The role of the vestibular system in value attribution to positive and negative reinforcers

Somatic inputs originating from bioregulatory processes can guide cognition and behavior. One such bodily signal, mostly overlooked so far, is represented by visuo-vestibular coupling and its alteration, which in extreme cases may result in motion sickness. We argued that the inherently perturbed interoceptive state that follows can be a powerful determinant of human motivated behavior, resulting in a blunted response to appetitive stimuli and an exaggerated response to noxious ones. We sought to assess such differential impact of visuo-vestibular mismatches on value through a task involving conflict monitoring. We therefore administered to 42 healthy participants a modified version of the Flankers task, in which distractors (arrows, pointing in either a congruent or incongruent direction) signaled the availability of monetary incentives (gains, losses, or neutral trials). While performing the task, participants received either galvanic vestibular stimulation (GVS), or sham stimulation. We have found impaired behavioral performances when value, which was attached to task-irrelevant information, was at stake. Gains and losses, interestingly, dissociated, and only the latter caused enhanced interference costs in the task, suggesting that negative incentives may be more effective in capturing human attention than positive ones. Finally, we have found some weak evidence for GVS to further increase the processing of losses, as suggested by even larger interference costs in this condition. Results were, however, overall ambiguous, and suggest that much more research is needed to better understand the link between the vestibular system and motivation. Highlights Visuo-Vestibular mismatches may be important somatic markers affecting the evaluation of reinforcers; When attached to distractors, value information impairs behavioral performance for the task at hand; Trials in which potential losses were at stake were associated with larger interference costs arising from conflicting information between the target and the flankers; GVS (Right-Anodal) may further increase the interference caused by losses, but the evidence in this respect was ambiguous and inconclusive;


Interoception and motivation
Human decision-making processes can hardly be understood in full without taking into account one individual's physiological state and needs (Berridge & Robinson, 2003;Craig, 2002;Critchley et al., 2004;Damasio, 1996;Gray & Critchley, 2007;Namkung et al., 2017;Naqvi et al., 2007;Naqvi & Bechara, 2010;Paulus, 2007). "Somatic markers" originating from bioregulatory processes can bias the subjective desirability of stimuli, thus guiding cognition and behavior (Craig, 2002;Damasio, 1996;Mayer, 2011;Morton et al., 2006;Namkung et al., 2017;Paulus, 2007;Seth, 2013). For example, think how desirable would be one's preferred food in normal conditions versus while experiencing mild nausea. A global interoceptive state is defined by the balance and integration of visceral, autonomic, somatosensory, motor and vestibular inputs, wherein the insula is known to represent a key structure in the integration of these signals (Craig, 2002;Namkung et al., 2017). The incentive value of any stimulus can be encoded in an abstract fashion, but also in relation to the expected effect on physiological homeostasis (Berridge & Robinson, 2003;Palminteri et al., 2012), before being exploited for guiding human choices via connections with other structures such as limbic areas and the dorsal striatum (Berridge & Robinson, 2003;Namkung et al., 2017;Palminteri et al., 2012). This double coding of one stimulus (i.e. abstract vs. homeostasis-related) is therefore capable to explain why the same stimulus can assume different motivational values under different interoceptive states.

Figure 2: A)
The nominal power for our design is depicted as a function of a range of a priori sample and effect sizes (partial eta squared). We planned to enroll a maximum of 42 participants, which allowed us to have high power for effect sizes in the small range (distributed around that reported in Blini et al., 2018a). However, interim analyses were planned at N= 30, and thus corrections for sequential analyses were applied. B) Graphical representation and time course of the modified Flankers Task (FT). Each trial started with a fixation cross appearing at the center of the screen. Then, a set of five arrows was presented on screen. The task consisted in indicating the direction of the central arrow (left, right). The four flanking arrows could point towards the same (congruent condition) or opposite direction 1 3 GVS conditions, inducing different (polarity dependent) effects. Electrodes were placed over the mastoid processes symmetrically (that is, in the Left-Anodal montage the cathode was placed over the right mastoid bone, and vice versa for the Right-Anodal montage). Left-Anodal stimulation activates mainly right hemisphere structures (Lopez et al., 2012;zu Eulenburg et al., 2012), whereas Right-Anodal activates comparatively more left hemisphere structures. A sham condition was also included, with electrodes placed symmetrically about 5 cm below the mastoids, above the neck, and distant from the trapezoidal muscles yielding proprioceptive signals (Lenggenhager et al., 2007). The sham condition was included to control for unspecific factors of electrical stimulation (e.g., arousal, discomfort). The anode was placed in this case on the left side (Ferrè et al., 2013). Participants performed the behavioral tasks three times, on three different days, under each GVS condition (Left-Anodal, Right-Anodal and sham). The order of GVS type administration was counterbalanced across subjects. Within each session, active stimulations were delivered for a maximum of 30 minutes in order to minimize side effects.

Apparatus and Behavioral Tasks
Participants were tested in a dimly lit, quiet room. Their head was restrained by a chinrest, facing a 17 inches large screen at a distance of approximately 57 cm. The open-source software OpenSesame (Mathôt et al., 2011) was used to display experimental stimuli on the screen and record the subjects' response. Participants provided responses by means of keyboard presses (on a standard QWERTY keyboard) using the index and middle fingers of their dominant hand. During the stimulation, they performed two tasks. One control task (Subjective Visual Vertical, SVV) was administered during the first 5 minutes of the stimulation. SVV required rotating visual segments until they appeared to be in a vertical position. It was meant to provide independent evidence for a successful vestibular stimulation, given that displacements occur towards the site of anodal GVS stimulation (Mars et al., 2001;Saj et al., 2006). Then, a modified version of the Flankers Task (FT) (Eriksen & Eriksen, 1974;Eriksen, 1995) probed interference costs arising from conflict and their modulation by potential rewards or losses. Practice trials for both tasks were administered at the beginning of each session, before the onset of the stimulation, and were not considered in the analyses. One brief evaluation of subjective feelings and sensations experienced during GVS was administered soon after the stimulation end (see Supplementary Materials). This was meant to monitor participants' distress and task compliance across the different days of the experiment and GVS protocols. study, we did not consider fatal the lack of this quality check, but only in the presence of a meaningful modulation of motivational assets by GVS. For example, Scenario 1 in Figure 2C predicts large interference costs for the sham condition in face of smaller costs with concurrent GVS: in this case it would be conceivable that the two-way interaction Congruency by Condition (thus collapsed across GVS levels) may fail to reach our significance criterion, precisely because moderated by GVS; we would therefore observe only a three-way interaction, suggesting that motivational assets are indeed deployed, though modulated by GVS.

Test
Invalidate results? Outcome The vestibular system is effectively perturbed SVV task: the main effect of GVS or the GVS by Starting Side interaction is significant. At least one contrast indicates the following pattern: Left-Anodal < sham < Right-Anodal

Figure 3: A) Subjective Visual
Participants also reported, after Left-Anodal GVS, having perceived their body moving forward in space. Albeit this is compatible with a vestibular stimulation, the direction of this illusion is unusual for a bilateral montage. Displacements along the anterior-posterior axis may be expected for antero-posterior montages (Aedo-Jury, Cottereau, Celebrini, & Severac Cauquil, 2019), rather than for lateral ones. This illusion may have been genuinely prompted by the (mixed) vestibular signals that participants experienced, or could simply reflect difficulties in accurately discriminating or recall (after the stimulation) the sensations experienced. That said, all other items indicated that GVS was not different from sham when assessing potential confounds that may follow electric brain stimulations, such as overall discomfort or painful sensations. This is important because it allowed us to circumscribe the experimental effects more precisely to a vestibular perturbation. Finally, we have found, in the FT task, all the requisites for us to assess a meaningful modulation by GVS: conflict costs, in terms of impaired responses in incongruent trials (Congruency effect), and motivational effects. The latter ones, interestingly, caused an impaired performance when value was at stake (less accurate responses). This is coherent with the view that distractors may gain in salience when value is attached to them, and hamper the performance to the relevant task.
However, PL and PG conditions seem to dissociate when assessing response times: PG trials were overall faster, suggesting a motivational performance boost; PL trials, on the other hand, were not faster than neutral trials, but they interacted instead with Congruency such that the observed conflict costs were enhanced (i.e. responses were faster in congruent trials but slower in incongruent trials, Figure 3C). This accurately reflects the pattern depicted in Figure 1C as an increase, with respect to neutral trials, of interference costs in PL (punishment) conditions; with respect to our a priori predictions, however, we have found no evidence of such increase in PG (rewarded) conditions.

The vestibular system, conflict, and motivation.
This section reports the results of the main tests of interest, namely the three-way interaction Congruency by Condition by GVS for both accuracy ( Figure 4A) and RTs ( Figure 4B).
Descriptive statistics are reported in Table 2.
When assessing accuracy, the interaction was not significant according to our pre-defined criteria (χ 2 (4) = 9.62, p fdr = 0.063, Besides the tests reported under "Outcome-neutral quality checks", no other effect or interaction was proven significant when assessing either accuracy or RTs (all p fdr > 0.14).

Figure 4:
Results of the main tests of interest, namely the three-way interaction Congruency by Condition by GVS for both accuracy (A)) and RTs (B)). The leftmost plots of each panel show interference costs for each GVS and Reward condition, depicted as in Figure 1C. In the rightmost plots of each panel, instead, we use interference costs in Neutral trials (dashed gray line) as a baseline to assess how different Reward conditions and GVS can modulate them.
We have found weak evidence that: Right-Anodal GVS increases, when assessing accuracy, the interference costs associated with PL trials; Left-Anodal GVS increases, when assessing RTs, the interference costs associated with PL trials (with respect to Right-Anodal GVS).
Results should, however, be taken with caution because: effects sizes were very small (about half the expected size); a range of exploratory robustness checks (presented in section 3.3) failed to fully corroborate the results from this main, pre-registered analysis. Error bars depict within-subjects standard errors of the mean (Morey, 2008). . We therefore proceeded with normalizing the data to a more stable baseline: for each subject, we first computed the interference costs (the difference between congruent and incongruent trials) for each GVS and Condition, and then referenced these costs to the Neutral condition (situation depicted in the rightmost plots of Figure 4). Thus, larger values in this context indicate larger enhancement of interference costs when value was at stake, with respect to a Neutral baseline. We probed the same post-hoc contrasts which resulted significant in the preregistered approach, which was based on mixed models, through paired ttests on these values. When assessing accuracy, we have found that Right-Anodal GVS yielded, in PL trials, larger costs than both sham (t (41) = 2.32, p= 0.026) and Left-Anodal GVS (t (41) = 2.09, p= 0.043), though only at the uncorrected significance level. When assessing RTs, instead, the difference between Right-Anodal and Left-Anodal GVS when potential losses were at stake was not significant (t (41) = 1.73, p= 0.091).

Exploratory analyses
Drift Diffusion Models. Another way to link accuracy rate and response time via one theorygrounded framework is through Drift Diffusion Models (DDM, Ratcliff, 1978). DDM assume that information is sampled continuously from the environment and, in the case of a two forced-choices task, one response is produced when a critical threshold is reached. Thus, at least two parameters are of interest: the drift rate, indexing the direction and speed of sensory accumulation, and the boundary separation parameter, indexing the amount of information that is necessary in order to produce a response. The first one has been classically associated with task difficulty -e.g. the amount of cognitive load, which hampers the speed by which information is gathered -whereas the second is thought to refer to speed-accuracy tradeoffs and impulsivity -larger boundary separation suggesting a more conservative threshold and response style. DDM approaches for the Flankers task have been discussed before ( when value was at stake, potential gain trials being associated with the most liberal response bias.

Figure 5:
We fitted a Drift Diffusion Model (DDM) to the data, and recovered the drift rate parameter, indexing the speed at which sensory information is gathered from the environment.
Such speed was slower for incongruent trials and when value was at stake, showing that irrelevant information was indeed causing distraction and hampering the performance to the relevant task. These interference costs were larger for PL trials, suggesting that potential losses were more effective in capturing attention. We have additionally found some evidence for this effect to be more pronounced with Right-Anodal GVS. The leftmost plot shows interference costs for each GVS and Reward condition, depicted as in Figure 1C. In the rightmost plot, instead, we use interference costs in Neutral trials (dashed gray line) as a baseline to assess how different Reward conditions and GVS can modulate them. Error bars depict within-subjects standard errors of the mean (Morey, 2008).

Summary of results
All the quality controls planned for the study were fulfilled, enabling us to interpret the results from the FT task across GVS conditions. One first result was that PL and PG trials dissociated. While both had detrimental effects on behavioral performances, as assessed by accuracy, only PL trials led to enhanced interference costs when assessing RTs. Thus, value can have detrimental effects on performance, when it is attached to task-irrelevant information, possibly as result of increased attentional salience of the distractors. Potential losses, in this respect, seem to be more effective than potential gains, suggesting some prioritization. The main objective of this study, however, was to assess whether this interaction was further modulated by GVS. We have found some evidence in this regard: Right-Anodal marginally increased interference costs for PL trials when assessing accuracy, whereas Left-Anodal GVS was seemingly more effective when assessing response times. It must be said, however, that several post-hoc robustness checks could not corroborate the results of the main analyses, the p-values being close (~0.1) to significance but not comparable to LMEM results. Importantly, the effect size for the three-way interaction was rather small (η p 2 = ~0.05), half of what hypothesized for the a priori power analysis. Post-hoc power for our design equals to only 50% for such small effects (given our set of p-value corrections While this pattern roughly matches our predictions of an increased saliency -and therefore interference costs -for PL trials during visuo-vestibular mismatches, we found no evidence for a reduction of such costs for PG trials, which was also expected.

Motivation, valence and conflict
Motivation -experimentally provided, for example, in the form of monetary rewards -can act by invigorating one's action or willingness to exert an effort (Chong, Bonnelle, & Husain, 2016;Chong et al., 2015;Muhammed et al., 2016), or by sharpening cognitive processes such 0 as attention and memory (Abe et al., 2011;Anderson et al., 2011;Chelazzi et al., 2013;Della Libera & Chelazzi, 2006, 2009Engelmann & Pessoa, 2007;Hickey et al., 2010 In the present study, we report data showing both facets of motivation. The invigorating component was exemplified by overall faster response times when potential gains were at stake. The detrimental component, on the other hand, had a larger share, and we observed both decreased accuracy when gains and losses were at stake and a peculiar increase in interference costs specifically for potential losses. While the bulk of brain circuits coding for rewards and punishments appears to be highly overlapping, signals associated with gains or losses are also differentially represented in the brain (Camara et al., 2009;Liu et al., 2011;Yacubian et al., 2006). Specialised brain areas originate a signal coding for relative reward magnitudes, which is thought to be exploited for in enhanced interference costs, thus, may indicate that the reported overlap is not merely evaluative in its functions. Rather, potential losses may consolidate and magnify neural signals related to conflict, and thus the associated behavioral signatures. There are at least two possible mechanisms that may frame this possibility. First, potential losses may be more effective than potential gains in capturing attention: the processing of task-irrelevant features (the flankers' direction) would also be enhanced as well as a consequence, causing a more pronounced interference. Second, one additional source of conflict may arise between an intrinsic tendency to associate losses with avoidance and the need to provide a response as 1 fast as possible (Carsten et al., 2019); gains, evoking approach tendencies instead, would not be associated with such additional mismatch. We tend, however, to favor the first option as it best accommodates our finding that incongruent trials were, indeed, slower, whereas congruent ones were slightly faster than the respective neutral condition. An unspecific source of conflict and avoidance prompted by losses should, instead, equally affect congruent and incongruent trials, causing response times to be overall slower. Drift diffusion models have shown, indeed, that responses were generally less impulsive when potential losses were at stake with respect to potential gains, which is compatible with this view. However, potential losses were associated, on the contrary, to more impulsive responses with respect to neutral trials, and thus impulsivity alone is unlikely to explain the increased interference costs only observed for losses. Marginally faster response times for congruent trials, instead, are compatible with the attentional account because the enhanced processing of the distractors' direction would, in this specific condition, hamper comparatively less the response to the primary task. In other words, part of the sensory information that is gathered from the distractors could, in a congruent condition, be more easily "recycled" and transferred to the task-relevant response. Indeed, this interaction between losses and congruency is best captured by the drift rate parameter, indexing the speed at which sensory information is gathered from the environment.
The notion that potential losses may be more effective than potential gains in driving contrast with potential gains, aversive motivation was associated with impaired performance in the task which, like in the present study, was counterproductive for participants as it resulted in increased occurrence of electric shocks; in addition, interference costs were increased for incongruent trials, though in a manner qualitatively similar to that observed in potential gain blocks (Liao et al., 2020).
All in all, thus, our results seem to corroborate the notion that potential losses may drive decision-making more powerfully than potential gains, the inconsistencies found in literature being explained by a different sensitivity of the behavioral tasks or by methodological constraints (e.g., intermixed presentation of PG and PL trials versus in separate blocks).

The vestibular system, value, and conflict
The role of the vestibular system in modulating motivational and monitoring resources remains, at state, elusive. The present study could only provide weak and ambiguous evidence for GVS to further qualify the interaction between value and conflict. If we were to rigidly apply our pre-registered criteria in evaluating this three-way interaction, we should conclude that Left-Anodal GVS may increase interference costs for potential losses (when assessing response times). It would be therefore tempting to discuss results in terms of hemispheric however, any conclusion would be premature as a number of post-hoc analyses failed to support this finding. In addition, in our previous study, using a spatial cueing paradigm (Blini et al., 2018a), Left-Anodal GVS showed an effect in the same direction as that of Right-Anodal GVS, and thus decreased sensitivity to rewards as well, though to a lesser extent.
Finally, in the present study we have also found hints suggesting that Right-Anodal GVS may enhance interference costs for potential losses. While robustness checks were, also in this case, far from providing a clear cut answer, this latter result was seemingly more robust internally. First, LMEM and robustness checks were roughly comparable, though the latter yielded post hoc contrasts that were only significant at the uncorrected level. Second, drift diffusion models, and the drift rate parameter specifically, could highlight a specific effect for Right-Anodal, but not Left-Anodal GVS, in the processing of losses. This pattern roughly matched our predictions (depicted in Scenario 3, Figure 2C) of an increased saliency -and therefore interference costs -for PL trials during visuo-vestibular mismatches, although the effect was much smaller than what foreseen; on the other hand, we have found no evidence for a reduction of such costs for PG trials, which was also expected across all a priori scenarios.
It should be noted that, unlike losses, gains did not yield an increase of interference costs; one possibility, thus, is that this supposed interaction is not something that could have been meaningfully modulated in first place. Rather, one may wonder why GVS did not modulate the response time advantage for gains, which we discussed above as reflecting an increase in response vigor due to the promise of rewards. One first possibility is that our previous study may simply report a false positive finding. The study was pre-registered as well: this process seems to work as intended in increasing the likelihood for null findings to be published (Allen & Mehler, 2019), and therefore mitigate publication bias, but it is not by itself a miracle cure for false positives, which are intrinsic to inferential statistics based on error rate probabilities.
That said, the response times advantage reported in the present study is much more subtle than the one we previously described, obtained through a different task (a spatial cueing paradigm, Blini et al., 2018a). Furthermore, the Flankers task adopted in this study was explicitly meant to probe an instance in which value actually impairs behavioral performance; indeed, in the FT, gains were also decreasing performances in terms of discrimination accuracy. One could therefore simply argue that the effect of potential gains was, in the current context, too weak and ambiguous to be effectively modulated.

Limitations of the present study
In the present study, value information was attached to distractors, which were presented on screen simultaneously to targets. Participants' responses occurred, on average, around 340-370 ms post-stimulus. This is a very short time window, which probably only allowed us to frame effects that are more likely to be related to very low-level, perceptual differences induced by the processing of value. Having observed such differences, for example in the processing of gains vs losses, in such a limiting setting is remarkable, and indicates that value can affect (and hamper) human performance very quickly, and starting with very basic sensory processes. Indeed, our findings were best captured by a difference in the speed of accumulation of evidence (the drift rate parameter), which is thought to reflect precisely sensory processing of the environment. On the other hand, we may have missed effects originating from late processes or components. In our previous study (Blini, Tilikete, et al., 2018a), the spatial cue signaling reward preceded the target of about 700 ms. More high-level (e.g. strategical) effects of reinforcers, in addition to perceptual ones, may be unveiled by allowing a more thorough processing of value. This, in turn, may increase the likelihood to induce and observe modulations of value-specific effects, for example by GVS.
Another aspect worth of consideration is the choice of the time threshold imposed to participants. One such threshold is important as it helps to highlight effects of value that are only present when the outcome of a trial is not certain, and the reinforcers are more likely to be obtained if comparatively more effort is deployed for the response. Differently from our previous approach (Blini et al., 2018a), this threshold was not fixed (e.g. to 500 ms), but adaptive, and it was meant to provide similar levels of task difficulty across all subjects.
Participants were continuously pushed by the algorithm to produce responses faster than 75% of their overall responses. This context was overall stricter, and, in hindsight, may have caused participants to reach the limits of their performances (i.e. floor effect) sooner during each session. In addition, this could have introduced some powerful extrinsic (though orthogonal to all GVS and Conditions) source of motivation, of such a large magnitude to partly conceal that induced by reinforcers.
Another remark worth mentioning is that we used, for the present task, fixed color-value associations. Considering that the task did not involve statistical learning (i.e., color-value associations were explicit), we sought to minimize any potential conflict arising at the semantic level, as to avoid conditions of uneven strength of the associations. As the mesolimbic reward system may be differentially tuned to different color wavelength (Hu et al., 2020), however, this may have introduced perceptual biases effectively defeating the purpose.
Last but not least, here we choose to present potential loss and potential gain trials in an intermixed, balanced fashion. Compared to their administration in separate blocks (e.g., Liao et al., 2020), this may have prompted a competition, more or less explicit, between the two valences, resulting in one being favored and prioritized over the other, perhaps depending on individual biases and personality traits. Measuring and accounting for such individual susceptibilities may help clarify this issue in future studies.

Conclusions
In this paper, we report an asymmetry in the processing of gains and losses, with the latter seemingly more effective in capturing human attention. When negative reinforcers are attached to distractors, these stimuli appear to be processed more thoroughly, and the interference costs arising in presence of conflicting information are consequently enhanced.
We also report some evidence for Right-Anodal galvanic vestibular stimulation to further increase the salience of losses, as suggested by even larger interference costs in this condition.
This effect appears, within the present study, very weak, and thus of dubious relevance, (ANR-14-CE13-0005-1). The study was performed within the framework of the LABEX CORTEX (ANR-11-LABX-0042) of the University of Lyon within the program "Investissements d'Avenir" (ANR-11-IDEX-0007) operated by the ANR. Funders have no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.   H  i  c  k  e  y  ,  C  .  ,  C  h  e  l  a  z  z  i  ,  L  .  ,  &  T  h  e  e  u  w  e  s  ,  J  .  (  2  0  1  0  )  .  R  e  w  a  r  d  C  h  a  n  g  e  s  S  a  l  i  e  n  c  e  i  n  H  u  m  a  n  V  i  s  i  o  n  v  i  a  t  h  e   A  n  t  e  r  i  o  r  C  i  n  g  u  l  a  t  e  .  T  h  e  J  o  u  r  n  a  l  o  f  N  e  u  r  o  s  c  i  e  n  c  e  ,  3  0  (  3  3  ) , 1 1 0 9  a  n  g  u  a  g  e  a  n  d  e  n  v  i  r  o  n  m  e  n  t  f  o  r  s  t  a  t  i  s  t  i  c  a  l  c  o  m  p  u  t  i  n  g  .  R  F  o  u  n  d  a  t  i  o  n  f  o  r   S  t  a  t  i  s  t  i  c  a  l  C  o  m  p  u  t  i  n  g  ,  V  i  e  n  n  a  ,  A  u  s  t  r  i  a  .  h  t  t  p  s  :  /  /  w  w  w  .  r  -p  r  o  j  e  c  t  .  o  r  g  /   T  h  e  e  u  w  e  s  ,  J  .  ,  &  B  e  l  o  p  o  l  s  k  y  ,  A  .  V  .  (  2  0  1  2  )  .  R  e  w  a  r  d  g  r  a  b  s  t  h  e  e  y  e  :  O  c  u  l  o  m  o  t  o  r  c  a  p  t  u  r  e  b  y  r  e  w  a  r  d  i  n