1. Neuroscience

Dissociable roles of reward prediction error in the contrasting mood dynamics of depression and anxiety

  1. Zhihao Wang
  2. Ting Wang
  3. Tian Nan
  4. Jiahua Xu
  5. André Aleman
  6. Yuejia Luo
  7. Bastien Blain
  8. Yunzhe Liu  Is a corresponding author
  9. Pengfei Xu  Is a corresponding author
  1. Center for Neurocognition and Social Behavior, Institute of Artificial Intelligence, Shenzhen University of Advanced Technology, China
  2. CNRS - Centre d'Economie de la Sorbonne, Panthéon-Sorbonne University, France
  3. Institute for brain research and rehabilitation, South China Normal University, China
  4. The State Key Lab of Cognitive and Learning, Faculty of Psychology, Beijing Normal University, China
  5. State Key Laboratory of Cognitive Neuroscience and Learning, IDG/McGovern Institute for Brain Research, Beijing Normal University, China
  6. Chinese Institute for Brain Research, China
  7. Faculty of Psychology and Neuroscience, Maastricht University, Netherlands
  8. Institute for Neuropsychological Rehabilitation, University of Health and Rehabilitation Sciences, China
  9. School of Psychology, South China Normal University, China
  10. Faculty of Health and Wellness, City University of Macau, China
  11. Beijing Key Laboratory of Applied Experimental Psychology, National Demonstration Center for Experimental Psychology Education (BNU), Faculty of Psychology, Beijing Normal University, China
https://doi.org/10.7554/eLife.110631.3
Share this article
Cite this article
  1. Zhihao Wang
  2. Ting Wang
  3. Tian Nan
  4. Jiahua Xu
  5. André Aleman
  6. Yuejia Luo
  7. Bastien Blain
  8. Yunzhe Liu
  9. Pengfei Xu
(2026)
Dissociable roles of reward prediction error in the contrasting mood dynamics of depression and anxiety
eLife 15:RP110631.
https://doi.org/10.7554/eLife.110631.3
  2. Download BibTeX
  3. Download .RIS
4 figures, 11 tables and 1 additional file

Figures

Figure 1 with 6 supplements
Download asset Open asset
Experimental protocol.

(A) Study outline. This study examines how depression and anxiety influence mood fluctuations. The first experiment assesses the bifactor structure that disentangles shared and specific components of depression and anxiety in the psychometric dataset (N=901). The second experiment tests associations between depression- and anxiety-specific traits and RPE-induced mood fluctuations using the questionnaire battery and a gambling task with momentary mood ratings in the laboratory dataset (N=44). The third and fourth experiments replicate the second experiment in online samples (online dataset 1, N=747; online dataset 2, N=235). The fifth experiment tests whether these findings generalize to a clinical dataset of patients with affective disorders (N=61). (B) Mapping between the three factors and 128 items in the bifactor model. (C) Factor loadings of items on the general factor, anxiety-specific factor, and depression-specific factor. (D) Orthogonality among the general, anxiety-specific, and depression-specific factor scores across datasets with complete questionnaire data. (E) Mean correlations between factor scores and questionnaire scores. Overall, the general factor showed high correlations with all questionnaires. The depression-specific factor correlated most strongly with TAIdep and MASQad, whereas the anxiety-specific factor correlated most strongly with the remaining questionnaires. These correlational results supported the bifactor structure of anxiety and depression. (F) Gambling task design. On each trial, participants were asked to choose between a certain option and a gamble option. Once an option was selected, the corresponding outcome was displayed in the center of the screen. The cumulative score was always shown in the upper-right corner. Every two or three trials, participants were asked to complete a self-paced rating of “How happy are you at this moment?” on a slider scale ranging from 0 (very unhappy) to 100 (very happy). (G) Temporal dynamics of happiness ratings for representative individuals with high (top 25%) and low (top 75%) mood variation, healthy datasets, and the clinical dataset. (H) Momentary mood model. (I) Results of momentary mood model for each dataset. MASQaa, the subscale of anxious arousal in the Mood and Anxiety Symptoms Questionnaire; TAIanx, the subscale of anxiety in the Trait Anxiety Inventory; CESD, Center for Epidemiologic Studies Depression Scale; BDI, Beck Depression Inventory; BFIn, the subscale of neuroticism in the Big Five Inventory; PSWQ, Penn State Worry Questionnaire; MASQad, the subscale of anhedonic depression in the Mood and Anxiety Symptoms Questionnaire; TAIdep, the subscale of depression in the Trait Anxiety Inventory; CR, certain reward; EV, expected value; RPE, reward prediction error.

Figure 1—figure supplement 1
Download asset Open asset
Gambling rate in each dataset.

All datasets showed the same gambling rate pattern, as well as in line with literature.

Figure 1—figure supplement 2
Download asset Open asset
Model-agnostic mood results for all datasets.

All datasets showed the same initial mood overall before the task, mean mood, and mood variation across the task.

Figure 1—figure supplement 3
Download asset Open asset
Replication of the classic mood effects.

(A) Happier after winning than losing. (B) Mood drift across the task.

Figure 1—figure supplement 4
Download asset Open asset
Parameter recovery for the winning model (M1) for the healthy and clinical datasets.

Parameter recovery analysis showed high correlations between the matched simulated and real parameters and low correlations between the mismatched simulated and real parameters.

Figure 1—figure supplement 5
Download asset Open asset
Replication of Rutledge et al., 2017’s findings using BDI (N=1026).

Depression symptom measured by BDI was negatively correlated with the baseline mood parameter. Data was combined from laboratory dataset, online dataset 1, and online dataset 2.

Figure 1—figure supplement 6
Download asset Open asset
Correlations of factor scores from the winning model (the bifactor model) with questionnaire scores for each dataset.
Figure 2
Download asset Open asset
Dissociable associations of depression and anxiety with mood fluctuations.

Correlations of depression- and anxiety-specific factor scores with mood variation and RPE-related mood sensitivity (βRPE) for the laboratory dataset (A, E), online dataset 1 (B, F), online dataset 2 (C, G), and the combined dataset (N=1026; D, H). (I) Reduced RPE-related mood sensitivity statistically mediated the association between depression-specific scores and reduced mood fluctuations. (J) Increased RPE-related mood sensitivity statistically mediated the association between anxiety-specific scores and increased mood fluctuations. (K) Among the mood sensitivity parameters, the depression-specific factor was selectively associated with decreased RPE-related mood sensitivity. (L) Among the mood sensitivity parameters, the anxiety-specific factor was selectively associated with increased RPE-related mood sensitivity. Regression coefficients are shown as bootstrapped mean ± SE. CR, certain reward; EV, expected value; RPE, reward prediction error. *p<0.05.

Figure 3
Download asset Open asset
Clinical validation of reduced RPE-related mood sensitivity in depression.

(A, B) Correlations of depression and anxiety factor score with mood variation and mood parameter of RPE (βRPE) for the clinical dataset. (C) The mediation model among depression, βRPE, and mood variation in the clinical population. The regression coefficients were represented by mean ± SE, which were estimated by bootstrap. RPE, reward prediction error; *p<0.05.

Figure 4
Download asset Open asset
Consistent negative associations of the common factor with mood baseline and gain-domain risk attitude across datasets.

(A) Negative association between the common factor and the baseline mood parameter. (B) Negative association between the common factor and gain-domain risk attitude. Regression coefficients are shown as mean ± SE. *p<0.05.

Tables

Table 1
Basic demographic details.
Psychometric dataset
(N=901)		Laboratory dataset
(N=44)		Online dataset 1
(N=747)		Online dataset 2
(N=235)		Clinical
dataset
(N=116)
Gender (female)6761850014389
Age22.04±2.1020.05±1.7020.90±2.4121.67±2.5216.40±3.86
MASQaa24.70±8.1322.98±6.1424.16±7.6023.92±7.6243.51±10.51
TAIanx16.06±4.8615.00±4.9515.69±4.9715.99±5.3423.35±5.44
CESD35.28±10.7233.30±10.7334.82±10.7435.06±11.4155.64±13.04
BDI11.38±9.357.91±8.1110.04±8.9410.26±9.1822.66±17.91
BFIn33.99±8.8230.52±8.2632.82±9.3732.79±9.8143.75±9.31
PSWQ48.28±11.6146.89±12.5248.25±12.1548.36±13.2959.18±12.87
MASQad61.49±14.0060.23±16.6358.07±15.3059.57±15.7177.16±15.39
TAIdep29.39±5.4727.32±6.1627.61±6.2727.85±6.3435.97±4.91
Data collection periods2021.6.22–2021.6.282022.6.2–2022.7.52022.4.24–2022.5.222022.7.22–2022.7.262023.3.16–2023.8.4

  1. Descriptive data are presented as mean ± SD.

  2. MASQaa, the subscale of anxious arousal in the Mood and Anxiety Symptoms Questionnaire; TAIanx, the subscale of anxiety in the Trait Anxiety Inventory; CESD, Center for Epidemiologic Studies Depression Scale; BDI, Beck Depression Inventory; BFIn, the subscale of neuroticism in the Big Five Inventory; PSWQ, Penn State Worry Questionnaire; MASQad, the subscale of anhedonic depression in the Mood and Anxiety Symptoms Questionnaire; TAIdep, the subscale of depression in the Trait Anxiety Inventory.

Table 2
Exploratory analysis for the common factor.
ModelsHealthy datasets (n=1026)Clinical dataset (n=116)
BICResultsBICResults
β0 ~ common–830.23b=–0.03, t=–6.54, p<0.001, 95% CI = [-0.042–0.022]–113.74b=–0.03, t=–2.05, p=0.043, 95% CI = [-0.052–0.001]
β0 ~ common + common2–823.74–110.91
αgain ~ common551.76b=–0.03, t=–3.08, p=0.002, 95% CI = [-0.049–0.011]101.7b=–0.11, t=–3.24, p=0.002, 95% CI = [-0.172–0.041]
αgain ~ common + common2553.73106.12

  1. Note: For healthy datasets, we added datasets as random variables. β0, the mood baseline parameter; αgain, risk attitude for gain.

Appendix 1—table 1
Psychometric model comparisons.
ModelsCFITLIRMSEASRMR
Bifactor model0.9890.9890.0460.061
Three-factor model0.7940.7890.2000.205
High-order model0.9810.9810.0600.072
Appendix 1—table 2
Mood model comparisons.
Model number# of parametersΔ BIC
Laboratory dataset N=44Online dataset 1 N=747Online dataset 2 N=235Clinical dataset N=116
Model #160000
Model #2543.99918.62228.0041.57
Model #38122.131815.15586.31549.73
Model #47155.222662.46814.28424.10
Model #5782.24709.23204.32151.56
Model #6647.762514.71743.17601.16

  1. Δ BIC, Bayesian information criterion relative to the winning model (Model #1).

Appendix 1—table 3
Objective vs. subjective happiness models.
# of parametersΔ BIC
Healthy dataset N=1026Clinical dataset N=116Vanhasbroeck et al., dataset N=49Rutledge’s dataset N=46,204
Model #160000
Model #653305.64601.16118.004997.29
Model #1 vs 6 EP-1.001.001.001.00

  1. Δ BIC, Bayesian information criterion relative to the winning model (Model #1); EP, exceedance probability; EP greater than 0.95 was considered significant.

Appendix 1—table 4
Time effect on happiness.
Model number# of parametersΔ BIC
Healthy dataset N=1026Clinical dataset N=116Vanhasbroeck et al., dataset N=49Rutledge’s dataset N=46,204
Model #160000
Model #761883.3984.8898.154152.61
Model #861487.7966.3761.732295.40
Model #961064.2259.5374.401458.34

  1. Δ BIC, Bayesian information criterion relative to the winning model (Model #1).

Appendix 1—table 5
Contributions to model fit.
Model number# of parametersMean R2
Healthy dataset N=1026Clinical dataset N=116Vanhasbroeck et al., dataset N=49Rutledge’s dataset N=46,204
M1: CR +EV + RPE60.680.470.580.69
M10: No CR50.620.420.520.60
M11: No EV50.590.400.540.60
M12: No RPE50.420.290.340.50
Appendix 1—table 6
Interaction between different wins and risk preference for gain on z-scored happiness.
DatasetsTrial types
Certain trialsGamble trials (better outcomes; nonzero)
Healthy dataset N=1026beta = –0.00; t=–0.07; p=0.943 95% CI = [–0.01, 0.01]beta = –0.00; t=–0.24; p=0.807 95% CI = [–0.00, 0.00]
Clinical dataset N=116beta = –0.01; t=–0.64; p=0.524 95% CI = [–0.04, 0.02]beta = –0.00; t=–2.13; p=0.034 95% CI = [-0.01,–0.00]
Vanhasbroeck et al., dataset N=49beta = 1.41; t=0.51; p=0.609 95% CI = [–4.00, 6.82]beta = 0.38; t=0.55; p=0.581 95% CI = [–0.96, 1.71]
Rutledge’s dataset N=46,204beta = 0.10; t=0.84; p=0.402 95% CI = [–0.14, 0.34]beta = 0.03; t=0.82; p=0.412 95% CI = [–0.04, 0.10]
Appendix 1—table 7
Correlations of the depression-specific factor with mood variation and βRPE.
Laboratory datasetOnline dataset 1Online dataset 2Clinical dataset
Mood variationr=–0.309 p=0.041r=–0.191 p<0.001r=–0.135 p=0.039r=–0.239 p=0.009
βRPEr=–0.352 p=0.019r=–0.142 p<0.001r=–0.189 p=0.004r=–0.216 p=0.020
Mediationa×b = –0.306 95% CI: [-0.547,–0.065] p=0.015a×b = –0.096 95% CI: [-0.145,–0.047] p<0.001a×b = –0.126 95% CI: [-0.212,–0.040] p=0.004a×b = –0.141 95% CI: [-0.261,–0.038] p=0.021
Appendix 1—table 8
Clinical characteristics of patients with affective disorders.
Clinical variablesN=116
Diagnosis (MDD/AD/MDD and AD/others)51/22/32/11
Illness duration (months)19.57±19.22
Medications (yes)
SSRI96
Antipsychotics57
BZDs38
Mood stabilizer14

  1. Note that mood stabilizer refers to Lithium in this dataset.

Appendix 1—table 9
Choice model comparisons.
Model number# of parametersΔ BIC
Laboratory dataset N=44Online dataset 1 N=747Online dataset 2 N=235Clinical dataset N=116
cM11320.9210959.493818.362368.27
cM2462.703655.221030.07620.85
cM360000

  1. Δ BIC, Bayesian information criterion relative to the winning model (cM3).

Additional files

MDAR checklist
https://cdn.elifesciences.org/articles/110631/elife-110631-mdarchecklist1-v1.pdf

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Zhihao Wang
  2. Ting Wang
  3. Tian Nan
  4. Jiahua Xu
  5. André Aleman
  6. Yuejia Luo
  7. Bastien Blain
  8. Yunzhe Liu
  9. Pengfei Xu
(2026)
Dissociable roles of reward prediction error in the contrasting mood dynamics of depression and anxiety
eLife 15:RP110631.
https://doi.org/10.7554/eLife.110631.3