Task relevance and negative reward modulate the disengagement deficit of patients with spatial neglect

Though motivational value is a recognized trigger of approach and avoidance behavior, less is known about the potential of reward to capture attention. We here explored whether positive or negative reward modulates the characteristic deficit of patients with left spatial neglect to disengage attention from an ipsilesional distracter. We built our study on recent observations showing that the disengagement deficit is exaggerated for distracters with target-defining features, indicating that task-relevance captures attention. Patients with left neglect and matched healthy controls were asked to react to lateralized, colored targets preceded by a peripheral cue. Crucially, the cue either possessed the color of the target and was thus task-relevant, or was followed by a positive, negative, or neutral symbolic reward. Neglect patients only exhibited a disengagement deficit when cues were task-relevant or were followed by a negative reward. This finding indicates that attentional selection is driven by task-relevance and negative reward, possibly through interactions between limbic and attention networks.


Introduction
In order to adjust approach or avoidance behavior flexibly humans must perceive physical saliency or perceived relevance of surrounding information. The power of this mechanism is probably best illustrated by the fact that human observers fail to detect conspicuous and seemingly obvious environmental changes if they are not relevant to the current task (Folk et al., 1992;Gibson and Jiang, 1998;Simons, 2000). A bias in the assignment of priority to surrounding stimuli may underlie spatial neglect, a neuropsychological syndrome characterized by the inability to process or react to sensory stimuli appearing on one side of space (Bartolomeo, 2021). Neglect typically follows right temporo-parietal damage and is characterized by severe unawareness of stimuli presented in contralesional hemispace, despite relatively intact sensory function (Hillis, 2006;Karnath and Rorden, 2012;Mort et al., 2003;Pedrazzini and Ptak, 2020). Using a spatial cuing paradigm, Posner and co-workers (Posner et al., 1984) reported that neglect patients exhibit disproportionally slow RTs to contralesional targets preceded by ipsilesional cues, the so-called disengagement deficit. As later studies confirmed, the disengagement deficit is a stable marker of neglect and thus reflects a major functional component of the disorder (Bartolomeo et al., 2001). Since the disengagement deficit is greatest at short time intervals between cue and target some authors have proposed that neglect may be explained by impaired exogenous/automatic capture of attention, while more top-down or endogenous processes may be relatively spared (Bartolomeo et al., 2001;Bourgeois et al., 2012). While the processing of task-relevant information is crucial for the adaptation of behavior in many situations, humans are also highly sensitive to affective and motivational information, and thus prioritize stimuli associated with positive or negative outcomes (Bourgeois et al., 2016;Vuilleumier, 2005). Thus, several studies demonstrated that positive and negative reward may affect decisions of healthy participants and distract their attention from task-relevant stimuli (Anderson, 2016;Bourgeois et al., 2017;Bourgeois et al., 2016). Similarly, spatial deficits in neglect may be modulated by emotional (e.g., affective faces) or motivational (e.g., stimulus value) information (Bourgeois et al., 2018;Li et al., 2016;Lucas et al., 2013;Lucas and Vuilleumier, 2008;Malhotra et al., 2013). For example, Lucas et al. (2013) demonstrated that associating left-sided targets with a reward progressively biased visual exploration leftward, both in healthy participants and neglect patients. Bourgeois et al. (2018) observed that during visual search neutral stimuli previously associated with a positive reward induced a cost on attentional orienting towards left targets in neglect patients, even though those stimuli were no longer task-relevant. These studies indicate that reward may diminish neglect when paired with stimuli that patients search for, and increase neglect when paired with distracting information. Thus, assigning motivational valence to sensory stimuli may change the allocation of attention in healthy subjects (Anderson et al., 2011;Della Libera and Chelazzi, 2009) as well as in patients with a severe attentional disorder.
While these findings suggest that reward may bias attention in neglect, the cognitive mechanisms behind this effect are far from clear. The three studies reporting a modification of spatial neglect by rewarding stimuli used visual search tasks (Bourgeois et al., 2018;Lucas and Vuilleumier, 2008;Malhotra et al., 2013). Performance on these tasks is not only determined by spatial attention, but may also strongly be affected by adjustments of effort or alertnesswhich are themselves modifiable through reward. For this reason, it is unclear to what extent some previous results reflect slow, strategic or even volitional processes. Another problem is that studies with neglect patients only used positive reward to guide or modify attention; it is therefore unknown whether similar effects also apply to negative reward.
Previous studies have proposed a variety of cognitive mechanisms that could explain attentional selection, such as decreased perceptual noise (Lu and Dosher, 1998), increased target saliency (Arcizet et al., 2011) or increased neural gain (i.e., narrowing of the bandwidth of neural responses to a given stimulus; Hillyard et al., 1998). In addition to these pure 'bottom-up' models, several authors posit that spatial attention reflects dynamic adjustments within a spatiotopic priority map, which is located within dorsal fronto-parietal areas (Bourgeois et al., 2020;Chelazzi et al., 2014;Fecteau and Munoz, 2006;Ptak, 2012;Ptak and Fellrath, 2013). Priority can be stimulus-driven or influenced by motivational factors and behavioral goals of the observer. In this conception, reward would affect priority in a manner very similar to 'bottom-up' factors, such as stimulus contrast or abrupt onset, and 'top-down' factors such as task-relevance. Previous studies lend support to the priority model of neglect by showing that stimulus-driven (i.e., physical saliency) and goal-driven (task-relevance) processes affect the spatial bias of neglect patients comparably (Bays et al., 2010). Of particular interest is the finding that an ipsilesional (mostly right-sided) distracter captures attention of patients with spatial neglect if it shares a task-relevant characteristic with the target (Ptak and Pedrazzini, 2021;Ptak and Schnider, 2006). This finding was observed with a spatial cueing paradigm, in which cues were briefly shown in the left or right hemifield prior to appearance of a colored target. Though cues were unpredictive of target presence or position, they impeded the detection of contralesional targets when sharing their color, as patients were both slower and made more omissions in this condition. Importantly, their performance was unaffected when cues were of a different color than the target, i.e. were task-irrelevant. This finding suggests that sharing the target-defining characteristic increases the priority of the cue and thus results in attentional capture.
In sum, while previous reports observed a modulation of neglect symptoms by task-relevance or reward, no study has examined these factors within the same paradigm. We here employed a spatial cueing paradigm to study whether the disengagement deficit is modulated by cues that are either task-relevant or associated with a positive or negative reward. The rationale of our study was to examine whether -as predicted by the priority map theory -attentional and motivational factors have similar, or even indistinguishable, effects on spatial attention. An additional motivation was to compare positive and negative reward directly, which could give clues about the use of different types of feedback in the rehabilitation of neglect.
Participants were required to press the space bar as fast and as precisely as possible when a unique colored target (e.g., a blue T-shape) was shown in the visual periphery. The target was preceded by an unpredictive cue, which in half of all trials was on same side as the target (valid trials) and in the other half appeared on the opposite side (invalid trials). Furthermore, cues either shared the task-relevant color (e.g., blue cueblue target), or were associated with a positive reward (a gain of 100 points), a negative reward (a loss of 100 points), or no reward (neutral cue). We expected to find a disengagement deficit, i.e. that ipsilesional cues would slow down the processing of contralesional targets. The critical question was whether cue characteristics such as task-relevance or reward-value may modulate the disengagement deficit of patients with neglect.

Participants
Twelve patients with a first-ever right hemispheric stroke suffering from spatial neglect participated in this study (5 women; mean age, 64 years, range 53-79; see Table 1 for demographical and clinical data). Patients were evaluated on average 53 days post-stroke onset (range 11-115). Five patients out of 12 had a partial left peripheral visual field defect, assessed through confrontation testing or perimetry, with sufficient sparing of the central visual field to allow perception of the experimental display. All neglect patients manifested behavioral signs of visual neglect such as unawareness of people or objects placed contralesionally and difficulty with grooming or wheelchair navigation, as well as objective signs of neglect assessed with a paper-and-pencil neglect battery including line bisection (Ronchi et al., 2012), target cancellation tasks (Azouvi et al., 2006;Gauthier et al., 1989;Ptak et al., 2007), drawing (Azouvi et al., 2006), and reading . Experimental testing occurred within a few days after the clinical evaluation. We additionally recruited a control group of 11 age-and sex-matched healthy subjects without any history of neurological or psychiatric disease. All participants had normal or corrected vision and reported no deficits of color vision.
Patients were recruited and tested at the Division of Neurorehabilitation of the University Hospitals of Geneva, while controls were recruited through advertisements and tested in the same conditions at the hospital. Written informed consent was obtained from each participant, according to procedures approved by the ethical committee of the Canton of Geneva (Geneva, Switzerland). Fig. 1A shows the rendering of the patient's lesions in an overlap plot. Lesions were visualized by creating a volumetric mask for each patient by delineating her/his lesion manually on the available clinical T1-or T2-weighted MRI scan using MRIcron software (Rorden et al., 2007). Structural images were then normalized to the MNI space with the Clinical Toolbox . Normalization produced images of 2 mm isometric voxel size. We observed a maximal overlap (six patients) in the superior temporal gyrus, the supramarginal gyrus, as well as in the Heschl gyrus and rolandic operculum, which is in good agreement with the typical anatomical correlates of neglect Pedrazzini and Ptak, 2020).
Cancellation tasks: for the Bells cancellation (Azouvi et al., 2006;Gauthier et al., 1989; cut-off left or right omissions >2), the inverted T cancellation (Ptak et al., 2007) and the letter A cancellation (Mesulam, 1985), patients were required to cross out respectively bells among several different shapes, inverted Ts among upright Ts, or A letters among different letters that are scattered on a A4 sheet of paper.
Line bisection (Ronchi et al., 2012;cut-off, 6.45 mm of leftward or rightward deviation): participants were asked to mark the midpoint of six horizontal black lines (all were 2 mm in width; lengthwise, two measured 10 cm, two 15 cm, and two 25 cm). The score was the percentage of the mean deviation of the participants' marks from the objective midpoint (measured to the nearest mm). A positive score denoted a rightward bias, a negative score a leftward bias.
Reading : the reading test consisted of 40 compound words presented in 5 columns on an A4 sheet of paper.
Clock drawing (Azouvi et al., 2006; cut-off scores <8): patients were required to place the 12 h in a circle drawn by the examiner. NA, not available; I, ischemic stroke, H, Hemorrhage, TR, tumour resection.
Patient 9 (indicated in italic) was discarded from final analysis.

Apparatus, stimuli and procedure
A PC Dell Optiplex 760 running E-prime software (Schneider et al., 2002) controlled the presentation of stimuli, timing and data collection. Participants were seated at a distance of approximately 57 cm from the monitor. The paradigm consisted of a spatial cuing task to probe attentional orienting towards a target preceded by a peripheral cue, that was associated with a positive, a negative or no reward. Cues were composed of two vertically aligned, parallel bars of 1 • × 3 • and separated by 4.5 • (Fig. 1). Targets and distractors were composed of 2 bars of 1 • × 3 • forming an L-or a T-shape, rotated by 90 • , 190 • or 180 • . Stimuli were red (RGB values: 222,80,80), green (0, 180, 0), blue (10,150,250) or yellow (255, 255, 0). One of these colors was randomly selected for each participant as the target color. The crucial variable was the color of the cue, which defined either its task-relevance or its association with a specific reward: The RELEVANT cue condition was associated with a null reward (+0) and was the only condition in which the cue had the same color as the target. The other three conditions were 'irrelevant', but reward-related, in the sense that the cue had a different color than the target, but was paired with a reward. In the POSITIVE and NEGA-TIVE reward condition, the cue color was either associated with a gain of 100 points (+100), or with a loss of 100 points (− 100), regardless whether the participants' response was correct or incorrect. In the NEUTRAL reward condition, a null reward was given (+0) (Fig. 1B and C).
Participants were instructed to press the space bar as fast as possible when a unique target-colored L or T shape was presented (go trial, 75% of the trials) irrespective of its location, and to withhold a response if the target color was not present (no-go trial, 25% of the trials). They were explicitly reminded to disregard and never to react to the parallel lines (cue) even if they had the target color. Patients were also instructed to try to maximize the number of points that they could earn across trials, but were not informed of reward contingencies. Instructions were given orally and repeated during a practice run composed of approximately 20 trials. Each trial started with a white central fixation cross, displayed against a grey background and presented for 1000 ms. This was followed by a cue for 300 ms and presented 5 • to the left or right of the fixation cross, replaced by a blank (100 ms) and finally the target display containing two colored shapes: either the target together with one distractor (go-trial) or two distractors (no-go trial). The cue presentation time was chosen to assure that neglect patients perceived the color of cues correctly even if the latter were shown in their weak hemifield (Bartolomeo et al., 2001;Ptak and Schnider, 2006). Since previous studies observed disengagement deficits even at long SOAs (>500 ms), we were confident to observe the deficit at the chosen SOA. Cues were not informative about target location: the target could be presented either at the same location as the cue (VALID condition; 50% of trials) or the opposite location (INVALID condition; 50% of trials). The target disappeared immediately after a response was given or remained on the screen for max. 3000 ms. Finally, visual feedback appeared for 2000 ms informing participants what reward they had earned during the trial (the number 0, +100 or − 100 appearing in the middle of the screen) ( Fig. 1B and C).
In sum, the experimental design had three factors: Target position (left, right), Cue validity (valid, invalid) and Cue condition (RELEVANT, POSITIVE, NEGATIVE, NEUTRAL), and the entire experiment consisted of eight blocks of 40 trials (32 go and 8 no-go trials). The experiment was performed in two sessions of approximatively 30 min on two separate days within the same week for brain-damaged patients (mean, 3 days; range 1-7) and in a single 1-h session for healthy participants. At the end of the experiment, we asked participants whether they had noticed any systematic relationship between their response and the reward. Participants who explicitly stated that a specific color was associated with a particular value were considered being aware of the color-reward relationship.

Results
Statistical analysis was performed with Statistica software (StatSoft Inc., version 14.0.0.15). Only correct responses with RTs less than 2 SDs from the participant's mean were included in the analysis. Incorrect responses accounted for 0.41% in healthy participants and in 2.45% in neglect patients. Trials less than 2 SDs from the participant's mean accounted for 4.26% of trials. One neglect patient (Patient 9) was excluded from the analysis due to an insufficient number of trials per condition, leading to a final sample size of n = 11 healthy participants and n = 11 neglect patients. Breaking down the four-way interaction potentially requires a large number of simple comparisons. However, our research question outlined in the introduction was whether task relevance and reward modulate the disengagement deficit of neglect patients. This exploratory question can be answered by comparing RTs to contralesional targets following valid and invalid cues. We therefore focused on responses of neglect patients to targets in the left hemifield and compared effects of valid and invalid cues for each cue condition (RELEVANT, POSITIVE, NEGATIVE, NEUTRAL) separately. These comparisons revealed only two significant findings: first, in the RELEVANT condition neglect patients were slower to respond to targets in the left hemifield following invalid compared to valid cues (798 ms vs 701 ms respectively; p < .001). Second, patients also exhibited a significant effect of cue validity for cues associated with a negative reward (invalid: 790 ms; valid: 728 ms; p = .002). In contrast, the validity effects for the POSITIVE and NEUTRAL conditions were not significant (all ps > .108).
Though our research question did not entail a direct comparison between cue conditions we wondered whether the effects of RELEVANT or NEGATIVE cues were strong enough to significantly differ from POSITIVE or NEUTRAL cues. We performed two-way rANOVAs on contralesional RTs with the factors Validity and Cue type. These comparisons only revealed a significant interaction for the RELEVANT - In order to better understand the effects of Cue condition on attentional orienting, we performed a similar analysis on the validity effect, which was calculated as the difference between relative RTs for invalid and valid trials (Fig. 2B). A positive validity effect indicates a performance cost, while a negative effect shows a benefit following invalid cues. This analysis focuses on differences between conditions and the pattern of responses, and therefore is better suited for a comparison between two groups that strongly differ regarding average response speed. The mixed ANOVA with Group, Target  condition and Group, [F(3,60) = 2.76, p = .050, η p 2 = 0.12]. Bonferroni post-hoc comparisons indicated that neglect patients were slower to respond to left-sided targets compared to right-sided targets in the RELEVANT condition (p < .001). The other comparisons did not reach significance (p > .19). In addition, we performed one-tailed t-tests to examine in which condition the validity effect was significantly different from zero. This was the case only for the RELEVANT (t = 3.51, p = .007) and the NEGATIVE condition (t = 2.42, p = .039).
Our group of brain-damaged patients was composed of seven neglect patients with preserved visual fields and four patients with a partial visual field defect (either hemianopia or quadrantanopia). In order to control for a potential effect of visual field impairment, we performed a mixed ANOVA only on the data of neglect patients with the withinsubjects factors Target position, Cue validity and Cue condition, as well as presence/absence of a visual field defect as categorical factor. As expected from the previous analyses, this analysis yielded a significant interaction between Target position, Cue validity and Cue condition [F (3,27) = 3.88, p = .020, η p 2 = 0.30], but no main effect and no interaction including the categorical factor (all ps > .33).
Finally, we examined whether participants were aware of the different cue-reward associations. None of our participants (whether healthy controls or patients) guessed that different amounts of reward were associated with a specific cue color.

Discussion
Our study shows that ipsilesional cues slow down the processing of contralesional targets in neglect patients particularly if they share a relevant characteristic with the target (here, color), or if they are associated with a negative reward. These findings can be understood with regard to a modulation of the so-called disengagement deficit, one of the most prominent signs of spatial neglect (Bartolomeo and Chokron, 2002). The disengagement deficit -defined as exaggerated cost to respond to contralesional targets after the presentation of an ipsilesional cue (i.e. for invalid trials) -qualitatively distinguishes attention dynamics observed in neglect from those of healthy participants. While validity effects (i.e., faster RTs following valid than invalid cues) are also observed in healthy participants, the temporal dynamics of such effects differ. Neglect patients show a disengagement deficit even with cue-target intervals of 500 ms or more, while in controls the validity effect reverses and becomes negative. Given that we used cue-target intervals of 400 ms in the present study, it is not surprising that our controls showed no benefit from valid cues. However, this finding was expected and was a trade-off of our focus on the modulation of the disengagement effect by different cue types.
Observing a disengagement deficit with task-relevant cues is in good agreement with previous findings with several groups of patients suffering from neglect (Losier and Klein, 2001;Ptak and Pedrazzini, 2021;Ptak andSchnider, 2006, 2010;Schnider et al., 2011). In one of our previous studies, we found a disengagement deficit when cues shared a perceptual characteristic with the target (e.g., a blue cue when the target was blue) or when the relation was purely semantic (e.g., the cue was the word 'red' and the target was red; Ptak & Schnider (2006)). This finding shows that capture of attention by the cue cannot be exclusively explained by perceptual mechanisms, but must be due to higher-order interference. Such higher-order effects suggest effects at the level of the priority map, a putative attentional representation established in parietal cortex that does not depend on specific perceptual features (Gottlieb et al., 1998;Ptak and Fellrath, 2013).
The principal finding of our study is that both negative reward and task-relevance modulate spatial attention sufficiently to produce a significant disengagement deficit in neglect. Only in these two conditions was the effect of cue validity significant, while it fell short of significance in the POSITIVE and NEUTRAL condition. Though the direct comparison of validity effects in the NEGATIVE and NEUTRAL conditions was not significant, negative reward yielded results that were indistinguishable from task-relevant cues. Given that a disengagement deficit was present in all conditionsat least to some degreeit would probably be too ambitious or require a much larger sample size to expect a significant difference in all comparisons.
Previous evidence demonstrated that stimuli with particular emotional or motivational values strongly shape perception and attention in healthy subjects (Anderson et al., 2011;Bourgeois et al., 2016;Della Libera and Chelazzi, 2009;Hickey et al., 2010;Theeuwes and Belopolsky, 2012) and in brain-damaged patients suffering from neglect (Bourgeois et al., 2018;Li et al., 2016;Lucas et al., 2013;Lucas and Vuilleumier, 2008;Malhotra et al., 2013;Neppi-Modona et al., 2020). However, none of these studies assessed the effect of positive and negative reward in the same paradigm and with the same participants. Bourgeois et al. (2018) observed increased cost of attentional orienting when neglect patients responded to contralesional targets accompanied by a distractor previously paired with a positive reward. Attentional capture by reward-associated stimuli occurred even when those stimuli were no longer task-relevant. Value-related stimuli may thus represent a sufficiently strong bottom-up signal to bias attentional priority outside of awareness and volitional control (Bourgeois et al., 2016).
The effects of reward on attentional selection have been observed with stimuli paired with a positive reward (Anderson, 2016;Bourgeois et al., 2016). In the current study, cues paired with a positive reward did not affect attentional selection differently than neutral cues, suggesting that only negative reward may represent a sufficiently strong signal to bias attentional priority. While this is the first study involving patients with neglect showing distinctive effects of negative reward on the shifting of attention, our findings are compatible with the observation that negative affective stimuli (such as fearful faces) tend to capture attention (Eastwood et al., 2001; but see Hedger et al., 2019 for alternative results; Vuilleumier, 2005Vuilleumier, , 2015. Several studies have reported that neglect patients detect negative affective stimuli (e.g., spiders or emotionally charged scenes) presented in contralesional hemispace faster than neutral stimuli (Grabowska et al., 2011;Vuilleumier and Schwartz, 2001). In contrast, while positive reward has a motivational effect in inducing these patients to explore more thoroughly contralesional space (Bourgeois et al., 2018;Lucas and Vuilleumier, 2008;Malhotra et al., 2013), previous studies did not observe better detection of positive affective stimuli compared to neutral items. Thus, while positive reward may affect motivational factors in neglect (such as increased effort during a difficult search task), negative reward may bias attention away from the weaker visual field.
An important question arising from our findings, is whether neural signals associated with task-relevance and negative reward converge in anatomical structures necessary for the engagement or shifting of attention. Several authors have emphasized the role of a putative parietal priority map in attentional selection. Evidence from functional imaging and single neuron studies places the priority map at the posterior parietal cortex (PPC) (Corbetta et al., 2008;Shulman, 2002, 2011;Gottlieb et al., 1998). This region is part of a dorsal frontoparietal network connecting the PPC with the premotor cortex (in particular, the frontal eye fields), and the lateral prefrontal cortex. The PPC also receives strong inputs from the visual cortex, which places this region at the crossroad of a feedforward stream carrying stimulus-related signals, and a feedback stream that conveys goal-related information from the prefrontal cortex. While this convergence of inputs makes the PPC the ideal candidate for the representation of priority, this region does not seem to be directly responsible for attention shifting. Indeed, fMRI findings have shown that the appearance of salient, unexpected or infrequent distracters (which is precisely what happens in the event of an invalid cue) activates the right temporo-parietal junction (rTPJ) more than the PPC (Corbetta et al., 2000;Corbetta and Shulman, 2002;Kincade et al., 2005). This finding is supported by lesion studies indicating that the disengagement deficit observed in patients with neglect is primarily associated with damage to the rTPJ (Friedrich et al., 1998). We recently expanded this finding by showing in a much larger group of patients that damage to the rTPJ results in pathologically increased attentional capture (i.e., the disengagement deficit) only for task-relevant distracters (Ptak and Pedrazzini, 2021). Importantly, task-irrelevant distracters had no significant effect on performance, indicating that task-relevance is the primary cause of the disengagement deficit. While we do not provide in the present study physiological data and our sample is too small for a lesion-symptom analysis our finding that negative reward and task-relevance capture attention is compatible with the presence of functional interactions between the limbic salience network (including the anterior cingulate gyrus, insula and amygdala) and the fronto-parietal attention network. A further moderating factor of the effect of reward may originate from dopaminergic pathways involving the dorsal striatum (Anderson et al., 2017;Hikosaka et al., 2014;Hikosaka et al., 2006). For instance, Malhotra et al. (2013) observed that positive reward may improve search performance in neglect, and absence of this effect is associated with damage to the striatum.
In conclusion, answering our research questions formulated in the introduction, we found that task-relevance and negative reward capture attention, and by doing so amplify the disengagement deficit characterizing spatial neglect. Our findings contribute to a better understanding of the modulatory control of attention by placing motivational value alongside task-relevance as powerful factors affecting attentional selection.

Funding
This work was supported by a grant of the Swiss National Science Foundation, grant 32003B-184702 to RP and grant no. 320030_175,472 to AS.