The specificity of Pavlovian regulation is associated with recovery from depression

Background Changes in reflexive emotional responses are hallmarks of depression, but how emotional reflexes make an impact on adaptive decision-making in depression has not been examined formally. Using a Pavlovian-instrumental transfer (PIT) task, we compared the influence of affectively valenced stimuli on decision-making in depression and generalized anxiety disorder compared with healthy controls; and related this to the longitudinal course of the illness. Method A total of 40 subjects with a current DSM-IV-TR diagnosis of major depressive disorder, dysthymia, generalized anxiety disorder, or a combination thereof, and 40 matched healthy controls performed a PIT task that assesses how instrumental approach and withdrawal behaviours are influenced by appetitive and aversive Pavlovian conditioned stimuli (CSs). Patients were followed up after 4–6 months. Analyses focused on patients with depression alone (n = 25). Results In healthy controls, Pavlovian CSs exerted action-specific effects, with appetitive CSs boosting active approach and aversive CSs active withdrawal. This action-specificity was absent in currently depressed subjects. Greater action-specificity in patients was associated with better recovery over the follow-up period. Conclusions Depression is associated with an abnormal influence of emotional reactions on decision-making in a way that may predict recovery.

SUPPLEMENTARY MATERIAL S1 DETAILED TASK DESCRIPTION INSTRUMENTAL TRAINING By design, the instrumental task was difficult, with few instrumental stimulus repetitions, interleaved presentation and highly noisy reinforcement signals. Weakly reinforcing the instrumental tendency allowed for strong Pavlovian CS effects in the PIT part of the task (below). Specifically, the instrumental task was framed in terms of collecting and sorting mushrooms. There were two counterbalanced blocks: approach and withdrawal ( Figure 1A,E). Each block contained six mushrooms. They were randomly selected from a pool of twelve and each presented 10 times, resulting in two blocks of 60 training trials. Subjects received reinforcement after every choice ("+20 cents" or "-20 cents") with probabilities of 0.75 or 0.25 depending on the stimulus. For half the mushrooms (three per block), a go response was rewarded with probability 0.75 and a nogo response with probability 0.25; for the other half, this reward contingency was reversed. Thus, each of the four types of action (approach go and nogo; and withdrawal go and nogo) were favoured by a greater chance of reward than punishment for three stimuli. Withdrawal was thus defined purely in terms of the intrinsically negative relationship of the action to the stimulus (moving away from it), and not in terms of any associated aversive outcomes (Roelofs et al., 2009). The design therefore controls for any possible associations of activation or approach/withdrawal with predictions of reward versus punishment.
Thirty-one of the control subjects came from the pool of subjects reported in (Huys et al., 2011), where the impact of deterministic and probabilistic feedback, and the impact of different types of withdrawal actions were tested. Overall, 35 controls experienced probabilistic feedback, as was the case for patients, and 5 deterministic feedback. Twenty-four used the throwaway action like patients. Sixteen performed the experiment with a different withdrawal action, which involved releasing the mouse button rather than moving the mouse cursor away from the mushroom. In the original experiment, detailed analyses revealed no difference between the two sorts of withdrawal action, or between deterministic and probabilistic instrumental feedback. We aggregated all the data. We note that the differences in instrumental reinforcement could, at worst, influence the overall strength of PIT effects, but it could not affect any differential influences of appetitive vs aversive CSs. Likewise, the variations in withdrawal actions could, at worst, influence the difference between approach and withdrawal, but could only do so in the controls. Neither of these variations can influence the pattern seen within the patients, or the association of depressive symptoms with the behaviour seen within the patient group.

PAVLOVIAN TRAINING
Five compound Pavlovian conditioned stimuli (CSs), consisting of a fractal visual stimulus and a pure tone were presented 12 times each in the first block, and 6 times in the second block ( Figure 1B). Each stimulus was deterministically followed by one particular outcome (-1e, -0.1e, 0, +0.1e or +1e). Every fifth trial was a forced choice trial: subjects were asked to choose between two Pavlovian stimuli in nominal extinction (no feedback; Figure 1D).

PAVLOVIAN-INSTRUMENTAL TRANSFER
In the third part of each block, the instrumental task was presented whilst the Pavlovian stimuli tiled the background of the display ( Figure 1C,F). Subjects were told to continue with the instrumental task, again in nominal extinction. It is critical to note that the Pavlovian stimulus was presented over the entire background, and, as such, could not by itself modulate the directionality of actions.

FURTHER EXPERIMENTS
Further experiments on helplessness, including a task (Huys et al., 2009) and questionnaires, will be reported elsewhere once analysed.

Follow-up
At follow-up, patients performed the task with a different, counterbalanced, stimulus set. FIGURE S1: Distribution of educational status between patients and controls. A two-sample Kolmogorov-Smirnov test identified no significant difference (p = 0.94). 60% of the controls and 73% of the patients were currently students.
From the total pool of 55 subjects, we extracted the 40 that matched the sample of 40 patients best for age by first sorting patients in terms of average squared age difference to all controls. Starting with the patient with highest average age discrepancy, each patient was assigned the control subject with minimum age difference. This yielded a sample matched for age, sex and IQ ( Figure S1).

S3 ACQUISITION OF PAVLOVIAN AND INSTRUMENTAL TRAINING
To test acquisition of Pavlovian conditioning, subjects intermittently chose amongst two Pavlovian stimuli in extinction (in the absence of feedback; Figure 1C). Two MDD patients performed no better than chance (binomial test, p < .05) and were excluded, resulting in a final sample of 40 controls, 25 patients at T1 (24 MDD, 1 DTH) and 23 patients with follow-up data at T2 (22 MDD, 1 DTH). Table 1 displays the group characteristics of this final sample.
Patients learned better at T2 (72 ± 8.6% correct) than at T1. This difference was significant (p = 0.04, t(22) = −2.18) and the linear learning trends during T2 were faster than during T1 (p = 0.055, t(23) = 2.02). Including the difference in learning trends as a covariate in a multiple regression abolished the session effect (p = 0.16), suggesting subjects' better performance better at T2 was due to faster learning.

SYMPTOM GROUPS
To examine whether action specificity might predict improvement in particular symptom domains, we examined four different subscores of the BDI (Table S1). Action specificity was most strongly associated with improve-  ment in anhedonic symptoms. It was also associated with improvement in somatic, affective and (at trend level) cognitive symptoms. The correlation with anhedonic score improvement was significantly larger than the correlation with cognitive score improvement (p = 0.03).

S5 MEDICATION EFFECTS
To maximise power, we examined the effect of medication in the entire depression + GAD sample. First, we examined whether T1 action specificity depended on medication. A linear multiple regression analysis with the predictors group and medication showed a significant effect of group p = 0.006 but not of medication (p = 0.3).
Similarly, in a separate analysis, the significant effect of the factor improver/nonimprover remained (p = 0.01) when including the medication factor; and this itself did not reach significance (p = 0.08), suggesting that medication did not explain the relationship between action specificity and improvement. Examining different types of medications (SSRIs, non-SSRI serotonergic medications or other medications (Lithium, mood stabilisiers, antipsychotics)) also failed to yield significant medication effects (all p > 0.1) and did not abolish the significant group effect at T1. Finally, we examined whether medication might account for the relationship between action specificity and BDI score at T2 by also regressing out medication status. The correlation between action specificity and the resulting residual BDI scores at T2 was still significantly negative ρ = −0.3924, p = 0.03.

S6 ACTION-SPECIFICITY: CHANGES OVER TIME
Action specificity was present in patients at T2 at a trend level (p = 0.08, signed rank). It no longer differentiated improvers from non-improvers (p = 0.7, signed rank; S2) and correlated neither with BDI scores at T2, nor with the change in BDI scores (both p > .2, rank correlation). Action specificity at T2 was driven by withdrawal PIT (p = 0.003, signed rank), with both improvers and non-improvers showing a significant effect at T2 (p = 0.02 and p = 0.04, respectively, signed rank tests). Effects of time on action specificity or withdrawal PIT did not reach significance in the depressed group (both p > 0.1).
We note that repetition effects may contribute to the behavioural findings at T2 in patients, and this needs to be ruled out using test-retest data in controls. However, that the ability to predict recovery is achieved based entirely on behavioural data at T1 only. Patients showed stronger instrumental behaviour at T2 and this may have negatively influenced our ability to see PIT effects at T2, but cannot by itself lead to false positives, particularly not at T1.   None  40  17  17  SSRI  0  5  4  5HT  0  5  5  Other  0  8 6 * significantly (p < 0.05) different from controls. † significantly (p < 0.05) different from T1. 1 trend difference from T1 (p < .1). All other comparisons are non-significant (p > 0.2). BAI = Beck Anxiety Index; BDI = Beck Depression Index II; HamD = Hamilton Depression score; SSRI = only selective serotonin inhibitor; 5HT = on one mainly serotonergic medication; Other = Lithium, antipsychotic medication, benzodiazepines or combination of multiple treatments. One patient with comorbid DTH+GAD was excluded due to poor performance on the Pavlovian forced choice trials. One patient with MDD+GAD was unavailable for follow-up. All tests are as described in the text, either Wilcoxon signed-rank or Mann-Whitney U tests. Results in column 'Depression only' are as reported in the main text. One patient with comorbid DTH+GAD was excluded due to poor performance on the Pavlovian forced choice trials. One patient with MDD+GAD was unavailable for follow-up.