Neuromodulators are chemicals released in the brain that affect the activity of brain cells. Serotonin is a neuromodulator with the most complicated role: it is released in most brain regions and affects behavior in diverse ways. Serotonin is implicated in the regulation of mood, anxiety, impulsivity and learning. Moreover, most medications for depression target serotonin. A lack of motivation is an important symptom of depression, but exactly how serotonin affects motivation still remains unclear.

Meyniel et al. studied how increasing the amount of serotonin in the brain affects motivation in healthy people. The volunteers in the experiments squeezed a handgrip: the longer they squeezed, the more money they got as a reward. Before the experiment, some of the volunteers received an antidepressant drug that increases the amount of serotonin surrounding their brain cells, while others received a placebo.

The experiments revealed that, compared to the people who had the placebo, those who received the drug put in more effort to get a reward. More serotonin could increase motivation by reducing the perceived cost of putting in more effort, or by making people value the reward more. A mathematical model of the results showed that the increased motivation in the antidepressant group was more consistent with serotonin reducing the cost of putting in an effort, rather than increasing how much the reward was valued.

Combined with previous findings, these results suggest that serotonin affects the processing of cost associated with tasks – be that the amount of effort required, delays in getting a reward, or a punishment. Further experiments are now required to understand if the same mechanism operates in people with depression, and if so, whether it can be altered to promote recovery. It will also be important to better understand the interaction between serotonin and other neuromodulators such as dopamine.

Selective serotonin reuptake inhibitors (SSRI) are the most prescribed medications for major depressive episodes (Bauer et al., 2008). The effects of SSRI on improving mood and reducing anxiety have been well documented (Trivedi et al., 2006; Stahl, 2008; Cipriani et al., 2009) and many experimental studies showed that serotonin modulates emotional processing even in healthy volunteers (Harmer et al., 2009; Serretti et al., 2010). Loss of interest or pleasure in daily activities is also a key symptom of depression (American Psychiatric Association, 2013). However, the impact of SSRI on the motivation deficit (apathy) remains rather controversial (Papakostas et al., 2006; Weber et al., 2009).

Indeed, several studies reported that SSRI treatments do not reduce apathy as much as other symptoms (Fava et al., 2014) or can even induce apathy in patients (Barnhart et al., 2004; Sansone and Sansone, 2010; Padala et al., 2012). However, other studies also reported that SSRI treatments increase motivation or at least the sensitivity to reward (Tang et al., 2009; Stoy et al., 2012; Yuen et al., 2014). The related animal literature is also contradictory: some studies on cost/benefit trade-off showed reduced effort expenditure with SSRIs (Yohn et al., 2016) or no effect with serotonin blockers (Denk et al., 2005), while other studies reported increased motor activity induced by SSRI (Weber et al., 2009) or optogenetic stimulation of the dorsal raphe nucleus (Warden et al., 2012).

One issue with clinical studies is that SSRI effects may be confounded by an interaction with the pathological state: an SSRI treatment was reported to decrease apathy in late-life depression (Yuen et al., 2014) and to increase it in Parkinson’s disease (Zahodne et al., 2012). Another issue is that apathy may not be a simple construct. It is typically captured as a loss of behavioral activation (Cléry-Melin et al., 2011; Treadway et al., 2012), which could arise from different causes at the cognitive level: for instance a diminished sensitivity to reward attractiveness, or alternatively an exacerbated sensitivity to action cost. The ambition of the present study therefore was two-fold: we aimed (1) to clarify the effects of an SSRI on the motivation of effortful behavior in healthy volunteers, unconfounded by psychopathology, and (2) to reconcile these effects with a more general role of serotonin in the sensitivity to action costs vs. benefits.

Although experimental findings seem diverse at first sight, they might suggest a general role for serotonin in the behavioral adaptation to action costs. Previous studies suggested that serotonin is implicated in the motor aspects of action production and also the valuation processes that motivate the action. Notably, serotonin was found to condition impulse control and the capacity for behavioral inhibition (Cools et al., 2005; Crockett et al., 2009; Warden et al., 2012; Guitart-Masip et al., 2014) and to promote patience for delayed rewards (Schweighofer et al., 2008; Miyazaki et al., 2014; Worbe et al., 2014). Serotonin also impacts the determinants of actions, for instance, how positive and negative outcomes guide learning in humans (Chamberlain et al., 2006; Crockett et al., 2009; Faulkner and Deakin, 2014), monkeys (Clarke et al., 2004) and rodents (Bari et al., 2010). Several attempts have been made previously to capture serotonin function in a coherent computational theory. Key ideas are that serotonin (1) regulates the impact of action costs such as punishments (Daw et al., 2002; Niv et al., 2007) or delay (Niv et al., 2007; Cools et al., 2011; Miyazaki et al., 2014) and (2) adjusts the propensity to activate versus inhibit the action (Boureau and Dayan, 2011; Cools et al., 2011).

Borrowing from these models, we retain the general working hypothesis that serotonin regulates behavioral activation by modulating the weight of action costs rather than benefits.

We further suggest that the notion of action cost could be more general than initially envisaged. Indeed, in the literature quoted above, actions are mostly implemented as binary responses (e.g., approach vs. avoidance) and their costs or benefits are manipulated through the valence of their outcome (e.g., monetary gain or loss). Such tasks therefore over-simplify the real-life situation, where actions may often require more or less effort, depending on their intensity and duration.

Here, we tested the specific hypothesis that serotonin regulates the weight of effort cost in action production, as opposed to the weight of expected benefit. To assess the impact of changing cerebral serotonin levels and how it unfolds over time, we compared different groups of healthy subjects treated under double blind conditions with escitalopram (an SSRI) or placebo during eight weeks. Participants performed a previously published task (Meyniel et al., 2013), during which they allocated physical effort over time in order to maximize a monetary payoff that increased with effort duration (see Figure 1A). The task thus involves trading a physical effort cost against distinct levels of monetary incentives. Although changes in cost and benefit were independently manipulated in the task, they may result in intricate effects at the behavioral level. To disentangle between potential effects of serotonin on cost or benefit estimates, or both, we used a formal model of how decisions are generated by a hidden level of computations in our task, as previously described (Meyniel et al., 2013, 2014). Based on this computational analysis of the behavior, we could pinpoint the specific effect of serotonin on cost-related parameters, and track the predicted effects of such a computational change onto the experimental measures.

Figure 1

Download asset Open asset

Figure 1—source data 1

The MATLAB data file contains the payoff earned by each participant at each visit, in the placebo and escitalopram groups.

https://doi.org/10.7554/eLife.17282.004

Download elife-17282-fig1-data1-v1.mat

Our behavioral results in healthy volunteers show that the SSRI escitalopram improves global performance and hence payoff in a task that involves trading effort cost against monetary benefit. Taking advantage of the independent manipulation of cost and benefit, our computational analysis characterized the effect of escitalopram as a specific diminution of effort cost.

Together with previous findings, our results support the hypothesis that cost may be a general functional domain for serotonin (Boureau and Dayan, 2011; Dayan, 2012). For instance, interventions targeting serotonergic transmission during probabilistic and reversal learning paradigms in rodents, monkeys and humans suggested that serotonin impacts sensitivity to negative rather than positive feedback (Clarke et al., 2004; Chamberlain et al., 2006; Crockett et al., 2009; Bari et al., 2010; Cohen et al., 2015). The reduction of serotonergic transmission in humans following acute tryptophan depletion (for a review, see Faulkner and Deakin, 2014) reduced information sampling in a decision-making task, but crucially, only when sampling had a financial cost (Crockett et al., 2012). Serotonin may also impact moral costs, such as unfairness in social decision-making (Crockett et al., 2008). Furthermore, serotonin may control the cost of delay in reward delivery in rodents, such that a higher firing rate of serotonergic neurons correlates with an increased ability to wait for bigger rewards (Miyazaki et al., 2014), without affecting the sensitivity to the reward itself (Fonseca et al., 2015). Conversely, low serotonin levels, e.g. after acute tryptophan depletion, were suggested to exacerbate sensitivity to the cost of waiting, which could result in impulse control disorders (Crockett et al., 2009; Cools et al., 2011; Dayan, 2012). Thus, it is tempting to build a parsimonious computational theory on the idea that serotonin is involved in processing all kinds of action costs. This is compatible with the notion that different types of costs are processed by distinct neural circuits, since neuromodulators such as serotonin can affect many brain regions. However, such a generalization would necessitate assessing the putative role of each mono-amine and their interactions with different types of costs within the same study. Previous attempts in rodents have observed effects that are inconsistent with the present results: in cost/benefit trade-off tasks, serotonin blockers impacted choices only when costs were delays, and not physical efforts (Denk et al., 2005), whereas dopamine blockers affected both types of costs. Such discrepancies may call for caution when translating the results of pharmacological studies in rodents into medication effects in humans.

A general role for serotonin in cost processing also seems compatible with the effects on apathy and impulsivity reported in healthy and pathological conditions (Cools et al., 2005; Papakostas et al., 2006; Schweighofer et al., 2008; Crockett et al., 2009; Weber et al., 2009; Warden et al., 2012; Guitart-Masip et al., 2014; Miyazaki et al., 2014; Worbe et al., 2014). Indeed, according to our interpretation, SSRIs might reduce the perceived cost of performing actions, which would promote behavioral activation, hence the alleviation of apathy. As SSRIs might also reduce the perceived cost of delaying action, they would improve response control and reduce impulsivity (Cools et al., 2005; Cools et al., 2011; Crockett et al., 2009; Dayan, 2012; Bari and Robbins, 2013; Miyazaki et al., 2014; Fonseca et al., 2015).

However, it is important to note that the SSRI effects obtained here relate to effort cost and not to time discounting. Although time is central in our task, it departs from delay and waiting paradigms in several respects. First, the overall task duration was fixed and its pace was independent from how subjects allocated their effort within trials. Second, the reward was gained instantaneously, concomitantly to effort production, without delay. Therefore, the SSRI effect on effort duration (how long subjects sustain an effort) cannot be explained by a change in temporal discounting (how long subjects wait for a reward).

The direction of the effect obtained in the present study suggests that higher serotonin level alleviates the effort cost. This interpretation is based on the assumption that the primary effect of SSRI is to increase serotonin level in the synaptic cleft (Nutt et al., 1999; Stahl, 2008). However, the net effect of SSRI might not be so straightforward to interpret at the molecular level for several reasons. First, a high tonic concentration could have the paradoxical effect of reducing the sensitivity to phasic serotonergic signals (Faulkner and Deakin, 2014). Second, the ubiquitous negative feedback regulation by auto-receptors can initially revert the response expected from high extracellular levels of serotonin (Stahl, 2008; Fischer et al., 2014) and also produce non-linear dose-response functions for cognitive performance (Bari et al., 2010). Third, much evidence suggests that different serotonin projections to different forebrain systems mediate varying acute and chronic adaptive responses to aversive events (Hale et al., 2013; Faulkner and Deakin, 2014). The effect of serotonin also depends on the location in the brain and on the type of receptors, for instance, 5-HT1A and 5-HT2A receptors respectively inhibit and excite motoneurons (Jacobs and Azmitia, 1992). Thus, the superficial interpretation that serotonin helps overcoming effort costs will need to be refined by addressing the complexity of SSRI effects at the molecular level. We nevertheless note that the superficial interpretation is consistent with demonstrations that serotonin also helps overcoming other costs such as delay in humans (Schweighofer et al., 2008).

Refinement is also needed at the computational level. Our model does not specify how exactly serotonin could attenuate the impact of effort cost on action production. An indirect effect through an increase in muscular capacity can be excluded since there was no difference in maximal force between the placebo and SSRI groups at any visit. The computational analysis simply suggests that a slower accumulation of effort cost (lower Se_m) under SSRI prolonged effort duration. It does not distinguish between down-regulation of the cost signal itself, or down-regulation of the weight this cost has in the decision to produce an effort. The former (perceptual) view would be in line with an analgesic effect of escitalopram, which is consistent with the findings that nociception and/or somatosensory perception are both modulated by serotonin at central levels (Jacobs and Azmitia, 1992), that serotonin is targeted by common drugs modulating pain like acetaminophen (Smith, 2009) and that serotonin modulates the inhibitory feedback loop that allows muscular fatigue to down-regulate the motor command (Gandevia, 2001; Cotel et al., 2013). The latter (decisional) view would be more generalizable to the capacity of overcoming other types of costs.

As the serotonin and dopamine systems strongly interact (Dremencov et al., 2009; Cools et al., 2011; Schilström et al., 2011; Fischer et al., 2014), the SSRI effect observed here might also be at least in part mediated by dopamine. Indeed, the beneficial effect of SSRI shown here is reminiscent of previous reports about dopaminergic manipulations in mesolimbic structures, which affected the propensity to choose high reward – high effort over low reward – low effort options in rodents (Cousins et al., 1996; Salamone et al., 2007). However, if SSRIs antagonize dopamine release (Dremencov et al., 2009; Cools et al., 2011), escitalopram should have reduced effort production in our task (but see Schilström et al., 2011). Moreover, the fact that escitalopram did not modulate incentive effects on parameters such as the amplitude or dissipation slope argues against a participation of dopamine. Indeed dopaminergic manipulations in humans have been repeatedly shown to affect reward processing, not only in learning contexts (Frank et al., 2004; Pessiglione et al., 2006; Palminteri et al., 2009) but also in reward/effort trade-off paradigms (Wardle et al., 2011; Treadway et al., 2012; Le Bouc et al., 2016). It might be that serotonin and dopamine have complementary roles in promoting action production (and alleviating apathy): the former by reducing effort cost and the latter by enhancing the incentive value of potential rewards.

Previous studies suggest that we should distinguish between acute and chronic effects of SSRIs (Fischer et al., 2014). We did not find an interaction with time in our results but only a main effect of treatment group. Both parametric and non-parametric statistics showed that such a group difference is very unlikely to arise from chance in sampling the population (under the null hypothesis). General psychological assessment at baseline was also similar in the two groups. However, we acknowledge that our shortest time since treatment onset (3 days) may already depart from the acute regime, that ceiling effects in the SSRI group could in principle mask an interaction with time, and that our study may lack the statistical power to reveal such an effect. This absence of an interaction contrasts with the apparently delayed clinical effect of SSRIs in depressed patients, which usually take weeks to improve mood (Stahl, 2008). However, at the molecular level, serotonin release is boosted by SSRIs from treatment onset (Kobayashi et al., 2008) and SSRI effects on emotional processing also occur with little or no delay in healthy subjects (Harmer et al., 2003). In pathological conditions, time may be needed to adjust to the new, less negative perception of costs, as well as the reduced emotional bias, and convert these implicit changes into a conscious subjective improvement that can be reported to the practitioner (Harmer et al., 2009; Cools et al., 2011). This idea has been formalized in a recent study showing how positive and negative outcomes can shape mood on the long run (Eldar and Niv, 2015). Further studies in depressed patients are needed to assess whether an early detection of effort cost attenuation could be used to predict long-term treatment effects on clinical symptoms.

Finally, our results also indicate that non-trivial behavioral effects can be accounted for by a change in a single computational parameter (Se_m): a specific modulation of cost can result in both longer efforts and an increased sensitivity of effort duration to potential benefits in our results. Intuitively, this effect on reward sensitivity corresponds to the idea that when perceived costs are too high, a change in reward prospect will have little effect. Thus, our analysis shows that the behavioral consequences of cost and benefit estimates are intricate but computationally tractable. Our paradigm and model could provide experimental and conceptual tools to refine the description of motivational disorders. Distinct dysfunctions, such as amplification of effort cost vs. flattening of reward prospects, might call for different treatments: if drugs modulating serotonin only affect cost estimates, then other drugs (possibly dopaminergic) should aim to impact the valuation of potential benefits.

1. Book

   1. American Psychiatric Association

   (2013)

   Diagnostic and Statistical Manual of Mental Disorders (DSM-5)

   American Psychiatric Pub.

   • Google Scholar
2. 1. Balasubramani PP
   2. Chakravarthy VS
   3. Ravindran B
   4. Moustafa AA

   (2015) A network model of basal ganglia for understanding the roles of dopamine and serotonin in reward-punishment-risk based decision making

   Frontiers in Computational Neuroscience 9:76.

   https://doi.org/10.3389/fncom.2015.00076

   • PubMed
   • Google Scholar
3. 1. Bari A
   2. Robbins TW

   (2013) Inhibition and impulsivity: behavioral and neural basis of response control

   Progress in Neurobiology 108:44–79.

   https://doi.org/10.1016/j.pneurobio.2013.06.005

   • PubMed
   • Google Scholar
4. 1. Bari A
   2. Theobald DE
   3. Caprioli D
   4. Mar AC
   5. Aidoo-Micah A
   6. Dalley JW
   7. Robbins TW

   (2010) Serotonin modulates sensitivity to reward and negative feedback in a probabilistic reversal learning task in rats

   Neuropsychopharmacology 35:1290–1301.

   https://doi.org/10.1038/npp.2009.233

   • PubMed
   • Google Scholar
5. 1. Barnhart WJ
   2. Makela EH
   3. Latocha MJ

   (2004) SSRI-induced apathy syndrome: a clinical review

   Journal of Psychiatric Practice 10:196–199.

   https://doi.org/10.1097/00131746-200405000-00010

   • PubMed
   • Google Scholar
6. 1. Bauer M
   2. Monz BU
   3. Montejo AL
   4. Quail D
   5. Dantchev N
   6. Demyttenaere K
   7. Garcia-Cebrian A
   8. Grassi L
   9. Perahia DG
   10. Reed C
   11. Tylee A

   (2008) Prescribing patterns of antidepressants in Europe: results from the Factors Influencing Depression Endpoints Research (FINDER) study

   European Psychiatry 23:66–73.

   https://doi.org/10.1016/j.eurpsy.2007.11.001

   • PubMed
   • Google Scholar
7. 1. Boureau YL
   2. Dayan P

   (2011) Opponency revisited: competition and cooperation between dopamine and serotonin

   Neuropsychopharmacology 36:74–97.

   https://doi.org/10.1038/npp.2010.151

   • PubMed
   • Google Scholar
8. 1. Chamberlain SR
   2. Müller U
   3. Blackwell AD
   4. Clark L
   5. Robbins TW
   6. Sahakian BJ

   (2006) Neurochemical modulation of response inhibition and probabilistic learning in humans

   Science 311:861–863.

   https://doi.org/10.1126/science.1121218

   • PubMed
   • Google Scholar
9. 1. Cipriani A
   2. Furukawa TA
   3. Salanti G
   4. Geddes JR
   5. Higgins JPT
   6. Churchill R
   7. Watanabe N
   8. Nakagawa A
   9. Omori IM
   10. McGuire H
   11. Tansella M
   12. Barbui C

   (2009) Comparative efficacy and acceptability of 12 new-generation antidepressants: a multiple-treatments meta-analysis

   The Lancet 373:746–758.

   https://doi.org/10.1016/S0140-6736(09)60046-5

   • Google Scholar
10. 1. Clarke HF
    2. Dalley JW
    3. Crofts HS
    4. Robbins TW
    5. Roberts AC

    (2004) Cognitive inflexibility after prefrontal serotonin depletion

    Science 304:878–880.

    https://doi.org/10.1126/science.1094987

    • PubMed
    • Google Scholar
11. 1. Cléry-Melin ML
    2. Schmidt L
    3. Lafargue G
    4. Baup N
    5. Fossati P
    6. Pessiglione M

    (2011) Why don't you try harder? An investigation of effort production in major depression

    PLoS One 6:e23178.

    https://doi.org/10.1371/journal.pone.0023178

    • PubMed
    • Google Scholar
12. 1. Cohen JY
    2. Amoroso MW
    3. Uchida N

    (2015) Serotonergic neurons signal reward and punishment on multiple timescales

    eLife 4:e06346.

    https://doi.org/10.7554/eLife.06346

    • Google Scholar
13. 1. Cools R
    2. Blackwell A
    3. Clark L
    4. Menzies L
    5. Cox S
    6. Robbins TW

    (2005) Tryptophan depletion disrupts the motivational guidance of goal-directed behavior as a function of trait impulsivity

    Neuropsychopharmacology 30:1362–1373.

    https://doi.org/10.1038/sj.npp.1300704

    • Google Scholar
14. 1. Cools R
    2. Nakamura K
    3. Daw ND

    (2011) Serotonin and dopamine: unifying affective, activational, and decision functions

    Neuropsychopharmacology 36:98–113.

    https://doi.org/10.1038/npp.2010.121

    • PubMed
    • Google Scholar
15. 1. Cotel F
    2. Exley R
    3. Cragg SJ
    4. Perrier JF

    (2013) Serotonin spillover onto the axon initial segment of motoneurons induces central fatigue by inhibiting action potential initiation

    PNAS 110:4774–4779.

    https://doi.org/10.1073/pnas.1216150110

    • Google Scholar
16. 1. Cousins MS
    2. Atherton A
    3. Turner L
    4. Salamone JD

    (1996) Nucleus accumbens dopamine depletions alter relative response allocation in a T-maze cost/benefit task

    Behavioural Brain Research 74:189–197.

    https://doi.org/10.1016/0166-4328(95)00151-4

    • PubMed
    • Google Scholar
17. 1. Crockett MJ
    2. Clark L
    3. Robbins TW

    (2009) Reconciling the role of serotonin in behavioral inhibition and aversion: acute tryptophan depletion abolishes punishment-induced inhibition in humans

    Journal of Neuroscience 29:11993–11999.

    https://doi.org/10.1523/JNEUROSCI.2513-09.2009

    • PubMed
    • Google Scholar
18. 1. Crockett MJ
    2. Clark L
    3. Smillie LD
    4. Robbins TW

    (2012) The effects of acute tryptophan depletion on costly information sampling: impulsivity or aversive processing?

    Psychopharmacology 219:587–597.

    https://doi.org/10.1007/s00213-011-2577-9

    • PubMed
    • Google Scholar
19. 1. Crockett MJ
    2. Clark L
    3. Tabibnia G
    4. Lieberman MD
    5. Robbins TW

    (2008) Serotonin modulates behavioral reactions to unfairness

    Science 320:1739.

    https://doi.org/10.1126/science.1155577

    • PubMed
    • Google Scholar
20. 1. Daunizeau J
    2. Adam V
    3. Rigoux L

    (2014) VBA: a probabilistic treatment of nonlinear models for neurobiological and behavioural data

    PLoS Computational Biology 10:e1003441.

    https://doi.org/10.1371/journal.pcbi.1003441

    • PubMed
    • Google Scholar
21. 1. Daw ND
    2. Kakade S
    3. Dayan P

    (2002) Opponent interactions between serotonin and dopamine

    Neural Networks 15:603–616.

    https://doi.org/10.1016/S0893-6080(02)00052-7

    • PubMed
    • Google Scholar
22. 1. Dayan P

    (2012) Twenty-five lessons from computational neuromodulation

    Neuron 76:240–256.

    https://doi.org/10.1016/j.neuron.2012.09.027

    • PubMed
    • Google Scholar
23. 1. Denk F
    2. Walton ME
    3. Jennings KA
    4. Sharp T
    5. Rushworth MF
    6. Bannerman DM

    (2005) Differential involvement of serotonin and dopamine systems in cost-benefit decisions about delay or effort

    Psychopharmacology 179:587–596.

    https://doi.org/10.1007/s00213-004-2059-4

    • PubMed
    • Google Scholar
24. 1. Doya K

    (2002) Metalearning and neuromodulation

    Neural Networks 15:495–506.

    https://doi.org/10.1016/S0893-6080(02)00044-8

    • PubMed
    • Google Scholar
25. 1. Dremencov E
    2. El Mansari M
    3. Blier P

    (2009)

    Effects of sustained serotonin reuptake inhibition on the firing of dopamine neurons in the rat ventral tegmental area

    Journal of Psychiatry & Neuroscience 34:223–229.

    • PubMed
    • Google Scholar
26. 1. Eldar E
    2. Niv Y

    (2015) Interaction between emotional state and learning underlies mood instability

    Nature Communications 6:6149.

    https://doi.org/10.1038/ncomms7149

    • PubMed
    • Google Scholar
27. 1. Faulkner P
    2. Deakin JF

    (2014) The role of serotonin in reward, punishment and behavioural inhibition in humans: insights from studies with acute tryptophan depletion

    Neuroscience & Biobehavioral Reviews 46:365–378.

    https://doi.org/10.1016/j.neubiorev.2014.07.024

    • PubMed
    • Google Scholar
28. 1. Fava M
    2. Ball S
    3. Nelson JC
    4. Sparks J
    5. Konechnik T
    6. Classi P
    7. Dube S
    8. Thase ME

    (2014) Clinical relevance of fatigue as a residual symptom in major depressive disorder

    Depression and Anxiety 31:250–257.

    https://doi.org/10.1002/da.22199

    • PubMed
    • Google Scholar
29. 1. Fischer AG
    2. Jocham G
    3. Ullsperger M

    (2015) Dual serotonergic signals: a key to understanding paradoxical effects?

    Trends in Cognitive Sciences 19:21–26.

    https://doi.org/10.1016/j.tics.2014.11.004

    • Google Scholar
30. 1. Fonseca MS
    2. Murakami M
    3. Mainen ZF

    (2015) Activation of dorsal raphe serotonergic neurons promotes waiting but is not reinforcing

    Current Biology 25:306–315.

    https://doi.org/10.1016/j.cub.2014.12.002

    • PubMed
    • Google Scholar
31. 1. Frank MJ
    2. Seeberger LC
    3. O'reilly RC

    (2004) By carrot or by stick: cognitive reinforcement learning in parkinsonism

    Science 306:1940–1943.

    https://doi.org/10.1126/science.1102941

    • PubMed
    • Google Scholar
32. 1. Gandevia SC

    (2001)

    Spinal and supraspinal factors in human muscle fatigue

    Physiological Reviews 81:1725–1789.

    • PubMed
    • Google Scholar
33. 1. Guitart-Masip M
    2. Economides M
    3. Huys QJ
    4. Frank MJ
    5. Chowdhury R
    6. Duzel E
    7. Dayan P
    8. Dolan RJ

    (2014) Differential, but not opponent, effects of L -DOPA and citalopram on action learning with reward and punishment

    Psychopharmacology 231:955–966.

    https://doi.org/10.1007/s00213-013-3313-4

    • PubMed
    • Google Scholar
34. 1. Hale MW
    2. Raison CL
    3. Lowry CA

    (2013) Integrative physiology of depression and antidepressant drug action: implications for serotonergic mechanisms of action and novel therapeutic strategies for treatment of depression

    Pharmacology & Therapeutics 137:108–118.

    https://doi.org/10.1016/j.pharmthera.2012.09.005

    • PubMed
    • Google Scholar
35. 1. Harmer CJ
    2. Bhagwagar Z
    3. Perrett DI
    4. Völlm BA
    5. Cowen PJ
    6. Goodwin GM

    (2003) Acute SSRI administration affects the processing of social cues in healthy volunteers

    Neuropsychopharmacology 28:148–152.

    https://doi.org/10.1038/sj.npp.1300004

    • PubMed
    • Google Scholar
36. 1. Harmer CJ
    2. Goodwin GM
    3. Cowen PJ

    (2009) Why do antidepressants take so long to work? A cognitive neuropsychological model of antidepressant drug action

    The British Journal of Psychiatry 195:102–108.

    https://doi.org/10.1192/bjp.bp.108.051193

    • PubMed
    • Google Scholar
37. 1. Harmer CJ
    2. Shelley NC
    3. Cowen PJ
    4. Goodwin GM

    (2004) Increased positive versus negative affective perception and memory in healthy volunteers following selective serotonin and norepinephrine reuptake inhibition

    American Journal of Psychiatry 161:1256–1263.

    https://doi.org/10.1176/appi.ajp.161.7.1256

    • PubMed
    • Google Scholar
38. 1. Jacobs BL
    2. Azmitia EC

    (1992)

    Structure and function of the brain serotonin system

    Physiological Reviews 72:165–229.

    • PubMed
    • Google Scholar
39. 1. Kobayashi T
    2. Hayashi E
    3. Shimamura M
    4. Kinoshita M
    5. Murphy NP

    (2008) Neurochemical responses to antidepressants in the prefrontal cortex of mice and their efficacy in preclinical models of anxiety-like and depression-like behavior: a comparative and correlational study

    Psychopharmacology 197:567–580.

    https://doi.org/10.1007/s00213-008-1070-6

    • PubMed
    • Google Scholar
40. 1. Le Bouc R
    2. Rigoux L
    3. Schmidt L
    4. Degos B
    5. Welter ML
    6. Vidailhet M
    7. Daunizeau J
    8. Pessiglione M

    (2016) Computational dissection of dopamine motor and motivational functions in humans

    Journal of Neuroscience 36:In press.

    https://doi.org/10.1523/JNEUROSCI.3078-15.2016

    • PubMed
    • Google Scholar
41. 1. Meyniel F
    2. Safra L
    3. Pessiglione M

    (2014) How the brain decides when to work and when to rest: dissociation of implicit-reactive from explicit-predictive computational processes

    PLoS Computational Biology 10:e1003584.

    https://doi.org/10.1371/journal.pcbi.1003584

    • PubMed
    • Google Scholar
42. 1. Meyniel F
    2. Sergent C
    3. Rigoux L
    4. Daunizeau J
    5. Pessiglione M

    (2013) Neurocomputational account of how the human brain decides when to have a break

    PNAS 110:2641–2646.

    https://doi.org/10.1073/pnas.1211925110

    • Google Scholar
43. 1. Miyazaki KW
    2. Miyazaki K
    3. Tanaka KF
    4. Yamanaka A
    5. Takahashi A
    6. Tabuchi S
    7. Doya K

    (2014) Optogenetic activation of dorsal raphe serotonin neurons enhances patience for future rewards

    Current Biology 24:2033–2040.

    https://doi.org/10.1016/j.cub.2014.07.041

    • PubMed
    • Google Scholar
44. 1. Niv Y
    2. Daw ND
    3. Joel D
    4. Dayan P

    (2007) Tonic dopamine: opportunity costs and the control of response vigor

    Psychopharmacology 191:507–520.

    https://doi.org/10.1007/s00213-006-0502-4

    • PubMed
    • Google Scholar
45. 1. Nutt DJ
    2. Forshall S
    3. Bell C
    4. Rich A
    5. Sandford J
    6. Nash J
    7. Argyropoulos S

    (1999) Mechanisms of action of selective serotonin reuptake inhibitors in the treatment of psychiatric disorders

    European Neuropsychopharmacology 9:S81–S86.

    https://doi.org/10.1016/S0924-977X(99)00030-9

    • PubMed
    • Google Scholar
46. 1. Padala PR
    2. Padala KP
    3. Monga V
    4. Ramirez DA
    5. Sullivan DH

    (2012) Reversal of SSRI-associated apathy syndrome by discontinuation of therapy

    Annals of Pharmacotherapy 46:e8.

    https://doi.org/10.1345/aph.1Q656

    • PubMed
    • Google Scholar
47. 1. Palminteri S
    2. Lebreton M
    3. Worbe Y
    4. Grabli D
    5. Hartmann A
    6. Pessiglione M

    (2009) Pharmacological modulation of subliminal learning in Parkinson's and Tourette's syndromes

    PNAS 106:19179–19184.

    https://doi.org/10.1073/pnas.0904035106

    • Google Scholar
48. 1. Papakostas GI
    2. Nutt DJ
    3. Hallett LA
    4. Tucker VL
    5. Krishen A
    6. Fava M

    (2006) Resolution of sleepiness and fatigue in major depressive disorder: A comparison of bupropion and the selective serotonin reuptake inhibitors

    Biological Psychiatry 60:1350–1355.

    https://doi.org/10.1016/j.biopsych.2006.06.015

    • PubMed
    • Google Scholar
49. 1. Pessiglione M
    2. Seymour B
    3. Flandin G
    4. Dolan RJ
    5. Frith CD

    (2006) Dopamine-dependent prediction errors underpin reward-seeking behaviour in humans

    Nature 442:1042–1045.

    https://doi.org/10.1038/nature05051

    • PubMed
    • Google Scholar
50. 1. Salamone JD
    2. Correa M
    3. Farrar A
    4. Mingote SM

    (2007) Effort-related functions of nucleus accumbens dopamine and associated forebrain circuits

    Psychopharmacology 191:461–482.

    https://doi.org/10.1007/s00213-006-0668-9

    • PubMed
    • Google Scholar
51. 1. Sansone RA
    2. Sansone LA

    (2010)

    SSRI-Induced Indifference

    Psychiatry (Edgmont) 7:14–18.

    • PubMed
    • Google Scholar
52. 1. Schilström B
    2. Konradsson-Geuken A
    3. Ivanov V
    4. Gertow J
    5. Feltmann K
    6. Marcus MM
    7. Jardemark K
    8. Svensson TH

    (2011) Effects of S-citalopram, citalopram, and R-citalopram on the firing patterns of dopamine neurons in the ventral tegmental area, N-methyl-D-aspartate receptor-mediated transmission in the medial prefrontal cortex and cognitive function in the rat

    Synapse 65:357–367.

    https://doi.org/10.1002/syn.20853

    • PubMed
    • Google Scholar
53. 1. Schweighofer N
    2. Bertin M
    3. Shishida K
    4. Okamoto Y
    5. Tanaka SC
    6. Yamawaki S
    7. Doya K

    (2008) Low-serotonin levels increase delayed reward discounting in humans

    Journal of Neuroscience 28:4528–4532.

    https://doi.org/10.1523/JNEUROSCI.4982-07.2008

    • PubMed
    • Google Scholar
54. 1. Serretti A
    2. Calati R
    3. Goracci A
    4. Di Simplicio M
    5. Castrogiovanni P
    6. De Ronchi D

    (2010) Antidepressants in healthy subjects: what are the psychotropic/psychological effects?

    European Neuropsychopharmacology 20:433–453.

    https://doi.org/10.1016/j.euroneuro.2009.11.009

    • PubMed
    • Google Scholar
55. 1. Smith HS

    (2009)

    Potential analgesic mechanisms of acetaminophen

    Pain Physician 12:269–280.

    • PubMed
    • Google Scholar
56. 1. Stahl SM

    (2008)

    Stahl’s Essential Psychopharmacology. Neuroscientific Basis and Practical Applications

    Chapter 12 - Antidepressants, Stahl’s Essential Psychopharmacology. Neuroscientific Basis and Practical Applications, (2nd ed), Cambridge University Press.

    • Google Scholar
57. 1. Stephan KE
    2. Penny WD
    3. Daunizeau J
    4. Moran RJ
    5. Friston KJ

    (2009) Bayesian model selection for group studies

    NeuroImage 46:1004–1017.

    https://doi.org/10.1016/j.neuroimage.2009.03.025

    • PubMed
    • Google Scholar
58. 1. Stoy M
    2. Schlagenhauf F
    3. Sterzer P
    4. Bermpohl F
    5. Hägele C
    6. Suchotzki K
    7. Schmack K
    8. Wrase J
    9. Ricken R
    10. Knutson B
    11. Adli M
    12. Bauer M
    13. Heinz A
    14. Ströhle A

    (2012) Hyporeactivity of ventral striatum towards incentive stimuli in unmedicated depressed patients normalizes after treatment with escitalopram

    Journal of Psychopharmacology 26:677–688.

    https://doi.org/10.1177/0269881111416686

    • PubMed
    • Google Scholar
59. 1. Tang TZ
    2. DeRubeis RJ
    3. Hollon SD
    4. Amsterdam J
    5. Shelton R
    6. Schalet B

    (2009) A placebo-controlled test of the effects of paroxetine and cognitive therapy on personality risk factors in depression

    Arch Gen Psychiatry 66:1322–1330.

    https://doi.org/10.1001/archgenpsychiatry.2009.166

    • Google Scholar
60. 1. Treadway MT
    2. Bossaller NA
    3. Shelton RC
    4. Zald DH

    (2012) Effort-based decision-making in major depressive disorder: a translational model of motivational anhedonia

    Journal of Abnormal Psychology 121:553–558.

    https://doi.org/10.1037/a0028813

    • PubMed
    • Google Scholar
61. 1. Trivedi MH
    2. Rush AJ
    3. Wisniewski SR
    4. Nierenberg AA
    5. Warden D
    6. Ritz L
    7. Norquist G
    8. Howland RH
    9. Lebowitz B
    10. McGrath PJ
    11. Shores-Wilson K
    12. Biggs MM
    13. Balasubramani GK
    14. Fava M
    15. STAR*D Study Team

    (2006) Evaluation of outcomes with citalopram for depression using measurement-based care in STAR*D: implications for clinical practice

    American Journal of Psychiatry 163:28–40.

    https://doi.org/10.1176/appi.ajp.163.1.28

    • PubMed
    • Google Scholar
62. 1. Warden MR
    2. Selimbeyoglu A
    3. Mirzabekov JJ
    4. Lo M
    5. Thompson KR
    6. Kim SY
    7. Adhikari A
    8. Tye KM
    9. Frank LM
    10. Deisseroth K

    (2012) A prefrontal cortex-brainstem neuronal projection that controls response to behavioural challenge

    Nature 492:428–432.

    https://doi.org/10.1038/nature11617

    • PubMed
    • Google Scholar
63. 1. Wardle MC
    2. Treadway MT
    3. Mayo LM
    4. Zald DH
    5. de Wit H

    (2011) Amping up effort: effects of d-amphetamine on human effort-based decision-making

    Journal of Neuroscience 31:16597–16602.

    https://doi.org/10.1523/JNEUROSCI.4387-11.2011

    • Google Scholar
64. 1. Weber M
    2. Talmon S
    3. Schulze I
    4. Boeddinghaus C
    5. Gross G
    6. Schoemaker H
    7. Wicke KM

    (2009) Running wheel activity is sensitive to acute treatment with selective inhibitors for either serotonin or norepinephrine reuptake

    Psychopharmacology 203:753–762.

    https://doi.org/10.1007/s00213-008-1420-4

    • PubMed
    • Google Scholar
65. 1. Worbe Y
    2. Savulich G
    3. Voon V
    4. Fernandez-Egea E
    5. Robbins TW

    (2014) Serotonin depletion induces 'waiting impulsivity' on the human four-choice serial reaction time task: cross-species translational significance

    Neuropsychopharmacology 39:1519–1526.

    https://doi.org/10.1038/npp.2013.351

    • PubMed
    • Google Scholar
66. 1. Yohn SE
    2. Collins SL
    3. Contreras-Mora HM
    4. Errante EL
    5. Rowland MA
    6. Correa M
    7. Salamone JD

    (2016) Not all antidepressants are created equal: differential effects of monoamine uptake inhibitors on effort-related choice behavior

    Neuropsychopharmacology 41:686–694.

    https://doi.org/10.1038/npp.2015.188

    • PubMed
    • Google Scholar
67. 1. Yuen GS
    2. Gunning FM
    3. Woods E
    4. Klimstra SA
    5. Hoptman MJ
    6. Alexopoulos GS

    (2014) Neuroanatomical correlates of apathy in late-life depression and antidepressant treatment response

    Journal of Affective Disorders 166:179–186.

    https://doi.org/10.1016/j.jad.2014.05.008

    • PubMed
    • Google Scholar
68. 1. Zahodne LB
    2. Bernal-Pacheco O
    3. Bowers D
    4. Ward H
    5. Oyama G
    6. Limotai N
    7. Velez-Lago F
    8. Rodriguez RL
    9. Malaty I
    10. McFarland NR
    11. Okun MS

    (2012) Are selective serotonin reuptake inhibitors associated with greater apathy in Parkinson’s disease?

    Journal of Neuropsychiatry and Clinical Neurosciences 24:326–330.

    https://doi.org/10.1176/appi.neuropsych.11090210

    • Google Scholar

Thank you for submitting your article "A specific role for serotonin in overcoming effort cost" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by Joshua Gold as the Reviewing Editor and Sabine Kastner as the Senior Editor. The following individuals involved in review of your submission have agreed to reveal their identity: Read Montague (Reviewer #1); Roshan Cools (Reviewer #3).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

This study used a physical effort/incentive-based task with two groups of subjects – one taking escitalopram, the other placebo – to test whether SSRIs, and putatively serotonin, have an effect on the amount of effort human subjects will expend to get a reward. Several of the authors had previously developed a rise-to-bound computational model to account for behavior on this task in terms of a "cost evidence" variable that rises during effort and falls during rest. Here they used the model to show that overall better performance of the SSRI group (more reward) is associated with reduced effort costs and not the weight in monetary incentives. Thus, 5HT's role in action costs might be more general than hitherto thought, extending from the regulation of punishment and delay to that of effort costs.

The reviewers all agreed that this is an excellent paper: interesting, well done, and well written. The role of serotonin in behavioral control and decision-making is complex, and this paper makes a solid contribution to shrinking some of this complexity by sharpening our understanding of it in relation to effort. The ubiquity of SSRI use – in depression, Parkinson's Disease, and other disorders – and the lack of model-based understanding of what these drugs do in terms of information processing makes the contribution important. It is also commendable from a clinical perspective that chronic rather than just acute effects of the drug were assessed.

Essential revisions:

1) Several reviewers thought that the model needed to be explained better, and the modeling results analyzed more thoroughly.

A) Specifically, the text only briefly refers to specific parameters ("amplitude, accumulation and dissipation slopes of the cost-evidence decision variable") that a reader without thorough familiarity of the earlier papers will have a hard time following. It would be useful to describe the model in the text. Likewise, the model parameters are listed ("A_i, Se_m, etc.) without being explained. For more context, it might also be useful to relate their model more directly to the RL model in Daw et al. 2002 referenced in the paper, and also the model in Balasubramani et al. 2015, Frontiers of Computational Neuroscience.

B) More substantively, there were questions about the specificity of the effect from the modeling analyses. It is stated that interactions between treatments and pairs of computational variables (Se_m versus each other parameter) were always significant. The implication of this result depends on the sign of the parameters. The cumulative payoff effect might represent, for example, enhanced amplitude modulation (A_i) and/or reduced cost-evidence accumulation (Se_m). How were computational parameters coded when entered into the ANOVA? To assess specificity, one might want to reverse the coding for A and Sr. More generally, to what extent were the parameters independent of each other in the model? Also, might it be possible to define a different model space that would allow the use of Bayesian stats, and model comparison to estimate the drug effects on the various parameters? The distinct advantage of such an approach would be that it allows the capturing of drug effects despite the presence of individual differences of no interest.

C) Is any background 'reference' neuropsychological data available to confirm that there were no group differences of no interest? If so, please report.

2) There were questions about the interpretation of the effects in terms of the role of time in task performance. The main effects coming out of the modeling analysis appear to be related to effort duration. But longer effort durations are also longer overall durations, which brings up the literature on 5-HT and delay costs (some of the delay discounting papers with 5-HT manipulations are actually cited in the manuscript). The authors appear to favor the view that different types of costs can be reduced to one process that is related overall to multiple types of costs. There is evidence both for and against this, and the authors are of course free to interpret their results in this way. In fact, this discussion contributes to the field in an interesting way. However, how much of the effect they are seeing is related to time per se? The authors should discuss this.

3) The Introduction might benefit from mentioning the implication of 5HT not just in apathy, but also in the other end of the motivation continuum, that is impulsivity. And in the Discussion, how do the findings relate to observations that 5HT is key for impulse control? More generally, the paper might come full circle, by linking, in the Discussion section, the obtained results to these various clinical hypotheses (involvement in apathy, impulsivity).

[Editors' note: further revisions were requested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "A specific role for serotonin in overcoming effort cost" for further consideration at eLife. Your revised article has been favorably evaluated by Sabine Kastner (Senior editor), a Reviewing editor (Joshua Gold), and one new reviewer, who offered input from a statistical perspective. The manuscript has been improved but there are some remaining issues that need to be addressed before acceptance, as outlined by the reviewer:

Reviewer #4:

I have reviewed this paper from a statistical perspective as this is a registered trial.

This manuscript presents the analysis of a single randomised placebo controlled and double blinded study. It is not a trial of an intervention but rather a controlled study to understand the mechanism of action that the specific drug takes. The trial has been registered with a target sample size of 128 and it has a single hypothesis statement that aims to 'assess the effects of' agolmelatine and escitalopram on emotional binding, emotional processing and motivation'. Each participant is measured over several tasks during the treatment period of 9 weeks.

I realise that it is not in the eLife style to report numerical estimates in the Abstract but as this report analyses a single controlled experiment I think they should put the strength of the evidence in the Abstract to back up their statement that 'they show that serotonin also regulates other types of action costs such as effort'. As there is no replication they should report this in a more guarded manner.

The authors have completed the Consort statement but they need to provide more complete information within the manuscript so they can say yes to more of the checklist.

The trial registration lists a compound aim/hypothesis and also lists three primary outcomes listed which I think have all been used in this analysis but the main question for the manuscript is 'does serotonin regulate the weight of effort cost as opposed to the weight of expected benefit?'

I would like to see a clearer link between the main research question and the forms of analysis. A sample size of 58 is not that large even with repeated measurement. Was there really any power to test for interactions? The Consort checklist says 'no' to sample size but surely the size was justified originally? I feel a bit concerned that sometimes Bonferonni correction is used to conclude there were no differences when this just reduces power.

Essential revisions:

1) Several reviewers thought that the model needed to be explained better, and the modeling results analyzed more thoroughly.

A) Specifically, the text only briefly refers to specific parameters ("amplitude, accumulation and dissipation slopes of the cost-evidence decision variable") that a reader without thorough familiarity of the earlier papers will have a hard time following. It would be useful to describe the model in the text. Likewise, the model parameters are listed ("A_i, Se_m, etc.) without being explained. For more context, it might also be useful to relate their model more directly to the RL model in Daw et al. 2002 referenced in the paper, and also the model in Balasubramani et al. 2015, Frontiers of Computational Neuroscience.

We agree that the model should have been presented with more details in the main text, instead of being confined to the Methods section. We have revised the model description in the Results section, which now appears as follows:

“The better performance observed in the escitalopram group could be underpinned by different mechanisms: alleviation of effort costs or inflation of incentive values, or both. […] To capture all the effects, the cost-evidence accumulation model therefore necessitates five free parameters: the mean slope of cost-evidence accumulation (Se_m) and its steepening for higher difficulty levels (Se_d); the mean slope of cost-evidence dissipation during rest (Sr_m) and its steepening for higher incentives (Sr_i); and the expansion of the cost-evidence amplitude as higher incentives push the bounds back (A_i).”

The referees also ask for a more detailed contextualization of our model, in particular with respect to the reinforcement learning framework (RL).

We would like to start by pointing important conceptual distinctions with respect to RL. First, RL deals with mapping stimuli to responses and outcomes across time. The parallel with our task is limited since here, external stimuli are mostly restricted to cues that have an explicit meaning (coins indicating the reward rate) and outcomes are deterministically and immediately coupled to actions: the cumulative payoff is displayed continuously as a function of the current force and incentive level. Second, the dynamics in RL reflects the learning of stimulus-response-outcome mapping across trials, whereas in our task, the dynamics reflects the allocation of effort within and between trials. Third, RL usually comes with a discretization of events into trials, with transient prediction errors whereas in our task, actions and feedback are continuous in time. Last, in RL, “cost” refers to negative outcomes that thereby vary along the same axis as “benefit” with a negative valence. In contrast, the effort cost is instrumental to get reward in our task. In total, the processes that we intended to model here do not involve much RL, as the mapping is clearly specified in the instructions. These processes are meant to solve a choice problem: knowing all the contingencies, how should I allocate my effort over time?

The referees also point to Balasubramani et al. 2015 who provide a detailed model for the implementation of RL in the basal ganglia, adding the notion of risk to standard RL. This level of detail is far beyond the sort of description that is needed to capture the behavior in our task. We acknowledge that the uncertainty about one's ability to sustain effort in the future could be seen as similar to the notion of risk defined as the variance in potential outcomes. However, such a link is extremely speculative for the moment.

Therefore, we prefer to avoid making a direct comparison to RL models. However, we intended to contextualize our model with respect to relevant notions that emerged in the field of RL, such as the notions of opponency, optimization, time scale and opportunity cost. We added the following paragraph in the Methods (with reference to Balasubramani et al., 2015):

“This computational model can be interpreted as follows, for a more complete discussion see (Meyniel 2013, 2014). The accumulation of cost evidence may reflect the build-up of physiological fatigue: the instantaneous effort cost increases as it becomes gradually more difficult to sustain a given force level. […] This opportunity cost seems to be taken into account since the cost-evidence dissipation is speeded up for higher incentives (which is captured by Se_i), so that effort can be resumed more quickly.”

B) More substantively, there were questions about the specificity of the effect from the modeling analyses. It is stated that interactions between treatments and pairs of computational variables (Se_m versus each other parameter) were always significant. The implication of this result depends on the sign of the parameters. The cumulative payoff effect might represent, for example, enhanced amplitude modulation (A_i) and/or reduced cost-evidence accumulation (Se_m). How were computational parameters coded when entered into the ANOVA? To assess specificity, one might want to reverse the coding for A and Sr. More generally, to what extent were the parameters independent of each other in the model? Also, might it be possible to define a different model space that would allow the use of Bayesian stats, and model comparison to estimate the drug effects on the various parameters? The distinct advantage of such an approach would be that it allows the capturing of drug effects despite the presence of individual differences of no interest.

Parameters were entered in their native format in the ANOVA reported in the submitted manuscript. We agree that the direction of the effect matters. After flipping the sign of Se_m, the interaction between variable types (Se_m vs. other variables) and groups (Placebo vs. Escitalopram) becomes:

Se_m:A_i p=0.295

Se_m:Se_d p<0.001

Se_m:Sr_m p=0.68

Se_m:Sr_i p=0.27

However, flipping signs is rather post-hoc and motivated by seeing the actual results. Since this approach is questionable, we adopted, instead, a Bayesian model selection (BMS) approach. Indeed, as suggested by the reviewers, BMS may be more suited to address the issue of specificity, irrespective of the direction of changes induced by the treatment. We therefore replaced the paragraph with the ANOVA result by a model comparison.

In short, this comparison shows that there is a specific modulation of the latent parameters of cost-evidence accumulation by the SSRI treatment. Among single-modulation models, a modulation of Se_m outperforms modulation of any other parameter. Among double-modulation models, modulations of Se_m and Se_d outperform any other combination. These best models are also much better than a null model (without any modulation).

We replaced the paragraph with the ANOVA results by the following paragraph:

“To further test the specificity of SSRI effect on cost-evidence accumulation, we performed another Bayesian model selection that contrasted the two groups of subjects. [...] A version with modulations of both Se_m and Se_d was slightly more likely than a version with only a modulation of Se_m (Δ=3) and much more likely than any other combination (Δ>5.9).”

We added the Table 3 to the revised manuscript.

We also added the following paragraph in the Methods section:

“Specificity of the treatment effect

We also used Bayesian model comparison to test the specificity of treatment effect on computational parameters. […] All models were fitted with the variational Bayesian procedure by Daunizeau et al. 2014. Values of log model evidence are reported in Table 3 for model comparison.”

Regarding whether model parameters are independent from one another, the equation of the model suggests that they are not. For instance, the sensitivity of effort duration to incentives is controlled by the parameter A_i and scaled by Se_m. However, the model is identifiable: for a given behavior, there is only one best-fitting set of parameter values. The critical issue for our pharmacological study is whether our model and fitting procedure can detect specific changes induced by the treatment. We quantified the specificity and sensitivity of our method with simulations (Figure 2C). We unpack this aspect further and the related paragraph now appears as:

“We checked the sensitivity and specificity of our model fitting procedure in detecting a treatment effect through simulations. A 20% change in any given parameter was reliably detected and the difference recovered by model fitting was significant in a 96% of simulations at least for A_i, Se_m, Sr_m and Sr_i and in a 77% of simulations for Se_d. Importantly, a change in one single parameter was recovered without propagating to other parameters and the false alarm rate was below 5% for all parameters (Figure 2C).”

C) Is any background 'reference' neuropsychological data available to confirm that there were no group differences of no interest? If so, please report.

Indeed, participants were scored on psychometric scales. We included the following paragraph in the Methods:

“We also used psychological tests to assess differences between groups of subjects at baseline, before treatment. […] Therefore, psychological variables were not significantly different at baseline when correcting for multiple comparisons (at p=0.05/9) and the one passing the uncorrected threshold p=0.05 was not significantly different during the testing phase.”

2) There were questions about the interpretation of the effects in terms of the role of time in task performance. The main effects coming out of the modeling analysis appear to be related to effort duration. But longer effort durations are also longer overall durations, which brings up the literature on 5-HT and delay costs (some of the delay discounting papers with 5-HT manipulations are actually cited in the manuscript). The authors appear to favor the view that different types of costs can be reduced to one process that is related overall to multiple types of costs. There is evidence both for and against this, and the authors are of course free to interpret their results in this way. In fact, this discussion contributes to the field in an interesting way. However, how much of the effect they are seeing is related to time per se? The authors should discuss this.

We agree that time is an essential dimension in our task. However, the cost is about effort, not delay. This is because prolonging effort duration does not postpone reward delivery. Although the effect of SSRI accords well with the general idea that serotonin helps with overcoming cost, it cannot be explained by a change in temporal discounting. We intended to clarify our interpretation by adding the following paragraph to the Discussion:

“It is important to note that the SSRI effects obtained here relate to effort cost and not to time discounting. Although time is central in our task, it departs from delay and waiting paradigms in several respects. [...]Therefore, the SSRI effect on effort duration (how long subjects sustain an effort) cannot be explained by a change in temporal discounting (how long subjects wait for a reward).”

3) The Introduction might benefit from mentioning the implication of 5HT not just in apathy, but also in the other end of the motivation continuum, that is impulsivity. And in the Discussion, how do the findings relate to observations that 5HT is key for impulse control? More generally, the paper might come full circle, by linking, in the Discussion section, the obtained results to these various clinical hypotheses (involvement in apathy, impulsivity).

We agree that impulsivity is a core concept in the literature on serotonin function. The Introduction implicitly referred to impulse control when quoting effects on behavioral inhibition and patience for delayed reward. We have now included this keyword in the text.

In the Discussion, we also integrated a reference to impulse control in the paragraph mentioning the effects of serotonin in waiting tasks. We also included a new paragraph to link our hypothesis and results to apathy and impulsivity:

“A general role for serotonin in cost processing also seems compatible with the effects on apathy and impulsivity reported in healthy and pathological conditions (Cools et al. 2005; Papakostas et al. 2006; Schweighofer et al. 2008; Crockett et al. 2009; Weber et al. 2009; Warden et al. 2012; Guitart-Masip et al. 2014; Miyazaki et al. 2014; Worbe et al. 2014). […] As SSRIs might also reduce the perceived cost of delaying action, they would improve response control and reduce impulsivity (Cools et al. 2005, 2011; Crockett et al. 2009; Dayan 2012; Bari and Robbins 2013; Miyazaki et al. 2014; Fonseca et al. 2015).”

[Editors' note: further revisions were requested prior to acceptance, as described below.]

The manuscript has been improved but there are some remaining issues that need to be addressed before acceptance, as outlined by the reviewer:

Reviewer #4:

I have reviewed this paper from a statistical perspective as this is a registered trial.

This manuscript presents the analysis of a single randomised placebo controlled and double blinded study. It is not a trial of an intervention but rather a controlled study to understand the mechanism of action that the specific drug takes. The trial has been registered with a target sample size of 128 and it has a single hypothesis statement that aims to 'assess the effects of' agolmelatine and escitalopram on emotional binding, emotional processing and motivation'. Each participant is measured over several tasks during the treatment period of 9 weeks.

I realise that it is not in the eLife style to report numerical estimates in the Abstract but as this report analyses a single controlled experiment I think they should put the strength of the evidence in the Abstract to back up their statement that 'they show that serotonin also regulates other types of action costs such as effort'. As there is no replication they should report this in a more guarded manner.

We let the eLife editorial board decide whether this change is compatible with their style.

The authors have completed the Consort statement but they need to provide more complete information within the manuscript so they can say yes to more of the checklist.

We revised several aspects of the manuscript (see details below), which allowed us to check the following items of the Consort list: 7a, 8a, 9, 10, 14.

We also report the dates of the trial (item 14) in the section “Drug and Testing Schedule”: “Data were collected between January 2012 and July 2013.”

The trial registration lists a compound aim/hypothesis and also lists three primary outcomes listed which I think have all been used in this analysis but the main question for the manuscript is 'does serotonin regulate the weight of effort cost as opposed to the weight of expected benefit?'

I would like to see a clearer link between the main research question and the forms of analysis. A sample size of 58 is not that large even with repeated measurement. Was there really any power to test for interactions? The Consort checklist says 'no' to sample size but surely the size was justified originally? I feel a bit concerned that sometimes Bonferonni correction is used to conclude there were no differences when this just reduces power.

The trial registration lists three points. Point #1 and #2 are about tests on emotions performed by other research groups, not reported here. Point 3 lists goals for three “motivation” tasks, one being the Effort Allocation Task presented in this article, and the two others being probabilistic learning tasks that involved other researchers. Trial registration forms are highly stereotyped, and do not really fit the kind of exploratory trial that we participated in, with many tasks from various researchers. Therefore, our study addresses a much more specific question than the entire trial.

The sample size was proposed by the coordinators of the clinical trial, without formal calculation. This number seemed reasonable given typical studies in the field of cognitive neuroscience resorting to drugs in healthy subjects.

We added the following sentences in the Methods: “Given that the clinical trial was exploratory and also included tests from other research groups, the sample size was not selected specifically for our study; however, it appeared reasonable given typical studies in the field. Indeed, published between-subject comparisons of placebo and anti-depressant treatments were made on a lower sample size per group (e.g. N=14 in Harmer et al. 2014, N=20 in Chamberlain et al. 2006 and N=30 in Guitart-Masip et al. 2014), and a single visit per subject (while we have three)”.

We modified the Consort checklist, checking the corresponding item (7a) and referring to the relevant page of the manuscript.

Regarding the Bonferonni correction, we believe that one must correct for multiple comparisons to take into account the unwanted inflation of type I error. To this end, we used a Bonferonni correction, which is the most conservative one. The reviewer suggests that a Bonferonni correction may be used to conveniently discard marginally significant effects of little interest. However, note that it imposes an equally stringent test for the effects of interest. In any case, we treated all effects equally, by reporting non-corrected significance levels, and when there were multiple comparisons, by highlighting which significance levels passed the Bonferonni-corrected threshold. We believe it is both transparent and statistically correct.

We nevertheless agree with the reviewer that a lack of power is an issue for marginally or non-significant results. We added the following sentence in the Discussion section, regarding the interaction with time that we could not detect: “… our study may lack the statistical power to reveal such an effect”.

Author details

1. Florent Meyniel

   1. Motivation, Brain and Behavior team, Institut du Cerveau et de la Moelle épinière, Groupe Hospitalier Pitié-Salpêtrière, Paris, France
   2. INSERM UMRS 1127, CNRS UMR 7225, Université Pierre et Marie Curie (UPMC-P6), Paris, France
   3. Cognitive Neuroimaging Unit, NeuroSpin Center, INSERM U992, Institut d'Imagerie Biomédicale, Direction de la recherche fondamentale, Commissariat à l'énergie atomique et aux énergies alternatives, Gif-sur-Yvette, France

   Contribution

   FM, Conception and design, Analysis and interpretation of data, Drafting or revising the article

   For correspondence

   florent.meyniel@gmail.com

   Competing interests

   FM: Supported by The French Ministère de la Recherche and the European Union Seventh Framework Programme (FP7/2007-2013) under grant agreement no. 604102 (Human Brain Project)

   "This ORCID iD identifies the author of this article:" 0000-0002-6992-678X

2. Guy M Goodwin

   1. Department of Psychiatry, University of Oxford, Oxford, United Kingdom
   2. Warneford Hospital, University of Oxford, Oxford, United Kingdom

   Contribution

   GMG, Conception and design, Drafting or revising the article

   Competing interests

   GMG: Holds shares in P1vital Ltd and has served as consultant, advisor or CME speaker in the last 12 months for AstraZeneca, Cephalon/Teva, Convergence, Eli Lilly, GSK, Lundbeck, Medscape, Otsuka, Servier, Sunovion, Takeda

3. JF William Deakin

   Neuroscience and Psychiatry Unit, The University of Manchester, Manchester, United Kingdom

   Contribution

   JFWD, Conception and design, Drafting or revising the article

   Competing interests

   JFWD: Currently advises or carries out research funded by Autifony, Sunovion, Lundbeck, AstraZeneca and Servier; all payment is to the University of Manchester; he has share options in P1vital Ltd

   "This ORCID iD identifies the author of this article:" 0000-0002-2750-962X

4. Corinna Klinge

   1. Department of Psychiatry, University of Oxford, Oxford, United Kingdom
   2. Warneford Hospital, University of Oxford, Oxford, United Kingdom

   Contribution

   CK, Collected all data together with CMF from the Oxford center, Acquisition of data

   Competing interests

   No competing interests declared.

5. Christine MacFadyen

   1. Department of Psychiatry, University of Oxford, Oxford, United Kingdom
   2. Warneford Hospital, University of Oxford, Oxford, United Kingdom

   Contribution

   CM, Collected all data together with CK from the Oxford center, Acquisition of data

   Competing interests

   No competing interests declared.

6. Holly Milligan

   Neuroscience and Psychiatry Unit, The University of Manchester, Manchester, United Kingdom

   Contribution

   HM, Collected all data together with EM from the Manchester center, Acquisition of data

   Competing interests

   No competing interests declared.

7. Emma Mullings

   Neuroscience and Psychiatry Unit, The University of Manchester, Manchester, United Kingdom

   Contribution

   EM, Collected all data together with EM from the Manchester center, Acquisition of data

   Competing interests

   No competing interests declared.

8. Mathias Pessiglione

   1. Motivation, Brain and Behavior team, Institut du Cerveau et de la Moelle épinière, Groupe Hospitalier Pitié-Salpêtrière, Paris, France
   2. INSERM UMRS 1127, CNRS UMR 7225, Université Pierre et Marie Curie (UPMC-P6), Paris, France

   Contribution

   MP, Interpretation of the result, Conception and design, Drafting or revising the article

   Contributed equally with

   Raphaël Gaillard

   Competing interests

   No competing interests declared.

9. Raphaël Gaillard

   1. Centre Hospitalier Sainte-Anne, Service de Psychiatrie, Paris, France
   2. Faculté de Médecine Paris Descartes, Université Paris Descartes, Sorbonne Paris Cité, Paris, France
   3. Laboratoire de Physiopathologie des maladies Psychiatriques, Centre de Psychiatrie et Neurosciences, INSERM U894, Université Paris Descartes, Sorbonne Paris Cité, Paris, France
   4. Human Histopathology and Animal Models, Department of Infection and Epidemiology, Institut Pasteur, Paris, France

   Contribution

   RG, Interpretation of the results, Conception and design, Drafting or revising the article

   Contributed equally with

   Mathias Pessiglione

   Competing interests

   RG: Received compensation as a member of the scientific advisory board of Janssen, Lundbeck, Roche, Takeda; he has served as consultant and/or speaker for Astra Zeneca, Pierre Fabre, Lilly, Otsuka, SANOFI, Servier and received compensation, and he has received research support from Servier

• 5,878

  Page views

• 707

  Downloads

• 50

  Citations

Article citation count generated by polling the highest count across the following sources: Scopus, Crossref, PubMed Central.