.g., Heller et al., 1997; Heller & Nitschke, 1998; Mathersul et al., 2008; Nitschke et al., 1999; Nusslock et al., 2015). Anxious apprehension, characterized by worry, is thought to be left-lateralized as a function of overactive verbal repetition, engaging brain regions involved in language processing such as inferior frontal gyrus (IFG) (Carter et al., 1986; Engels et al., 2007; Hofmann et al., 2005; Smith et al., 2016). In contrast, anxious arousal, characterized by panic, is thought to be right-lateralized as a function of excessive fear and heightened signals of physiological alarm (Engels et al., 2010; Stewart et al., 2008), although this asymmetry may be more robust across samples within parietal as opposed to frontal regions (Heller, 1993; Tucker, 1984; see Stewart et al. 2011 and Bruder et al. 2017 for reviews). Generalized anxiety disorder (GAD) is marked by excessive worry consistent with heightened anxious apprehension (e.g., Brown et al., 1992; Dugas et al., 1998), whereas panic disorder (PD) and posttraumatic stress disorder (PTSD), and to some extent, social anxiety disorder (SAD), are marked by excessive panic consistent with heightened anxious arousal (e.g., Brown & McNiff, 2009; Brown et al., 2016). Indeed, greater right than left frontal EEG activity characterizes: (1) higher anxious arousal symptoms (Stewart et al., 2008); (2) PD during rest and exposure to a fearful stimulus (Wiedemann et al., 1999); (3) heightened symptom severity in PTSD (Kemp et al., 2010; see Butt et al., 2019); and (4) SAD during speech anticipation (Davidson et al., 2000) and exposure to faces (Myllyneva et al., 2015). Moreover, greater anxious arousal symptoms are associated with greater right than left frontal asymmetry measured by fMRI signals during an emotion-word Stroop task (Engels et al., 2010). Patients with comorbid MDD and SAD show greater right than left frontal EEG activity at rest than CTL and pure MDD patients (Bruder et al., 1997), whereas patients with comorbid MDD and anxious apprehension exhibit greater left than right frontal EEG activity at rest than CTL and pure MDD patients (Nusslock et al., 2018). However, other studies report null frontal EEG asymmetry findings, both at rest (e.g., Kentgen et al., 2000) or during tasks (e.g., reward anticipation, Gorka et al., 2015); again modest sample sizes, clinical comorbidity, sex and/or methodological EEG differences may play a role.

A final issue involves the degree of replicability of frontal asymmetry patterns in MDD and anxiety disorders across neuroimaging modalities that differ substantially in temporal and spatial resolution. Most frontal lateralization studies have recorded EEG from fewer than 32 scalp electrodes, limiting the ability to accurately localize these electrical signals; however, recent work has localized frontal EEG alpha asymmetry signals to middle frontal gyrus (MFG) in both MDD and CTL samples (Auerbach et al., 2015; Smith et al., 2018) as well as superior frontal gyrus (SFG), orbitofrontal cortex (OFC), and medial prefrontal regions across CTL and MDD with and without comorbid PD (Gorka et al., 2015). As functional magnetic resonance imaging (fMRI) possesses higher spatial resolution than EEG, localization of frontal asymmetry to particular regions within prefrontal cortex as a function of MDD and type of anxiety may improve the development of more precise screening methods as well as intervention targets.

The capability model of individual differences argues that brain activity recorded during cognitive-emotional tasks may produce larger effect size differences between clinical and CTL groups than activity recorded at rest because relevant motivational systems thought to drive frontal asymmetry will likely be more engaged (Coan et al., 2006; Stewart et al., 2014). Paradigms relevant to MDD, anxiety disorders, and frontal asymmetry would be those involving both approach and withdrawal motivations/emotions (e.g., processing of wins and losses). Although fMRI research suggests that MDD patients differ from CTL in frontal blood oxygen-level-dependent (BOLD) signal during win/loss paradigms, the particular region, hemisphere, and task condition varies by study. During win anticipation, MDD show lower BOLD signal than CTL within left SFG (Knutson et al., 2008), left OFC (Rothkirch et al., 2017), bilateral MFG (Smoski et al., 2009), right IFG (Smoski et al., 2009), and right OFC (Smoski et al., 2011; Ubl et al., 2015). Additionally, remitted MDD is linked to attenuated left SFG BOLD signal during loss anticipation (Schiller et al., 2013) but heightened right MFG BOLD signal during win anticipation (Dichter et al., 2012), suggesting frontal differences may vary as a function of valence or perhaps arousal. Other fMRI studies demonstrate that individuals with mood and/or anxiety disorders exhibit lower, often bilateral, prefrontal cortex recruitment than CTL during emotion paradigms (e.g., Ball et al., 2013; Picó-Pérez et al., 2017; Zilverstand et al., 2017). Task context be a moderating factor of frontal asymmetry, as patients with MDD exhibit greater right than left MFG BOLD signal during emotional judgements, with attenuated left MFG linked to positive valence, but heightened right MFG linked to depressive symptom severity (Grimm et al., 2008).

It is still unclear whether MDD show differential frontal BOLD signal differences as a function of comorbid anxiety disorders due to limited research (Gorka et al., 2014). Even more crucially, few fMRI studies have explicitly tested hemispheric BOLD signal differences in frontal regions as a function of pure versus comorbid MDD. Testing of hemispheric differences is needed to evaluate whether brain asymmetry indexing a particular clinical phenotype is marked by an increase or decrease within one hemisphere, as opposed to an imbalance between hemispheres; divergent findings as a function of pure versus comorbid MDD and/or anxiety subtype could point to distinct impairments in function (e.g., exaggerated word repetition in anxious apprehension), leading to phenotype-specific treatment targets.

The present project focuses on secondary data analysis of the Tulsa 1000 (T1000) study, a project evaluating potential biobehavioral markers of clinical symptoms in 1000 individuals, comparing treatment-seeking individuals with mood, anxiety, eating, and/or substance use disorders to individuals without a history of psychiatric disorder (Victor et al., 2018). Data are currently available for the first 500 participants to allow us to test initial hypotheses, while data from the second 500 participants will eventually be made available, allowing us to attempt to replicate findings gleaned from the first 500. Within the first 500 participants, rates of lifetime MDD were 3:1 women to men, limiting statistical power to detect differences in pure versus comorbid MDD within men; given limitations on statistical power, we therefore focused on asymmetry differences as a function of MDD comorbidity within women. This sex difference in MDD/anxiety disorder prevalence within our sample is consistent with research indicating that adult women experience higher lifetime prevalence of anxiety disorders and endorse greater MDD comorbidity with anxiety disorders than adult men (Eaton et al., 2015; McLean et al., 2011).

Within the first 500 T1000 participants, we therefore categorized right-handed women into five groups and compared them on frontal BOLD signal elicited by a monetary incentive delay (MID) task during fMRI recording. Four groups with lifetime MDD differed as a function of comorbid lifetime anxiety disorders and were compared to each other as well as CTL: (1) MDD = no anxious apprehension or anxious arousal; (2) MDD-AnxApp = comorbid anxious apprehension (GAD); (3) MDD-AnxAro = comorbid anxious arousal (SAD, PD, and/or PTSD); and (4) MDD-Both = comorbid anxious apprehension and anxious arousal. Planned contrast analyses evaluated group, hemisphere, and MID anticipation condition (win, loss, none) differences for four frontal regions of interest (ROIs: SFG, MFG, IFG, OFC). These four frontal ROIs were selected because previous research demonstrates that: (1) depression/anxiety groups differ in fMRI BOLD signal from CTL in these regions (SFG: Knutson et al., 2008; Schiller et al., 2013; MFG: Smoski et al., 2009; IFG: Engels et al., 2007, Smoski et al., 2009; OFC: Rothkirch et al., 2017; Smoski et al., 2011; Ubl et al., 2015); and/or (2) EEG alpha asymmetry scores have been localized to these regions (SFG/OFC: Gorka et al., 2015; MFG: Auerbach et al., 2015; Smith et al., 2018). On the basis of EEG and fMRI literatures, we hypothesized that: (1) CTL would exhibit greater left than right SFG and MFG activity, marked by higher left hemisphere BOLD signal, than all four MDD groups during win anticipation, given that depression is characterized by reduced approach motivation previously linked to left prefrontal function; (2) MDD-AnxApp would exhibit greater left than right IFG activity, marked by higher left hemisphere BOLD signal, than MDD, MDD-AnxAro, and MDD-Both groups during loss anticipation, consistent with heightened language processing within the context of withdrawal-relevant negative stimuli; and (3) MDD-AnxAro and MDD-Both would show greater right than left SFG and MFG activity, marked by higher right hemisphere BOLD signal, than MDD during loss anticipation, consistent with a heightened aversive arousal state. Group differences in personality and clinical symptoms were also explored to aid in interpretation of group differences in fMRI findings.

Methods

Participants

Participants were a subsample (N = 150) of the first 500 individuals recruited into the T1000 study consisting of treatment-seeking individuals and healthy comparison subjects (Victor et al., 2018). This study was approved by the Western Institutional Review Board and implemented in accordance with The Code of Ethics of the World Medical Association. Participants were recruited via advertisements and gave informed consent. Individuals orally consented to complete a study eligibility screening. Eligible subjects were scheduled for a clinical interview session, wherein trained staff administered the MINI International Neuropsychiatric Interview (version 6.0 or 7.0) (Sheehan & Lecrubier, 2010; Sheehan et al., 2015) to assess lifetime mental disorders via Diagnostic and Statistical Manual of Mental Disorders, 4^th Edition or 5^th Edition (DSM-5; American Psychiatric Association, 2013) criteria.

Figure 1 illustrates the T1000 group selection and elimination process for this analysis. More specifically, Figure 1 shows that out of 500 participants, only 16 men met criteria for pure MDD, and only 36 men met criteria for MDD with any comorbid anxiety disorders (prior to further potential exclusion for left-handedness and issues with fMRI data quality). In contrast, 48 women met criteria for pure MDD and 100 women met criteria for comorbid MDD and anxiety disorders, providing greater statistical power to test hypotheses involving four distinct MDD subgroups within women than within men. Therefore, for the present analysis, participants were limited to right-handed women with complete fMRI MID task data (126 with MDD and 24 CTL) who were classified into five groups based on lifetime MINI diagnoses: (1) MDD (n=40): MDD only; (2) MDD-AnxApp (n=26): MDD and GAD; (3) MDD-AnxAro (n=34): MDD and at least one: SAD, PTSD and/or PD; (4) MDD-Both (n=26): MDD, GAD, and at least one: SAD, PTSD, and/or PD; and (5) CTL (n=24): healthy comparison subjects. Exclusion criteria were: (1) positive urine screen for substances of abuse; (2) lifetime bipolar, schizophrenia spectrum, obsessive compulsive, eating, or substance use disorders; (3) active suicidal ideation with intent/plan; (4) moderate-to-severe traumatic brain injury, defined as ≥ 30 min loss of consciousness (Department of Defense & Department of Veterans Affairs, 2009); (5) severe/unstable medical issues; (6) antisocial personality disorder; (7) anxiety disorders without MDD; (8) fMRI contraindications; and (9) unusable fMRI data (e.g., outliers, excessive motion, etc.)

Figure 1.

The first 500 of Tulsa 1000 participants were categorized into five groups for analysis (total N = 150), highlighted in dashed colored boxes. Groups were confined to women due to limited sample sizes for men. Healthy comparison subjects (CTL) did not meet criteria for any lifetime DSM-IV/DSM-5 diagnoses. Participants assigned to one of four patient groups met lifetime criteria for major depressive disorder: (1) alone (MDD); (2) with comorbid anxious apprehension (MDD-AnxApp), meeting lifetime criteria for generalized anxiety disorder (GAD); (3) with comorbid anxious arousal (MDD-AnxAro), meeting lifetime criteria for social anxiety disorder (SAD), panic disorder (PD) and/or posttraumatic stress disorder (PTSD); and (4) with both anxious apprehension and anxious arousal, MDD-Both. Grey boxes depict participants who were excluded for various diagnoses (see key within the figure for acronym definitions).

Procedure

Participants completed sessions within a two-week time span, on average (Victor et al., 2018). The ClinicalTrials.gov identifier for the clinical protocol associated with data published in the current paper is NCT02450240, “Latent Structure of Multi-level Assessments and Predictors of Outcomes in Psychiatric Disorders”.

Clinical interview session.

Participants completed the MINI interview, provided demographic information, and completed the following measures associated with depression, anxiety, trauma, pleasure, motivation, and emotion: (1) PROMIS Depression and Anxiety scales (Pilkonis et al., 2011); (2) Patient Health Questionnaire (PHQ-9) for depression (Kroenke et al., 2001); (3) Anxiety Sensitivity Index (ASI; Taylor et al., 2007); (4) State Trait Anxiety Inventory (STAI; Spielberger et al., 1983); (5) Childhood Trauma Questionnaire (CTQ; Bernstein & Fink, 1998); (6) Ruminative Responses Scale (RRS; Treynor et al., 2003); (7) Temporal Experiences of Pleasure Scale (TEPS; Gard et al., 2006); (8) Behavioral Inhibition/Behavioral Activation (BIS/BAS) Scales (Carver & White, 1994); (9) Toronto Alexithymia Scale (TAS; Bagby et al., 1994); and (10) Positive and Negative Affect Scales, Expanded Form (PANAS-X; Watson & Clark, 1999).

Neuroimaging session.

Participants completed two runs (562 s and 45 trials each, for a total of 90 trials) of the MID task (Knutson et al., 2001) during fMRI recording. On each trial, participants were given a cue indicating potential win (circle), loss (square), or no win/loss (circle or square). To earn a win or avoid a loss, subjects were required to press a button within a certain duration of time (adapted for individual reaction times, or RT) following onset of a white square (target cue). Task difficulty was based on RT collected during practice prior to the fMRI scan and was calibrated so that subjects would succeed on two-thirds of trials. Degree of potential win or loss varied as a function of the location of a horizontal line within a cue and was also indicated with text: (1) line at the bottom of the queue = no win/loss; (2) line in the middle of the shape = intermediate win/loss; and (3) line at the top of the shape = highest win/loss. Subjects could gain or lose points and earned $30 on average, which they were paid post-scan. Images were acquired with a GE MRI750 3T scanner. MID task parameters consisted of 281 contiguous echo-planar imaging (EPI) volumes [TR/TE = 2000/27ms, FOV/slice = 240/2.9 mm, 128 × 128 matrix, 39 axial slices]. High-resolution structural images were acquired through a 3D axial T1-weighted magnetization-prepared rapid acquisition with gradient-echo sequence [TR/TE = 5/2.012 ms, FOV/slice = 240 × 192/0.9mm, 186 axial slices]. Mean RT and hit rate per condition were extracted for further group analyses.

Neuroimaging Data Processing

Neuroimaging data analyses were performed with analysis of functional neuroimaging (AFNI, http://afni.nimh.nih.gov) software (Cox, 2012). The first 3 EPI volumes were discarded for signal equilibrium and scanning noise adaptation. Data were then despiked, slice-timing corrected, co-registered to anatomical volumes, motion corrected, smoothed (4 mm3 full width at half maximum), and normalized to standard Montreal Neurological Institute space (resampling voxel size = 2 × 2 × 2 mm³). A general linear model was employed to analyze single-subject data. Boxcar regressors defined for each participant modeled the blood-oxygen-level dependent (BOLD) response to the anticipation window (4 s) in six conditions (15 trials per condition per run): large win, small win, no win, large loss, small loss, and no loss. BOLD signal change from baseline were extracted for: (1) large wins; (2) large losses; and (3) the average of no wins and no losses. Prefrontal cortex regions of interest (ROIs) illustrated in Figure 2 were extracted from the Brainnetome Atlas (Fan et al., 2016). Left and right hemisphere ROIs were created for four regions: (1) SFG: average of A8m, A8dl, A91, A9m, and A10m; (2) MFG: average of a946d, IFJ, A46, a6vl, and a10l (including dorsolateral prefrontal cortex); (3) IFG: average of a44d, a45c, a45r, a44op, and a44v; and (4) OFC: average of al247o, al1l, and al247l. Averages weighted each Brainnetome ROI subregion equally.

Figure 2.

Left and right hemisphere regions of interest (ROIs) were created for each of three monetary incentive delay (MID) task conditions via extractions from Brainnetome Atlas (Fan et al., 2016): (A) superior frontal gyrus (SFG): the average of A8m, A8dl, A9l, A9m, and A10m, shown in the axial figure above, top Z = 7, bottom Z = 42; (B) middle frontal gyrus (MFG): the average of a946d, IFJ, A46, a6vl, and a10l depicted in the sagittal figure above, top X = 33, bottom Z = −33; (C) inferior frontal gyrus (IFG): the average of a44d, a45c, a45r, a44op, and a44v shown in the axial figure above, top Z = 2, bottom Z = 24; and (D) orbitofrontal cortex (OFC): the average of a1247o, a11l, and a1247l depicted in the axial figure above, top Z = −24, bottom Z = −12.

Statistical Analysis

Analyses were performed in R Studio (RStudio Team, 2015), version 1.1463, using packages car, ggplot2, gmodels, nlme, psych, and stats, guided by Field and colleagues (2012). Histograms and boxplots were examined to evaluate normality, whereas Levene’s test evaluated homogeneity of variance.

Demographics.

Chi-square/Fisher Exact tests evaluated group differences in education (three categories: at least some high school/general education diploma [GED], some college, or graduated college), whereas an ANOVA tested for group differences in age.

Patient groups.

Chi-square/Fisher Exact tests compared differences among the clinical groups on major depressive episode (MDE) frequency, MDE status (current versus partial/lull remission), and medication status. Differences in questionnaire scores were tested: (1) within the clinical groups using one-way ANOVAs (as homogeneity of variance assumptions were upheld); and (2) between the clinical groups and CTL using Kruskal-Wallis rank sum (H) tests, as variances for CTL tended to be non-homogeneous with those of the clinical groups. Within the ANOVA framework, three orthogonal planned contrasts between clinical groups were computed: (1) apprehension versus other MDD (MDD = 1, MDD-AnxApp = −3, MDD-AnxAro = 1, MDD-Both = 1); (2) pure MDD versus arousal (MDD = −2, MDD-AnxApp = 0, MDD-AnxAro = 1, MDD-Both = 1); and (3) arousal versus both (MDD = 0, MDD-AnxApp = 0, MDD-AnxAro = −1, MDD-Both = 1).

MID behavior.

Linear mixed models were computed using maximum likelihood (ML) estimation, one for each of two behavioral dependent variables: mean RT and correct hits (%). Group was the between-subjects factor, whereas MID anticipation condition (averaged no win/no loss, win, and loss) was the within-subjects factor. Random intercepts were specified as a function of participant as well as condition nested within each participant.

MID frontal ROIs.

Linear mixed models were computed using ML estimation, one for each region: SFG, MFG, IFG, and OFC. Group was the between-subjects factor, whereas MID anticipation condition (averaged no win/no loss, win, and loss) and hemisphere were within-subjects factors. Random intercepts were specified as a function of participant, hemisphere nested within each participant, and condition nested within hemisphere. Main effects and all interactions were included in each model. Percent BOLD signal change from baseline was extracted for the three conditions of interest to be compared directly using planned contrasts.

Planned contrasts for MID data.

To test group differences, four orthogonal planned contrasts were evaluated: (1) control versus MDD (CTL = −4, MDD = 1, MDD-AnxApp = 1, MDD-AnxAro = 1, MDD-Both = 1); (2) apprehension versus other MDD (MDD = 1, MDD-AnxApp = −3, MDD-AnxAro = 1, MDD-Both = 1); (3) pure MDD versus arousal (MDD = −2, MDD-AnxApp = 0, MDD-AnxAro = 1, MDD-Both = 1); and (4) arousal versus both (MDD = 0, MDD-AnxApp = 0, MDD-AnxAro = −1, MDD-Both = 1). To test condition differences, two orthogonal planned contrasts were computed: (1) arousal (averaged no win/no loss = −2, win = 1, loss = 1); and (2) valence (averaged no win/no loss = 0, win = 1, loss = −1). For ROI analyses, the hemisphere contrast compared left (−1) and right (1) hemispheres.

To test our first hypothesis that CTL would differ from all MDD groups in the left hemisphere during win anticipation, we expected a significant control versus MDD*valence*hemisphere interaction for SFG and MFG ROIs. To test our second hypothesis that MDD-AnxApp would differ in the left hemisphere from the other three MDD groups during loss anticipation, we expected a significant apprehension versus other MDD*valence*hemisphere interaction for the IFG ROI. To test our third and final hypothesis that MDD-AnxAro and MDD-Both would differ in the right hemisphere from MDD during loss anticipation, we expected a significant arousal versus both*valence*hemisphere interaction.

Results

Demographics

Groups did not differ in education, χ²(8) = 13.00, p = .11, or age, F(4, 144) = 1.43, p = .23. On average, participants were 35.37 years old (SD = 12.00; range = 18-56) and 35% graduated college, 50% attended some college, and 15% completed at least some high school or attained a GED. With respect to race/ethnicity the sample was: 69% White, 12% Native American, 8% Black, 5% Other, 4% Hispanic, 1% Asian, and 1% unknown.

Patient groups

Table 1 illustrates diagnostic characteristics of the clinical groups, who were comprised of participants with current and/or past MDD who did not differ in MDE frequency/status, or medication status. Supplementary Table 1 shows questionnaire differences among the five groups. All four clinical groups (MDD, MDD-AnxApp, MDD-AnxAro, and MDD-Both) endorsed higher PROMIS depression/anxiety, PHQ-9, ASI, CTQ, STAI, RRS, BIS, TAS, and PANAS-X negative affect scores but lower PANAS-X positive affect scores than CTL. In addition, the pure MDD versus arousal contrast indicated that MDD-AnxAro and MDD-Both groups reported higher PROMIS Anxiety, ASI, STAI, BIS, and PANAS-X negative affect scores than the pure MDD group; in contrast, the apprehension versus other MDD contrast was non-significant for these measures. Groups did not differ on CTQ, BAS, or TEPS scores.

Table 1.

Diagnostic and medication characteristics of the four clinical groups (n = 126 women).

	MDD (n = 40)	MDD-AnxApp (n = 26)	MDD-AnxAro (n = 34)	MDD-Both (n = 26)
Lifetime DSM Diagnoses	Count	Count	Count	Count
Major Depressive Disorder	40 (100%)	26 (100%)	34 (100%)	26 (100%)
Generalized Anxiety Disorder	0	26 (100%)	0	26 (100%)
Social Anxiety Disorder	0	0	13 (38%)	9 (35%)
Panic Disorder	0	0	9 (26%)	16 (62%)
Posttraumatic Stress Disorder	0	0	20 (59%)	6 (23%)
MDE Frequency	Count	Count	Count	Count	Statistics
Single	6 (15%)	5 (19%)	8 (24%)	5 (19%)	χ2(3) = 0.87, p = .83
Recurrent	34 (85%)	21 (81%)	26 (76%)	21 (81%)
MDE Status^	Count	Count	Count	Count	Statistics
Current Episode	31 (78%)	19 (73%)	23 (68%)	15 (58%)	Fisher Exact p = .26
Partial Remission*	4 (10%)	4 (15%)	9 (26%)	9 (35%)
Full Remission**	5 (12%)	3 (12%)	2 (6%)	2 (7%)
Medication Status	Count	Count	Count	Count	Statistics
Medicated	27 (68%)	18 (69%)	22 (65%)	18 (69%)	χ2(3) = 0.19, p = .98
Unmedicated	13 (32%)	8 (31%)	12 (35%)	8 (31%)
Type of Medication
SSRI	18 (45%)	11 (42%)	9 (26%)	10 (38%)
SNRI	6 (15%)	3 (12%)	5 (15%)	3 (12%)
Atypical Antidepressants	7 (18%)	5 (19%)	8 (24%)	8 (31%)
Anticonvulsants	4 (10%)	0	1 (3%)	2 (8%)
Benzodiazepines	5 (13%)	5 (19%)	8 (24%)	7 (27%)
Opioids	5 (13%)	3 (12%)	2 (6%)	4 (15%)
Other Anxiolytics	1 (3%)	3 (12%)	4 (12%)	2 (8%)
Other Muscle Relaxants	5 (13%)	2 (8%)	3 (9%)	2 (8%)
Stimulants	3 (8%)	4 (15%)	0	0
Lithium	1 (3%)	0	1 (3%)	0
Atypical Antipsychotics	0	0	2 (6%)	1 (4%)

Note.

^{^}Assessed during the MINI clinical interview.

^*Defined by DSM-5 in two ways: (1) some MDE symptoms are still present, but full MDE criteria are no longer met; or (2) there are no longer any significant MDE symptoms, but the period of remission has been less than two months.

^**Defined by DSM-5 as a period of at least two months in which there are no significant MDE symptoms.

MDE = major depressive episode. MDD = individuals with major depressive disorder but no anxious apprehension or anxious arousal. MDD-AnxApp = individuals with MDD and anxious apprehension. MDD-AnxAro = individuals with MDD and anxious arousal. MDD-Both = individuals with MDD, anxious apprehension, and anxious arousal. SSRI = selective serotonin reuptake inhibitor. SNRI = selective norepinephrine reuptake inhibitor.

Behavioral Results

Supplementary Figure 1 shows that for mean RT, a main effect of condition emerged (F(2, 290) = 74.64, p < .001), and the relevant arousal contrast indicated that wins and losses elicited faster RTs than no wins/no losses (t(290) = −5.55, p < .001).

Frontal ROI Analysis

SFG.

Main effects of group (F(4, 145) = 3.11, p = .02) and condition (F(2, 580) = 162.74, p < .001) were qualified by a group*condition interaction (F(8, 580) = 4.20, p < .001). A significant arousal contrast indicated that wins and losses evoked greater BOLD signal than no wins/no losses, whereas a significant valence contrast showed that wins evoked greater BOLD signal than losses (see Supplementary Table 2). A control versus MDD main effect (t(145) = −2.51, p = .01) was qualified by a control versus MDD*arousal interaction, highlighted in Figure 3, indicating that CTL displayed higher BOLD signal than the four MDD groups in the high arousal (win and loss) conditions (t(580) = −1.98, p < .05; Cohen’s d for each clinical group versus CTL: MDD = .66, MDD-AnxApp = .48, MDD-AnxAro = .67, and MDD-Both= .76).

Figure 3.

Superior frontal gyrus (SFG) blood oxygen-level-dependent (BOLD) signal change from baseline as a function of group and monetary incentive delay (MID) task anticipation condition. Results for planned contrasts indicated that healthy comparison subjects (CTL) exhibited higher BOLD signal to high arousal (the average of win and loss anticipation) conditions than the four groups with major depressive disorder (MDD, MDD-AnxApp, MDD-AnxAro, and MDD-Both). This difference was greater for high arousal (win/loss) than low arousal (no win/no loss) conditions. Error bars reflect ± 1 standard error.

MFG.

Main effects of hemisphere (F(1, 145) = 6.81, p = .01) and condition (F(2, 580) = 87.54, p < .001) were qualified by a hemisphere*condition interaction (F(2, 580) = 4.22, p = .02). An arousal*hemisphere contrast showed that wins/losses evoked larger right MFG BOLD signal than no wins/no losses. A significant valence contrast showed that wins evoked greater BOLD signal than losses (see Supplementary Table 2). A main effect of group (F(4, 145) = 2.57, p = .04) was qualified by a group*condition interaction (F(8, 580) = 3.26, p < .01). Inspection of relevant planned contrasts indicated that a control versus MDD main effect (t(145) = −2.76, p < .01) was qualified by a control versus MDD*arousal interaction; Figure 4 illustrates that CTL displayed higher bilateral MFG BOLD signal than the four MDD groups, a difference exaggerated in high arousal (win and loss) conditions (t(580) = −2.37, p = .02, Cohen’s d for each clinical group versus CTL: MDD = .61, MDD-AnxApp = .53, MDD-AnxAro = .85, and MDD-Both = .76).

Figure 4.

Middle frontal gyrus (MFG) blood oxygen-level-dependent (BOLD) signal change from baseline as a function of group and monetary incentive delay (MID) task anticipation condition. Results for planned contrasts indicated that healthy comparison subjects (CTL) exhibited higher BOLD signal to high arousal (the average of win and loss anticipation) conditions than the four groups with major depressive disorder (MDD, MDD-AnxApp, MDD-AnxAro, and MDD-Both). This difference was greater for high arousal (win/loss) than low arousal (no win/no loss) conditions. Error bars reflect ± 1 standard error.

IFG.

Main effects of hemisphere (F(1, 145) = 9.26, p < .01) and condition (F(2, 580) = 61.78, p < .001) were qualified by a hemisphere*condition interaction (F(2, 580) = 5.99, p < .01). An arousal*hemisphere interaction indicated that wins/losses evoked larger right IFG BOLD signal than no wins/no losses; moreover, a significant valence*hemisphere interaction indicated that wins evoked greater right IFG BOLD signal than losses (see Supplementary Table 2). A group main effect (F(4, 145) = 2.75, p = .03) was qualified by a group*condition interaction (F(8, 580) = 2.05, p = .04). A control versus MDD contrast illustrated in Figure 5 demonstrated that CTL exhibited higher bilateral IFG BOLD signal than the four MDD groups (t(145) = −2.38, p = .02, Cohen’s d for each clinical group versus CTL: MDD = .61, MDD-AnxApp = .29, MDD-AnxAro = .68, and MDD-Both = .57).

Figure 5.

Inferior frontal gyrus (IFG) blood oxygen-level-dependent (BOLD) signal change from baseline as a function of group and monetary incentive delay (MID) task anticipation condition. Results for planned contrasts indicated that healthy comparison subjects (CTL) exhibited higher BOLD signal averaged across conditions than the four groups with major depressive disorder (MDD, MDD-AnxApp, MDD-AnxAro, and MDD-Both). Error bars reflect ± 1 standard error.

OFC.

As main effects of hemisphere (F(1, 145 = 16.58, p < .001) and condition (F(2, 580) = 67.90, p < .001) accompanied a group*condition interaction (F(8, 580) = 3.08, p < .01), planned contrasts relevant to these effects were inspected. Supplementary Table 2 illustrates that: (1) a significant arousal contrast indicated that wins and losses evoked greater BOLD signal than no wins/no losses; (2) a significant hemisphere contrast demonstrated that OFC BOLD signal was larger in the left than the right hemisphere; and (3) a valence contrast indicated that wins were associated with higher BOLD signal than losses. Finally, an arousal versus both*arousal interaction (t(580) = 2.85, p < .01) indicated that BOLD signal increases from low to high arousal conditions were larger within MDD-Both (d = .69) than within MDD-AnxAro (d = .40); however, MDD-AnxAro and MDD-Both did not differ from each other on OFC BOLD signal to low arousal (d = .27) or high arousal (d = .17) conditions separately (see Supplementary Figure 2).

Follow-Up Analyses: Current Symptom Severity

Across participants from the four MDD groups with PROMIS data (n = 124), a linear regression was computed for SFG, MFG, and IFG regions differentiating clinical and CTL groups. Predictors entered simultaneously within each model were PROMIS Depression and Anxiety scales. The dependent variable was BOLD signal change from baseline for high arousal (win and loss anticipation averaged). As PROMIS Depression and Anxiety were positively correlated but not quite collinear (r = .51, R² = .26, p < .01), regressions enabled us to evaluate the contribution of each scale to BOLD signal responses. Supplementary Table 3 indicates that none of the models approached significance. Analogous regressions were computed with PHQ-9 depression and STAI state anxiety scales as predictors, but no models approached significance (all p > .13).

Discussion

We attempted to replicate and extend the frontal EEG asymmetry literature by evaluating hemispheric differences in frontal fMRI BOLD signal among four MDD groups varying in the presence versus absence of comorbid anxious apprehension and anxious arousal, in addition to CTL. We tested three specific hypotheses using planned contrasts. First, we predicted that CTL would exhibit greater SFG and MFG BOLD signal within the left hemisphere than the four MDD groups specifically during win anticipation; this hypothesis was not supported. Instead CTL displayed greater SFG, MFG, and IFG BOLD signal than the four MDD groups across hemispheres as well as across win and loss anticipation conditions, consistent with a bilateral arousal effect as opposed to a valence effect. These results suggest that women with lifetime MDD, regardless of comorbid anxiety, show bilateral frontal attenuations while anticipating both positive and negative stimuli, consistent with work showing that: (1) women with MDD exhibit blunted bilateral frontal EEG activity at rest (Jaworska et al., 2012); (2) MDD patients show bilateral MFG BOLD signal attenuations during anticipation of wins (Smoski et al., 2009) and changes in emotional stimulus intensity (Strigo et al., 2010); and (3) individuals with MDD display lower bilateral MFG and IFG BOLD signal than CTL in the presence of emotional distractors (Wang et al., 2008). Our findings of blunted bilateral prefrontal cortex responses across positive and negative stimuli and lifetime MDD subgroups, regardless of comorbid lifetime anxiety disorders, is consistent with the emotion context insensitivity theory (e.g., Rottenberg, Gross & Gotlib, 2005; Rottenberg & Hindash, 2015), which posits that MDD is characterized by blunted emotional reactivity to both appetitive and aversive stimuli (Bylsma et al., 2008). Although there is an extensive literature suggesting that MDD is characterized by an attentional bias toward negatively valenced information (Peckham et al., 2010), an increased bias of attention toward negative stimuli is not necessarily at odds with attenuated reactivity to these stimuli. Moreover, it is likely that the various measures relied upon for studies of negative attentional bias versus emotional reactivity are using tasks that probe slightly different constructs (Bylsma et al., 2008; Peckham et al., 2010). Although beyond the scope of the present analysis, it would be beneficial for future work to evaluate attentional bias and reactivity to emotional stimuli to further clarify relationships between constructs in MDD samples. As bilateral attenuations were not strongly correlated with current depression and anxiety symptoms, these findings may reflect a trait (as opposed to state) MDD marker. Additional research is warranted to determine whether this bilateral frontal attenuation is evident across multiple paradigms, and whether these results generalize to men as well as women.

Second, we hypothesized that MDD-AnxApp would exhibit greater IFG BOLD signal within the left hemisphere than the other three MDD groups specifically during loss anticipation; this prediction was also not supported, as MDD-AnxApp did not differ from the other MDD groups in any frontal region. However, evaluation of effect size differences between the four MDD groups and CTL suggests that MDD-AnxApp showed the greatest similarity to CTL in IFG BOLD signal to high arousal stimuli. Our findings of no hemispheric differences between MDD-AnxApp and CTL are also consistent with recent resting EEG work comparing these two groups (Nusslock et al., 2018); it may be the case that left hemisphere lateralization is stronger in individuals with pure anxious apprehension without comorbid MDD. Third, we forecasted that MDD-AnxAro and MDD-Both would show greater SFG and MFG BOLD signal within the right hemisphere than the pure MDD group specifically during loss anticipation; this prediction, based upon EEG asymmetry findings, was also not supported with fMRI data. Our findings suggest that frontal lateralization patterns in MDD are contingent upon the neuroimaging modality used in analysis; as fMRI BOLD signal is an indirect measure of neuronal activity, it is to be expected that there is not a 1:1 mapping to EEG signal, a direct measure of neuronal processing.

The present study benefited from the four clinical groups having similar distributions of single-episode versus recurrent MDE frequency, current versus remitted MDE symptoms, and medication use, as well as similar levels of rumination and depression symptom severity. Planned contrasts indicated that the two MDD groups with comorbid anxious arousal reported higher anxiety symptoms and greater behavioral inhibition (PROMIS anxiety, ASI, STAI, and BIS) than the pure MDD group, consistent with expectations. Although it would theoretically be expected that the four MDD groups would report lower approach motivation (BAS) and anticipatory and consummatory pleasure (TEPS) than CTL, no group differences emerged; null findings may be related to our all-female, as opposed to a mixed sex, sample (see Gheza et al., 2019). The current analysis benefitted from clear apriori hypotheses that were easily testable using planned contrasts, omitting the need for post-hoc tests.

Limitations

Despite benefits, the current study possesses multiple limitations. First, findings were based on ROIs averaged across anatomical subregions of SFG, MFG, IFG, and OFC to limit excessive testing (i.e. type I error inflation) and facilitate parsimonious explanations of group differences. Averaging across these subregions may have limited our ability to replicate frontal EEG asymmetry findings with fMRI data (e.g., Grimm et al., 2008; Herrington et al., 2010). Moreover, we did not include whole-brain voxelwise fMRI analyses of the MID task in this analysis, but acknowledge that regions outside of prefrontal cortex are important in MDD pathophysiology. Second, results are based on cross-sectional data mainly comprised of women diagnosed with recurrent MDD exhibiting symptoms of a current MDE; longitudinal data are needed to determine whether bilateral frontal attenuations index vulnerability to future MDD. Third, the three comorbid MDD and anxiety groups were categorized on the basis of lifetime GAD, SAD, PD, and PTSD diagnosis, which may not reflect participants’ current diagnostic status; although we collected a questionnaire measure of anxious arousal (ASI), we did not include a scale to quantify anxious apprehension symptoms (e.g., Penn State Worry Questionnaire) that may have allowed us to detect more nuanced differences between anxiety groups. Fourth, due to our limited number of men in Tulsa 1000 with MDD and anxiety, we restricted our sample to women; it is important for future research to directly test sex and hemisphere differences in frontal lateralization to generalize study findings. Fifth, in an ideal world we would relate simultaneously collected frontal EEG alpha power for each hemisphere to fMRI BOLD signal for the same task; although the T1000 study collected simultaneous EEG-fMRI during the MID task, we are in the early stages of preprocessing the EEG data so that we can incorporate it into fMRI analysis. We see the current analysis as the first step in this multi-step process. Sixth, although our targeted analyses were limited to prefrontal cortex, we acknowledge that brain asymmetry deviations as a function of MDD and anxiety are also present over parietal cortex, with pure MDD showing left parietal lateralization thought to reflect underarousal, and MDD with anxiety/anxious arousal symptoms showing right parietal lateralization thought to reflect hyperarousal (e.g., Bruder et al., 2017; Stewart et al., 2011); we plan to analyze parietal differences in these groups as a next step to determine whether parietal fMRI asymmetry patterns align with those seen with EEG.

Finally, is important to note that the degree of response to positive and negative stimuli in MDD appears to depend upon the paradigm employed to probe these responses, according to findings from multiple meta-analyses (Bylsma et al., 2008; Diener et al., 2012; Groenewold et al., 2013; Joyal et al., 2019; Zhang et al., 2013). For instance, for tasks probing changes in emotional responses, MDD is characterized by reduced emotional reactivity to positive and negative stimuli, indexed by self-report, behavior, and physiology, but the overall effect size is twice as large for positive as opposed to negative stimuli (Bylsma et al., 2008). In contrast, within the context of cognitive control tasks, MDD is characterized by greater interference to negative but not positive stimuli (Joyal et al., 2019). With respect to prefrontal brain function, meta-analyses are also inconsistent with respect to hypo- versus hyper-activations, as MDD has been linked to: (1) reduced left MFG BOLD signal to negative stimuli but increased OFC BOLD signal to positive stimuli across visual tasks probing emotional distraction, faces, words, or images, or monetary reward (Groenewold et al., 2013); (2) heightened right IFG/SFG BOLD signal during working memory load and negative emotional processing (Diener et al., 2012); and (3) heightened bilateral MFG or left SFG BOLD signal to rewards (Zhang et al., 2013). Although the current findings across win and loss anticipation conditions are suggestive of an arousal effect, the current data are not able to address potential arousal differences across valence. Thus, it could be the case that gains and losses of the same magnitude may differ in arousal level. Future research examining the MID task employing methodologies well-suited to differentiating arousal and valence effects (e.g., startle-reflex modulation; Sege et al., 2014) would be helpful to clarify this literature. Furthermore, research is warranted to determine whether hemispheric BOLD signal differences in regions of prefrontal cortex in various emotional tasks identify MDD symptom severity or predict treatment outcome.

Conclusions

Within a sizable sample of women, our findings suggest that lifetime MDD, irrespective of lifetime anxiety disorder, is characterized by attenuated bilateral superior, middle, and inferior frontal brain signals elicited by anticipation of wins and losses. These findings are consistant with the emotion context insensitivity theory of MDD, wherein blunted responsivity to emotional stimuli is linked to dysphoria. Further research is warranted to determine whether it is present across paradigms and modifiable with targeted therapeutic interventions.

Highlights.

• Four depression groups with presence/absence of anxious apprehension and arousal

• Depression linked to bilateral prefrontal attenuations during win/loss anticipation

• Comorbid anxiety did not moderate the link between depression and attenuations

• Categorical but not dimensional capture of depression related to attenuations