Ambivalence-dependent alterations in brain function during self-evaluation in adults with depressive symptoms

Abstract

Although patients with depression frequently report distorted self-related beliefs, such as all-or-nothing thinking, little is known about disruptions in behavioural and brain processes that occur when adults with depression make such self-evaluations. Here, we examined 117 participants ranging in depression severity (from non- to severely depressed) as they rated their belief in statements about themselves or famous people during fMRI. We used mixed-effects models to examine task characteristics (e.g. self vs other) and depression severity to test our hypotheses that during self-evaluation, adults with depression would show increased all-or-nothing thinking (i.e. decreased responding in the middle of the scale), slower reaction times (RTs), and increased brain activation and connectivity in cortical regions involved in self-evaluation. Greater depression severity was associated with increased ambivalence (decreased all-or-nothing thinking) overall. Furthermore, RTs during low-ambivalence judgements increased with depression severity. Depression severity was also linked to altered brain function, including decreased activation during low-ambivalence self-evaluation (vs other) in the medial prefrontal cortex, superior frontal cortex, and the perigenual anterior cingulate cortex (pgACC). Additionally, the pgACC displayed increased activation during high-ambivalence self-evaluation (vs other). Our findings clarify how adults with depression evaluate self-related beliefs, which may inform novel treatments to target distortions in these beliefs.

Introduction

Depression is a prevalent, debilitating disorder, affecting more than 17 million adults in the USA alone and costing billions of dollars annually (Malhi and Mann 2018, Greenberg et al. 2021). One hallmark feature of unipolar depression is distorted self-evaluation, such as negative bias when evaluating traits and abilities, yet the mechanisms underlying these distortions are poorly understood. In non-depressed individuals, knowledge and beliefs about the self appear to have special status in the brain (Klein and Lax 2010, Wagner et al. 2012): they are accessed rapidly, resistant to many forms of brain-related illness, more positive than negative, and subserved by a relatively constrained ventral region of the medial prefrontal cortex (MPFC). In contrast, individuals who are depressed report more negative self-evaluations, which seemingly activate a larger network of brain regions when accessed (Lemogne et al. 2009, Fournier and Price 2014, Philippi et al. 2018). Addressing dysfunctional self-related beliefs is a focus of common psychotherapies, including cognitive therapy (e.g. Beck and Greenberg 1984). While cognitive therapy is a highly effective treatment for depression and is superior to acute antidepressant medications in preventing relapse (Hollon et al. 2005, Vittengl et al. 2007, Beshai et al. 2011), its efficacy could be improved, as ∼40% of patients do not respond to acute cognitive therapy, and ∼30% of those who do respond are expected to relapse within a year (Hollon et al. 2005, 2006, Vittengl et al. 2007, Beshai et al. 2011). Furthermore, many individuals in remission from depression continue to experience altered self-evaluation (Allison et al. 2023), increasing their risk of relapse (Lye et al. 2020). This pattern of continued dysfunction in remitted patients with few residual symptoms suggests there are elements of self-evaluation that are not adequately targeted and altered with treatment.

Normative self-evaluation

Previous studies demonstrate that self-related information is prioritized in the brain. Non-depressed individuals answer self-related questions faster than questions unrelated to the self, an effect referred to as the self-reference speed advantage (Bargh and Tota 1988, Prentice 1990, Gillihan and Farah 2005, Lou et al. 2010, Christian et al. 2014, Tacikowski and Ehrsson 2016, Sui et al. 2023). Although it is difficult to determine if this advantage is unique to self-related information, or simply reflects a familiarity effect (Gillihan and Farah 2005), self-related information appears to be more readily accessible than information about others (Christian et al. 2014).

Additional evidence of specialized self-related information processing comes from studies with patients who, despite having neurological impairments affecting knowledge acquisition and retention, possess robust self-related information. Multiple studies have shown that although patients with injury-related amnesia cannot explicitly recall memories, they are nevertheless able to describe their traits and personality accurately (Klein and Lax 2010, Picard et al. 2013). Additionally, patients with developmental amnesia—who from birth cannot form episodic memories—appear able to accurately describe what they are like to others, suggesting intact self-related information (Klein and Lax 2010, Picard et al. 2013). Furthermore, the ability to accurately describe oneself—despite other knowledge impairments—also remains in patients with dementia and Alzheimer’s disease, though there is evidence that disease may prevent self-related information from updating after a certain stage (Klein and Lax 2010). Together, these cases highlight the resilience of self-related information, suggesting it may be differentiated from other types of information behaviourally and neurally.

Self-evaluation appears to have dedicated neural resources. For example, rapid self-evaluation tends to be associated with activation in the ventral portion of the MPFC—including portions of Brodmann Areas 10 and 32—thought to be pivotal for self-evaluation (Amodio and Frith 2006, Wagner et al. 2012, 2019). Localization of MPFC activation has been shown to differentiate between self and others, even close others, with self-related activation located more ventrally than other-related activation (Wagner et al. 2012). By contrast, social cognition (i.e. thinking about others) activates a wider range of regions in healthy adults, including regions within the default mode network (temporoparietal junction and precuneus), cognitive control network [dorsolateral prefrontal cortex (DLPFC) and ventrolateral prefrontal cortex], and salience network [anterior cingulate cortex (ACC) and the superior temporal sulcus/gyrus] (Zink et al. 2008, Van Overwalle 2009, Apps et al. 2016, Li et al. 2018).

Self-evaluation in depression

Unlike healthy individuals and those with the neurological disorders described above, individuals with depression demonstrate altered processing of self-related beliefs (Telles-Correia and Marques 2015, DeRubeis and Strunk 2017, Kendler 2020). Whereas never-depressed participants express mildly positive views of themselves, patients with depression routinely exhibit a negative bias when considering their traits and abilities (Beck and Greenberg 1984, Swallow and Kuiper 1988, Hards et al. 2020, Zell et al. 2020). Perhaps equally importantly, some studies show that adults with depression do not show the same self-reference speed advantage as those without depression. It is unclear, though, from these studies if the increased reaction time (RT) is due to general psychomotor slowing in depressed patients (Swallow and Kuiper 1988, Delaveau et al. 2016, Li et al. 2017).

Furthermore, in individuals with depression, neural activation appears to expand beyond the MPFC during self-evaluation into portions of social-cognitive regions within the default mode and cognitive control networks, including the precuneus and SFG (Lemogne et al. 2009, Pizzagalli 2011, Fournier and Price 2014, Philippi et al. 2018, PizzagalLi et al. 2018, Peng et al. 2021, Ibrahim et al. 2022), and the salience network (e.g. ACC, anterior insula; Fournier and Price 2014, Guo et al. 2015). Using a classic trait-judgement task, Lemogne et al. (2009) reported that participants with depression showed increased activation in the left inferior frontal gyrus (IFG) and dorsomedial frontal gyrus when making self (vs ‘other’) evaluations compared to non-clinical controls. Patients with depression also displayed increased connectivity between the medial frontal gyrus and the insula, ACC, SFG, and left IFG compared to non-depressed controls. The authors hypothesize that this increased activation and connectivity could be due to increased self-focused thought, requiring greater cognitive control and conflict monitoring. They suggest that participants with depression may be comparing themselves to an inner standard and finding themselves lacking. They argue that the salience of that standard and the conflict between these representations may account for the increased ACC activation and connectivity, and that cognitive control regions (e.g. the SFG) may be upregulated to either resolve the discrepancy or to avoid self-focus.

Shifting distorted self-related beliefs with psychotherapy

The cognitive model of depression posits that patients who are depressed often experience cognitive distortions when thinking about themselves, such as all-or-nothing thinking, and tend to endorse absolute statements like ‘I’ll never be enough’ or ‘I always mess up’ (Beck 1979, Beck 2021). In Cognitive Behavioural Therapy (CBT), patients are taught to identify dysfunctional thoughts and beliefs about themselves and to challenge those thoughts. For example, the strategy of cognitive restructuring involves examining the evidence for, and utility of, distorted thoughts and developing more accurate and/or useful alternatives (Beck and Greenberg 1984, Beck 2021). CBT is an effective therapy, equivalent to antidepressant medications (Hollon et al. 2005, Vittengl et al. 2007, Beshai et al. 2011), with approximately 40%–50% or more achieving remission after acute phase treatment (Casacalenda et al. 2002, Kennard et al. 2009, Johnco et al. 2024). However, although CBT relapse outcomes are equivalent to or better than pharmacotherapy alone, nearly one-third of acute CBT-treated patients with depression relapse in the first year (compared to >50% following acute pharmacotherapy) when both are discontinued (Hollon et al. 2005, Vittengl et al. 2007, Beshai et al. 2011, DeRubeis and Strunk 2017). Furthermore, there is evidence that distorted self-related beliefs persist into remission for some patients despite treatment with cognitive therapy, elevating relapse risk (Segal et al. 2006, Lau et al. 2012, Adler et al. 2015). For instance, those with remitted depression, compared to never-depressed controls, are slower to reject negative self-referential stimuli and show neural differences when responding to positive stimuli (Allison et al. 2023), display increased dysfunctional attitudes about themselves following negative mood induction (Weissman and Beck 1978, Lau et al. 2012), and continue to endorse implicit maladaptive beliefs despite shifts in explicit dysfunctional attitude ratings during cognitive therapy (Adler et al. 2015). Additionally, the degree to which negative self-evaluation can be contextually reactivated in remitted patients predicts relapse risk over 18 months (Segal et al. 2006). Together, these data suggest current treatments may not sufficiently treat the mechanisms associated with altered self-evaluation in depression, leaving patients vulnerable to returning symptoms.

To clarify the mechanisms of distorted self-related beliefs that are central to depression, the current study examines how self-related beliefs are processed and evaluated in adults across a range of depression severity. We build on the previous literature by exploring self-evaluation as a function of symptom severity (rather than group comparison), by using more naturalistic task stimuli, and by allowing self-evaluation to vary in ambivalence (rather than ‘Yes/No’ extreme responses). We assessed how—if at all—increased depression severity relates to behavioural ambivalence in belief ratings, RTs, and neural metrics when making self-evaluations. We hypothesized that greater depression severity would be associated with (i) decreased ambivalence (i.e. increased endorsement of all-or-nothing beliefs) during self-evaluation, (ii) increased RTs during self-evaluation, and (iii) increased ambivalence-dependent neural activity and connectivity during self-evaluation, both within and between brain regions involved in self-evaluation, social cognition, salience detection, and cognitive control.

Methods and materials

Participants

All procedures were approved by The Ohio State University Institutional Review Board. We recruited N = 135 participants (ages 18–40) from the community, representing a range of depressive symptoms from none to severe [Quick Inventory of Depressive Symptoms (QIDS)—range = (0,19) (Rush et al. 2003), additional details in Table S1, see online supplementary material for a colour version of this table]. To ensure adequate representation of individuals with depression, we oversampled participants with at least mild depressive symptoms, defined as a QIDS score >6 (full exclusion/inclusion details in online supplementary material). Eighteen participants were excluded for the following reasons after recruitment: excessive motion during scanning (6), incomplete or otherwise unusable behavioural data (5: 2 did not follow directions and 3 missed too many trials), acquisition problems (3), anatomical abnormality (1), drug test failure (1), and participant request to end study early (2), for a total N = 117 (69/117 QIDS >6, 59% with elevated depressive symptoms) (Table 1).

Table 1.

Sample characterization (N = 117).

Variable	M (SD)/N %
Age	28.11 (6.27)
Sex at birth (female/total)	66.7%
Race
American Indian/Alaskan Native	0.9%
Asian	15.4%
Black	7.7%
White	68.4%
More than one	7.7%
Hispanic or Latino	6.8%
QIDS	7.98 (5.56)
HAMD	9.32 (7.71)
MADRS	12.10 (11.18)

Eligible participants completed a battery of depression severity assessments, including the Hamilton Depression Rating Scale (HAMD) (Hamilton 1960), Montgomery–Åsberg Depression Rating Scale (MADRS) (Montgomery and Asberg 1979), and QIDS. Previous research has shown that clinician-rated and self-report measures capture different levels of depression severity, and there is added benefit to including both in depression studies (Olino et al. 2012, Uher et al. 2012). Accordingly, HAMD, MADRS, and QIDS scores were standardized and aggregated into a depression severity composite score.

Self/other evaluation task

In our Self/Other Evaluation task, participants rated the strength of their belief in statements from the Automatic Thoughts Questionnaire (ATQ), referencing either themselves or others (Hollon and Kendall 1980, Ingram et al. 1995), adapted for fMRI. This paradigm is similar to the Self-Reference Encoding Task used by others (Derry and Kuiper 1981, Dainer-Best et al. 2018) but differs in the stimuli and rating prompt used. Rather than simple traits or adjectives as stimuli, we used ATQ items reflecting real-world thoughts, whose frequency in daily life has been shown to distinguish between adults with and without depression (Hollon and Kendall 1980, Ingram et al. 1995, Du et al. 2015). Participants rated how much they believed the ATQ statement about a given target, either themselves or a famous other, on a scale from 1 (Not at all) to 5 (Completely), with 3 (Somewhat) in the middle. These stimuli allow us to more directly assess participants’ strength of beliefs, which may be more informative than ‘Yes/No’ agreement ratings. Participants identified five famous ‘others’ familiar to them prior to the scan. Twenty negative and 20 positive statements were presented, in the format shown in Fig. 1, and matched for sentence length, phoneme length, and word frequency. Each of the 40 statements was shown twice in a semi-random order, once referencing the self and once referencing one of the five famous others, for a total of 80 stimuli. We collected both participant ratings as well as their RT for each trial for behavioural and fMRI analysis.

Figure 1.

Self/Other paradigm. Example trials of our Self/Other evaluation paradigm, demonstrating the elements of Target, Valence, and Ambivalence. Coloured boxes are displayed here for clarity but were not included in the stimuli shown to participants.

Coding of behavioural responses

We recoded the behavioural responses to reflect ambivalence. Coding extreme values—those consistent with all-or-nothing thinking (1 and 5)—as ‘Low Ambivalence’ responses and middle responses (2, 3, and 4) as ‘High Ambivalence’ responses gave us sufficient trials per subject (Fig. S1, see online supplementary material for a colour version of this figure); thus, we performed subsequent analyses using these binarized ‘High’ and ‘Low’ ambivalence designations. In secondary analyses, we examined the two ‘Low Ambivalence’ response categories (1 and 5) separately (see the online supplementary material).

MRI protocol

Participants completed the Self/Other Evaluation task at either The Ohio State University or the University of Pittsburgh, both with a 3T Siemens Prisma scanner. Full scan parameters may be found in the online supplementary material. For both the region of interest and whole-brain analyses, scanner site was added as a covariate, and statistical models allowed variance/covariance parameters to differ across sites (see below).

Behavioural analysis

Response ambivalence and RT

Behaviourally, we examined differences in ambivalence and RT with increasing depression severity. We employed a mixed-effects logistic model with ambivalence as the outcome and depression severity, target (Self/Other), and valence (positive/negative) as predictors of interest. We likewise modelled the RT data using a mixed-effects model with depression severity, target, valence, and ambivalence level (high/low) as fixed effects. We cleaned the RT data to remove responses <300 ms. For the RT model, we used the Kenward–Roger approach to approximate degrees of freedom (Kenward and Roger 1997). To capture the structure of our repeated-measures data, each mixed-effects model included a random intercept, nested within subject ID, and was estimated using restricted maximum likelihood estimation. Age, sex, and scanner site were included as covariates for both behavioural models. Age, RT, and depression severity were standardized.

fMRI analysis

Preprocessing

We used fMRIPrep (23.0.1) (Esteban et al. 2019)—an automated pipeline intended to streamline and standardize preprocessing across datasets—to preprocess our data. Full preprocessing details can be found in the online supplementary material. Additionally, participants with a mean framewise displacement >0.5 and those with >20% of their volumes identified as motion outliers were excluded. All fMRIPrep output was visually examined to identify quality concerns. Following preprocessing, all functional files were smoothed using an 8 mm full width half maximum (FWHM) kernel.

First-level analysis

We performed first-level analysis using SPM12 (2024), using the canonical haemodynamic response function with an additional basis function representing temporal derivatives. Due to our interest in the response-dependent (i.e. ambivalence-dependent) activation and connectivity, we used the combination of each response type (i.e. ‘High/Low’ Ambivalence) in conjunction with each target (i.e. ‘Self/Other’) to form the four conditions that we included as our parameters of interest (e.g. ‘SelfHigh’, etc.). We included 24 motion regressors, 5 aCompCor regressors, and up to 5 tCompCor physiological noise regressors computed for each subject by fMRIPrep (details in online supplementary material), as well as the scanner drift regressors, as nuisance regressors. We used the resulting (standardized) betas from each of the four conditions for second-level analyses instead of examining contrasts (e.g. Edmiston et al. 2024), given the low reliability of subtraction-based contrasts (Hajcak et al. 2017). For our analyses of neural function, including valence, resulted in insufficient power (e.g. too few volumes per bin to support analyses of valence-by-target-by-ambivalence effects; Fig. S1, see online supplementary material for a colour version of this figure), so we collapsed across valence and focused the analyses on the target of the statement (Self/Other).

Second-level analysis

We conducted both activation and functional connectivity analyses at the ROI and whole-brain voxel-wise levels. ROI analyses were conducted by extracting averaged beta values across the ROI voxels. To capture the repeated measures structure of data across conditions, we used generalized linear models (using the gls package from nlme in R), in which we modelled the residual variance/covariance structures, which were allowed to vary by scanner site. A compound symmetry variance/covariance structure provided the best model fit (Table S2, see online supplementary material for a colour version of this table). Whole-brain models were estimated using the Neuropointillist (0.0.0.9000) package in R (Madhyastha et al. 2018). We used the Benjamini–Hochberg method for controlling the FDR (Benjamini and Hochberg 1995).

ROI definition

Our seven ROIs of interest—MPFC, Insula, Precuneus, ACC, SFG, and IFG—were defined as 10 mm-radius spheres centred around the peak coordinates of previous cited work that examined self- vs other evaluation (Fig. 2). Specifically, ventral MPFC and precuneus coordinates were taken from the results of trait evaluation tasks utilized by Courtney and Meyer (2020), whereas the insula, ACC, SFG, and left IFG coordinates came from Lemogne et al. (2009). Because the ACC and IFG were only reported in one hemisphere, and because we had no a priori laterality hypotheses for these regions, we mirrored these coordinates to create another sphere in the opposite hemisphere for bilateral coverage. These spheres were combined into a ROI for the precuneus, ACC, SFG, and IFG regions. Right and left insula spheres were retained as separate ROIs, as previous research has specifically implicated the right insula in the context of depression (Fournier and Price 2014, Fournier et al. 2017, 2022, PizzagalLi et al. 2018, Ibrahim et al. 2022).

Figure 2.

Visualization of a-priori ROIs. ROIs were defined as 10 mm spheres using coordinates from previous studies. MPFC = 1, PCC/Precuneus = 2, ACC = 3, IFG = 4, left insula = 5, right insula = 6, SFG = 7.

Multiple comparison correction

Following the recommendations from Woo et al. (2014), we used AFNI’s (Analysis of Functional NeuroImages, version 21.10) (Cox 1996, Cox and Hyde 1997) 3dClusterize tool for cluster-based multiple comparison correction. Using smoothness estimates from first-level model residuals and setting an uncorrected threshold of P < .001 resulted in an FDR-corrected (P < .05) cluster size of 38 voxels. For display and interpretation purposes, values were extracted from significant clusters to visualize the directionality of observed effects.

Functional connectivity

We calculated generalized psychophysiological interaction (gPPI) connectivity (McLaren et al. 2012) between the MPFC (seed) and our ROIs using SPM12. gPPI models were constructed using the individual condition regressors, as described in our first-level analysis. As an exploratory analysis, we also examined whole-brain voxel-wise gPPI connectivity from the MPFC using Neuropointillist as above.

Results

Behavioural

With all-or-nothing thinking/ambivalence as the outcome, our mixed-effects logistic regression model yielded a significant three-way interaction between target, valence, and symptom severity (x² = 12.14, P < .001, OR = 1.43). Contrary to our hypothesis, participants with greater depressive symptoms were more likely to answer with high ambivalence (i.e. less likely to respond with all-or-nothing evaluations at the extremes of the scale) when making self-evaluations (vs others) compared to participants with fewer depressive symptoms, with a stronger effect when judging negative compared to positive statements (Neg: z13.80, P < .001, OR = 2.88; Pos: z = 9.76, P < .001, OR = 2.01) (Fig. 3).

Figure 3.

Probability of responding with high ambivalence (i.e. low all-or-nothing thinking) by depression severity, target, and valence. A significant interaction between target, valence, and depression severity (x² = 12.14, P < .001) revealed that effects were driven by self-evaluation trials.

Regarding our mixed-effects RT model, we observed a significant three-way interaction between ambivalence level, target, and symptom severity [F(1,9168) = 24.76, P < .001, β = 0.19]. In support of our hypothesis, post-hoc comparisons revealed a significant increase in RTs with increasing depression severity during self-evaluation (vs other) (z = 7.80, P < .001, β = 0.21) for low-ambivalence (Fig. 4 and Fig. S2, see online supplementary material for a colour version of this figure). Importantly, the slower RTs cannot be attributed to general slowing, as participants with greater depression showed no differences from those with lower depression in their RTs to self- vs other-evaluations for high-ambivalence responses (z = 0.72, P = .47, β = 0.019).

Figure 4.

Visualization of the three-way Target × Ambivalence × Depression Severity interaction on reaction time. To visualize differences in target over levels of ambivalence (i.e. all-or-nothing thinking) and depression severity, we calculated the differences in target (i.e. Self/Other) reaction times and displayed them as a function of ambivalence level, and depression severity [F(1, 9168) = 24.76, P < .001)].

Neural activation

Two of our a priori ROIs, the MPFC and the SFG (Fig. 5), exhibited a three-way interaction between target, ambivalence, and depression severity [MPFC: F(1,457) = 6.68, P = .035, β = 0.30; SFG: F(1,457)=7.87, P = .035, β = 0.29]. This was driven by activation differences during low-ambivalence responses in both the MPFC [t(455)-2.25, P = .025, β = −0.19] and SFG [t(455) = −2.29, P = .022, β = −0.17], where increasing depression severity was associated with decreased self-related activation compared to other-related activation (for a visualization of each condition separately, see Fig. S3, see online supplementary material for a colour version of this figure). A post-hoc analysis for the SFG ROI revealed that left SFG [F(1,457) = 10.47, P = .0013], but not the right [F(1,457)=3.02, P = .083], demonstrated this three-way interaction.

Figure 5.

Visualization of the three-way Target × Ambivalence × Depression Severity interaction on BOLD activation in our significant ROIs [a priori ROIs: MPFC = F(1,457) = 6.68, P = .035; SFG = F(1,457) = 7.87, P = .035; whole-brain ROI: pgACC = F(1,457) = 16.40, P = .0001]. To visualize differences in target over levels of ambivalence (i.e. all-or-nothing thinking) and depression severity, we calculated the differences in target (i.e. Self/Other) BOLD activation and displayed them as a function of ambivalence level and depression severity. Post-hoc analyses indicated that the interaction was driven by differences in low-ambivalence activation in the MPFC (Low: t(455) = −2.25, P = .025) and SFG (Low: t(455) = −2.29, P = .022), whereas in the pgACC, both low- and high-ambivalence activation differed significantly from zero (Low: t(455) = −3.09, P = .0021; High: t(455) = 2.63, P = .0087).

Whole-brain analysis identified a three-way interaction [F(1,457)=16.40, P = .0001, β = 0.46] between symptom severity, target, and ambivalence in a cluster in the perigenual anterior cingulate cortex (pgACC) (size = 42 voxels, peak MNI: −6.5, 50.5, 14.5, peak intensity = 15.35) (Fig. 5). This cluster partially overlapped with, but was located more medially than, the left portion of the SFG ROI (Dice coefficient = 0.11; Dice 1945; Fig. S4, see online supplementary material for a colour version of this figure). This effect was driven by differences in activation between self- and other-evaluation during both high and low ambivalence trials. When making low-ambivalence responses, activation during self-evaluation compared to other-evaluation decreased with depressive severity [t(455) = −3.09, P = .0021, β = −0.25]. Additionally, when making high-ambivalence responses, activation during self-evaluation (compared to other-evaluation) increased with depressive severity [t(455) = 2.63, P = .0087, β = 0.21]. Recognizing that total endorsement vs total rejection of a statement may be associated with different patterns of response (Wagner et al. 2012), we also performed a supplementary analysis with the two low-ambivalence responses (‘Not at all’ and ‘Completely’) modelled as separate conditions (see online supplementary material for details and visualization, Fig. S5, see online supplementary material for a colour version of this figure). Our results show that both low-ambivalence response options (‘Not at all’ and ‘Completely’) were associated with similar neural response patterns that were different from that of the high-ambivalence responses, suggesting it is ambivalence, rather than the type of extreme response, that may be disrupted in depression.

Additionally, we conducted a post-hoc analysis investigating the impact of RT on neural activation during low-ambivalence evaluations. To do this, we ran first-level analyses as before, but including RT as a parametric regressor. Within each of our eight ROIs (7 a priori + 1 whole-brain), we conducted a one-sample t-test evaluating whether the mean RT-dependent activation in that region differed significantly from zero. In the MPFC, longer RTs were associated with significantly decreased activation, aligning with both our behavioural RT findings and our MPFC fMRI results [t(215) = −2.51, P = .013, Fig. S6, see online supplementary material for a colour version of this figure].

Neural connectivity

We did not observe any significant associations between depression, ambivalence, and target in either our ROI or whole-brain gPPI analyses (all Ps > .05; Table S3, see online supplementary material for a colour version of this table).

Discussion

Given the centrality of distorted self-related beliefs to depression, identifying the behavioural and neural mechanisms underlying this dysfunction could point to ways to refine our treatments. The current study probed how depression severity was related to one hypothesized mechanism—low ambivalence/all-or-nothing thinking—using an adapted Self/Other evaluation fMRI task. We found that depression severity is associated with greater ambivalence in behavioural ratings and slower RTs as well as an increased activation difference between low-ambivalence self- and other-evaluations in the MPFC, SFG, and pgACC.

The results we observed are somewhat surprising considering our hypothesis that depression would be related to decreased ambivalence—in line with increased all-or-nothing thinking frequently observed among adults with depression (e.g. Beck 1979, 2021, Disner et al. 2011). Our data show an increase in ambivalent responses (i.e. fewer all-or-nothing responses) with greater depression severity. This is not the first study to report this pattern. In their 1988 review, Swallow and Kuiper concluded that several smaller studies had shown that patients with depression exhibited decreased confidence and uncertainty in their own attributes and suggested that those with depression may lack a ‘consolidated self’. More recently, others have shown that individuals with depression exhibit decreased confidence in their judgements and abilities in general (Szu-Ting Fu et al. 2012, Hoven et al. 2019) and argue that these abnormalities in confidence may underlie many psychiatric disorders, including depression. Our results are consistent with the hypothesis that adults with depression show decreased confidence in their own attributes, which could manifest as increased ambivalence during self-evaluation.

We also found that participants with greater depression severity did exhibit a loss of the self-reference speed advantage, as expected, but these longer RTs were restricted to low-ambivalence responses. This, again, may point to increased uncertainty when making self-evaluations. Although these results appear to contradict the widely accepted belief that dysfunctional thoughts experienced by patients (which the ATQ stimuli were designed to capture) occur rapidly and are automatically accepted as true, it is important to note that the stimuli examined in the current study were not internally generated. Rather, participants saw common positive and negative automatic thoughts and were asked to evaluate them explicitly. The increased ambivalence and RTs we observed in those with elevated depression suggest that when adults with depression are asked to explicitly reflect on statements about themselves, the process is slower and results in more ambivalence than it does for those with lower depression symptoms.

One explanation for the observed pattern of results is that individuals with depression have difficulty resolving conflicting representations about themselves. This could be, as Lemogne et al. (2009) suggest, a conflict between the perceived self and an internal standard, a concept similarly captured by self-discrepancy theory (Strauman and Eddington 2017), which predicts that when a person perceives a discrepancy between their current self and their ideal self, they experience dysphoria. It may also be that subjects with depression have competing positive and negative representations that require resolution. When asked to explicitly evaluate a statement about themselves, some individuals with depression may need to resolve conflicting internal representations, culminating in both increased RTs and higher levels of ambivalence.

Our fMRI findings further highlight the ambivalence-dependent differences in self- and other-evaluation with increasing depression severity. In the MPFC and SFG, participants with higher depressive severity showed a larger activation difference between self- and other-evaluation during low-ambivalence trials; the more depressed a participant was, the lower the predicted self-related activation (compared to other-evaluation). The MPFC is critically involved in the representation of self and others (Wagner et al. 2012), and it may be that decreased activation in the MPFC during low-ambivalence self-evaluation reflects impaired access to self-related information. Due to the SFG’s role in working memory and higher-order processing, decreased activation during low-ambivalence self-evaluations in the SFG may indicate impaired cognitive control when attempting to resolve self-related questions, leading to increased ambivalence and longer RTs (Lemogne et al. 2009).

Whole-brain analysis additionally identified a cortical midline cluster in the pgACC exhibiting differential activation for target and ambivalence in participants with depression. Like the MPFC and SFG, this region showed decreased self-related activation associated with low-ambivalence responses. Our secondary analysis suggested that this decrease in activation was present regardless of which low-ambivalence response option the participant chose (i.e. ‘Not at all’ or ‘Completely’), suggesting that it is the process of responding with low ambivalence itself, not the presence or absence of endorsement, that may be aberrant. In recent years, the pgACC has been identified in multiple studies examining the pathophysiology of depression. The pgACC shows increased connectivity with the DLPFC (an area of cognitive control) in patients with depression (Davey et al. 2012), as well as increased resting-state connectivity with the middle frontal gyrus and precuneus, which was associated with negative self-focused thought (Philippi et al. 2018). Interestingly, Wagner et al. (2015) found increased pgACC activation during negative self-evaluation using a similar paradigm to ours, which the authors attributed to an increased need for conflict monitoring during those trials. Though their results seem opposed to our pgACC findings, we note that these authors did not examine response-dependent neural activity, as we did. Of particular interest is the recent finding that shifts in excitatory/inhibitory metabolite concentrations in the pgACC are associated with decreased self-awareness within the context of alexithymia, which may be due to low reflective functioning and high uncertainty about the self (Kühnel et al. 2020). The pgACC results that we observed, in conjunction with our ambivalence and RT findings, are consistent with the hypothesis that adults with depression experience decreased certainty when evaluating their own attributes.

Interestingly, the pgACC also showed significant differences between self- and other-evaluation when participants made high-ambivalence responses. This suggests that there may be depression-related alterations in social cognition as well as self-cognition. Previous work has shown that social comparison—involving both the self and others—is abnormal in depression (Furnham and Brewin 1988, Swallow and Kuiper 1988), and that lonelier individuals show increasingly differentiated neural patterns between themselves and others (Courtney and Meyer 2020). The differentiation we observed is consistent with these findings, although further work is needed to confirm these effects. We note that altered activation in these regions during high-ambivalence responses may also be associated with increased rumination in depression. Previous work has shown that increased rumination in subjects with depression is associated with increased activation in core regions of the DMN, including the MPFC, as well as altered functional connectivity between the MPFC and orbitofrontal regions (Satyshur et al. 2018, Zhou et al. 2020). It is certainly possible that by asking participants to reflect on themselves, the task may have triggered rumination. But in the context of the current study alone, it is difficult to determine the role that ruminative processes may have played. Future work, which specifically probes rumination during self-evaluation, will be needed to determine the potential role of rumination in the neural patterns we observed.

Our results suggest that the process of rapidly accessing self-related information may be disrupted in depression, whether through decreased access to existing self-representations, increased competition between contradicting representations, or some other mechanism. These disruptions may provide insight into why distortions in self-related beliefs continue despite current treatments and may inform new therapeutic approaches. Further research could explore whether the brain-behaviour alterations seen here, such as the ambivalence in self-evaluations or alterations in activation, persist and shape real-world daily experiences in individuals in remission from depression. Novel interventions may be necessary to target these altered aspects of self-related processing in depression.

Limitations

First, despite observing both behavioural and neural activation differences with increased depression severity, we did not identify any connectivity differences between the MPFC and our ROIs or at the whole-brain level. This is somewhat surprising, considering that other studies have observed connectivity between these areas associated with depressive symptomology (Sheline et al. 2010, Pizzagalli 2011, Jing et al. 2020). It could be that we are underpowered to capture existing connectivity differences for such a complex relationship (i.e. a three-way interaction). This may be particularly true if it turns out that the evaluation of both self- and other-related stimuli triggers patterns of connectivity with the MPFC that are quite similar, potentially reflecting a more general self/social process. Including different comparison conditions, e.g. blocks of unstructured rest, or trials requiring non-self or social salience detection or cognitive manipulation, could help to reveal more nuanced differences in patterns of connectivity associated with self- vs other-evaluation. Finally, it is possible that different seed regions may have revealed different connectivity patterns. This is the first study of which we are aware to examine neural function as it relates to ambivalence in self-ratings in depression. As this research continues to develop, connectivity with seeds in alternative network hubs (e.g. in the salience or cognitive control networks) could emerge as important contributors to differences in self-evaluation in depression.

Second, we are limited in what we can conclude about the precise reasons participants with higher depression severity showed higher ambivalence; we cannot differentiate with our paradigm whether the increased ambivalence we observed is indicative of true ambivalence (i.e. an equal weighting of positive and negative evidence), uncertainty/low confidence, or a kind of specific self-related apathy in their response to the prompt. Further work would benefit from addressing participants’ subjective certainty and confidence when making self- and other-related evaluations. To start, it would be informative to determine whether measures of ambivalence, uncertainty/low confidence, and self-directed apathy tend to cohere together across participants, whether one of these constructs is more prevalent for most patients, or whether there are individual differences among adults with depression in the prominence of each of these processes. Determining which of these processes (true ambivalence, uncertainty, or apathy) is responsible for the patterns we observed will be important for future treatment development work, as each would suggest potentially different underlying mechanisms. True ambivalence might, e.g. suggest the presence of a unitary self-representation that does not have the positively biased valence that is often observed among non-depressed individuals; uncertainty might suggest the presence of two or more conflicting self-representations and the engagement of conflict-monitoring neural systems; and self-specific apathy might suggest an aberrant reward-related process associated with self-relevant information. That said, it appears unlikely that the increased ambivalence that we observed in subjects with greater depression severity reflects a general process, like general slowing or general apathy, as the behavioural effects we observed were selective for self-evaluation in subjects with depression. No behavioural differences emerged as a function of depression during the evaluation of others.

Third, although our results suggest differences in the conscious evaluation of self-related beliefs, we cannot say whether these differences extend to the spontaneous evaluation of self-related thoughts in everyday life. Furthermore, we do not know how these effects may shift with mood state, over time, or with treatment. Finally, we note that our decision to collapse across responses in our scale (to have sufficient trials in each bin) may have masked more nuanced responding patterns for middle-ground responses (i.e. differences in responses of 2, 3, or 4 on the scale).

Conclusions

We observed clear behavioural and neural differences in the way individuals with greater depression severity process beliefs about themselves and others compared to less- and non-depressed individuals. Specifically, we observed increased ambivalence (i.e. fewer all-or-nothing evaluations) in adults with depression, as well as slower response times and differential ambivalence-dependent activation within the MPFC, SFG, and pgACC. These alterations may point towards opportunities for improved treatments and may help explain why distorted beliefs continue despite treatment.

Author contributions

Athena Biggs (Data curation [lead], Formal analysis [lead], Methodology [equal], Visualization [lead], Writing—original draft [lead]), Nikki Puccetti (Data curation [supporting], Investigation [equal], Methodology [equal], Software [supporting], Writing—review & editing [supporting]), Neil Patrick Jones (Conceptualization [supporting]), Patrick Da Silva (Data curation [equal], Software [lead], Writing—review & editing [supporting]), Henry Chase (Conceptualization [supporting], Writing—review & editing [supporting]), Mary L. Phillips (Conceptualization [supporting], Writing—review & editing [supporting]), and Jay C. Fournier (Conceptualization [lead], Formal analysis [supporting], Funding acquisition [lead], Investigation [lead], Methodology [supporting], Project administration [lead], Resources [lead], Software [supporting], Supervision [lead], Visualization [supporting], Writing—review & editing [supporting])

Supplementary data

Supplementary data are available at SCAN online.

Conflict of interest: None declared.

Funding

This work was supported by the National Institute of Mental Health [MH112758, MH122674]. The funder did not participate in conducting the research, collecting, analysing, interpreting the data, writing the manuscript, or deciding to submit the article for publication.

Data availability

The raw data used for the current study can be found in collections C2945 and C3506 at the National Data Archive; additional information will be shared upon reasonable request to the corresponding author.