Multilevel Growth Curve Analyses of Behavioral Activation for Anhedonia (BATA) and Mindfulness-Based Cognitive Therapy Effects on Anhedonia and Resting-State Functional Connectivity: Interim Results of a Randomized Trial

Abstract

Background:

The neural mechanisms associated with anhedonia treatment response are poorly understood. Additionally, no study has investigated changes in resting-state functional connectivity (rsFC) accompanying psychosocial treatment for anhedonia.

Methods:

We evaluated a novel psychotherapy, Behavioral Activation Therapy for Anhedonia (BATA, n=38) relative to Mindfulness-Based Cognitive Therapy (MBCT, n=35) in a medication-free, transdiagnostic, anhedonic sample in a parallel randomized controlled trial. Participants completed up to 15 sessions of therapy and up to four 7T MRI scans before, during, and after treatment (n=185 scans). Growth curve models estimated change over time in anhedonia and in rsFC using average region-of-interest (ROI)-to-ROI connectivity within the default mode network (DMN), frontoparietal network (FPN), salience network, and reward network. Changes in rsFC from pre- to post-treatment were further evaluated using whole-network seed-to-voxel and ROI-to-ROI edgewise analyses.

Results:

Growth curve models showed significant reductions in anhedonia symptoms and in average rsFC within the DMN and FPN over time, across BATA and MBCT. There were no differences in anhedonia reductions between treatments. Within-person, changes in average rsFC were unrelated to changes in anhedonia. Between-person, higher than average FPN rsFC was related to lesser anhedonia across timepoints. Seed-to-voxel and edgewise rsFC analyses corroborated reductions within the DMN and between the DMN and FPN over time, across the sample.

Conclusions:

Reductions in rsFC within the DMN, FPN, and between these networks co-occurred with anhedonia improvement across two psychosocial treatments for anhedonia. Future anhedonia clinical trials with a waitlist control group should disambiguate treatment versus time-related effects on rsFC.

Introduction

Anhedonia (i.e., impairments in pleasure and motivation) is a transdiagnostic symptom that predicts a more severe course of illness in mood, anxiety, and psychotic disorders (1–4). Most medication and cognitive-behavioral interventions do not sufficiently address anhedonia (5) and identifying biomarkers of treatment response remains an unmet clinical need. Examining changes in brain resting-state functional connectivity (rsFC) that accompany anhedonia treatment may improve our understanding of the functional neuroanatomy of anhedonia and the neural mechanisms of therapeutic response. In the present investigation we evaluated four intrinsic neural networks of relevance to anhedonia: the reward network (RN), default mode network (DMN), salience network (SN), and frontoparietal network (FPN). Each network has been shown to be associated with hedonic capacity (6) or aberrant connectivity in disorders such as major depressive disorder (MDD) (7, 8), post-traumatic stress disorder (PTSD) (9, 10), and schizophrenia (11, 12). No study to date has evaluated changes in rsFC among these four canonical networks as a function of anhedonia treatment.

Projections from the ventral tegmental area to the striatum and prefrontal cortex comprise a RN which has largely been the focus of mechanistic studies of anhedonia. Preclinical research demonstrates that motivational aspects of reward are primarily mediated by dopaminergic pathways (13). However, direct evidence of dopamine dysfunction in anhedonia from human neuroimaging is limited and inconsistent (14, 15). A more robust finding with respect to the RN is that nucleus accumbens (NAc) activation during reward anticipation and outcomes is attenuated in MDD relative to controls (16). In terms of rsFC, individual differences in connectivity between the NAc and targets in the RN, SN, and DMN have shown to predict concurrent and future anhedonia severity. Greater anhedonia has been shown to be associated with weaker positive NAc connectivity with the caudate nucleus and anterior cingulate cortex (ACC) in adolescents (17), with the posterior cingulate cortex (PCC) in young adults (18), and with the ventromedial prefrontal cortex in adults with MDD or bipolar disorder (6). Conversely, anhedonia has been shown to be associated with stronger positive NAc connectivity with the medial frontal gyrus in socially-anxious adults (18), and the right dorsomedial prefrontal cortex among trauma-exposed adults (19). A study of adolescents also found stronger connectivity between the left ventral striatum and nodes distributed throughout the RN predicted a depressive episode at 3 year follow up (20).

The relevance of the DMN, SN, and FPN to anhedonia derives from a triple network model of psychopathology (21) which proposes that abnormalities in these networks are observed across psychiatric disorders. The DMN supports self-referential processes (22), the SN coordinates responses to homeostatically relevant stimuli (23), and the FPN (also known as the central executive network) regulates working memory and cognitive control (24). In a study of individuals with remitted psychosis, attenuated rsFC between the left insula and other SN regions was related to residual negative symptoms, including anhedonia (25). Among depressed adolescents, dorsomedial prefrontal cortex hypoconnectivity with the PCC and hyperconnectivity with the ACC predicted lower capacity for subjective anticipatory pleasure (26). Another study of adolescents found higher levels of mind-wandering and stronger DMNSN connectivity among an anhedonic cohort as compared to a healthy control group (27). Lastly, hyperconnectivity within the anterior DMN was shown to be related to reward responsivity in MDD, bipolar disorder, schizophrenia, and relatives without psychosis (28).

These rsFC correlates of hedonic capacity suggest anhedonia is characterized by dysconnectivity across multiple networks. If this is indeed the case, effective treatments may improve hedonic function by impacting disparate network targets. The purpose of our investigation was to evaluate this premise by examining the impact of two psychosocial anhedonia treatments on connectivity within the RN, DMN, SN, and FPN.

Mindfulness-based and behavioral activation interventions are psychosocial treatments that are empirically supported for MDD, a disorder commonly characterized by anhedonia (29, 30). Despite substantial research into the efficacy of these interventions for MDD, few studies have examined changes in rsFC as a function of treatment. Lifshitz and colleagues (31) reported that two weeks of intensive mindfulness therapy resulted in reduced rsFC between a dorsolateral prefrontal cortex seed and the fusiform and angular gyrus. Likewise, Yokoyama and colleagues (32) observed decreases in rsFC within the anterior DMN using an independent component analysis following five weeks of behavioral activation treatment for young adults with subthreshold depression. Neither of these studies examined how changes in rsFC related to anhedonia reductions.

In the current investigation, we examined the extent to which changes in rsFC accompanied treatments for anhedonia in a transdiagnostic sample. These analyses occurred in the context of an ongoing clinical trial (ClinicalTrials.gov Identifier NCT02874534) comparing a novel intervention, Behavioral Activation Therapy for Anhedonia (BATA), to individually administered Mindfulness-Based Cognitive Therapy (MBCT). Three analytic methods were used to characterize rsFC. Our primary outcomes were average ROI-to-ROI connectivity within the RN, DMN, FPN, and SN using a standard atlas (see Methods for details). This metric was selected to probe distributed effects throughout these canonical networks independent of specific seed selections. We estimated the changes in average network connectivity over time across up to four timepoints using multilevel models. Our primary analyses were extended by two additional rsFC metrics that could identify more topographically specific changes within, as well as between, and outside, networks of interest. Seed-to-voxel analyses examined changes between the mean time-series of all ROIs within a network and voxels across the whole brain. An edgewise ROI-to-ROI analysis evaluated changes between ROIs across the whole brain (i.e., agnostic to network membership).

We had two broad hypotheses. First, we predicted BATA would increase rsFC within the RN and SN compared to MBCT. Behavioral activation interventions encourage frequent engagement with positive reinforcers (33) and BATA explicitly teaches participants to savor pleasant moments, be open-minded to novel types of reward, and attend to the instrumental associations between activities and emotional responses. Theoretically, this would enhance the salience of potentially rewarding stimuli over time and sensitize RN function. Second, we predicted MBCT would attenuate rsFC within the DMN and increase rsFC within the FPN. Mindfulness-based treatments teach participants to disengage from habitual modes of self-reference through repeated redirection of attention and greater present-moment awareness (34). Studies in non-clinical samples have shown the DMN and FPN to be impacted by mindfulness training, although both increases (35, 36) and decreases (37) have been demonstrated, and network reconfiguration likely depends on the total amount of practice (38). Given hyperconnectivity within the DMN and hypoconnectivity in the FPN are often characteristic of MDD (39) we expected a remediation of this pattern over the course of MBCT treatment.

Methods

Participants

In this parallel randomized controlled trial, 73 participants were randomized via computerized allocation performed by the study statistician (38 randomized to BATA, 35 randomized to MBCT) and attended at least one therapy session (see Table 1). Assignment to groups was stratified by relatively high (≥30) versus low (20–30) scores on the Snaith Hamilton Pleasure Scale using the ordinal scoring of Franken and colleagues (40). Participants completed ultrahigh field (7T) MRI scans pre- and post-treatment as well as one or two times during treatment (n=185 scans). Inclusion criteria were age 18–50 years old, no psychotropic medication use for >30 days, Snaith Hamilton Pleasure Scale (41) scores > 20 (40) and Clinical Global Impressions-Severity scores > 3 indicative of a clinical impairment. Exclusion criteria were MRI contraindications; concurrent psychotherapy; prior behavioral activation or mindfulness-based psychotherapy; history of moderate or severe substance use disorder, eating disorder, or psychotic / bipolar disorder assessed via the Structured Clinical Interview for DSM-5, Research Version (SCID-5) (42).

Table 1.

Demographics and sample characteristics.

	BATA (n = 38)	MBCT (n = 35)	t (71)	χ2	p
Age (years), mean (SD)	27.9 (8.8)	31.8 (9.2)	1.85		.07
Sex (female), n (%)	27 (71%)	24 (69%)		0	1
Annual Income, mean (SD)	$85,945 (77,446)	$59,280 (51,120)	1.7		.09
Baseline SHAPS, mean (SD)	35.7 (3.8)	36.9 (5.3)	1.12		.27
Attrition, n (%)	8 (21%)	14 (40%)		2.27	.13
Race, n				5.2	.27
Asian	9	5
Black	8	7
Native American	0	1
Other	0	3
White	21	19
Ethnicity, n				5.7	.06
Hispanic or Latino	1	6
Not Hispanic or Latino	37	28
Prefer Not to Answer	0	1
Number of scans per time point, n
Pre-treatment	37	33
Week 8	22	12
Week 12	13	9
Post-treatment	33	26
Primary DSM-5 Diagnosis, n
Major Depressive Disorder	13	11
No Diagnosis	11	6
Generalized Anxiety Disorder	7	4
Post-Traumatic Stress Disorder	1	6
Persistent Depressive Disorder	3	2
Social Anxiety Disorder	1	1
Other Specified Anxiety Disorder	0	2
Other Specified Traumatic Stress Disorder	1	0
Attention Deficit Hyperactivity Disorder	0	2
Adjustment Disorder	1	0

Note. BATA – Behavioral Activation Therapy for Anhedonia, MBCT – Mindfulness Based Cognitive Therapy, SHAPS – Snaith Hamilton Pleasure Scale

^*p<.05.

Procedures

This protocol was approved by the IRBs at the University of North Carolina-Chapel Hill (UNC) and the Duke University Health System. Diagnoses were made using the SCID-5 by research assessors trained to 100% diagnostic reliability with a standard rater over a minimum of three training interviews. Participants were randomized to weekly 45-minute sessions of either BATA or MBCT, with a crossed-therapist design (i.e., all therapists delivered both treatment modalities). Therapy was considered complete after 15 weeks or after a minimum of 8 sessions plus the therapist’s assessment that participants obtained maximum benefit. Anhedonia severity was measured weekly via the Snaith Hamilton Pleasure Scale (SHAPS) at each therapy and neuroimaging session. Scans were acquired at the UNC Biomedical Research Imaging Center within one-week pre-treatment and one-month following the final therapy session. Participants enrolled during the first two years of the trial also completed mid-treatment scans at weeks 8 and 12. The registered primary endpoint of the first phase of this trial was neural responses to reward measured with fMRI and the secondary endpoint was changes in anhedonia. The current analysis of rsFC is an exploratory analysis. The present analysis includes all participants enrolled in the trial from trial initiation in January, 2017 to March, 2020 when recruitment was interrupted by COVID-19 precautions. The CONSORT Flow Diagram is provided in Figure 1.

Figure 1.

The CONSORT Flow Diagram is provided in Figure 1.

Behavioral Activation Therapy for Anhedonia (BATA)

BATA is a modification of Behavioral Activation Treatment for Depression (43) and was developed to treat anhedonia transdiagnostically. BATA entails psychoeducation of reward and learning processes to frame experiences with anhedonia in a neuroscientific perspective. Patients are encouraged to engage in activities that increase contact with personally relevant, values-congruent reinforcers. While achievable goal-setting is often impaired in anhedonia, therapists use motivational techniques to elicit goal-directed behaviors. Differences from traditional behavioral activation treatment include a focus on initiating new behaviors outside the established behavioral set (i.e., “dabbling”) and moment-savoring exercises to enhance consummatory reward experiences. Increased positive affect and decreased negative affect are theorized to result from reduced behavioral avoidance and consequently increased contact with potential reinforcers.

Mindfulness-Based Cognitive Therapy (MBCT)

MBCT was administered in an individual format (modeled on 34, 44). This format retains the primary components of traditional group MBCT including didactic instruction, guided in-session meditations, between-session home practice, and inquiry of subjective experience. Mindfulness is presented as a means to facilitate flexible cognitive-emotional responses to events and reduce habitual reactions. Practices emphasize core meditation skills of attention, decentering, and nonjudgmental acceptance through exercises such as body scans, mindful movement, and focused awareness of breath. Guided inquiry is geared toward awareness of the interrelations among thoughts, emotions, and sensations, as well as the recognition of habitual patterns that can interfere with the experience of pleasure.

Snaith Hamilton Pleasure Scale (SHAPS)

Anhedonia was measured by the SHAPS, a 14-item self-report questionnaire that assesses the degree to which a person has the capacity to experience pleasure, or the anticipation of a pleasurable experience “in the last few days” (41). Each item has four possible responses – strongly disagree, disagree, agree, or strongly agree – for a total score ranging between 14 – 56 using the ordinal scoring of Franken, Rassin (40), with higher values reflecting greater hedonic impairment.

MRI Acquisition Parameters

Scans were acquired using a 7T Magnetom MR system with a 32-channel head coil. T1-weighted images used these parameters: echo time (TE) = 2.78 ms, repetition time (TR) = 2200 ms, inversion time (TI) = 1050 ms, flip angle = 7°, and voxel size of 1 × 1 × 1 mm³. Resting-state data were acquired during an 8-minute (480 volume) gradient-echo echoplanar imaging sequence with the following parameters: TE = 22.2 ms, TR = 1000 ms, flip angle = 45°, and voxel size of 1.6 × 1.6 × 1.6 mm³. Participants were instructed to keep their gaze fixed on a cross hair and not to fall asleep.

MRI Preprocessing

T1 images were bias-corrected with FSL v5.0.9 to enhance FreeSurfer v6.0 removal of non-brain tissue. Remaining steps proceeded in CONN toolbox v19c (45) and SPM12 (46) including 1) realignment and motion estimation, 2) slice-timing correction, 3) volume outlier detection (framewise displacement > .5 mm or global signal intensity change > 3 z-scores) 4) direct normalization to MNI space, and 5) smoothing 4 mm FWHM (only applicable for seed-to-voxel analyses). Brain masks were segmented into white matter (WM) and cerebrospinal fluid (CSF), normalized, and eroded to minimize partial volume effects. Linear regression accounted for variance due to motion estimates (6 parameters and their 1^st order derivatives), WM and CSF (mean time-series plus 4 principal components), and outlier volumes. Residual time-series were band-pass filtered (.008–.09 Hz). Runs with >20% total outlier volumes were discarded.

Analytic Plan

Growth Curve Model Building Procedures

Treatment effects on SHAPS and on average ROI-to-ROI network connectivity over time were evaluated using longitudinal multilevel models (i.e., growth curves) in R’s nlme package. Histograms were plotted to confirm normal distribution of variables. Univariate outliers were identified by values more than 1.5 times the interquartile range.

First, the intraclass correlation coefficients (ICCs) were calculated to characterize the degree of variance in outcomes attributable to between- vs within-participant factors. Second, linear and quadratic time effects were evaluated, allowing random variation in intercepts between participants. Time was coded as a continuous variable in weeks, centered by the date of the first therapy session. Third, random slopes were examined, allowing the rectilinear and quadratic effects of time on outcomes to vary between participants. Fourth, serially correlated residuals were examined using an autoregressive-1 error structure (47).

Once optimal unconditional growth curves were selected, participant level (i.e., level 2) factors – age, sex and treatment condition – were included in the model to explain variation in intercepts. A treatment-by-time cross-level interaction also examined differences in time-outcome slopes between BATA and MBCT. Lastly, in models predicting SHAPS outcomes, average ROI-to-ROI network connectivity was decomposed into between- and within-person predictor variables to evaluate within- and between-participants effects, as follows. The between-person (trait-like, static across all time points) connectivity variable was calculated as the person’s mean connectivity across all time points (i.e., the “person-mean”), and was included in the model after grand-mean centering. The within-person (state-like, fluctuating across time points) network connectivity variable was calculated by subtracting the person-mean connectivity from the raw value at each time point; after this “person-centering”, the within-person variable was included in the model as a time-varying predictor (48).

Models were compared using likelihood ratio tests and changes in Akaike Information Criterion (AIC) (49). Parameters were estimated using full maximum likelihood to compare models differing in fixed effects. Otherwise, restricted maximum likelihood was used to compare differing random effects and to explore final model parameters.

Average ROI-to-ROI Network Connectivity

Average ROI-to-ROI network connectivity was calculated with CONN’s ROI-to-ROI analyses. A 300 ROI functional atlas developed by Seitzman, Gratton (50) was used to probe the DMN (65 ROIs), FPN (36 ROIs), SN (13 ROIs), and RN (8 ROIs). This atlas largely maintains the network structure of the 264 ROI set by Power, Cohen (51) while expanding subcortical and cerebellar coverage. Spheres with 4 mm radii were centered at specified MNI coordinates (https://greenelab.wustl.edu/data_software) to avoid overlapping ROIs. Correlations between the mean time-series of each ROI pair within a network were Fisher transformed. This metric represents a mean value among 2080 correlations for the DMN, 360 for the FPN, 78 for the SN, and 28 for the RN.

Seed-to-Voxel Analyses

Networks that showed change in average ROI-to-ROI network connectivity were probed using CONN’s seed-to-voxel analyses. The mean time-series across all ROIs within a network was used as a seed with voxels across the whole-brain as targets. Differences in connectivity pre- to post-treatment were investigated, controlling for age and sex. Unlike multilevel models, these analyses included only participants with pre- and post-treatment scans, including those who withdrew from treatment but agreed to complete follow up MRI (n=56). In the absence of a treatment-by-time interaction effect, connectivity changes were examined across all participants agnostic to treatment condition. Cluster-level inferences were based on Threshold Free Cluster Enhancement (TFCE) statistics using 5,000 permutations of the original data (52).

Edgewise ROI-to-ROI Analyses

A whole-brain ROI-to-ROI analysis was conducted within CONN to examine edgewise changes in connectivity pre- to post-treatment, controlling for age and sex. As in seed-to-voxel analyses this model included only two time points and participants without missing data (n=56). In the absence of a treatment-by-time interaction effect, changes were examined across all participants. The ROI-to-ROI connectivity matrix was sorted using optimal leaf ordering for hierarchical clustering based on spatial proximity and functional similarity of the ROIs (53). Cluster-level TFCE statistics and peak-level family wise error corrected p-values were computed to identify edges with reliable change over time.

Results

The treatments did not cause any adverse events.

Treatment Effects on Anhedonia

Fifty-one participants completed treatment and 22 withdrew or were lost-to-follow-up (see Procedures for a definition treatment completers). The rates of attrition in MBCT (40%) and BATA (28.6%) did not differ, χ² (1) = 2.27, p = .13.

ICCs revealed that approximately 56% of the variance in SHAPS scores was due to between-participant differences. The unconditional growth curve model intercept (implied mean value) of SHAPS at therapy onset was 35.7. A linear effect of time indicated SHAPS decreased about .57 points every week of therapy across the sample (p<.001). Model comparison favored incorporating random slopes (χ² (2) = 245.5, p<.001) indicating significant variability in SHAPS rate of change between participants. An autoregressive error structure also improved model fit (χ² (1) = 54, p<.001) indicating that observations closer in time were more similar (ρ =.32). There was a weak negative correlation between the random effects of intercepts and slopes (r=−.07) such that participants with higher baseline SHAPS improved slightly faster over time (i.e., regression to the mean). Incorporating between-person (level-2) predictors in the model revealed no significant effects of age, sex, or treatment condition, and no significant treatment-by-time interaction. In other words, there were no differences between MBCT and BATA in baseline SHAPS or change over time, with both groups improving. Using last observations carried forward, the overall mean difference in SHAPS scores from pre- to post-treatment across the sample was 7.25 points.

MRI Motion and Artifacts

Eight scans were excluded from analyses due to excessive motion (5 MBCT, 2 BATA) and artifact (1 MBCT). Among the remaining 185 scans, there were no group differences in the mean number of volumes scrubbed (BATA=21.2 (19.5), MBCT=19 (15.5), t(182.69)=.85, p=.39), amounting to ~4% of the data. However, the BATA group had greater mean framewise displacement (BATA=.19 (.07) mm, MBCT=.16 (.05) mm, t(182.69)=2.93, p=.003). Including mean framewise displacement as a time-varying covariate of connectivity did not affect any results.

Treatment Effects on Average ROI-to-ROI Network Connectivity

The model parameters of the optimal unconditional and conditional growth curves for each of the four networks are reported in Tables 2 & 3 respectively. Only the DMN and FPN showed change over time and are subsequently described.

Table 2.

Final unconditional growth curve models of average region-of-interest (ROI)-to-ROI connectivity within four resting-state networks as defined by the Seitzman-300 ROI functional atlas. Parameters were estimated using restricted maximum likelihood.

Predictors	Default Mode Network		Frontoparietal Network		Salience Network		Reward Network
	Estimates	95% Confidence	Estimates	95% Confidence	Estimates	95% Confidence	Estimates	95% Confidence
Intercept	.1717 ***	.1615 – .1819	.1651 ***	.1530 – .1773	.2349 ***	.2169 – .2530	.0597 ***	.0507 – .0688
Linear Effect of Time	−0.0045 ***	−0.0067 – −0.0023	−0.0027 *	−0.0052 – −0.0001	.0002	−0.0012 – .0015	.0007	−0.0002 – .0016
Quadratic Effect of Time	.0002 **	.0001 – .0004	.0002 *	.0000 – .0004
Random Effects
σ2	.0011	.0014	.0035	.0017
τ00	.0011	.0016	.0034	.0003
Intraclass Correlation	.5083	.5366	.4915	.1305
Marginal R2 / Conditional R2	.067 / 0.541	.013 / 0.543	.000 / 0.492	.012 / 0.141

Note.

^*p<.05

^**p<.01

^***p<.001.

Table 3.

Conditional growth curve models of average region-of-interest (ROI)-to-ROI connectivity within four resting-state networks as defined by the Seitzman-300 ROI functional atlas. Parameters were estimated using restricted maximum likelihood. In all cases likelihood ratio tests indicated superior fit for the simpler unconditional growth curves.

	Default Mode Network		Frontoparietal Network		Salience Network		Reward Network
Predictors	Estimates	95% Confidence	Estimates	95% Confidence	Estimates	95% Confidence	Estimates	95% Confidence
Intercept	.1762 ***	.1614 – .1911	.1732 ***	.1550 – .1913	.2292 ***	.2023 – .2562	.0641 ***	.0506 – .0776
Linear Effect of Time	−.0056 ***	−.0084 – −.0028	−.0019	−.0053 – .0014	.0006	−.0047 – .0058	−.0015	−.0050 – .0019
Quadratic Effect of Time	.0003 ***	.0001 – .0006	.0002	−.0001 – .0004	−.0001	−.0005 – .0003	.0001	−.0001 – .0004
Treatment (MBCT)	.0051	−.0150 – .0253	−.0062	−.0308 – .0183	.0189	−.0177 – .0554	−.0091	−.0277 – .0096
Age	.0006	−.0004 – .0016	.0016 *	.0004 – .0029	−.0012	−.0031 – .0006	.0002	−.0006 – .0010
Sex (Male)	−.0209 *	−.0406 – −.0012	−.0184	−.0427 – .0058	−.0111	−.0467 – .0245	−.0006	−.0167 – .0156
Treatment × Linear Time	.0029	−.0014 – .0074	−.0020	−.0072 – .0032	−.0002	−.0083 – .0079	.0027	−.0025 – .0080
Treatment × Quadratic Time	−.0003	−.0006 – .0000	.0001	−.0003 – .0005	.0002	−.0004 – .0008	−.0001	−.0005 – .0003
Random Effects
σ2	.0011		.0014		.0035		.0017
τ00	.0011		.0015		.0031		.0003
Intraclass Correlation	.5002		.5210		.4715		.1474
Marginal R2 / Conditional R2	.112 / .556		.084 / .561		.064 / .506		.037 / .179

Note.

^*p<.05

^**p<.01

^***p<.001

There were two outliers each for DMN and FPN connectivity. Removal of these did not affect any results; thus, all values were retained. ICCs indicated that 42% of the variance in connectivity within the DMN, and 53% of the variance within the FPN was attributable to stable between-participant differences. The mean Z-score of connectivity at baseline was .17 for both networks. Significant linear effects of time indicated connectivity decreased .005 standard deviations every week within the DMN, and decreased .003 standard deviations every week within the FPN. A significant quadratic effect of time indicated more rapid decreases in connectivity earlier in treatment (with subsequent slowing) within both networks. Although AIC values were lower incorporating random slopes to model DMN connectivity, likelihood ratio tests were not significant, χ² (2) = 4.7, p=.1, and this term was dropped. Estimation difficulties precluded incorporation of random slopes for FPN connectivity, likely due to low power (see limitation section). Initially the model did not converge unto a solution using the nlme package. Estimating the same model with the lme4 package resulted in singular fit (i.e., an element of the variance-covariance matrix was estimated at exactly zero) which suggested inferential test on these model parameters may be inaccurate. Thus, random slopes were dropped from the FPN unconditional model.

Although some between-participant predictors of connectivity emerged, incorporating these fixed effects did not improve model fit. With respect to the DMN, men showed approximately 11% lower connectivity than women. With respect to the FPN, each one-year increase in age was associated with about .002 standard deviations lower connectivity. There were no baseline differences in connectivity between BATA and MBCT, and no interactions between treatment and time in either network. The optimal unconditional growth curves are plotted alongside the raw data in Figure 2.

Figure 2.

Spaghetti plots depicting raw changes in mean ROI-to-ROI connectivity (i.e., Z-scores) within (A) the DMN and (C) the FPN. Unconditional growth curve model predictions incorporating random-intercepts plus linear and quadratic fixed effects of time for (B) average DMN and (D) FPN connectivity. Time was coded in weeks, centered at the date of the first therapy session. Within both (B) the DMN and (D) FPN there was a small negative linear effect of time and a smaller positive quadratic effect of time. This pattern indicates connectivity decreased over time and that changes occurred more rapidly earlier in therapy. Notably, conditional growth curve models incorporating a cross-level interaction between treatment and time were not significant indicating no difference in connectivity change between groups. Note. BATA – Behavioral Activation Therapy for Anhedonia, DMN – Default Mode Network, FPN – Frontoparietal Network, MBCT – Mindfulness-Based Cognitive Therapy, ROI – Region of Interest.

Average ROI-to-ROI Network Connectivity as a Time-Varying Covariate of Anhedonia

Although SN and RN connectivity did not change over time, all four networks were examined as time-varying covariates of SHAPS. The only predictor of anhedonia was the between-participant effect of FPN connectivity. For every 0.1 standard deviation above the sample mean of FPN connectivity between-participants, SHAPS was about 2.2 points lower (β=−2.23, SE=1.06, p=.04). There were no significant within-participant effects of connectivity changes on SHAPS.

Seed-to-Voxel Analyses

Mean time-series across DMN and FPN ROIs respectively were examined in seed-based voxelwise models. There were no treatment-by-time interactions for either network seed. Paired-sample t-tests across all participants, controlling for age and sex, showed widespread decreases in connectivity between mean DMN signal and targets in the superior and middle frontal gyrus, paracingulate gyrus, precuneus, and angular gyrus – regions associated with the FPN and DMN (see Table 4 and Figure 3). There were no changes observed with the mean FPN signal over time.

Table 4.

Seed-to-voxel analyses (including only participants pre- and post-treatment scans; n=56) showing clusters with significant decreases in connectivity pre- to post-treatment across all participants controlling for age and sex. Seed time-series was derived from the mean of 65 ROIs labeled as part of the Default Mode Network in the Seitzman-300 ROI atlas.

MNI (x,y,z)	Brain Region(s)	Cluster Size	Num. Peaks	TFCE	p-FWE	peak-FDR	peak p-uncorrected
−18, 30, 40	Left Superior Frontal Gyrus, Left Middle Frontal Gyrus, Left Frontal Pole	1699	69	1579.8	0	0.000018	0
−40, −70, 34	Left Lateral Occipital Cortex, Left Angular Gyrus	1062	33	1246.95	0.0018	0.005154	0.000008
6, 38, −4	Left Frontal Pole, Frontal Medial Cortex, Bilateral Paracingulate Gyrus, Right Frontal Pole, Anterior Cingulate Gyrus	2068	119	1208.42	0.002	0.005154	0.000012
−8, −54, 20	Precuneus Cortex, Posterior Cingulate Gyrus	1171	53	991.99	0.0094	0.005154	0.000078
12, 34, 20	Right Paracingulate Gyrus	94	3	762.08	0.0348	0.009896	0.00039
20, 64, 20	Right Frontal Pole	101	3	713.69	0.0468	0.012461	0.000544
−2, 40, 4	Anterior Cingulate Gyrus	3	1	710.22	0.048	0.012486	0.000558

Note. FDR – False Discovery Rate, FWE – Family Wise Error, MNI – Montreal Neurological Institute, ROI – Region of Interest, TFCE – Threshold Free Cluster Enhancements

Figure 3.

Seed-to-voxel analyses (including only participants pre- and post-treatment scans; n=56) of connectivity change over time across all participants, controlling for age and sex, and corrected for multiple comparison using TFCE. (Top) Graphical illustration of the seed-to-voxel approach. The seed ROI time-series was calculated as the mean signal across 65 ROIs labeled as DMN within the Seitzman-300 ROI atlas. (Bottom) Results show significant decreases in connectivity pre- to post-treatment mostly within areas typically considered as part of the DMN.

Edgewise ROI-to-ROI Analyses

There were no treatment-by-time interactions observed across the whole-brain ROI-to-ROI connectivity matrix. A paired-sample t-test across all participants, controlling for age and sex, showed four significant clusters, comprised of 31 edges total, with reliable change over time. Within each cluster there was attenuated connectivity. Notably, all 31 edges involved ROIs in the DMN and only 5 of these involved ROIs outside the DMN (see Table 5 and Figure 4).

Table 5.

Edgewise ROI-to-ROI connectivity analyses (including only participants pre- and post-treatment scans; n=56) showing clusters of edges with decreases in connectivity pre- to post-treatment across all participants controlling for age and sex. Results represent significant clusters among 44,850 possible connections in the Seitzman-300 ROI atlas. Numeric labels reflect the order in which ROIs appear listed in the atlas information text file (https://greenelab.wustl.edu/data_software).

	MNI Coordinates (x,y,z)	TFCE	T(54)	p-uncorrected	p-FDR	p-FWE
Cluster 1/4		102.43		.000001	.003523	.004
UN.9 — DM.108	(65, −24, −19) – (−3, 44, −9)		4.48	.00004	.268293
DM.89 — PM.138	(−44, −65, 35) – (−2, −35, 31)		4.17	.000112	.268293
UN.9 — DM.119	(65, −24, −19) – (11, −54, 17)		4.16	.000114	.268293
DM.104 — DM.99	(−7, −55, 27) – (−11, −56, 16)		4.16	.000114	.268293
DM.97 — DM.119	(−17, 29, 53) – (11, −54, 17)		4.08	.00015	.292326
DM.89 — DM.99	(−44, −65, 35) – (−11, −56, 16)		3.98	.000209	.315444
DM.97 — DM.99	(−17, 29, 53) – (−11, −56, 16)		3.96	.000218	.315444
DM.89 — DM.108	(−44, −65, 35) – (−3, 44, −9)		3.64	.000602	.404922
DM.89 — DM.106	(−44, −65, 35) – (−3, −49, 13)		3.59	.000706	.404922
DM.89 — DM.92	(−44, −65, 35) – (−39, −75, 44)		3.52	.000892	.44055
DM.89 — DM.119	(−44, −65, 35) – (11, −54, 17)		3.4	.00129	.452504
UN.9 — DM.105	(65, −24, −19) – (−7, 51, −1)		3.39	.001307	.452504
DM.125 — DM.92	(23, 33, 48) – (−39, −75, 44)		3.38	.00134	.452504
DM.112 — DM.106	(6, −59, 35) – (−3, −49, 13)		3.29	.001759	.476628
UN.9 — DM.99	(65, −24, −19) – (−11, −56, 16)		3.24	.002035	.495236
DM.97 — DM.108	(−17, 29, 53) – (−3, 44, −9)		3.22	.002201	.501351
DM.100 — DM.92	(−10, 39, 52) – (−39, −75, 44)		3.21	.002267	.509138
DM.93 — DM.106	(−35, 20, 51) – (−3, −49, 13)		3.16	.002581	.518822
DM.93 — DM.108	(−35, 20, 51) – (−3, 44, −9)		3.11	.003025	.524086
UN.9 — DM.116	(65, −24, −19) – (8, 42, −5)		3.09	.003117	.524086
DM.115 — DM.92	(8, −48, 31) – (−39, −75, 44)		3.08	.003209	.524086
DM.100 — DM.119	(−10, 39, 52) – (11, −54, 17)		3.07	.003311	.524086
DM.97 — DM.92	(−17, 29, 53) – (−39, −75, 44)		3.07	.003366	.524086
DM.104 — DM.119	(−7, −55, 27) – (11, −54, 17)		3.06	.003488	.525095
DM.104 — DM.92	(−7, −55, 27) – (−39, −75, 44)		3.03	.003796	.529749
DM.130 — DM.106	(52, −59, 36) – (−3, −49, 13)		3.02	.003851	.529749
DM.104 — DM.107	(−7, −55, 27) – (−2, −37, 44)		3.01	.003943	.529749
Cluster 2/4		83		.000005	.005213	.026
DM.97 — DM.116	(−17, 29, 53) – (8, 42, −5)		4.05	.000167	.311434
Cluster 3/4		79.59		.000006	.005213	.032
DM.101 — DM.116	(−10, 55, 39) – (8, 42, −5)		3.73	.000458	.375425
DM.101 — DM.122	(−10, 55, 39) – (12, 36, 20)		3.48	.001009	.452504
Cluster 4/4		74.9		.000009	.005389	.045
DM.97 — DM.107	(−17, 29, 53) – (−2, −37, 44)		3.3	.001704	.476628

Note. DM – Default Mode, FDR – False Discovery Rate, FWE – Family Wise Error, PM – ParietoMedial, ROI – Region of Interest, TFCE – Threshold Free Cluster Enhancements, UN – Unassigned

Figure 4.

Edgewise ROI-to-ROI connectivity results ((including only participants pre- and post-treatment scans; n=56) depicting clusters of edges with significant decreases pre- to post-treatment across all participants, controlling for age and sex, and corrected for multiple comparison using TFCE. Results corroborate findings from multilevel models by highlighting reliable changes within the DMN over time. All 31 edges depicted involve at least one ROI labeled DMN and only 5 of these involve an ROI not labeled as DMN (see Table 5 for details).

Discussion

This was the first study to evaluate changes in anhedonia and rsFC as a function of psychosocial treatments in a transdiagnostic anhedonic sample. We found SHAPS scores improved over time in both the BATA and MBCT groups to a comparable extent, with a mean reduction of 7.25 points across the sample. Participants in both BATA and MBCT also showed indistinguishable changes in rsFC. Our most consistent finding was that connectivity within the DMN was attenuated over time across both treatment groups, an effect observed using three disparate analytic approaches. Decreases (i.e., lesser positive correlations in rsFC) over time were also observed between DMN and FPN regions using seed-to-voxel analyses, and within the FPN using average ROI-to-ROI network connectivity analyses. Interestingly, weaker FPN connectivity between-participants was related to greater anhedonia across timepoints, and we observed no change over time or relations with anhedonia with respect to the RN and SN. While this pattern of results did not support our broad hypotheses of rsFC changes specific to BATA and MBCT, null findings with respect to treatment differences may also be related to low power to detect effects (see limitations section for detail).

Within Default Mode Network Effects

Contrary to hypotheses, we observed similar attenuation within DMN in both treatment groups. Our original hypothesis was rooted in theory that mindfulness practice teaches participants to disengage from unhelpful self-reflective or ruminative processes. Indeed, studies in non-clinical samples show mindfulness training is associated with reconfiguration of the DMN, although both increases (35, 54) and decreases (38, 55) have been reported, even within the same study (37). It is therefore noteworthy that reductions across anterior and posterior nodes of the network were revealed by different analytic approaches.

Reductions in DMN rsFC with BATA are consistent with the findings of Yokoyama and colleagues (32) who reported decreases in anterior regions of the DMN following five behavioral activation sessions. Moreover, changes in DMN rsFC have been reported across multiple treatment modalities for disorders characterized by anhedonia. Serotonin norepinephrine reuptake inhibitors (56), selective serotonin reuptake inhibitors (57), and transcranial magnetic stimulation (58) diminish depression-related hyperconnectivity within the DMN. Transcranial magnetic stimulation has also been reported to decrease connectivity between the DMN and subgenual ACC in comorbid MDD and PTSD (59). Studies in schizophrenia, however, have shown both increases (60) and decreases in homogeneity within the precuneus following olanzapine treatment (61), as well as increased connectivity between the PCC and medial prefrontal cortex (62), or the PCC and left lingual gyrus (63) following risperidone treatment.

We did not observe relations between reductions in DMN rsFC and improvements in self-reported anhedonia. This dissociation between treatment-related changes in rsFC and symptom reduction is consistent with previous reports (56–59). One possibility is that the changes in rsFC and self-report operate on different timescales. While a time-lagged analysis may be better suited to capture DMN effects on mood, the non-uniformity between timepoints in the present study is not well-suited to such an analysis. Another possibility is that DMN modulation more directly accompanies changes in related constructs such as mind-wandering propensity (27) or rumination (64), thereby exerting benefits on anhedonia indirectly. While the SHAPS primarily assesses consummatory aspects of anhedonia (65) it is possible changes in other measures that assess anticipatory aspects of anhedonia may have revealed associations with treatment-related changes in rsFC. Additionally, it is possible that alterations in DMN rsFC occur in parallel but are wholly unrelated to changes in self-reported anhedonia.

Between Default Mode Network and Frontoparietal Network Effects

Seed-to-voxel analyses using the mean DMN time-series showed attenuation of rsFC with FPN clusters in the left superior and middle frontal gyrus (among other DMN clusters). These regions support working memory and high-level spatial processing (66). Post-hoc investigation found these changes resulted from decreases in positive connectivity between networks. A study of rumination-focused cognitive behavioral therapy (a treatment utilizing mindfulness-based strategies) in adolescents with MDD found that decreases in rsFC between the PCC and right inferior frontal gyrus were correlated with improvement in depression and rumination (67). It may be that changes in rsFC between networks in our sample reflected less frequent FPN suppression of the DMN that accompanies decreased engagement of unhelpful thought patterns over time – a pattern Bauer & colleagues have suggested develops with substantial mindfulness practice (38). However, this interpretation must be considered cautiously absent measures of cognitive engagement.

While our whole-brain edgewise analyses did not reveal changes between the DMN and FPN, this method examines a fundamentally separate question than seed-to-voxel analyses. The latter probes for subtle, widespread effects between aggregate DMN function and other brain regions, whereas edgewise analyses instead characterize strong localized effects between nodes. Thus, null-findings with respect to edgewise analysis do not conflict with seed-to-voxel results.

Within Frontoparietal Network Effects

Contrary to hypotheses, we observed reductions in average FPN connectivity over time across treatments. Attenuation within the FPN is unexpected in the context of a meta-analysis that found MDD is often associated with FPN hypoconnectivity (but see (7)). At least two studies have also observed decreases in FPN rsFC following MDD treatment, including a mindfulness-based intervention (31) and electroconvulsive therapy (68).

The finding that higher levels of average FPN connectivity between-participants were associated with lower levels of anhedonia severity was surprising given the pattern of FPN attenuation over time within-participants. A 59% greater magnitude of average FPN connectivity at baseline (i.e., 1.67 standard deviations above the mean) was associated with a SHAPS score 2.2 points lower. This finding may indicate that the changes observed within FPN connectivity over time reflect a pernicious effect (e.g., an increased vulnerability to future mood episodes) which neither treatment protected against. However, we find this interpretation unlikely given the symptomatic improvement across the sample.

Limitations

Findings must be considered in light of several limitations. First, the lack of a non-anhedonic comparison group precludes confirmation that patterns of rsFC change reflect a remediation of deficits associated with anhedonia specifically, especially in the absence of relations between rsFC and symptom severity. Second, either a waitlist condition or a pre-treatment observation period would be necessary to disambiguate treatment effects from time effects on rsFC. As it stands, we cannot rule out the possibility that rsFC changes observed reflect spontaneous remission or the passage of time alone. Third, whereas MBCT and BATA are substantively different treatments, BATA nevertheless includes instruction on present moment savoring during pleasant activities which may account, in part, for similar effects on rsFC (though MBCT encourages momentary awareness decontextualized from positive affective experiences). Also, the duration of resting-state scans (i.e., 8-minute sessions) may be sub-optimal to explore treatment effects (69), and future research should replicate and extend our findings using longer duration scans and/or a general functional connectivity approach with superior psychometric properties (70).

Finally, the sample size for this conditional growth curve analysis was modest. We conducted a power analysis to aid interpretation of null findings between treatments by running 500 simulations of our dataset using the parameter estimates from the DMN and FPN conditional growth curve models. The model was estimated using the R nlme package and the paramtest package was used for power estimation. The estimated power to detect treatment-by-time interaction effects (i.e., the proportion of significant results) was 5–6%. Thus, null findings between groups in average network connectivity may be due to low power and small effect sizes requiring larger samples to detect.

Conclusions

This study compared a novel adaptation of behavioral therapy (BATA) and MBCT in a transdiagnostic anhedonic sample. Given the nature of the sample, results are likely generalizable to patients with a range of psychiatric diagnoses. We found both treatments were associated with improvement in self-reported anhedonia and changes in brain rsFC over time. Widespread decreases were observed within and between networks that broadly support internal and external attention. DMN and FPN involvement was observed across multiple analytic approaches to defining resting-state networks. Surprisingly, these changes in rsFC were unrelated to improvements in anhedonia symptoms when modeled as time-varying covariates. While these findings suggest disparate psychosocial interventions may exert beneficial effects by impacting common resting-state networks, conclusions are limited by the lack of a waitlist control condition. Future research should examine whether other anhedonia interventions impact DMN and FPN rsFC to determine the extent to which these networks truly represent a common target of different anhedonia treatment modalities.