Brain Mechanisms of Rumination and Negative Self-Referential Processing in Adolescent Depression

Abstract

Background

Depression is linked to cognitive biases towards more negative and less positive self-relevant information. Rumination, perseverative negative thinking about the past and the self, may contribute to these biases.

Methods

159 adolescents (12–18 years), with a range of depression symptoms, completed the SRET during fMRI. Multiple regressions tested associations between conventional self-report and ecological momentary assessment (EMA) measured rumination, and neural and behavioral responses during a self-referent encoding task (SRET).

Results

Higher rumination (conventional self-report and EMA) was associated with more negative and fewer positive words endorsed and recalled. Higher self-reported (but not EMA) rumination was associated with higher accuracy in recognizing negative words and greater insula and dorsal anterior cingulate activity to negative versus positive words.

Limitations

The sample included mostly non-Hispanic White participants with household incomes above the national average, highlighting the need for replication in more diverse samples. Word endorsement discrepancies required fMRI analyses to model neural response to viewing negative versus positive words.

Conclusions

Adolescents with higher rumination endorsed and recalled more negative and fewer positive words and recognized more negative words during the SRET. Higher insula reactivity, a key region for modulating externally-oriented attention and internally-oriented self-referential processes, may contribute to links between rumination and negative memory biases. These findings provide insight into neurocognitive mechanisms underlying depression.

Introduction

Major depressive disorder is a common and costly mental illness (Marx et al., 2023). Rates of depression increase substantially during adolescence (Avenevoli et al., 2015), a trend that has accelerated in recent years (Twenge & Joiner, 2020; Weinberger et al., 2018). Cognitive theories of depression indicate that a cognitive style characterized by frequent, often automatic, negative self-judgments is a major risk factor for depression (Abela & Hankin, 2008; Disner et al., 2011). Indeed, ruminative thought processes have been consistently linked to depression onset, duration, and severity (Nolen-Hoeksema, 2000; Smith & Alloy, 2009). Additionally, some cognitive models posit that there may be a bidirectional relationship between depressogenic biases in information processing and rumination in the context of depression (DeJong et al., 2016; Hilt et al., 2014). Identifying cognitive and neural mechanisms that confer risk for adolescent depression could aid in early identification and treatment before the onset of full depressive episodes (Thapar et al., 2022).

Cognitive models of depression posit that individuals with depression hold negatively biased views or schemas about the world, the future, and themselves (Beck, 2019). The presence of negative self-schemas in depression has been shown to develop and peak during adolescence (Alloy et al., 2012; Butterfield et al., 2023; McArthur et al., 2019). The self-referent encoding task (SRET) is commonly used to assess self-schemas (Gotlib et al., 2004). In the SRET, participants are asked to make decisions about whether positive or negative adjectives do or do not describe them. The task also involves free recall and recognition components, which allow researchers to measure self-concept via endorsement of negative versus positive words, and memory biases via discrepancies in the recall or recognition of negative and positive words. Prior work using the SRET in adult samples has found that multiple symptoms of depression were linked to more negative self-views (Beevers et al., 2019; Hitchcock et al., 2023), whereas mindful awareness was linked to stronger positive self-views (Hitchcock et al., 2023). Recent work using computational modeling has also shown that depression is associated with memory deficits for positive words, which were driven by discrepancies in evidence accumulation for positive versus negative words (Cataldo et al., 2023).

Similar associations between negative information processing biases and depression symptoms have been found in adolescents. For example, adolescent girls with depression endorse more negative words and fewer positive words and recall and recognize fewer positive words than healthy controls (Auerbach et al., 2015; Speed et al., 2016). Furthermore, in depressed teens, event-related potentials in frontal attentional brain regions to negative SRET words was associated with negative self-view and self-criticism (Auerbach et al., 2015). In a community sample of early adolescents, more negative and fewer positive words endorsed and recalled were associated with concurrent and future increases in depression symptoms (Connolly et al., 2016). These findings suggest that across a range of depression severity, negative stimuli capture and sustain attention, particularly in the context of self-evaluation, which may in turn contribute to memory biases favoring negative information and self-concepts.

In a recent review of the neurobiological correlates of self-referential processing (broadly defined) in adolescent depression, Butterfield and colleagues (2023) present a model in which more negative self-concept is influenced by social factors and dysfunction in cortical midline brain regions during self-referential processing, which increases risk for depression (Butterfield et al., 2023). However, this review highlights the limited research in this area, as only a handful of studies have investigated self-referential processing in adolescents with depression, and only three used a similar task to the SRET. The few available studies suggest that adolescents with depression show increased activation in midline cortical regions, including the anterior cingulate and medial prefrontal cortex, during a variety of self-referential tasks, including when evaluating if negative phrases describe them during the SRET (Quevedo et al., 2017), peer rejection (Silk et al., 2014), and when prompted to ruminate about negative events (Burkhouse et al., 2017). Further underscoring the link between brain function, depression, and rumination, Kaiser and colleagues (2019) found that the amount of time that youth spent in the frontoinsular and default mode network during resting-state fMRI was associated with more severe depression, and that rumination mediated this association (Kaiser et al., 2019). Together, these findings suggest that negative stimuli are highly salient to youth with depression and that heightened reactivity in medial frontal and insular regions, particularly during self-referential processing of negative words/events, may increase risk for maladaptive rumination and negative processing biases. However, no study has explicitly examined links between rumination, negative processing biases, and brain activity during self-referential processing in youth across a range of depression severity.

Rumination may be an important mechanism by which negative self-concept and affective memory biases are linked to depression. Rumination refers to repetitive, negative self-referential thoughts, typically focused on one’s negative feelings and problems (Nolen-Hoeksema et al., 2008). Recently, using an ecological momentary assessment (EMA) paradigm, we found that when teens reported elevated rumination, they were also more likely to be thinking negative thoughts about themselves and the past (Pidvirny et al., 2023; Webb et al., 2022c). Rumination is posited to be maladaptive due to its prolonged focus on negative evaluations of the self, and contributes to sustaining negative mood states, and fuels depressogenic cognitive biases (Nolen-Hoeksema et al., 2008; Rude et al., 2007). For example, adults with higher levels of rumination have more negative attentional and memory biases on the SRET and other attentional tasks (Joormann et al., 2006), and more negative autobiographical memory biases (Lyubomirsky et al., 1998).

Most studies measure rumination using traditional, retrospective self-report questionnaires which ask participants to recall and summarize their thinking in general over a lengthy period (e.g., weeks). For example, a prominent measure of rumination in youth, the Children’s Response Styles Questionnaire (CRSQ) (Abela et al., 2004), requires participants to report on what they “generally” do when feeling sad. Recently, there has been increasing interest in using EMA surveys to acquire a more fine-grained and ecologically valid assessment of current or recent (i.e., since the last survey) affect, cognitions, and behaviors in daily life. EMA uses mobile phones to deliver brief surveys that participants can answer during their daily lives. Using EMA to probe rumination may be particularly useful in teens with depression symptoms, as EMA may be less prone to recall bias and have higher ecological validity than conventional self-report measures (Mofsen et al., 2019; Van den Bergh & Walentynowicz, 2016). However, recent work suggests that conventional self-report and EMA measures of rumination are only modestly correlated and that each may explain unique variance in depression symptom improvement during treatment (Webb et al., 2023a). Thus, assessing rumination through multiple modalities may provide unique insight into mechanisms sustaining depression.

The current study investigated the links between conventional and EMA-measured rumination, and neural reactivity during self-referential processing, and subsequent memory biases in youth with a range of depression severity. We hypothesized that youth with higher rumination, both on conventional self-report and EMA measures, would demonstrate a bias towards more endorsement, recall, and recognition of negative words versus positive words during the SRET. Given the links between depression, rumination, and altered brain function during self-referential tasks (Butterfield et al., 2023; Kaiser et al., 2018) and at rest (Kaiser et al., 2019), we hypothesized that higher rumination would be associated with greater activity in midline cortical and insular regions to negative versus positive words during the SRET. Analyses were not pre-registered.

Method

Participants

Youth aged 12–18 years old were recruited across a range of anhedonia (M_age[SD]=15.96[1.94]; Table 1) as a part of ongoing studies focused on adolescent anhedonia.(Murray et al., 2023; Webb et al., 2022a) At the initial study session, youth completed a diagnostic interview (K-SADS; Kaufman et al., 1997) and self-report measures, including the Snaith-Hamilton Pleasure Scale (SHAPS; Snaith et al., 1995). Teens with elevated depression symptoms (n=75: 55 female, 20 male) were defined based on experiencing elevated anhedonia on the SHAPS, but also showed elevated depression symptoms on the Center for Epidemiological Studies Depression (M[SD]=31.56[11.05]). Typically developing teens (n=84: 57 female, 26 male) were defined based on reporting no current anhedonia and no history of DSM-5 diagnosis (See Supplement for details). This recruitment strategy allowed for sufficient variance in rumination across a full range of depression severity. Eligible youth also completed an MRI scanning visit followed by a 5-day (Thursday-Monday) EMA protocol consisting of 2–3 surveys per day (2/day on weekdays 3/day on weekends), delivered via the MetricWire smartphone application (See Supplement for additional information). All procedures were approved by the Mass General Brigham IRB.

Table 1:

Sample Characteristics.

Elevated Depression Sx.	N	%	Typically Developing	N	%
Biological Sex			Biological Sex
Female	55	73.3	Female	57	68.7
Male	20	26.7	Male	26	31.3
Race			Race
American Indian or Alaska Native	0	0.0	American Indian or Alaska Native	0	0.0
Asian	8	10.7	Asian	10	12.1
Black or African American	7	9.3	Black or African American	7	8.4
Native Hawaiian or Other Pacific Islander	1	1.3	Native Hawaiian or Other Pacific Islander	1	1.2
White	48	64.0	White	57	68.7
Other	3	4.0	Other	1	1.2
More than one race	8	10.7	More than one race	7	8.4
Ethnicity			Ethnicity
Hispanic or Latino	9	12.0	Hispanic or Latino	6	7.2
Not Hispanic or Latino	66	88.0	Not Hispanic or Latino	77	92.8
Current Diagnoses (DSM-5)			Current Diagnoses (DSM-5)
Major Depressive Episode	28	37.3	Major Depressive Episode	0	0.0
Generalized Anxiety Disorder	15	20.0	Generalized Anxiety Disorder	0	0.0
Social Anxiety Disorder	9	12.0	Social Anxiety Disorder	0	0.0
Panic Disorder	2	2.7	Panic Disorder	0	0.0
Specific Phobia	1	1.3	Specific Phobia	0	0.0
Attention-Deficit / Hyperactivity Disorder	4	5.3	Attention-Deficit / Hyperactivity Disorder	0	0.0
Oppositional Defiant Disorder	2	2.7	Oppositional Defiant Disorder	0	0.0
Obsessive Compulsive Disorder	2	2.7	Obsessive Compulsive Disorder	0	0.0
Medication			Medication
SSRI	10	13.3	SSRI	0	0.0
	M	SD		M	SD
Age (in years)	15.9	2.0	Age (in years)	16.0	1.9
Family Income (dollars)	136,875	86,077	Family Income (dollars)	166,797	87,538
CESD Score	31.56	11.05	CESD Score	6.06	5.43
CRSQ Score	38.66	9.23	CRSQ Score	23.68	6.88
EMA Rumination	1.53	0.95	EMA Rumination	0.40	0.40
SRET Characteristics	M	SD	SRET Characteristics	M	SD
Endorse			Endorse
Positive	12.0	5.2	Positive	23.8	3.8
Negative	14.2	6.4	Negative	2.4	3.1
Recall			Recall
Positive	6.8	2.9	Positive	7.3	3.0
Negative	7.4	2.9	Negative	6.9	3.2
Positive recall & endorsed	3	2.2	Positive recall & endorsed	5.7	2.6
Negative recall & endorsed	3.9	2.5	Negative recall & endorsed	0.7	1.3
Positive recall proportion	0.11	0.09	Positive recall proportion	0.22	0.1
Negative recall proportion	0.15	0.09	Negative recall proportion	0.03	0.05
Recognition			Recognition
Positive	26.6	3.9	Positive	26.9	4.0
Negative	27.0	3.1	Negative	25.4	4.6
Positive d’	2.89	0.81	Positive d’	2.78	0.83
Positive c	0.11	0.34	Positive c	0.03	0.41
Negative d’	2.91	0.72	Negative d’	2.55	0.71
Negative c	0.09	0.31	Negative c	0.16	0.47

Measures

Rumination

Self-reported rumination was assessed at the screening visit using the rumination subscale of the Children’s Response Styles Questionnaire (CRSQ)(Abela et al., 2004). The CRSQ is a 25-item questionnaire assessing responses to depressive symptoms. Rumination items (n=13) included items such as “When I am sad, I think, ‘There must be something wrong with me or I wouldn’t feel this way’” and were summed to create a trait measure of rumination. Consistent with prior work (Abela et al., 2002), the rumination subscale of the CRSQ demonstrated excellent reliability (α=.94).

Rumination was also assessed during EMA. During each EMA survey, youth were asked to report on the most stressful or negative time since they completed the last survey. They then completed two rumination items adapted from (Ruscio et al., 2015) “After this stressful thing happened, I was dwelling on my mistakes, failures, or losses” and “After this stressful thing, I kept thinking about something negative that has happened.” Items were rated from 1 (“Very slightly or not at all”) to 5 (“Extremely”). Scores from the two items were averaged at each time point and then aggregated across the 5-day EMA period to create a measure of mean state rumination, as we have done previously (Webb et al., 2020; Webb et al., 2023b). Split-half reliability (r=.76, p<.001) was calculated by randomly splitting each participant’s EMA dataset into two, computing mean rumination in each subset for each subject and correlating these scores. A total of 5 subjects were missing CRSQ data, and 13 were missing EMA data (See Supplement).

Self-Referent Encoding Task

During the MRI scan, participants completed an event-related SRET (Gotlib et al., 2004), consisting of 60 trials in which a word was displayed in white text on a black background (2160ms). Thirty words were positively valenced (e.g., “friendly”) and thirty were negatively valenced (e.g., “terrible”) (Bradley & Lang, 1999). After a jittered ISI (720–5760ms), participants were instructed to indicate via button press whether each word described them or not. Trials were separated by a jittered ITI (720–5760ms).

Immediately following the fMRI scan (approximately 15-minutes after SRET), participants were instructed to write down as many words as they could remember. Participants were given three minutes to complete the recall task. Following recall, participants completed a recognition task in which they were shown a list of 120 words (60 novel words, 60 task words) and were instructed to indicate which words were shown during the SRET, regardless of whether they endorsed those words. Nine subjects were missing SRET endorsement data, and seven subjects were missing SRET word recognition and recall data (See Supplement).

MRI Collection and Processing

Imaging data was collected on two scanners. 12 participants were scanned using a Siemens Tim Trio 3.0 Tesla MRI equipped with a 32-channel coil, and 112 participants were scanned using a Siemens Prisma 3.0 Tesla MRI system equipped with a 64-channel coil. Functional images were collected using the following parameters, TR=720ms, TE=30ms, FOV=212mm, multiband accelerator factor=6, voxel size=2.5 × 2.5 × 2.5.

Analyses were conducted in SPM12. Data were grey matter segmented, realigned and unwarped with a field map, slice-time corrected, co-registered to high-resolution structural images, normalized to Montreal Neurological Institute (MNI) space, resampled to 2×2×2mm voxels, and smoothed with a 4mm FWHM gaussian filter. Artifact Detection Tools software (http://www.nitrc.org/projects/artifact_detect/) was used to identify movement outliers (>3 SD from mean intensity, or >1mm movement) and create a regressor in each subject’s first level model. Seven subjects (3 elevated depression, 4 typically developing) were excluded from MRI analyses because of excessive movement, eight subjects (6 elevated depression, 2 typically developing) were missing SRET endorsement data, and one typically developing subject’s imaging data was lost due to a scanner failure.

Statistical Analysis

SRET Behavioral Analyses

Multiple regression analyses were conducted in which conventional self-reported rumination and EMA rumination were included as predictors of interest (in the same model), and age and sex were included as covariates. Separate regressions were conducted for each of the following 8 SRET behavioral variables (dependent variables).

Endorsement.

Endorsement was measured using the number of positive words endorsed and number of negative words endorsed, separately. The negative endorsement variable was square-root transformed to address heteroscedasticity.

Recall.

Recall was measured by negative words endorsed and recalled divided by total words endorsed, and positive words endorsed and recalled divided by total words endorsed (i.e., the two recall variables). This approach to calculating recall is ideal because it controls for overall endorsement rates (Goldstein et al., 2015; Prieto et al., 1992; Speed et al., 2016).

Recognition.

Recognition accuracy was measured using d-prime (d’=z(H)-z(F), computed separately for positive and negative words), and memory bias was measured using c (−.5[z(H)+z(F)], separately for positive and negative words) (i.e., the 4 recognition variables) (Macmillan & Creelman, 2004; Stanislaw & Todorov, 1999) H and F refer to hit rate (correctly identifying words that were presented during the SRET task) and false alarm rate (false positives), respectively. To reduce the influence of outliers, we winsorized extreme values using the Winsorize function in the DescTools R package. To control for multiple testing, we used a Bonferroni-adjusted p-value (.05/8=.00625).

MRI Analyses

Individual contrast images were used to create second-level random-effects models using one-sample t-tests for the contrast of viewing negative > positive words. Whole brain analyses were corrected for multiple comparisons using the 3dttest++ function in AFNI with the Clustsim flag, using a voxel-wise correction of p<.001 and a cluster-wise α=.05. This produced 10,000 iterations of noise-only generated t-tests and determine cluster-level threshold (k=49) for regression analyses. Separate regressions tested the association between CRSQ or EMA rumination and whole-brain reactivity to negative words > positive words. Age, sex, and scanner type were included as covariates in all models. Additionally, to test whether associations between rumination and SRET were unique to rumination versus major depression, post hoc analyses included MDD diagnosis as a covariate. See the Supplement for main effects of the task, group differences (typically developing vs. elevated depression) in brain reactivity, and an analysis of an indirect effect of brain reactivity during SRET processing on the links between rumination and SRET recall and recognition.

Results

Behavioral Results

Word Endorsement:

Higher conventional self-reported rumination (b= −0.26, t(130)= −4.64, p<.001) and EMA rumination (b= −3.88, t(130)= −5.02, p<.001) was associated with fewer positive words endorsed. Similarly, higher self-reported rumination (b=0.08, t(130)= 11.06, p<.001) and EMA rumination (b=0.42, t(130)=4.89, p<.001) was associated with more negative words endorsed (see Table 2, including standardized betas and R² values). Endorsement results remained significant in post-hoc sensitivity tests controlling for MDD diagnosis.

Table 2.

Regression Models Comparing Self-Reported (CRSQ) and EMA Rumination as Predictors of SRET Behavioral Variables

Predictor	Model 1: Positive words endorsed	Model 2: Negative words endorsed	Model 3: Recall for positive words	Model 4: Recall for negative words	Model 5: d’ for positive words	Model 6: d’ for negative words	Model 7: c for positive words	Model 8: c for negative words
CRSQ rumination	−.38***	.64***	−.31**	.44***	.20	.30**	.10	−.07
EMA rumination	−.41***	.28***	−.27**	.30***	−.17	.00	−.02	−.05
Age	.15*	−.08	.25***	−.12	.04	.08	−.12	−.01
Sex	.10	.00	.35***	.07	.26**	.19*	.01	−.09
Multiple R2	.52	.76	.38	.50	.10	.16	.02	.02
F(df)	35.82*** (4, 130)	102.42*** (4, 130)	19.80*** (4, 127)	31.90*** (4, 127)	3.83** (4, 134)	6.52*** (4, 134)	0.77 (4, 134)	0.85 (4, 134)

Note. Standardized regression coefficients are presented. CRSQ = Children’s Response Styles Questionnaire. EMA = Ecological Momentary Assessment. The recognition variables, d’ and c, represent recognition accuracy and memory bias, respectively.

^*p < .05.

^**p < .01.

^***p < .001

Free Recall:

Greater self-reported rumination (b=−0.003, t(127)=−3.28, p=.001), and EMA rumination (b=−0.031, t(127)=−2.82, p=.005) was associated with lower recall of positive words. Only CRSQ remained significant in post-hoc sensitivity tests controlling for MDD diagnosis. Greater self-reported rumination (b=0.004, t(127)=5.12, p<.001), and EMA rumination (b=0.03, t(127)=3.55, p<.001) was associated with higher recall of negative words. This result remained significant in post-hoc sensitivity tests controlling for MDD diagnosis.

Recognition:

Greater self-reported rumination (b=0.02, t(134)=2.87, p=.005), but not EMA rumination (b=0.002, t(134)=0.02, p=.984), was associated with higher accuracy in recognizing negative words that were previously presented (d’). This result remained significant in post-hoc sensitivity tests controlling for MDD diagnosis. Neither self-reported nor EMA rumination were associated with accuracy in recognizing positive words (d’; ps>.07). No significant findings emerged for memory bias (c).

Rumination and Brain Reactivity to SRET

Self-reported rumination (CRSQ) was associated with greater reactivity to negative words versus positive words in three regions, the left anterior insula (T=5.25, k=307, x =−40, y=20, z=−4), dACC (T=5.08, k=224, x=−6, y=28, z=40), and left lingual gyrus (T=4.57, k=110, x=−4, y=−86, z=−8). This pattern of results held when controlling for MDD diagnosis, although only the activation in the anterior insula remained statistically significant. Additionally, based on recent work suggesting unique effects of self-reported versus EMA-measured rumination (Webb et al., 2023a), we conducted post-hoc sensitivity analyses controlling for EMA rumination, and only activation in the anterior insula remained significant. EMA rumination was not significantly related to brain reactivity to negative words versus positive words.

Discussion

In a sample of teens with a range of depressive symptoms, the current study characterized the relationships between conventional self-report and EMA measures of rumination, brain reactivity, and endorsement, recall, and recognition of negative versus positive self-referential words during the SRET. We found that teens who ruminate more endorse more negative words and fewer positive words. This effect was consistent across conventional self-report and EMA measures of rumination. Additionally, teens with higher rumination have higher activation in the anterior insula, dACC, and lingual gyrus when making assessments of whether negative words describe them versus positive words. We also found that teens with higher rumination recalled fewer positive and more negative self-referential words and are better able to recognize negative words that they saw previously (d’). These effects were robust to controlling for depression diagnosis, suggesting that rumination uniquely contributes to self-referential processing differences beyond those linked to depression diagnosis.

Consistent with our hypotheses, we found that youth with higher levels of rumination—as measured by the CRSQ, but not EMA—had higher activation in the dACC and anterior insula to negative versus positive words during the SRET. The AI and dACC are core nodes of the salience network, which plays an important role in affective processing, particularly in orienting attention to external, affectively salient stimuli (Corbetta et al., 2008). The salience network is also posited to serve as a dynamic switch between self-oriented processing, mediated by the default mode network, and externally-oriented and directed attention, mediated by the frontoparietal network (Schimmelpfennig et al., 2023). A recent review of adolescent brain functioning during self-referential processing (broadly defined) emphasized the link between depression and increased reactivity in medial cortical regions typically considered to be nodes of the default mode network (Butterfield et al., 2023). Consistent with the notion of the salience network being a mediator of externally-oriented and internally-oriented attention, Kaiser et al found that time spent in a frontoinsular state was related to rumination in teens (Kaiser et al., 2019), that higher depression scores were related to increased salience network activity during an affective Stroop task (Kaiser et al., 2015). Incorporating our findings with previous work, youth with higher levels of trait rumination may be especially attuned to negative, self-relevant cues in the environment and have higher levels of brain activation in the salience network to these cues. This pattern of brain reactivity may enhance the personal salience of negative stimuli and contribute to negative memory biases that ultimately reinforce ruminative cognitive styles.

It was somewhat surprising that although conventional self-reported rumination and EMA rumination were related (r=0.64, see Supplement), only the CRSQ was associated with brain reactivity and negatively biased recognition during the SRET. This may suggest that trait (CRSQ) and recent (EMA) rumination are related but not redundant measures of rumination and contribute unique variance to SRET performance. The CRSQ is considered a global measure of how youth believe that they respond to negative events in general (e.g., “When I am sad, I think about how sad I feel”), while the EMA items assessed how much they ruminated on a specific negative or stressful event since the last survey, several hours ago (“After this stressful thing happened I was dwelling on my mistakes, failures, or losses”). The SRET could also be considered a global measure of self-concept, as youth are asked to endorse whether a word does or does not describe them. It may be that youth who report more general ruminative thinking styles are also more likely to react and effectively encode negative self-referential stimuli, whereas youth who report higher recent rumination (but do not consider themselves to be high on general ruminative tendencies) do not show the same degree of reactivity and encoding of negative stimuli.

In support of the differentiation of trait versus recent rumination, a recent investigation of EMA and CRSQ rumination in the context of clinical interventions found only modest correlations between these assessment methods, yet change in each measure uniquely predicted depressive symptom improvement, suggesting that both measures capture clinically-meaningful, but distinct information (Webb et al., 2023a). If replicated, our findings suggest that researchers should carefully consider whether to incorporate relatively burdensome EMA measures in a study. For example, if a researcher is especially interested in assessing participants’ momentary (or recent) experiences (e.g., to test temporal relationships between ruminative thoughts and negative affective states), EMA may be valuable. However, if they are interested in more global internal representations of the self (e.g., “I’m a worrier”), then briefer, retrospective self-report measures may be sufficient.

Limitations

The study has several limitations. First, most of our subjects identify as non-Hispanic, White females with household incomes above the national average. Future work should replicate our findings in a more diverse sample. Second, to be mindful of limitations of smart phone use during school hours, our EMA protocol was restrained to 2 surveys per day on weekdays. Teens received 3 surveys per day on weekends. While similar sampling windows have been used previously in EMA studies with teens (Forbes et al., 2012; Price et al., 2016; Webb et al., 2022a), denser sampling may have led to more robust results for EMA rumination since our measure asked teens to report rumination from a recent stressful event. Third, due to fMRI scanning constraints, the recall and recognition portions of the SRET were completed outside the scanner approximately 15-minutes after the SRET endorsement task. In previous work, the recall and recognition tasks were completed immediately (Cataldo et al., 2023) or after a brief (e.g., 3-min) delay (Gotlib et al., 2004). Despite the longer delay, our results still support our hypotheses that individuals with higher levels of rumination demonstrate memory biases towards negative stimuli. Fourth, similar to prior work with the SRET in youth with and without depression (Speed et al., 2016), there were substantial differences on the number of positive and negative words endorsed, such that some typically developing teens endorsed few negative words, and some teens with elevated depression symptoms endorsed few positive words. While this discrepancy is consistent with cognitive theories of depression, it limited the kinds of analyses we could do. We were unable to examine brain reactivity during negative word endorsement versus positive word endorsement, as findings based on averages with relatively few trials would lack the power to detect true effects. Instead, we focused on brain reactivity when participants viewed the word, which immediately preceded their endorsement window. The SRET is a simple task, thus it is reasonable to expect that when viewing a word, youth were engaged with the task goals of evaluating whether the word described them. Additionally, during the SRET endorsement task, participants were not instructed to respond as quickly as possible, and we were unable to examine more complex computational modeling of SRET, such as drift-diffusion modeling (Cataldo et al., 2023; Hitchcock et al., 2023).

Conclusion

The current study is the first, to our knowledge, to examine the relationships between conventional self-reported rumination, EMA rumination, and neural and behavioral correlates of negative and positive self-referential processing in youth. In a sample enriched for high levels of depressive symptoms, we found that youth with higher levels of trait rumination, measured via conventional self-report, endorsed, recalled, and recognized more negative words and had higher activation in regions of the salience network to negative words relative to positive words. It is possible that increased brain reactivity to negative stimuli in regions involved in personally relevant external attention may be one potential reason why teens who ruminate more are more likely to remember negative self-relevant stimuli. The current findings are highly relevant for our understanding of the depression etiology, as they link salience network reactivity during self-referential processing to rumination and link rumination to negative information processing biases that increase depression risk. An important future direction of this work is to explore whether interventions that improve awareness and non-judgment of affective experiences (i.e., mindfulness) may be particularly helpful for teens with high levels of rumination (Webb et al., 2022b; Webb et al., 2021). as they may improve salience network reactivity to negative cues and reduce attentional biases towards negative stimuli.

Figure 1. Self-Reported Rumination and Brain Reactivity to SRET.

Self-Reported rumination (CRSQ) was linked to increased activity in the insula, dorsal anterior cingulate, and lingual gyrus. Figure displayed at the peak voxel of the insula cluster (top; t=5.25, k=307, x=−40, y=20, z=−4), peak voxel of the dorsal anterior cingulate cluster (middle; t=5.08, k=224, x=−6, y=28, z=40), and peak voxel of the lingual gyrus cluster (t=4.57, k=110, x=−4, y=−86, z=−8). Scatter plots display extracted mean activation from insula (top), dorsal anterior cingulate (middle), and lingual gyrus (bottom) clusters.

Highlights.

• Rumination may contribute to negative cognitive biases in depression

• Rumination is linked to more negative memory biases on a self-referential task

• Rumination is linked to more insula activity to negative self-referential words

• Brain response to negative self-referential content may facilitate rumination