The dynamical signature of anhedonia in major depressive disorder: positive emotion dynamics, reactivity, and recovery

Background Major Depressive Disorder (MDD) is the leading cause of disability worldwide. The cardinal features of MDD are depressed mood and anhedonia. Anhedonia is defined as a “markedly diminished interest or pleasure in all, or almost all, activities of the day”, and has generally been investigated on group-level using retrospective data (e.g. via questionnaire/interview). However, inferences based on group-level findings not necessarily generalize to daily life experiences within individuals. Methods We repeatedly sampled pleasurable experiences within individuals’ daily lives by means of Experience Sampling Methods, and compared how positive affect unfolded in the daily life of healthy controls versus patients diagnosed with MDD and anhedonia. We sampled Positive Affect (PA) and reward experiences on 10 semi-random time points a day, for seven days in the daily lives of 47 MDD patients with anhedonia, and 40 controls. Results Multilevel models showed that anhedonia was associated with low PA, but not to differences in PA dynamics, nor reward frequency in daily life. In reaction to rewards, MDD patients with anhedonia showed no difference in their increase in PA (i.e., PA reactivity), and showed no signs of a faster return to baseline thereafter (i.e., PA recovery). Conclusions Our results suggest that the dynamical signature of anhedonia in MDD can be described best as a lower average level of PA, and “normal” in terms of PA dynamics, daily reward reactivity and reward recovery. Preregistration: https://osf.io/gmfsc/register/565fb3678c5e4a66b5582f67. Preprint: https://osf.io/cfkts Electronic supplementary material The online version of this article (10.1186/s12888-018-1983-5) contains supplementary material, which is available to authorized users.


SUPPLEMENTARY MATERIAL 5
Level 2 (Equation 3b): β 0j = γ 00 + γ 01 (Group j ) + u 0j H5. To investigate whether MDD anhedonia is associated with more inert PA, we tested whether the groups differed in their autocorrelation of person-mean centered PA (Equation 4c; γ 11 (Group j )): Level 1 (Equation 4a): Level 2 (Equation 4b-d): β 0j = γ 00 + γ 01 + u 0j β 1j = γ 10 + γ 11 (Group j ) + u 1j β 2j = γ 20 + γ 21 (Group j ) + u 2j H6. Simultaneously, to investigate whether MDD anhedonia is associated with a different PA reaction to rewards, we tested whether the groups differed in the level of PA at i after a reward between i-1 and i had taken place (Equation 4d; γ 21 (Group j )), while controlling for person-mean centered PA at i-1. In case PA inertia did not differ by group, we reran the analyses while still controlling for the level of PA at i-1 but after omitting its non-significant cross-level interaction (i.e., Equation 5c): Level 1 (Equation 5a): Level 2 (Equation 5b-d): H7. To investigate whether anhedonia in MDD is associated with a faster decrease or recovery in PA after the initial increase in PA in reaction to a reward, we tested for group differences in PA recovery slope and duration.

Slope.
To investigate whether anhedonia in MDD is associated with a steeper slope, we tested whether the groups differed in their difference score of PA (i.e., P A i+1j − P A ij 3 ) after a reward (Equation 6d), while controlling for differences in person-mean centered PA reactivity to reward (operationalized here as P A ij − P A i−1j ; Equation 6b) as well as controlling for differences in minutes between i and i+1 ("Time" variable; Equation 6c): Level 1 (Equation 6a): Level 2 (Equation 6b-d): Duration. To investigate whether the time needed for PA recovery after rewards is shorter in MDD patients with anhedonia than in healthy controls, we tested whether the groups differed in the number of minutes participants needed to come back to baseline after a reward (i.e., the number minutes after i up until PA was equal or below the level of PA at i-1, the level before the reward was encountered) while controlling for the person-mean centered PA reactivity to reward (i.e., (P A ij − P A i−1,j )) in the following random intercepts model 4 : Level 1 (Equation 6a):

Familywise error rate control
In total, we tested 12 different but comparable models of PA reactivity (namely reactivity to psychological rewards and behavioral rewards; in a full and a trimmed model; followed by rerunning these four tests in the exploratory analyses with regard to PA2 and the newly developed ESM-item).
In order to maintain a familywise error rate of .05, we applied a VeffLi-Bonferroni-correction to the models of PA reactivity and PA recovery. The effective number of independent PA reactivity interactions were 9, making the significance threshold 4 It should be noted that we restricted the number of assessments in which the recovery could take place to i+5 (i.e., on average six hours after the reward was reported), and that we only calculated the number of minutes if no "new" reward was reported before PA was recovered. For the exact computations, please see our programming code in SPSS language in the supplementary material called "SPSS programming syntax".
With regard to PA recovery, we also tested 14 different models in total, namely: recovery after psychological rewards and behavioral rewards; with two alternative ways of modelling PA recovery; followed by rerunning the first two in the exploratory analyses with regard to PA2 and the newly developed ESM-item; complemented by two tests of the PA recovery duration after psychological rewards and behavioral rewards; and re-iterated for PA2. The effective number of independent PA reactivity interactions were 10, making the significance threshold that is required to keep Type I Error Rate at 5% with 10 independent variables 1 − (1 − 0.05) (1/10) = 0.01.

Supplementary material to H7a: Robustness of PA recovery slope model
We preregistrated that we would operationalized the PA Recovery slope as "P A i+1 − P A i |P E i , meaning that, if a Psychological Reward took place, a difference score is calculated between the first and second assessment of PA thereafter. To control for differences in participants' initial increase in PA (i.e., PA reactivity to PE, measured at assessment i), PA reactivity was operationalized as P A i − P A i−1 |Reward i , person-mean centered, and included as a covariate. In addition, to control for the different duration of assessment intervals, the time passed between assessments was used as a covariate." The time passed was measured as the number of minutes between t and t+1.
In addition to the model that we preregistrated, we came up with two additional models of how we could have modeled the PA recovery slope model to test our hypothesis that the PA recovery slope was steeper in individuals with MDD and anhedonia than in those individuals without such diagnoses.
To investigate whether the PA recovery slope decreases more steeply in MDD patients SUPPLEMENTARY MATERIAL 9 with anhedonia versus controls after reward experiences (p < .05), we controlled for the person-mean centered level of PA at time point i, the time point right after a reward was experienced (Reward or BR), P A i+1j reflects the amount of decrease in PA on one assessment after the initial increase in PA in reaction to experiencing a reward. Again, the slopes were separately modelled with regard to Psychological Rewards and Behavioral Rewards, that were reported on i (and thus experienced somewhere between i and i-1 ): Level 1 (Equation 4a): As shown in Supplementary STable 1, the first alternative option showed that, while controlling for PA at t, the experience of a Psychological Reward orBehavioral Reward between t and t-1 is unrelated to the level of PA on t+1 . Similarly, inalternative 2, while controlling for the person-mean centered initial increase in PA immediately thereafter (i.e., on t), the experience of a Psychological Reward between t and t-1 is unrelated to the level of PA on t+1. It should be noted, however, that these relationship coeficients are conditional on the value of Anhedonia. The coefficient for Psychological Reward reflects the effect of Psychological Reward in the control group (i.e., when Anhedonia=0); whereas the coefficient for anhedonia is the effect of anhedonia when there is no reward experienced (i.e., when PE=0).
As shown in Supplementary STable 2, the second alternative option showed that, while controlling for the person-mean centered initial increase in PA immediately thereafter on t, SUPPLEMENTARY MATERIAL 10 the experience of a Behavioral Reward between t and t-1 is linked to an increased level of PA on t+1 in the control group. Nevertheless, there is no difference in the effect of Behavioral Reward between patients with MDD and anhedonia and healthy controls.

Controlling for MDD severity
In a meta-analysis, the effect sizes of emotional reactivity in MDD showed considerable heterogeneity (Bylsma et al., 2008). In his recent review on emotions in depression, Rottenberg (2017) , 1996). After the introductory text "please checkmark the one response to each item that is most appropriate to how you have been feeling over the past 7 days", 16 statements followed on possible depressive states.
Answer categories on each item range from 0 to 3. To get the total score, symptom domains are summed. For domains that require more than one item, the highest score of the item relevant for each domain is taken. For example, if early insomnia is 0, middle insomnia is 1, late insomnia is 3, and hypersomnia is 0, the sleep disturbance domain is rated 3. The total score ranges from 0-27, and can be categorized as: 1-5 = No depression; 6-10 = Mild from what we preregistrated, we decided to ommit these follow-up analyses.

Influence of "Euphoria"
These exploratory analyses were not preregistered, and became of interest after accessing the data, plotting the data, and calculating its cronbachs alpha. Distribution plots of the PA items showed that feeling happy and relaxed where normally distributed, whereas feeling euphoric was heavily skewed towards zero (see SFigure 1). Cronbachs alpha improved from .03 to .44 when leaving out the euphoria item.
To investigate the impact of the "Euphoria"-item on our results, we reran the relevant analyses while omitting the scores on the euphoria item and thus using the average PA  In conclusion, results were all in the same direction when using PA calculated as a mean of happy and relaxed (instead of happy, relaxed, and euphoric).

Validation of newly developed ESM-item
These exploratory analyses were registered prior to accessing the data (see: Agreement between SCID, QIDS, and ESM-item. Next, we compared the agreement between our newly developed momentary ESM-item of anhedonia, and our two other measures of anhedonia: 1) anhedonia measured in retrospect by a trained clinian using the semi-structured interview SCID-I, and 2) the in retrospect self-reported anhedonia item measured by the QIDS-13.
As also described in the main article, anhedonia was assessed by the SCID-I question "Did you lose interest or pleasure in things you usually enjoyed? (What was that like?)" and, if yes, "When was that? Was that nearly every day? How long did it last? As long as two SUPPLEMENTARY MATERIAL 14 weeks?". Based on the participants' answers during the assessment of the SCID-I, a researcher who was a trained clinician rated anhedonia as "absent" "subthreshold", or "present".
For the 13th item of the QIDS questionnaire on general interest, participants could indicate (0) My normal interest in other people or activities has not changed; (1) I notice that I have less interest in other people or activities; (2) I am only interested in one or two of the activities I used to have; (3) I have practically no interest in activities that I used to have.
These exploratory analyses were registered prior to accessing the data (see: https://osf.io/gmfsc/). Because the QIDS was only assessed before ESM study and not thereafter, please note that we substituted step 2 by a different correlation than originally preregistrated. In addition, we used a slightly different statistical test than we preregistered.
The correlation between the patients' self-reported 13th item of the QIDS of loss of interest over the last two weeks, and the clinian-rated measure of loss of interest or pleasure in things the patient usually enjoyed over the last two weeks was r = .65, 95% CI [.51, .76], t(85) = 7.86, p < .001.
The correlation between the clinian-rated measure of loss of interest or pleasure in things the patient usually enjoyed over the last two weeks, and the median of patients' degree to which they found it difficult to experience pleasure in activities at ten semi-random momentary assessment in the two weeks that followed after baseline SCID diagnosis was r = .80, 95% CI [.71, .87], t(85) = 12.46, p < .001.

15
The correlation between the patients' self-reported 13th item of the QIDS of loss of interest over the last two weeks, and the median of patients' degree to which they found it difficult to experience pleasure in activities at ten semi-random momentary assessment in the two weeks that followed after baseline SCID diagnosis was r = .53, 95% CI  Note. Dependent variable is Positive Affect (PA) on time point t+1; PA is the average of feeling relaxed, happy, and euphoric; Time is the time past since the last assessment in minutes; laggedPA is the person-mean centered lagged variable of PA (i.e., PA on t-1); PAreactivity is the person-mean centered amount of increase in PA on time point t in comparison to t-1 (i.e., PA reactivity, but now modelled as a difference score: PA minus PA on t-1); Reward refers to positive event experienced somewhere between t-1 and t. Note. Dependent variable is Positive Affect (PA) on time point t+1; PA is the average of feeling relaxed, happy, and euphoric; Time is the time past since the last assessment in minutes; laggedPA is the person-mean centered lagged variable of PA (i.e., PA on t-1); PAreactivity is the person-mean centered amount of increase in PA on time point t in comparison to t-1 (i.e., PA reactivity, but now modelled as a difference score: PA minus PA on t-1); Reward refers to having experienced a positive event experienced somewhere between t-1 and t. Note. 'QIDS_SUM' refers to the sumscore of depression as measured by the QIDS; 'QIDS_13' refers to the score on the anhedonia item as measured by the QIDS; 'Anhedonia_ESM_med' refers to the within-person median of the momentary ESM item across the whole study period; PA is the average of the EMA items feeling relaxed, happy, and euphoric.