Abstract
Selection and integration of information based on current goals is a fundamental aspect of flexible goal-directed behavior. Motivation has been shown to improve behavioral performance across multiple cognitive tasks, yet the underlying neural mechanisms that link motivation and control processes, and in particular its effect on context-dependent information processing, remain unclear. Here, we used functional magnetic resonance imaging (fMRI) in 24 human volunteers to test whether reward motivation enhances the coding of behaviorally relevant category distinctions across the frontoparietal cortex, as would be predicted, based on previous experimental evidence and theoretical accounts. In a cued-detection categorization task, participants detected whether an object from a cued visual category was present in a subsequent display. The combination of the cue and the visual category of the object determined the behavioral status of the presented objects. To manipulate motivation, half of all trials offered the possibility of a substantial reward. We observed an increase with reward in overall activity across the frontoparietal control network when the cue was presented, reflecting cognitive effort when the context is set for a task. Multivariate pattern analysis (MVPA) further showed that behavioral status information for the objects presented was conveyed across the network. However, in contrast to our prediction, reward did not increase the discrimination between behavioral status conditions in the stimulus epoch of a trial when object information was processed depending on a current context. In the high-level general object visual region, the lateral occipital complex, the representation of behavioral status was driven by visual differences and was not modulated by reward. Our study provides another tier of evidence for the interaction between cognitive control and motivation to further inform computational models of reward and contribute to the understanding of the underlying neural mechanisms.
Introduction
A fundamental aspect of flexible goal-directed behavior is the selection and integration of information depending on a current goal to determine its relevance to behavior and lead to a decision. In non-human primates, single-cell data from the lateral prefrontal cortex, as well as parietal cortex, provide detailed evidence for the coding of task-relevant information. It has been shown that neural activity contains information about the context, also referred to as cue or task-set, as well as the integrated information of cue and a subsequent input stimulus, such as task-related categorical and behavioral decision (Freedman, Riesenhuber, Poggio, & Miller, 2001; Kadohisa et al., 2013; Kusunoki, Sigala, Nili, Gaffan, & Duncan, 2010; Mante, Sussillo, Shenoy, & Newsome, 2013; Stokes et al., 2013; Wallis, Anderson, & Miller, 2001). In the human brain, a network of frontal and parietal cortical regions, the ‘multiple-demand’ (MD) network (Fedorenko, Duncan, & Kanwisher, 2013; Mitchell et al., 2016), has been shown to be involved in information selection and integration, and more generally in control processes. This network is associated with multiple aspects of cognitive control, such as spatial and verbal working memory, math, conflict monitoring, rule-guided categorization and task switching, (Cole, Ito, & Braver, 2016; Fedorenko et al., 2013; Vergauwe & Cowan, 2015). The MD network spans the anterior-posterior axis of the middle frontal gyrus (MFG); posterior dorso-lateral frontal cortex (pdLFC); the anterior insula and frontal operculum (AI/FO); the pre-supplementary motor area and the adjacent dorsal anterior cingulate cortex (preSMA/ACC); and intraparietal sulcus (IPS) (Duncan, 2010). Multiple neuroimaging studies demonstrated that distributed patterns of activity across the MD network measured by functional magnetic resonance imaging (fMRI) reflected a variety of task-related information. These include task sets, behavioral relevance and task-dependent categorical decisions (Erez & Duncan, 2015; Li, Ostwald, Giese, & Kourtzi, 2007; Muhle-Karbe, Duncan, De Baene, Mitchell, & Brass, 2017; Wisniewski, Goschke, & Haynes, 2016; Woolgar, Hampshire, Thompson, & Duncan, 2011; Woolgar, Thompson, Bor, & Duncan, 2011; Woolgar, Williams, & Rich, 2015). In contrast, sensory areas such as the high-level general object visual region, the lateral occipital complex (LOC), as well as the primary visual cortex, contain information about the visual properties and categorization of stimuli, with weaker, or non-existing, task effects (Bugatus, Weiner, & Grill-Spector, 2017; Harel, Kravitz, & Baker, 2014; Hebart, Bankson, Harel, Baker, & Cichy, 2018).
With growing interest in recent years in the link between cognitive control and motivation, it has been proposed that motivation enhances control processes by sharpening representation of task goals and prioritizing task-relevant information across the frontoparietal network and other regions associated with cognitive control (Botvinick & Braver, 2015; Etzel, Cole, Zacks, Kay, & Braver, 2016; Kruglanski et al., 2002; Simon, 1967). In line with this idea, it has been shown that motivation, usually manipulated as monetary reward, increases task performance (Padmala & Pessoa, 2010, 2011). Neuroimaging studies linked increased activity with reward in frontoparietal regions across a range of tasks, including working memory (Pochon et al., 2002; Taylor et al., 2004), selective attention (Krebs, Boehler, Roberts, Song, & Woldorff, 2012; Mohanty, Gitelman, Small, & Mesulam, 2008), response inhibition (Padmala & Pessoa, 2011), and problem solving (Shashidhara, Mitchell, Erez, & Duncan, 2019).
Although the accumulating evidence at the behavioral and neural level in humans are consistent with this sharpening and prioritizing account (Braver, 2012; Chiew & Braver, 2014; Kruglanski et al., 2002; Pessoa, 2009; Simon, 1967; Wallace, 1960), they do not directly address the effect of motivation on the coding of task-related information and selection and integration processes. Some support for this idea comes from single-neuron data recorded from the prefrontal cortex of non-human primates: reward was associated with greater spatial selectivity, enhanced activity related to working memory and modulated task-related activity based on the type of reward (Kennerley & Wallis, 2009; Leon & Shadlen, 1999; Watanabe, 1996). A more direct evidence in humans was recently demonstrated by Etzel et al. (2016). They showed that reward enhances coding of task cues across the frontoparietal cortex, and suggested that task-set efficacy increases with reward. Subsequently, Hall-McMaster et al. (2019) used electroencephalogram (EEG) and demonstrated that this effect of reward on the coding of task cues was particularly evident when a switch in context was required (Hall-McMaster, Muhle-Karbe, Myers, & Stokes, 2019). They also provided some evidence that the representation of features relevant for a given task is enhanced when the reward level is high. However, since their results were obtained using EEG, the spatial specificity of such reward effects are limited. Therefore, it remains unclear whether the previously reported facilitative effect of reward on representation across the frontoparietal cortex is limited to preparatory cues, or whether reward also enhances the coding of behaviorally relevant information, when the cue and a subsequent stimulus are integrated, leading to the behavioral decision thus supporting goal-directed flexible behavior.
Following the sharpening hypothesis, in this study we asked whether reward motivation enhances the representation of behaviorally relevant information, as determined by the integration of cue and stimulus input. Furthermore, previous studies have associated reward with decreased conflict in interference tasks (Krebs, Boehler, Appelbaum, & Woldorff, 2013; Padmala & Pessoa, 2011; Stürmer, Nigbur, Schacht, & Sommer, 2011), suggesting that any effect of reward may be particularly important for high-conflict items, in other words, a conflict-contingent effect. We therefore also asked whether such facilitative effect of reward is selective for highly conflicting items. We recently showed that behaviorally relevant, but not irrelevant, category distinctions of objects were coded across the MD network (Erez & Duncan, 2015). In contrast, such differences were not observed in the LOC. Here, we used a similar cued detection categorization task while participants’ brain activity was measured using fMRI. Participants detected whether an object from a cued visual category (target category) was present or absent. On each trial, one of two categories was cued, and objects from those two categories could be either Targets, or nontargets with high behavioral conflict, as they could be targets on other trials (High-conflict nontarget). An additional category was never cued, serving as nontarget with low behavioral conflict (Low-conflict nontarget). This design created three levels of behavioral status (Targets, High-conflict nontargets, Low-conflict nontargets). Critically, following this integration process, the relevant information that is expected to be represented across the MD network is the behavioral status of a given category, rather than the visual category itself (Erez & Duncan, 2015). Therefore, the behaviorally relevant category distinctions were pairs of categories with different behavioral status. We used multivariate pattern analysis (MVPA) to measure representation of the behaviorally relevant category distinctions as reflected in distributed patterns of response in the a priori defined MD network. To manipulate motivation, on half of all trials a substantial monetary reward was offered. We tested whether the neural pattern discriminability between the behaviorally relevant category distinctions increased with reward, and whether this effect was selective for the distinction between Targets and High-conflict nontargets. A common view posits that top-down signals from the frontoparietal MD network to the visual cortex play an important role in the processing of task-related information. Therefore, to test for the effect of reward motivation on the representation of the behavioral status distinctions in the visual cortex, we conducted similar analyses in the high-level general object visual region, the lateral occipital complex (LOC).
Materials and Methods
Participants
24 participants (13 females), between the ages of 18-40 years (mean age: 25) took part in the study. Four additional participants were excluded due to large head movements during the scan (greater than 5 mm). The sample size was determined prior to data collection, as typical for neuroimaging studies and in accordance with counter-balancing requirements of the experimental design across participants. A similar sample size showed sufficient power to detect representation of behavioral status in a previous study (Erez & Duncan, 2015). All participants were right handed with normal or corrected-to-normal vision and had no history of neurological or psychiatric illness. The study was conducted with approval by the Cambridge Psychology Research Ethics Committee. All participants gave written informed consent and were monetarily reimbursed for their time.
Task Design
Participants performed a cued categorization task in the MRI scanner (Figure 1A). Our primary question concerned the representation during the stimulus epoch of a trial where cue and stimulus are integrated, and we therefore designed the task accordingly. At the beginning of each trial, one of three visual categories (sofas, shoes, cars) was cued, determining the target category for that trial. Participants had to indicate whether the subsequent object matched this category or not by pressing a button. For each participant, only two of the categories were cued as targets throughout the experiment. Depending on the cue on a given trial, objects from these categories could be either Targets (T), or nontargets with high conflict (as they could serve as targets on other trials). The third category was never cued, therefore objects from this category served as Low-conflict nontargets. This design yielded three behavioral status conditions: Targets, High-conflict nontargets and Low-conflict nontargets (Figure 1B). The assignment of the categories to be cued (and therefore serve as either Targets or High-conflict nontargets) or not (and serve as Low-conflict nontargets) was counter-balanced across participants.
To manipulate motivation, half of the trials were cued as reward trials, in which participants had the chance of earning £1 if they completed the trial correctly and within a time limit. To assure the incentive on each reward trial, four random reward trials out of 40 in each run were assigned the £1 reward. To avoid longer reaction times when participants try to maximize their reward, a response time threshold was used for reward trials, set separately for each participant as the average of 32 trials in a pre-scan session. The participants were told that the maximum reward they could earn is £24 in the entire session (£4 per run), and were not told what the time threshold was. Therefore, to maximize their gain, participants had to treat every reward trial as a £1 trial and respond as quickly and as accurately as possible, just as in no-reward trials.
Each trial started with a 1 s cue, which was the name of a visual category that served as the target category for this trial. On reward trials, the cue included three red pound signs presented next to the category name. The cue was followed by a fixation dot in the center of the screen presented for 0.5 s and an additional variable time of either 0.1, 0.4, 0.7 or 1 s, selected randomly, in order to make the stimulus onset time less predictable. The stimulus was then presented for 120 ms and was followed by a mask. Participants indicated by a button press whether this object belonged to the cued target category (present) or not (absent). Following response, a 1 s blank inter-trial interval separated two trials. For both reward and no-reward trials, response time was limited to a maximum of 3 s, after which the 1 s blank inter-trial interval started even when no response was made. For reward trials, an additional subject-specific response time threshold was used as mentioned above to determine whether the participants earned the reward or not, but this time threshold did not affect the task structure and was invisible to the participants.
We used catch trials to decorrelate the BOLD signals of the cue and stimulus phases. 33% of all trials included cue followed by fixation dot for 500 ms, which then turned red for another 500 ms indicating the absence of the stimulus, followed by the inter-trial interval.
Stimuli
Objects were presented at the center of the screen on a grey background. The objects were 2.95° visual angle along the width and 2.98° visual angle along the height. Four exemplars from each visual category were used. Exemplars were chosen with similar colors, dimensions, and orientation across the categories. All exemplars were used an equal number of times in each condition and in each run to ensure that any differences between the experimental conditions will not be driven by the variability of exemplars. To increase the task demand, based on pilot data, we added Gaussian white noise to the stimuli. The post-stimulus mask was generated by randomly combining pieces of the stimuli that were used in the experiment. The mask was the same size as the stimuli and was presented until a response was made or the response time expired.
Structure and Design
Each participant completed 6 functional runs of the task in the scanner (mean duration ± SD: 6.2 ± 0.13 min). Each run started with a response-mapping instructions screen (e.g. left = target present, right = target absent), displayed until the participants pressed a button to continue. Halfway through the run, the instructions screen was presented again with the reversed response mapping. All trials required a button response to indicate whether the target was present or absent, and the change of response mapping ensured that conditions were not confounded by the side of the button press. Each run included 104 trials. Out of these, 8 were dummy trials following the response mapping instructions (4 after each instructions screen), and were excluded from the analysis. Of the remaining 96 trials, one-third (32 trials) were cue-only trials (catch trials). Of the remaining 64 trials, 32 were no-reward trials and 32 were reward trials. Of the 32 no-reward trials, half (16) were cued with one visual category, and half (16) with the other. For each cued category, half of the trials (8) were Target trials, and half of the trials (8) were nontarget trials, to assure an equal number of target (present) and nontarget (absent) trials. Of the nontarget trials, half (4) were High-conflict nontargets, and half (4) were Low-conflict nontargets. There were 4 trials per cue and reward level for the High- and Low-conflict nontarget conditions, and 8 for the Target condition, with the latter split into two regressors (see General Linear Model (GLM) for the Main Task section below). A similar split was used for reward trials. An event-related design was used and the order of the trials was randomized in each run. At the end of each run, the money earned in the reward trials and the number of correct trials (across both reward and no-reward trials) were presented on the screen.
Functional Localizers
In addition to the main task, we used two other tasks in order to functionally localize MD regions and LOC in individual participants using independent data. These were used in conjunction with ROI templates and a double-masking procedure to extract voxel data for MVPA (See ROI definition for more details).
To localize MD regions, we used a spatial working memory task (Fedorenko et al., 2013). On each trial, participants remembered 4 locations (Easy condition) or 8 locations (Hard condition) in a 3X4 grid. Each trial started with fixation for 500 ms. Locations on the grid were then highlighted consecutively for 1 s (1 or 2 locations at a time, for the Easy and Hard conditions, respectively). In a subsequent two-alternative forced-choice display (3 s), participants had to choose the grid with the correct highlighted locations by pressing the left or the right button. Feedback was given after every trial for 250 ms. Each trial was 8 s long, and each block included 4 trials (32 s). There was an equal number of correct grids on the right and left in the choice display. Participants completed 2 functional runs of 5 min 20 sec each, with 5 Easy blocks alternated with 5 Hard blocks in each run. We used the contrast of Hard vs. Easy blocks to localize MD regions.
As a localizer for LOC we used a one-back task with blocks of objects interleaved with blocks of scrambled objects. The objects were in grey scale and taken from a set of 61 everyday objects (e.g. camera, coffee cup, etc.). Participants had to press a button when the same image was presented twice in a row. Images were presented for 300 ms followed by a 500 ms fixation. Each block included 15 images with two image repetitions and was 12 s long. Participants completed two runs of this task, with 8 object blocks, 8 scrambled object blocks, and 5 fixation blocks. The objects vs. scrambled objects contrast was used to localize LOC.
Scanning Session
The scanning session included a structural scan, 6 functional runs of the main task, and 4 functional localizer runs – 2 for MD regions and 2 for LOC. The scanning session lasted up to 100 minutes, with an average 65 minutes of EPI time. The tasks were introduced to the participants in a pre-scan training session. The average reaction time of 32 no-reward trials of the main task completed in this practice session was set as the time threshold for the reward trials to be used in the scanner session. All tasks were written and presented using Psychtoolbox3 (Brainard, 1997) and MatLab (The MathWorks, Inc).
Data Acquisition
fMRI data were acquired using a Siemens 3T Prisma scanner with a 32-channel head coil. We used a multi-band imaging sequence (CMRR, release 016a) with a multi-band factor of 3, acquiring 2 mm isotropic voxels (Feinberg et al., 2010). Other acquisition parameters were: TR = 1.1 s, TE = 30 ms, 48 slices per volume with a slice thickness of 2 mm and no gap between slices, in plane resolution 2 × 2 mm, field of view 205 mm, flip angle 62°, and interleaved slice acquisition order. No iPAT or in-plane acceleration were used. T1-weighted multiecho MPRAGE (van der Kouwe, Benner, Salat, & Fischl, 2008) high-resolution images were also acquired for all participants, in which four different TEs were used to generate four images (voxel size 1 mm isotropic, field of view of 256 × 256 × 192 mm, TR = 2530 ms, TE = 1.64, 3.5, 5.36, and 7.22 ms). The voxelwise root mean square across the four MPRAGE images was computed to obtain a single structural image.
Data and Statistical Analysis
The primary analysis approach was multi-voxel pattern analysis (MVPA), to assess representation of behaviorally relevant category distinctions with and without reward. An additional ROI-based univariate analysis was conducted to confirm the recruitment of the MD network. Preprocessing, GLM and univariate analysis of the fMRI data were performed using SPM12 (Wellcome Department of Imaging Neuroscience, London, England; www.fil.ion.ucl.ac.uk), and the Automatic Analysis (aa) toolbox (Cusack et al., 2014).
We used an alpha level of .05 for all statistical tests. Bonferroni correction for multiple comparisons was used when required, and the corrected p-values and uncorrected t-values are reported. All t tests that were used to compare two conditions were paired due to the within-subject design. A one-tailed t test was used when the prediction was directional, including testing for classification accuracy above chance level. All other t tests in which the a priori hypothesis was not directional were two-tailed. Additionally, effect size (Cohen’s dz) was computed. All analyses were conducted using custom-made MATLAB (The Mathworks, Inc) scripts, unless otherwise stated.
All raw data and code used in this study will be publicly available upon publication.
Pre-processing
Initial processing included motion correction and slice time correction. The structural image was coregistered to the Montreal Neurological Institute (MNI) template, and then the mean EPI was coregistered to the structural. The structural image was then normalized to the MNI template via a nonlinear deformation, and the resulting transformation was applied on the EPI volumes. Spatial smoothing of FWHM = 5 mm was performed for the functional localizers data only.
General Linear Model (GLM) for the Main Task
We used GLM to model the main task and localizers’ data. Regressors for the main task included 12 conditions during the stimulus epoch and 4 conditions during the cue epoch. Regressors during the stimulus epoch were split according to reward level (no-reward, reward), cued visual category (category 1, category 2), and behavioral status (Target, High-conflict nontarget, Low-conflict nontarget). To assure an equal number of target present and target absent trials, the number of Target trials in our design was twice the number of High-conflict and Low-conflict nontarget trials. The Target trials included two repetitions of each combination of cue, visual category and exemplar, with a similar split for reward trials. These two Target repetitions were modelled as separate Target1 and Target2 regressors in the GLM to make sure that all the regressors were based on an equal number of trials, but were invisible to the participants. All the univariate and multivariate analyses were carried out while keeping the two Target regressors separate to avoid any bias of the results, and they were averaged at the final stage of the results. Overall, the GLM included 16 regressors of interest for the 12 stimulus conditions. Each regressor was based on data from all correct trials in the respective condition in each run (up to 4 trials). To account for possible effects of reaction time (RT) on the beta estimates because of the varying duration of the stimulus epoch, and as a consequence their potential effect on decoding results, these regressors were modelled with durations from stimulus onset to response (Woolgar, Golland, & Bode, 2014). This model scales the regressors based on the reaction time, thus the beta estimates reflect activation per unit time and are comparable across conditions with different durations. Regressors during the cue epoch included both task and cue-only (catch) trials and were split by reward level and cued category, modelled with duration of 1 s. Cue regressors were based on 16 trials per regressor per run. As one-third of all trials were catch trials, the cue and stimulus epoch regressors were decorrelated and separable in the GLM. Regressors were convolved with the canonical hemodynamic response function (HRF). The 6 movement parameters and run means were included as covariates of no interest.
GLM for the Functional Localizers
For the MD localizer, regressors included Easy and Hard blocks. For LOC, regressors included objects and scrambled objects blocks. Each block was modelled with its duration. The regressors were convolved with the canonical hemodynamic response function (HRF). The 6 movement parameters and run means were included as covariates of no interest.
Univariate Analysis
We conducted an ROI analysis to test for the effect of reward on overall activity for the different behavioral status conditions and cues. We used templates for the MD network and for LOC as defined below (see ROI definition). Using the MarsBaR toolbox (http://marsbar.sourceforge.net; Brett et al. 2002) for SPM 12, beta estimates for each regressor of interest were extracted and averaged across runs, and across voxels within each ROI, separately for each participant and condition. For the MD network, beta estimates were also averaged across hemispheres (see ROI definition below). Second-level analysis was done on beta estimates across participants using repeated measures ANOVA. The data for the Target condition was averaged across the two Target1 and Target2 regressors, separately for the no-reward and reward conditions.
ROI Definition
MD network template
ROIs of the MD network were defined a priori using an independent data set (Fedorenko et al. 2013; see t-map at http://imaging.mrc-cbu.cam.ac.uk/imaging/MDsystem). These included the anterior, middle, and posterior parts of the middle frontal gyrus (aMFG, mMFG, and pMFG, respectively), a posterior dorsal region of the lateral frontal cortex (pdLFC), AI-FO, pre-SMA/ACC, and IPS, defined in the left and right hemispheres. The visual component in this template is widely accepted as a by-product of using largely visual tasks, and is not normally considered as part of the MD network. Therefore, it was not included in the analysis. The MD network is highly bilateral, with similar responses in both hemispheres (Erez & Duncan, 2015; Fedorenko et al., 2013). We therefore averaged the results across hemispheres in all the analyses.
LOC template
LOC was defined using data from a functional localizer in an independent study with 15 participants (Lorina Naci, PhD dissertation, University of Cambridge). In this localizer, forward- and backward-masked objects were presented, as well as masks alone. Masked objects were contrasted with masks alone to identify object-selective cortex (Malach et al., 1995). Division to the anterior part of LOC, the posterior fusiform region (pFs) of the inferior temporal cortex, and its posterior part, the lateral occipital region (LO) was done using a cut-off MNI coordinate of Y=-62, as previous studies have shown differences in processing for these two regions (Erez & Yovel, 2014; MacEvoy & Epstein, 2011).
Voxels selection for MVPA
To compare between regions within the MD network and between sub-regions in LOC, we controlled for the ROI size and used the same number of voxels for all regions. We used a dual-masking approach that allowed the use of both a template, consistent across participants, as well as subject-specific data as derived from the functional localizers (Fedorenko, Hsieh, Nieto-Castañón, Whitfield-Gabrieli, & Kanwisher, 2010; Shashidhara, Spronkers, & Erez, 2019). For each participant, beta estimates of each condition and run were extracted for each ROI based on the MD network and LOC templates. For each MD ROI, we then selected the 200 voxels with the largest t-value for the Hard vs. Easy contrast as derived from the independent subject-specific functional localizer data. This number of voxels was chosen prior to any data analysis, similar to our previous work (Erez & Duncan, 2015). For each LOC sub-region, we selected 180 voxels with the largest t-values of the object vs. scrambled contrast from the independent subject-specific functional localizer data. The selected voxels were used for the voxelwise patterns in the MVPA for the main task. The number of voxels that was used for LOC was smaller than for MD regions because of the size of the pFs and LO masks. For the analysis that compared MD regions with the visual regions, we used 180 voxels from all regions to keep the ROI size the same.
Multivoxel pattern analysis (MVPA)
We used MVPA to test for the effect of reward motivation on the discrimination between the task-related behavioral status pairs. Voxelwise patterns using the selected voxels within each template were computed for all the task conditions in the main task. We applied our classification procedure on all possible pairs of conditions as defined by the GLM regressors of interest during the stimulus presentation epoch, for the no-reward and reward conditions separately (Figure 1B). For each pair of conditions, MVPA was performed using a support vector machine classifier (LIBSVM library for MATLAB, c=1) implemented in the Decoding Toolbox (Hebart, Gorgen, & Haynes, 2015). We used leave-one-run-out cross-validation in which the classifier was trained on the data of five runs (training set) and tested on the sixth run (test set). This was repeated 6 times, leaving a different run to test each time, and classification accuracies were averaged across these 6 folds. Classification accuracies were then averaged across pairs of different cued categories, yielding discrimination measures for three pairs of behavioral status (Targets vs. High-conflict nontargets, Targets vs. Low-conflict nontargets, and High-conflict vs. Low-conflict nontargets) within each reward level (no-reward, reward). Because the number of Target trials in our design was twice the number of High-conflict and Low-conflict nontarget trials, each discrimination that involved a Target condition was computed separately for the two Target regressors (Target1 and Target2) and classification accuracies were averaged across them.
The Target and High-conflict nontarget pairs of conditions included cases when both conditions had an item from the same visual category as the stimulus (following different cues), as well as cases in which items from two different visual categories were displayed as stimuli (following the same cue). To test for the contribution of the visual category to the discrimination, we split the Target vs. High-conflict nontarget pairs of conditions into these two cases and the applied statistical tests accordingly.
Whole-brain searchlight pattern analysis
To test whether additional regions outside the MD network show change in discriminability between voxelwise patterns of activity of behavioral status conditions when reward is introduced, we conducted a whole-brain searchlight pattern analysis (Kriegeskorte, Goebel, & Bandettini, 2006). This analysis enables the identification of focal regions that carry relevant information, unlike the decoding based on larger ROIs, which tests for a more widely distributed representation of information. For each participant, data was extracted from spherical ROIs with an 8 mm radius, centered on each voxel in the brain. These voxels were used to perform the same MVPA analysis as described above. Thus, for each voxel, we computed the classification accuracies for the relevant distinctions, separately for the no-reward and reward conditions. These whole-brain maps were smoothed using a 5 mm FWHM Gaussian kernel. The t-statistic from a second level random-effects analysis on the smoothed maps was thresholded at the voxel level using FDR correction (p < 0.05).
Results
Behavior
RT for three behavioral status conditions, Target, High-conflict nontarget and Low-conflict nontarget, in the no-reward trials were 589 ± 98 ms, 662 ± 103 ms, and 626 ± 107 ms, respectively (mean ± SD); RTs for these conditions in the reward trials were 541 ± 99 ms, 614 ± 99 ms, 585 ± 97 ms, respectively (mean ± SD). A two-way repeated measures ANOVA with motivation (no-reward, reward) and behavioral status as within-subject factors showed a main effect of motivation (F1, 23 = 40.07, p < 0.001), with reward trials being shorter than no-reward trials, as expected from the experimental design in which response was required within a time limit to receive the reward. An additional main effect of behavioral status (F2, 23 = 50.97, p < 0.001) was observed, with no interaction between reward and behavioral status (F2, 23 = 0.63, p = 0.54). Subsequent post-hoc tests with Bonferroni correction for multiple comparisons showed that RTs for Target trials were faster than High-conflict and Low-conflict nontarget trials (t23 = 10.03, p < 0.001, dz = 2.05; t23 = 5.17, p < 0.001, dz = 1.06 respectively), and Low-conflict nontarget trials were faster than the High-conflict ones (t23 = 4.96, p < 0.001, dz = 1.01), as expected from a cued target detection task.
Overall accuracy levels were high (mean ± SD: 92.51% ± 0.08%). Mean and SD accuracy rates for the Target, High-conflict nontarget and Low-conflict nontarget conditions in the no-reward trials were 91.2% ± 5.8%, 89.1% ± 8.8%, and 96.6% ± 3.8%, respectively; and for the reward trials they were 94.2% ± 5.0%, 87.8%± 8.7%, 96.1% ± 4.4%, respectively. A two-way repeated measures ANOVA with motivation and behavioral status as within-subject factors showed no main effect of motivation (F1, 23 = 0.49, p = 0.49), confirming that the added time constraint for reward trials did not lead to drop in performance. There was a main effect of behavioral status (F2, 23 = 29.64, p < 0.001) and an interaction between motivation and behavioral status (F2, 23 = 5.81, p < 0.01). Post-hoc tests with Bonferroni correction for multiple comparisons showed larger accuracies for Low-conflict nontargets compared to Targets and High-conflict nontargets (Two-tailed t-test: t 23 = 5.64, p < 0.001, dz = 1.15; t23 = 5.50, p < 0.001, dz = 1.12 respectively) in the no-reward trials, as expected given that the Low-conflict nontarget category was fixed throughout the experiment. In the reward trials, performance accuracies were larger for Target compared to High-conflict nontarget (t23 = 4.45, p < 0.001, dz = 0.91) and Low-conflict nontarget compared to High-conflict ones (t23 = 5.92, p < 0.001, dz = 1.2) and similar between Targets and Low-conflict nontargets (t23 = 2.49, p = 0.06). Accuracies for Target trials were larger for the reward trials compared to no-reward (t23 = 2.92, p = 0.008, dz = 0.61), indicating a possible behavioral benefit of reward. There was no difference between reward and no-reward trials for High-conflict and Low-conflict nontargets (t23 < 1.1, p > 0.1, for both).
Activity across the MD network during the cue epoch
To address our primary research question, the analysis focused on the stimulus epoch. However, to get a full picture of the data and for comparability with previous studies that showed increase in cue information, we also report the results for the cue epoch here. The analysis focuses on the MD network only and not the LOC since no object stimuli were presented at this epoch of the trial.
We first tested for a univariate effect of reward during the cue phase (averaged across the β estimates of the two cues) across all MD regions. A two-way repeated measures ANOVA with reward (2: no-reward, reward) and ROI (7) as factors showed a main effect of reward (F1, 23 = 13.75, p = 0.001) with increased activity during the reward trials compared to the no-reward trials. There was also a main effect of ROI (F6, 138 = 6.44, p < 0.001) and an interaction of reward and ROI (F6, 138 = 6.67, p < 0.001). Post-hoc tests showed that all regions except aMFG showed increased activation for reward trials compared to no-reward trials (Two tailed, Bonferroni corrected for 7 comparisons: t23 > 3.06, p < 0.04, dz > 0.62 for all ROIs except aMFG; t23 = 2.38, p = 0.18, dz = 0.49 for aMFG). Overall, the MD network showed a strong univariate reward effect during the cue epoch.
We next asked whether the cues were decodable as measured using MVPA, and whether decoding levels increased with reward as has been previously reported (Etzel et al., 2016; Hall-McMaster et al., 2019). Decoding between the two cues separately for the two reward levels were computed in each of the MD ROIs. A two-way repeated measures ANOVA with reward (2) and ROI (7) as factors showed no main effects or interactions (F < 1.9, p > 0.08). Decoding levels averaged across all MD ROIs were (mean ± SD) 51.71% ± 4.39% and 50.55% ± 6.69% for the reward and no-reward conditions, respectively. There was no significant cue decoding above chance for both no-reward and reward conditions (One-tailed t-test, reward: t23 = 1.9, p = 0.07, dz = 0.39; no-reward: t23 = 0.4, p = 0.7, dz = 0.07). Overall, we found that cue information could not be decoded in any of the MD ROIs and in both no-reward and reward conditions.
Lastly, we conducted a complementary whole-brain searchlight analysis to test whether cue decoding was observed in other regions beyond the MD network. A second level random-effects analysis of cue decoding, separately for the no-reward and reward conditions, did not reveal any additional regions that showed cue decoding (FDR correction p < 0.05). An additional searchlight analysis was set to test for increase in cue decoding with reward, but similar results were obtained, with no voxels surviving FDR correction (p < 0.05). Altogether, our results show that despite substantial increases in overall univariate activity with reward during the cue epoch across the MD network, the cues in our study were not decodable in both no-reward and reward conditions.
Univariate activity in the MD network during the stimulus epoch
We started our analysis for the stimulus epoch by testing for the effect of reward motivation on the overall activity in MD regions, and whether such effect is different for the three behavioral status conditions. We used averaged β estimates for each behavioral status (Target, High-conflict nontarget, Low-conflict nontarget) and reward level (no-reward, reward) in each of the MD ROIs (Figure 2). A three-way repeated measures ANOVA with reward (2), behavioral status (3) and ROI (7) as within-subject factors showed no significant main effect of reward (F1, 23 = 3.37, p = 0.079). There was an interaction of reward level and ROI (F6, 138 = 5.02, p = 0.001), with only the AI/FO showing reward effect following post-hoc tests and Bonferroni correction for multiple (7) comparisons (t23 = 3.88, p = 0.005, dz = 0.79). The IPS showed a reward effect that did not survive multiple comparisons (t23 = 2.58, uncorrected p = 0.016, corrected p = 0.11, dz = 0.53 Importantly, there was no main effect of behavioral status (F2, 46 = 0.97, p = 0.57) and no interaction of reward and behavioral status (F2, 46 = 0.51, p = 0.61). Overall, the univariate results indicated similar levels of activity for the three behavioral status conditions. While we expected an increase in univariate activity in many MD regions with reward (Dixon & Christoff, 2012; Padmala & Pessoa, 2011; Shashidhara, Mitchell, et al., 2019), we observed such an increase only in the AI/FO.
Effect of reward motivation on discrimination of behaviorally relevant category distinctions in the MD network
Our main question concerned the representation of task-related behavioral status information across the MD network and its modulation by reward, and we used MVPA to address that. For each participant and ROI we computed the classification accuracy above chance (50%) for the distinctions between Target vs. High-conflict nontarget, Target vs. Low-conflict nontarget and High-conflict vs. Low-conflict nontargets, separately for no-reward and reward conditions (Figure 3). The analysis was set to test for discrimination between behavioral status conditions within each reward level, and whether these discriminations are larger when reward is introduced compared to the no-reward condition. A three-way repeated-measures ANOVA with reward (2), behavioral distinction (3) and ROI (7) as within-subject factors showed no main effect of ROI (F6, 138 =0.97, p = 0.45) or any interaction of ROI with reward and behavioral distinction (F < 1.16, p > 0.31). Therefore, the classification accuracies were averaged across ROIs for further analysis (Figure 3A). First, we looked at the overall discrimination of behavioral status pairs. Averaged across the three pairs of behavioral status, decoding accuracies were (mean ± SD) 51.4% ± 2.8% and 51.8% ± 3.5% for the no-reward and reward conditions, respectively. Decoding levels were above chance (50%) for both the no-reward and reward trials (one-tailed t-test against chance, corrected for 2 comparisons, reward: t23 = 2.5, corrected p = 0.02, dz = 0.5; no-reward: t23 = 2.34, corrected p = 0.03, dz = 0.48). The decoding levels above chance for the individual pairs of behavioral status for the no-reward and reward conditions are summarized in Table 1. Overall, our results show that on average behaviorally relevant categorical distinctions are represented across the MD network in both no-reward and reward conditions, with some differences between individual pairs of behavioral status.
In our critical analysis we tested for the modulatory effect of reward on the discriminability between pairs of behavioral status. In contrast to our prediction, a two-way repeated measures ANOVA with reward (2) and behavioral distinction (3) as within-subject factors showed no main effects of reward or behavioral distinction (F1, 23 = 0.26, p = 0.6; F2, 46 = 1.37, p = 0.26, respectively), and no interaction of the two (F2, 46 = 0.74, p = 0.48). To test for the specific prediction that reward might increase discrimination for the high conflict pair of conditions that may not have been picked up by the ANOVA, we compared decoding levels for the Target vs. High conflict nontarget for the no-reward and reward conditions. Further in contrast to our prediction, classification accuracy was not larger in the reward trials compared to the no-reward trials for the Target vs. High-conflict nontarget distinction (One-tailed paired t-test: t23 =1.07, p = 0.15, dz = 0.22). In summary, although the average decoding levels of behavioral status were above chance across the MD network, we did not find increases in decodability with reward.
To test for other brain regions that may have shown increased pattern discriminability with reward beyond the MD network, we conducted a complementary whole-brain searchlight analysis. In a second-level random-effects analysis of behavioral status classification maps (average across the three pairs of behavioral status) of reward vs. no-reward conditions, none of the voxels survived an FDR threshold of p < 0.05. A separate searchlight analysis for classification of Targets vs. High-conflict nontargets showed similar results, with no voxels surviving FDR correction (p < 0.05). Therefore, our data did not reveal any brain regions that showed the predicted increase in discriminability with reward.
Effects of reward motivation on behaviorally relevant category distinctions in LOC
It is widely accepted that the frontoparietal MD network exerts top-down control on visual areas, contributing to task-dependent processing of information. To test for reward effects on decoding of categorical information based on behavioral status in the visual cortex, we performed similar univariate and multivariate analyses during the stimulus epoch in the high-level general object visual region, the lateral occipital complex (LOC), separately for its two sub-regions, LO and pFs. We first conducted univariate analysis to test for an effect of reward and behavioral status on overall activity in LOC, which did not show a change in BOLD response with reward. A four-way repeated-measures ANOVA with reward (2), behavioral status (3), ROI (2), and hemisphere (2) as within-subject factors showed no main effect of reward (F1, 23 = 0.3 p = 0.6), no main effect of ROI (F1, 23 = 3.7 p = 0.07), or hemisphere (F1, 23 = 3.95 p = 0.06). There was an interaction of reward and ROI (F1, 23 = 7.3, p = 0.01), but post-hoc tests with correction for multiple (2) comparisons showed that activity was not larger for reward compared to no-reward trials in both LO and pFs (Two-tailed t-test: t23 = 1.45, p = 0.16, dz = 0.3; t23 = 0.94, p = 0.36, dz = 0.2; for LO and pFs, respectively). There was a main effect of behavioral status (F2, 46 = 8.73, p < 0.001), but no interaction of behavioral status and reward (F2, 46 = 0.56, p = 0.57). Altogether, the univariate results show that reward did not lead to increased activity in LOC for any of the behavioral status conditions.
We then tested for the representation of the task-related behavioral status conditions in LOC (Figure 4). Decoding levels averaged across all pairs of behavioral status and the two LOC ROIs were above chance for both no-reward and reward conditions (mean ± SD: 3.23% ± 3.64% and 3.76% ± 4.14% for the no-reward and reward conditions, respectively. One-tailed t-test against chance, corrected for 2 comparisons, reward: t23 = 4.45, corrected p < 0.001, dz = 0.9; no-reward: t23 = 4.34, corrected p < 0.001, dz = 0.9). Importantly, decoding levels were not larger for the reward conditions compared to the no-reward conditions for any of the behavioral status distinctions, with similar results for both LO and pFs. A four-way repeated-measures ANOVA with reward (2), behavioral distinction (3), ROIs (2) and hemispheres (2) as within-subject factors showed no main effect of reward (F1, 23 = 0.34, p = 0.56) or interaction of reward and ROI (F1, 23 = 1.14, p = 0.29). No other main effects or interactions were significant (F < 3.15, p > 0.05). Overall, these results demonstrate that reward did not modulate the coding of the task-related behavioral status distinctions in LOC.
Conflict-contingent vs. visual category effects
An important aspect of the Target and High-conflict nontarget conditions in this experiment was that they both contained the same visual categories, which could be either a target or a nontarget (Figure 1B). Therefore, the Target vs. High-conflict nontarget pairs of conditions in our decoding analysis included cases where the stimuli in the two conditions were items from different visual categories (e.g. shoe and sofa following a ‘shoe’ cue), as well as cases where the two stimuli were items from the same visual category (e.g. shoe following a ‘shoe’ cue and a ‘sofa’ cue). We further investigated whether the representation in the MD network and in the LOC was driven by the task-related high conflict nature of the two conditions or by the different visual categories of the stimuli, and whether there was a facilitative effect of reward which is limited to the representation of the visual categories. For each participant, the decoding accuracy for this behavioral status distinction was computed separately for pairs of conditions in which the stimuli belonged to the same visual category (different cue trials), and for pairs in which the stimuli belonged to different visual categories (same cue trials). This analysis was conducted by selecting 180 voxels for both MD and LOC ROIs, to keep the ROI size the same. For both MD and LOC regions, there was no interaction with ROI or hemisphere, therefore accuracy levels were averaged across hemispheres and ROIs for the MD network and LOC (repeated measures ANOVA with reward (2), distinction type (2, same or different visual category), ROIs (7 for MD, 2 for LOC) and hemispheres (2, just for LOC) as within-subject factors: F < 3.15 p > 0.05 for all interactions with ROI and hemisphere). Figure 5 shows Target vs. High-conflict nontarget distinctions separately for same and different visual categories for no-reward and reward conditions, for both the MD and LOC.
We next tested for the effect of reward and distinction type (same or different visual category) on decoding levels in each the two systems. In the MD network, a two-way repeated measures ANOVA with reward (2) and distinction type (2) as factors showed no main effect of reward (F1, 23 = 0.92, p = 0.35) and no effect of category distinction or their interaction (F1, 23 = 2.9, p = 0.1; F1, 23 = 0.1, p = 0.8, respectively). These results show that there was no effect of reward on high conflict items that may be specific for the distinction between visual categories. In contrast, a similar ANOVA for LOC showed a main effect of distinction type (F1, 23 = 25.9, p < 0.001) and no effect of reward or their interaction (F1, 23 = 0.05, p = 0.8; F1, 23 = 0.46, p = 0.50, respectively). Together, these results demonstrate that representation was driven by visual categories in LOC, but not in the MD network. To further establish this dissociation between the two systems, we used a three-way repeated measures ANOVA with distinction type (2, same or different visual category), reward (2), and brain system (2, MD or LOC) as within-subject factors. There was no main effect of brain system (F1, 23 = 1.2, p = 0.3), allowing us to compare between the two systems. An interaction between distinction type and system (F1, 23 = 16.7, p < 0.001) confirmed that decoding levels in the two systems were affected differently by visual category. Critically to our research question, reward did not lead to increased decoding levels in either of the systems (no main effect of reward or interactions with reward: F < 0.77, p > 0.39).
Discussion
In this study we used a cued target detection task to test for the effect of reward motivation on the coding of task-related behaviorally-relevant category distinctions in the frontoparietal MD network as reflected in distributed patterns of fMRI data. Participants detected whether an item belonged to a cued visual category. Two visual categories served as either targets or nontargets, depending on the cue. A third category was never cued and therefore was never a target, creating three levels of behavioral status: Targets, High-conflict nontargets, and Low-conflict nontargets. During the cue epoch of reward trials, activity across the MD network increased, possibly reflecting increased cognitive effort when a context is set for a trial. Using MVPA, we showed that information about the behavioral status during the subsequent stimulus epoch of the trial was represented across the MD network. However, in contrast to our prediction, motivation, in the form of monetary reward, did not enhance the distinctions between the three behavioral status conditions across the MD network. Additionally, we did not find evidence for a selective facilitative effect of reward on discriminability of highly conflicting items (competition-contingent effect). In the LOC, information about the behavioral status of the presented stimuli was primarily driven by visual categories and was not modulated by motivation.
Previous reports showed an enhancement effect of motivation on overall activity in the frontoparietal control network (Botvinick & Braver, 2015; Dixon & Christoff, 2012; Padmala & Pessoa, 2011). Recently, it was demonstrated that cue decoding increased with motivation (Etzel et al., 2016), and in particular when task rules change from one trial to another (Hall-McMaster et al., 2019). Whether reward also modulates the representation of task-related information that is processed while cue and stimulus information is integrated remained unclear. These two effects of reward are complementary to one another, and are both key aspects of cognitive control and essential when reaching a decision. If reward enhances cue coding, then it would be reasonable to hypothesize that it may also facilitate the integration process of the cue and the subsequent stimulus that leads to successful completion of the task. Furthermore, previous studies have suggested that motivation particularly affects conditions of high conflict. Padmala and Pessoa (2011) reported a decrease in interference with reward in response inhibition tasks. Reward also reduced incongruency effect in the Stroop task compared to non-rewarded trials (Krebs et al., 2013) and enhanced error monitoring (Stürmer et al., 2011). Thus, we predicted that reward motivation will enhance the representation of task information, and that this effect may be specific for highly conflicting items. Based on our previous work that showed representation of behavioral status information across the frontoparietal cortex (Erez & Duncan, 2015), we used three behavioral status levels and their distinctions to test these predictions. While the overall representation of behavioral status across the MD network replicated our previous findings (Erez & Duncan, 2015), our results did not show an increase in representation with reward, in contrast to our predictions. Additionally, we did not observe a selective increase in representation for the highly conflicting items, namely Targets vs. High-conflict nontargets. Recently, Hall-McMaster et al., (2019) showed some increases in task-relevant stimulus features information when reward levels were high, using distributed patterns in EEG data. We did not observe such changes in our data, and the difference in results may possibly be due to differences in the design, as well as the limited time window where such differences were observed in EEG that cannot be detected with the low temporal resolution fMRI data. More generally, several reasons can provide potential explanations for the results obtained in our study, showing no facilitative effect of reward on pattern discriminability of behavioral status. Indeed, it is possible that the effect of reward in limited to cue decoding, as has been previously demonstrated (Etzel et al., 2016), and does not extend to the stimulus phase when information is processed based on the cue. However, we cannot rule out other possible explanations, including that our reward manipulation was not sufficiently strong to make a difference to pattern discriminability, insufficient power, and multiple factors that result in overall low decoding levels across the frontoparietal cortex (Bhandari, Gagne, & Badre, 2018) making small effects hard to detect with current MVPA methods.
Our predictions were based on the sharpening and prioritization account, which postulates that motivation leads to a sharpened neural representation of relevant information depending on the current task and needs. Previous neurophysiological evidence provide support for this aspect: reward has been associated with firing of dopaminergic neurons (Bayer & Glimcher, 2005; Schultz, Dayan, & Montague, 1997), and dopamine has been shown to modulate tuning of prefrontal neurons and to sharpen their representations (Ott & Nieder, 2016; Thurley, Senn, & Lüscher, 2008; Vijayraghavan, Wang, Birnbaum, Williams, & Arnsten, 2007). The prioritization aspect can be related to the expected value of control (EVC) theory (Shenhav, Botvinick, & Cohen, 2013) and reward-based models for the interaction of reward and cognitive control, essentially a cost-benefit trade-off (Botvinick & Braver, 2015). Cognitive control is effortful and hence an ideal system would allocate it efficiently, with a general aim of maximizing expected utility. Despite the appeal of this account, our results did not show experimental support for this view. At the behavioral level, we observed some evidence for such a benefit of reward. Accuracy levels of performance in the task for Target trials were higher in the reward compared to the no-reward condition. Additionally, while in the no-reward condition Target trials were less accurate than Low-conflict nontargets, in the reward condition there were no differences between them. We did not observe a similar benefit in reaction times. This could be due to the time threshold that we used for reward trials, which reduced the reaction time on all the reward trials, and may have masked an interaction with reward.
Relatedly, we did not observe univariate differences in activity between targets and the two nontargets across the MD network, and in contrast to previously reported data (Hampshire, Duncan, & Owen, 2007; Hampshire, Thompson, Duncan, & Owen, 2009). In our design, the targets were twice as frequent as High and Low-conflict nontargets, which could have resulted in lower activity of the targets compared to the other two due to a frequency effect (Braver, Barch, Gray, Molfese, & Snyder, 2001; Hampshire et al., 2009). Another possible univariate effect could have been a higher activation for more demanding condition, the High-conflict nontargets. The target, frequency and difficulty effects together could have led to no univariate differences between the behavioral status conditions, as also seen in our previous work (Erez & Duncan, 2015).
The visual categorization aspect of our task allowed us to investigate effects of reward on representation in LOC compared to the MD network, and in particular whether there is a specific effect of reward that is driven by visual differences. In the MD network, decoding levels were similar between conditions with the same visual category and different visual category, and there was no modulation by reward in any of them. This does not mean that there is no visual information in the MD network (Stokes et al., 2013), but rather that the we did not observe it in our fMRI study. In contrast, the discrimination in LOC was driven by the visual categories, as expected in the visual cortex, with Targets and High-conflict nontargets being discriminable only when items belonged to two different visual categories. While it is widely agreed that the frontoparietal cortex exerts top-down effects on visual areas, there is no clear prediction as to whether any effects of reward should be observed in the visual cortex. Our results provide evidence that the effects of reward were not present in LOC. Although previous studies have shown differences in representations between pFs and LO (Harel et al., 2014; Jiang et al., 2007; Li et al., 2007), our results were similar for both regions.
Although our primary question addressed the representation of task-related information during the integration of stimulus and cue, we also tested for an effect of reward in the cue epoch. The use of catch trials ensured that the cue and stimulus GLM regressors were appropriately decorrelated. The overall univariate activity across the MD network increased with reward during the cue epoch, possibly reflecting an increase in cognitive effort due to the reward. However, we did not observe cue decoding above chance, in contrast to previous reports (Etzel et al., 2016; Hall-McMaster et al., 2019). One reason for this difference in decoding results may be related to the design of our task. We used words of the category names as cues, which appeared together at the same time with the no-reward/reward indication – the reward trial cues had additional red pound signs. Our cues allowed for high task performance (compared to using abstract symbols, which is more difficult) rather than maximizing the decodability between cues. The visually salient reward signal that was presented simultaneously with the cues may have masked the decoding of the cue. Other reasons for the different results may be differences in the areas that were used for the analysis (Etzel et al., 2016), and differences in patterns that can be identified using fMRI and EEG data (Hall-McMaster et al., 2019).
In summary, we asked whether reward motivation leads to increased representation of task-related information across the frontoparietal network and the LOC. We found that information about behavioral status was present across the MD network. However, in contrast to the prediction based on the sharpening and prioritization account of reward effects, we did not find an increase in representation levels with reward. In the LOC, we observed representation of behavioral status, but this was driven by visual category information and was not modulated by reward. With growing interest in the interaction between control processes and motivation, our study provides an important experimental evidence for the underlying neural mechanisms and the potential limitations of the effects of reward, thus contributes another tier to the accumulating knowledge in the field which is critical for the development of computational models.
The authors declare no competing financial interests.
Acknowledgements
This work was funded by a Royal Society Dorothy Hodgkin Research Fellowship (UK) to Yaara Erez (DH130100). Sneha Shashidhara was supported by a scholarship from the Gates Cambridge Trust, Cambridge, UK. This work was also supported by the Medical Research Council (UK) Intramural Program MC-A060-5. We thank John Duncan and Daniel Mitchell for fruitful discussions and advice throughout the study.