Introduction

Major depressive disorder

Major Depressive Disorder (MDD) is highly prevalent with a lifetime prevalence estimated at 13.5–21.2% of the general population (Kessler et al., 2005). Depression affects ~ 5% of the population globally, making it the leading cause of disability and ill-health worldwide (World Health Organization, 2017). MDD is characterized by impaired functioning and symptoms such as loss of energy, difficulties concentrating, fluctuations in weight, and feelings of worthlessness. Diminished interest or pleasure in daily activities is a key symptom of MDD, and is manifested by difficulties carrying out one’s daily routine (DSM-V; American Psychiatric Association, 2013).

Reward processing and motivated behavior

The motivational deficits observed in MDD are thought to be linked to the processes governing reward-seeking behaviors, including reward anticipation, attainment, and learning, which are collectively referred to as the Positive Valence System (PVS; National Institute of Mental Health, 2011; Olino, 2016). Another more detailed mechanistic parsing of motivational deficits was proposed by Husain and Roiser (2018), who suggested that a dysfunction in any one or more of the stages of effort-based decision-making to obtain a reward may underlie a-motivation in MDD and in other psychiatric and neurological conditions (Husain & Roiser, 2018). More broadly, the conceptualization of reward processing as a range, ensemble, or system has allowed for a more fine-grained characterization of the impairments in depression across various reward-related processes.

This differentiation of motivational deficits into sub-processes also helps reconcile the conflicting findings in the field. Although some studies have failed to provide robust evidence for reduced reward responsiveness and/or motivation in individuals with MDD (e.g., Moutoussis et al., 2018; Rothkirch et al., 2017; Winter et al., 2021), there is a growing consensus that MDD is linked to substantial impairment in some but not other facets of reward processing (e.g., Gradin et al., 2011; Huys et al., 2013; Pizzagalli et al., 2008; Proudfit, 2015).

Specifically, a recent meta-analysis by Halahakoon et al. (2020) drew two key conclusions that are directly related to the current study. The first is the substantial behavioral evidence supporting disrupted reward processing in MDD. The second is that this evidence varies substantially in strength across facets of reward processes. The strongest evidence points to less reward bias in MDD; i.e., a reduction in the adaptive preference for selecting more rewarding options. By contrast, there is scant evidence indicative of an association between MDD and impairments in response vigor, with the available evidence being extremely weak, if not altogether absent.

Different measures tap into the influence of different circuits of reinforcement

The Halahakoon et al. meta-analysis of 48 studies cited above included 12 studies that measured differences in response vigor, all of which involved monetary or social rewards and implemented variants on the Monetary Incentive Delay Task (Knutson et al., 2000). As noted in the overview, this measure of response vigor, when indexed by reaction time, captures response execution and not the decision to (and how to) act (Halahakoon et al., 2020).

Nevertheless, this presumed dissociation between intact reward processing at the response execution level and aberrant reward processing at the action selection level was challenged by a study showing that monetary value alone failed to consistently reduce reaction time in the general population when the action selection phase was held constant; in other words, when the action execution phase was isolated (Karsh et al., 2020). If rewards do not consistently increase the speed of response execution in general, it should come as no surprise that there is no convincing evidence for a reaction time difference between individuals with MDD and healthy controls in the rewarded versus unrewarded conditions on a response vigor task based on RT. Conversely, in the Karsh et al. (2020) study, the so-called pure effectiveness feedback, i.e., a response contingent but task-irrelevant perceptual change, did indeed consistently shorten response times in the general population.

Reinforcement as a function of desired outcomes and sensorimotor predictability

The finding that decision-uninformative feedback facilitates response execution is consistent with the theoretical and empirical differentiation between the well-documented modulation of behavior by desired outcomes (incentives or rewards) and another very rarely explored source of reinforcement from the confirmation of sensorimotor predictions (Bednark & Franz, 2014; Eitam et al., 2013; Hemed et al., 2020, Hemed et al., 2022; Karsh et al., 2016, 2020, 2021; Karsh & Eitam, 2015a; Penton et al., 2018; Tanaka et al., 2021; Wen et al., 2018; Wen & Imamizu, 2022). However, systematic experimentation supports the conclusion that control feedback reinforces behavior exclusively at the level of response execution, regardless of whether it is hedonically charged or valenced. Over 15 general population studies have shown that when responses are followed by an immediate and predictable own-response-contingent effect, response time is faster by about 40 ms than when no such feedback, or subtly lagged feedback, is delivered (Eitam et al., 2013; Karsh et al., 2016), or alternatively when (sensorially) the immediate feedback is unpredictable (Karsh et al., 2016).

Currently, the general assumption in the literature is that this facilitation reflects the workings of a modular reinforcement mechanism that is independent of the appetitive, valence, or hedonic value of a stimulus or abstract cognition. Rather, this mechanism is activated when an incoming sensation confirms a sensorimotor prediction. These predictions are generated by the system that controls motor movements and are produced unconsciously by mental processes that allow volitional movement whenever a voluntary movement is initiated (Eitam et al., 2013; Hemed et al., 2020, 2022; Karsh & Eitam, 2015a, 2015b; Karsh et al., 2016). The hypothesized mechanism, depicted in Fig. 1, is referred to as Reinforcement from Sensorimotor Predictability or RSP. The robust empirical effect of sensory feedback documented in the general population, and RSP theory that links it to the value-free reinforcement of response execution, enables a novel test of whether facilitation of response vigor, as a function of a non-reward related reinforcement, is aberrant in MDD.

Fig. 1
figure 1

Copyright 2016 by the American Psychological Association

RSP as a novel form of primary reinforcement that exclusively affects motor programing but not abstract response (action) representations. Note: Schematic illustration of the control-based response selection framework. The framework highlights the role of the Reinforcement from Sensorimotor Predictability (RSP) in reinforcing specific motor response parameters, as represented by the rhombuses (r1,…,rn). However, RSP does not influence the selection of abstract actions, denoted by the upper rectangles (R1,…,Rn), or the allocation of effort, both of which are decision-based processes. Dashed arrows indicate the flow of relevant information for explicit judgments of an action's effectiveness. Dotted arrows depict the generation of a forward model, which predicts the anticipated effect of a motor program on the environment and compares it to the perceived motor-program-related effect. Adapted from Karsh et al. (2016)

Full size image

Is reinforcement by sensorimotor predictability intact in MDD?

The present study examined whether modulation of response vigor by a reinforcement process that is not driven by rewards is intact in MDD. To do so, we tested whether individuals with mild–moderate MDD would display a weaker modulation of response vigor due to confirmation of sensorimotor predictions (the RSP process) on the Effect Motivation Task. If MDD is associated with a lack of modulation of response vigor, individuals with MDD should be less sensitive to sensorimotor feedback than in the general population, and there should be no response time facilitation. Conversely, finding that sensorimotor feedback modulates response vigor in MDD would be strong evidence that action execution is sensitive to reinforcement in this condition, at least for this type of reinforcement. Since in our theoretical framework this type of reinforcement does not rely on reward sensitivity as manifested in action selection or decision-making (Karsh et al., 2016, 2020), we expected that individuals with MDD would exhibit a modulation of response vigor due to RSP. Specifically, the presence (vs. absence) of sensorimotor feedback was predicted to facilitate response time.

Method

Overview

Data were collected over a 3-year period (2016–2018), as part of the training and active phases of an ongoing randomized controlled trial involving supportive expressive therapy for depression (Zilcha-Mano et al., 2018, 2021). The experiment described below examined the trial, pre-treatment phase. Patients were recruited through advertisements offering free treatment at the Psychotherapy Research Lab’s Clinic. Applicants interested in participating attended two pretreatment assessments, after providing written informed consent. The present study was conducted during the second pretreatment assessment.Footnote 1

This assessment included the administration of the Hamilton Rating Scale for Depression (HRSD; Hamilton, 1960), the MINI—International Neuropsychiatric Interview (Sheehan et al., 1998), and the Beck Depression Inventory BDI-II (Beck et al., 1996). In addition, structured interviews were conducted, and the participants also completed a baseline battery of questionnaires covering demographic information, concurrent medication, and personality characteristics.

The Effect Motivation Task (EMT; see Fig. 2) was administered after this battery (approximately 1.5 h into the second pretreatment assessment). Participants then filled in a post-task questionnaire on their perceived control and effort during the task.

Fig. 2
figure 2

The Effect Motivation Task (EMT). Note: Participants are instructed to respond with a keypress, given an imperative cue (a colored circle moving down the screen on a designated path). In the Sensorimotor-Feedback condition (B), a valid keypress turns the colored circle white mid-fall for 100 ms and then vanishes. In the No-feedback condition (A), a keypress results in no perceivable change on the screen and the circle descends uninterrupted to the bottom of the screen (Color figure online)

Full size image

Participants

The sample was composed of participants seeking treatment for depression as part of the training and active phase of an ongoing randomized controlled trial (N = 121, mean age = 30.69, SD = 8.47, range 19–59; 41% men). Only those who met the clinical criteria for depression were included in the analysis (see below), resulting in a final sample of 80 participants with MDD (mean age = 31.61, SD = 8.45, 39% men). Participants described themselves as being Jewish (n = 66; 82.5%), Christian (n = 8; 10%), Muslim (n = 3; 3.75%), or atheist (n = 2; 2.5%). Of the sample, 55.13% reported having a below-average income, 25.64% indicated an average income, and 19.23% an above-average income on the pre-treatment questionnaire. Most participants (63.75%) reported having 13–20 years of education; 32.5% had a high school education (12 years) and a small minority reported having fewer than 12 years or more than 20 years of education (3.75%), for an overall mean of 14.61 years of education (SD = 2.74, range 7–25). There were no significant differences between the experimental condition groups on any of the background variables (Table 1).

Table 1 Participants’ demographic data in the two experimental conditions
Full size table

The participants were considered to have MDD if they met the following criteria: (a) a diagnosis based on the structured clinical interview (MINI), and (b) a score above 14 on the 17-item HRSD during the two pretreatment assessments, administered 1 week apart. If they met only one criterion [HRSD score < 14 (8 participants) or did not meet the MINI diagnostic criteria (33 participants)], they were excluded from the main analyses.Footnote 2 However, these participants constituted the sub-clinical comparison group (see Fig. 6). The sociodemographic information and additional analyses pertaining to the participants in this group appear in the Supplementary Information.

All participants took the EMT after being randomly assigned to one of two conditions: Sensorimotor-feedback (N = 61; post-exclusion N = 40) or No-feedback (N = 60; post exclusion N = 40). The task was administered before treatment began.Footnote 3

Given the considerable challenges involved in data collection for this population, the experiment ran for a period of 3 years to sample as many participants as possible. A sensitivity analysis was conducted after the pool of available participants was exhausted using G*Power V.3.1.9.2 (Faul et al., 2009). It indicated that a fixed-effect linear regression with 38 participants per group (N = 76) would be sensitive to effects of Cohen’s d = 0.65 with 80% power (α = 0.05, two-tailed). This means the current study could not reliably detect effects smaller than Cohen’s d = 0.65.

The effect motivation task

Participants were administered the EMT while seated in front of a computer screen with their fingers positioned on the standard PC keyboard keys of ‘D’, ‘F’, ‘J’, and ‘K’. Each trial began with the presentation of a response cue (a colored circle; 53 pixels in size) in the upper section of the game window (9.5 cm × 10 cm), in one of four possible horizontal locations. On appearance, the circle descended smoothly (“fell”) vertically to the bottom of the game window. Participants were instructed to ‘stamp’ the circle as quickly and accurately as possible by pressing the relevant key (spatially and color-coded). In the Sensorimotor-feedback condition, a valid keypress always resulted in the circle immediately flashing white for 100 ms (ms) and then disappearing. In the No-feedback condition, a valid keypress had no perceivable effect and the circle continued its descent until it disappeared when it reached the bottom of the screen. The descent gave participants 850 ms to respond (until the circle vanished from the game window). Importantly, the time between the appearance of one circle and the next (stimulus onset asynchrony, SOA) was kept constant at 1500 ms regardless of participants' reaction times. Participants completed 180 experimental trials; the approximate total duration of the task was ~ 7 min. Reaction time and the accuracy of the responses were recorded.

Post-task questionnaire

Immediately after the experimental procedure, all the participants filled in a computerized self-report questionnaire that assessed their perceived control, enjoyment, perceived effort, and self-reported motivation to perform the task and be effective (to succeed on the EMT, be quick, accurate, and “make the circle flash”) on a slider scale from 1 (not at all) to 100 (very much).

Measures

Behavioral measures

(1) Reaction Time (RT). RT served to test the main prediction that when sensorimotor predictions are confirmed by incoming sensations, the RTs should become consistently shorter, thus indexing increased vigor in response execution; (2) Accuracy: the proportion of correct responses out of the total number of responses. In previous experiments using this task, the percent correct has generally been very high. Studies have occasionally found a reduction in accuracy levels that has been attributed to speeding up (potentially due to an increased ‘urge’ to respond), but this pattern has not been consistent across experiments, with some studies (3 out of 8) reporting facilitation of RTs without a reduction in accuracy (see Fig. 6 for details of previous studies).

Depression levels

The primary measure of the level of depression was the HRSD, a 17-item clinically administered measure of severity of depression. Total scores range from 0 to 54, with higher scores indicating greater severity of depression. The inter-rater reliability, as assessed by intraclass correlations, was .97. We also assessed self-reported symptoms using the BDI-II. Analysis indicated that choice of measure used to diagnose MDD had no effect on the key results.

Ethical considerations

All participants provided written informed consent before any of the study procedures, in accordance with the requirements for the ethical conduct of human subject research. The study design, procedure, and informed consent form were approved by the University’s Ethics Committee (approval#:118/15). Participants received detailed information about the study procedure both orally and in writing. Participants received complete information about the implications of their participation, including potential risks, inconveniences, and benefits, their right to halt their participation at any time without adverse consequences, issues of confidentiality, etc. A detailed, comprehensive data management plan ensured careful handling of confidential data throughout all stages of the research process.

Results

All statistical analyses were performed using Stata 16 (StataCorp, 2019) and JASP 0.9 (JASP Team, 2018); plotting was produced with Stata’s bischofschemes package (Bischof, 2017) and Python's Seaborn package (Waskom, 2021).

Data preparation and screening

To enable comparison to previous studies using the EMT, before analyzing the data, we applied five filters based on previous studies using the EMT. We rejected data in the following cases: (i) below the accuracy threshold: participants with more than 50%Footnote 4 incorrect responses (1 participant, ~ 1% of the observations), (ii) aberrant response times for this task: responses that were either above 800 ms or below 200 ms (~ 2%), (iii) incorrect responses (~ 5%), (iv) trials that deviated from their condition’s mean by at least 2 standard deviations in mean reaction time (~ 4%), and (v) participants who deviated from their condition’s mean by at least 2 standard deviations in percent correct (3 participants, ~ 4%). After applying these filters, the number of participants in both conditions was identical (N = 38 in the No-feedback condition and N = 38 in the Sensorimotor-feedback condition).

Depression levels

The HRSD scores ranged from 14 to 31, whereas the BDI-II scores ranged from 11 to 59. The distribution of depression levels did not significantly differ between conditions (Fig. 3). The HRSD and BDI-II scores in the No-feedback condition were MHRSD = 20.21 (SD = 4.00) and MBDI-II = 27.18 (SD = 8.90), respectively. These scores were not significantly different from the scores in the Sensorimotor-feedback condition, which were MHRSD = 20.97 (SD = 4.63) and MBDI-II = 29.22 (SD = 7.34); see Supplementary Information for a detailed analysis.

Fig. 3
figure 3

Hamilton Rating Scale for Depression (HRSD) and Beck Depression Inventory-II (BDI-II) scores in the No-feedback and Sensorimotor-feedback conditions. Note: Red and blue dots are individual scores on the HRSD and BDI-II No-feedback and Sensorimotor-feedback conditions, respectively (Color figure online)

Full size image

Behavioral measures

Reaction time

Given the relative ease of the task and generally very high accuracy of performance, reaction time (RT) is always taken as the key dependent measure on this task (Beierholm et al., 2013; Brown & Bowman, 1995; Griffiths & Beierholm, 2017; Haggard, 2008; Manohar et al., 2017; Pullman et al., 1988). A (fixed-effects) regression model was used to test the influence of sensorimotor feedback and depression level (and their interaction) on RT.Footnote 5 The first model included the HRSD score and Sensorimotor-feedback condition as predictors and was significant, explaining a substantial portion (25%) of the variance in RT (F(3,72) = 9.39, p < .001). Crucially, and fully replicating previous findings in the general population (Eitam et al., 2013; Hemed et al., 2020; Karsh et al., 2016, 2020; Karsh & Eitam, 2015a), there was a decisive main effect of sensorimotor feedback on RT (β =  − .51, p < .001, BF = 2449.00Footnote 6). As shown in Fig. 4, responses were faster [Cohen's d = 1.16; 95% CI (.67, 1.64)] in the Sensorimotor-feedback condition (M = 469.83, SD = 52.73) than in the No-feedback condition (M = 540.87, SD = 69.10).

Fig. 4
figure 4

Reaction time (RT) as predicted by the Sensorimotor-feedback factor and depression level (HRSD) Note: In the scatter plot, individual dots represent participants' scores on the Hamilton Rating Scale for Depression (HRSD) and mean reaction time (RT). The bands on the plot represent the bootstrapped 95% confidence intervals of the slopes. Red indicates the No-feedback condition; blue indicates the Sensorimotor-feedback condition (Color figure online)

Full size image

The apparent interaction in this model between sensorimotor feedback and HRSD was not significant when using the frequentist framework (β =  − .22, p = .15, BF = .84). However, the Bayes factor for a model with two main effects as compared to one that also included the interaction term (i.e., a test of the strength of an interaction effect) showed that the data were in fact insensitive to an interaction effect. Further examination of the robustness of the interaction effect using the log-transformed RT and range-restricted data to better capture the actual range of observed data (and avoid the possible hyper-influence of a few extreme observations) and the inconsistency between the two measures of depression regarding the interaction (see below) suggested that additional data would be required to determine whether there was an interaction effect that could reflect the modulation of potential sensorimotor retardation by sensorimotor feedback.

We then tested a second fixed-effects linear regression model that included the participants' BDI score (instead of the HRSD score used in the first model) to index level of depression, with sensorimotor feedback (and the interaction) as predictors of reaction time. This yielded a similar pattern, explaining 24% of the variance in RT (F(3,72) = 8.86, p =  < .001), where the Sensorimotor-feedback factor alone predicted RT (β =  − .52, p < .001, BF = 2020.51). As shown in Fig. 5, neither the BDI scores nor their interaction with Sensorimotor-feedback condition were associated with RT [(β = .05, p = .71, BF = .52) and (β = .08, p = .55, BF = .47) respectively]. The model that was best supported by the data only contained sensorimotor feedback as a predictor of RT (see Supplementary Information).

Fig. 5
figure 5

Response time (RT) as predicted by Sensorimotor feedback and Depression level (BDI) Note: The scatter plot shows individuals' mean RTs; bands represent the bootstrapped 95% CI of the slopes (Color figure online)

Full size image

Accuracy

Participants in the No-Feedback condition were significantly more accurate on average than participants in the Sensorimotor-feedback condition [t(74) = 4.35, p < .001, Cohen's d = 1.00; 95% CI (.52, 1.47), BF = 463.09]; see Table 2. Previous studies using a similar task in tests of the general population (Mean age = 22.98; SD = 4.38; 49% female) reported similar accuracy levels in both experimental conditions. As shown in Fig. 6c, d, the mean accuracy of 0.92 (SD = 0.06) in the No-feedback condition and a mean accuracy of 0.91 (SD = 0.05) in the Sensorimotor-feedback condition (Eitam et al., 2013, Experiments 1–3; Karsh et al., 2016, Experiments 1a, 2a, 2b; Karsh et al., 2020, Experiment 1b; Tanaka et al., 2021) were obtained.

Table 2 Means and standard deviations for accuracy in each condition
Full size table
Fig. 6
figure 6

The effect of sensorimotor feedback in MDD relative to the subclinical and general population samples Note: Mean reaction time [R and bootstrapped 95% confidence intervals (CI) for the Sensorimotor-feedback (light blue) and No-feedback (grey) conditions in Eitam et al., 2013, Experiments 1–3; Karsh et al., 2016, Experiments 1a, 2a, 2b; Karsh et al., 2020, Experiment 1b; and Tanaka et al., 2021, Experiment 1]. The results for the current study (clinical and sub-clinical samples) are represented in a darker hue (Box A). Across all studies, the mean RT was higher in the No-feedback condition. Box B shows the similarity in effect sizes between previous studies conducted on the general population and the current study involving individuals with MDD. Accuracy is shown in Box D, where the current findings exhibited relatively high accuracy overall, especially in the No-feedback condition. In particular, in the current study accuracy was significantly lower in the Sensorimotor-feedback condition than in the No-feedback condition. Box C presents a visual depiction of the variation in effect sizes for the differences in accuracy (Color figure online)

Full size image

In addition, a fixed-effects linear regression model was employed to examine the influence of sensorimotor feedback, depression level (HRSD score), and the interaction on accuracy. The results indicated that sensorimotor feedback significantly affected accuracy (β =  − .45, p < .001, BF = 513.01). Furthermore, the interaction between sensorimotor feedback and HRSD score also significantly affected accuracy (β =  − .39, p = .01, BF = 5.26). Depression level (HRSD score) did not independently affect accuracy, but instead in some cases slightly improved it (β = .27, p = .08, BF = 2.8). Specifically, accuracy was lower for participants in the Sensorimotor-feedback condition (M = 0.90, SD = 0.06) compared to participants in the No-feedback condition [M = 0.95, SD = 0.03) Cohen's d = 0.99; 95% CI (0.52, 1.47)]. The interaction between sensorimotor feedback and depression was such that as severity of depression increased, participants' relative accuracy in the Sensorimotor-feedback condition decreased.

Self-reported control, motivation, and performance

In previous studies using this task (Eitam et al., 2013; Hemed et al., 2020; Karsh & Eitam, 2015a; Karsh et al., 2016, 2020), RT was not consistently correlated with any of the self-reported measures. In the present study there was a correlation between participants' self-reported perceived control over the circle (i.e., the sensory effect) and RT solely in the Sensorimotor-feedback condition (r =  − .36, p = .03) as compared to the No-feedback condition (r =  − .16, p = .35). No correlation was found between the other self-report measures and RT, such as wanting the sensorimotor feedback, perceived effort, or intention to perform well (see Supplementary Information).

Accuracy and self-reported control over the flashing circle were found to be negatively and moderately correlated (r =  − .47, p < .001). There was also a weaker correlation between accuracy and three other self-reported measures; namely, enjoyment on the task (r =  − .24, p = .04), participants' evaluation of the enjoyment others could experience when doing the same task (r =  − .23, p = .04) and the urge to provoke the flashing circle (r =  − .24, p = .04). However, when examining the correlations by condition, accuracy was not related to any of the self-reported variables.

Comparison of current results and previous research on the general population and a sub-clinical comparison group

Compared to our previous work, the findings in the present study suggest that the confirmation of sensorimotor predictions effectively reinforced the response execution of individuals diagnosed with MDD, and increased the vigor of their responses to a similar extent. Since this study did not include a non-MDD control group, in what follows, as a proxy, we contrast the findings to previous experiments in which we used the EMT on general (student) populations to obtain a benchmark. We also examine the performance of the participants who failed to meet the criteria for a diagnosis of MDD as a sub-clinical comparison group (see SI).

Mean RT and effect size

To better estimate the effect of sensorimotor feedback on RT in the general population and facilitate a comparison to individuals diagnosed with MDD, we conducted a meta-analysis incorporating all (8) previously published studies that have utilized the EMT in the general population. The effect sizes, represented by Cohen's d, indicated faster response times in the Sensorimotor-feedback condition than in the No-feedback condition (Fig. 6). The combined effect size and confidence interval across all eight studies [Cohen's d = 0.75; 95% CI (0.54, 0.96)] were comparable to the effect size observed in the present study in both the MDD and Subclinical groups. In fact, the effect size in the current study on MDD patients was somewhat larger [Cohen's d = 1.16; 95% CI (0.67, 1.64)]. This finding further supports the claim that confirming sensorimotor predictions (RSP) reinforces individuals with MDD by increasing their response vigor (execution speed), as effectively as in the general population (Fig. 6), if not more so.Footnote 7

Accuracy

The pattern of accuracy in the current study is consistent with previous studies that have implemented the EMT, where lower accuracy rates were found in the Sensorimotor-feedback condition (as shown in Fig. 6c, d) in 5 out of the 8 studies. While this pattern was not systematically examined in our study, the theoretical framework (Fig. 1) positing that RSP influences motor programming but not action selection suggests that this outcome was not the result of a speed-accuracy tradeoff arising from a decision process. Rather, it appears to indicate a manifestation of motor “impulsivity” attributable to an urge to respond (or the propositional learning of S–R associations; cf. Tanaka et al., 2021). Hence, our current working hypothesis is that it constitutes a second independent effect of sensorimotor feedback on responding (but see Supplementary Information).

Discussion

The current study investigated the impact of sensorimotor feedback on response vigor in individuals diagnosed with Major Depressive Disorder (MDD). Participants with MDD were asked to respond to cues using a specific keypress, where their responses either resulted in immediate and highly predictable (i.e., constant) sensorimotor feedback or no visible feedback.

As shown in this study, sensory feedback significantly boosted the speed of responding in individuals with MDD, indicating the sensitivity of response vigor to the confirmation of sensorimotor predictions. Since this effect was observed to be as strong in clinically depressed individuals as reported in the general population, our answer to the question we set out to answer is that MDD does not appear to adversely influence the modulation of response vigor in response to reinforcement, at least of the RSP type.

These findings are generally consistent with our working model (see Fig. 1; Hemed et al., 2020; Karsh et al., 2016; Karsh & Eitam, 2015a) in which RSP is based solely on the motor-system comparator model (Blakemore et al., 2001) and is not modulated by cognitions such as participants’ control-relevant knowledge or self-causal attributions (for further empirical support see Hemed et al., 2020; Karsh et al., 2016, 2020; see also, Haggard, 2005; Mir et al., 2011). The dissociation observed between cognition and reinforcement resulting from the confirmation of sensorimotor predictions may provide an explanation for the notable resilience of this particular type of reinforcement.

The present findings provide strong evidence that sensorimotor prediction remains intact in individuals with mild to moderate MDD. As such, this finding sheds light on the robustness of sensorimotor processes to damage occurring in other (reward-related) circuits of reinforcement and points to the need for future, more fine-grained studies on the interactions between reinforcement and different levels of response planning (notably, action selection and response execution) to better understand and address depressive symptoms and a-volition in general.

The nature of the motivational deficit in MDD

Building on the current understanding of the mechanism underlying the effect of confirming sensorimotor predictions as reinforcing (for the latest version of the working model, see Hemed et al., 2020), the current results localize the set of mechanisms involved in reward processing in a potentially larger system of reinforcement that does not necessarily involve rewards or desired outcomes.

Speculating further, if the brain uses a central system for reinforcing representations, then such a system may have different inputs or subsystems; for example, one that is involved in assessment of the current value of a reward/utility and an independent one that operates within the motor system (i.e., RSP). Elsewhere, we have argued that the mental representations targeted by these different inputs are also different, with the former (desired outcomes/utility) targeting choice among more abstract (action) representations and the latter targeting the choice of motor programs driving the execution of the action in a specific context and time (Hemed et al., 2020).

Based on this terminology, the current findings can be restated and integrated with the literature on disrupted reward processing which has suggested that in MDD, reinforcement leading to effective motor execution (RSP) is intact but that the ability to perform effective motor execution is underutilized or inert because the ‘upper floor’; i.e., selection between actions for maximizing the probability of obtaining desired outcomes, is damaged.

Another limitation of the current study has to do with the lack of a healthy control group. However, we do not see this as a significant issue since the current findings were considered in relation to all previous published experiments that have tested the EMT on the general population. Furthermore, the sample of sub-clinical individuals exhibited comparable effects. Although the majority of the participants in the general population samples were undergraduate students, this was not the case for the sub-clinical sample. Hence the current findings constitute a significant step toward a finer-grained description of a-volition in individuals with MDD.