Abstract
Dopamine release in cortical and subcortical structures plays a central role in reward-related neural processes. Within this context, dopaminergic inputs are commonly assumed to play an activating role, facilitating behavioral and cognitive operations necessary to obtain a prospective reward. Here, we provide evidence from human fMRI that this activating role can also be mediated by task-demand-related processes and thus extends beyond situations that only entail extrinsic motivating factors. Using a visual discrimination task in which varying levels of task demands were precued, we found enhanced hemodynamic activity in the substantia nigra (SN) for high task demands in the absence of reward or similar extrinsic motivating factors. This observation thus indicates that the SN can also be activated in an endogenous fashion. In parallel to its role in reward-related processes, reward-independent activation likely serves to recruit the processing resources needed to meet enhanced task demands. Simultaneously, activity in a wide network of cortical and subcortical control regions was enhanced in response to high task demands, whereas areas of the default-mode network were deactivated more strongly. The present observations suggest that the SN represents a core node within a broader neural network that adjusts the amount of available neural and behavioral resources to changing situational opportunities and task requirements, which is often driven by extrinsic factors but can also be controlled endogenously.
Introduction
In our everyday behavior, we are guided by environmental events that interact with our internal goals. For external events, numerous studies have shown that increased processing resources are allocated to certain stimuli because of their inherent biological relevance. Most prominently, reward-predicting stimuli encourage animals to pursue cognitive and behavioral operations necessary to receive the reward, and successful operations are reinforced. These effects have been linked to responses in mesencephalic dopaminergic areas [i.e., substantia nigra (SN) and ventral tegmental area (VTA)] that influence processing in their target structures in the basal ganglia and in cortical control areas (Wise, 2004; Pessiglione et al., 2006; Schultz, 2007; Robbins and Arnsten, 2009).
Pharmacological and lesion studies have indicated that disruptions of the dopaminergic system interfere with these activational aspects, whereas consummatory behavior is not affected (for an overview, see Salamone, 2009; see also Zink et al., 2004). Therefore, the role of dopamine in the context of reward has been conceptualized as being related to activation or incentive salience (Wise, 2004; Berridge, 2007; Robbins and Everitt, 2007). Consistent with this view, Parkinson's disease (PD) in humans has been linked to apathy, which can be counteracted by dopaminergic medication (Czernecki et al., 2002). Moreover, PD patients display deficits in a host of effortful executive functions (for an overview, see Nieoullon, 2002).
In addition to reward-related responses, dopaminergic midbrain areas have also been reported to react to environmental stimuli that are novel or otherwise perceptually salient (Redgrave et al., 1999; Horvitz, 2000; Zink et al., 2003; Bunzeck and Düzel, 2006). Crucially, all of these stimulus characteristics can be subsumed as extrinsic. Despite the large body of evidence for dopamine-system involvement in extrinsically driven resource allocation, it is as yet unknown whether the dopaminergic system can also be recruited in the absence of extrinsic factors. Importantly, although such endogenously driven recruitment of this area has been suggested (Nieoullon, 2002; Salamone et al., 2005; Braver et al., 2007; Arnsten et al., 2009), very little direct evidence for this notion has been provided thus far (Robbins and Everitt, 2007).
To test the hypothesis that dopaminergic midbrain areas also play an activating role in situations that do not entail extrinsically motivating or salient factors, we performed an fMRI experiment in which human observers performed a cued visual discrimination task with two task-demand levels. In doing so, we tried to closely parallel typical reward experiments in humans, which have observed activations in the dopaminergic midbrain areas (Knutson et al., 2005; Wittmann et al., 2005; Adcock et al., 2006). Many of these experiments used variants of the well established monetary-incentive delay task, in which a precue predicts whether reward can be obtained for fast and accurate performance in a subsequent simple task (e.g., detecting a target) (Knutson et al., 2001). In contrast, precues in the present experiment only predicted whether the upcoming task would be relatively easy or difficult, thus conceptually replacing reward prediction with the prediction of task demands.
Materials and Methods
Participants.
Fifteen subjects participated in this experiment after providing written informed consent. Two subjects were excluded from further analysis because of excessive head motion (exceeding 1° and/or 0.5 mm in abrupt movements in more than half of the runs); and an additional subject was excluded because of insufficient anatomical coverage of the midbrain regions of the functional images, leaving 12 subjects for the main analyses (mean age, 26 years; five females). Note that the inclusion of female subjects might have introduced some variability, since dopamine levels are known to vary across the menstrual cycle (Fernández-Ruiz et al., 1991), which will be addressed in future studies. The experiment was approved by the ethics committee of the Otto-von-Guericke University, Magdeburg, Germany, and subjects were paid for participation.
Paradigm.
Each trial started with a central blue or green arrow indicating the task demands of the forthcoming trial (high vs low task demands; cue-colors were counterbalanced across subjects). Cues were presented for 200 ms, followed 1000 ms later by a stimulus array (Fig. 1A). The task display consisted of two symbols synchronously presented (30 ms) in a rectangle centered 4° below and 8° right of the fixation spot. The fixation spot and the rectangle were constantly visible throughout the experiment. The stimuli were taken from an array of seven geometrical figures that were either red or white (Fig. 1B). In 60% of the trials, two randomly chosen nonidentical white figures were presented together, which represented nontargets under both conditions. In 20% of the trials, the stimulus array consisted of two randomly chosen identical white figures (targets under the high-demand condition), and the remaining 20% of the trials consisted of two randomly chosen red stimuli (targets under the low-demand condition; no shape discrimination required). Thus, the low-demand task required detecting a deviant color, whereas the high-demand task required comparing the shape of two symbols, which requires a deeper level of processing. The task (high vs low task demands) alternated in short blocks (∼25 s) to minimize the task-switching load. This, however, was only done to facilitate cue interpretation; each trial was individually cued and data analysis was performed in a fully event-related fashion. Intertrial intervals (ITI) were varied pseudorandomly between 2 and 6 s through inclusion of null events (using a geometrical distribution biased toward smaller ITIs) to allow for the separation of the different trials in an event-related fashion (Dale, 1999; Hinrichs et al., 2000; Liu, 2004). For the design matrices used in this experiment, correlations between the conditions of interest were −0.48 (averaged across runs, ±0.02), indicating that whenever signals were high in one condition they were low in the other condition; thus good separation of trial types was ensured.
Both potential target types (relevant and irrelevant ones) appeared with equal probability after both cue colors, but subjects were only to respond to the targets of the currently cued task (right index finger button). The occurrence of both potential target types after both cue types ensured that subjects had to be aware of the preceding cue to perform the task correctly. This paradigm allowed us to compare fMRI activity in response to physically identical nontarget trials under different levels of task difficulty in the absence of any motor response. Additionally, eye movements were monitored using an MR-compatible infrared recording system (Kanowski et al., 2007). In general, subjects maintained very accurate fixation and no differential effects of eye-movement across conditions were observed.
fMRI data acquisition.
fMRI data were acquired on a Siemens 3-T Trio scanner using an eight-channel phased-array whole-head coil (Siemens); six runs with 169 volumes covering all but the uppermost part of the subjects' brains (MNI z-dimension after normalization, −30–56 mm) were recorded (repetition time, 2 s; echo time, 30 ms; flip angle, 80°; 128 × 104 × 24 at 2 × 2 × 3.6 mm). Additionally, an inversion-recovery EPI (inversion time, 1500 ms) and a proton-density-weighted image were acquired.
fMRI data analysis.
The first six volumes of each run were discarded from the analysis. The remaining functional images were analyzed with SPM5 (http://www.fil.ion.ucl.ac.uk/spm/). Preprocessing included slice-time acquisition correction, realignment, spatial normalization, and smoothing (6 mm full width half maximum). Normalization parameters were derived from the segmented inversion-recovery EPI using standard procedures with 12 linear and 16 nonlinear transformations and medium regularization. This deformation matrix was then applied to all other volumes. Since the accuracy of standard normalization procedures for midbrain regions has been questioned (D'Ardenne et al., 2008; but see Düzel et al., 2009), we interactively checked the alignment of each subject's individual SN in the normalized PD image with the SN in the PD template provided by SPM, which were consistently in good register. After preprocessing, all experimental conditions were modeled with the canonical hemodynamic response function plus time and dispersion derivatives in an event-related model. A high-pass filter of 128 s and an autoregressive integrated moving-average model for serial correlations (equivalent to a low-pass filter) were applied to account for high- and low-frequency artifacts in the data. Task-relevant and task-irrelevant target trials and false alarms were modeled separately. The contrast of interest was calculated (high-demand > low-demand nontarget trials) for each subject and entered into a random-effects group analysis (one-sample t test). Statistical analysis was performed using cluster-based thresholding (p < 0.05 corrected cluster-size threshold with an auxiliary voxel-level threshold of t > 4.5; p < 0.0005). Results within midbrain regions were visualized by overlaying the results on the average spatially normalized proton-density weighted image of the participants using MRIcron (http://www.cabiatl.com/mricro/), on which the SN is visible as a bright stripe of increased image intensity. Additionally, whole-brain renderings of the activity maps were produced using Surfrend (http://spmsurfrend.sourceforge.net/), which were overlaid on an inflated whole-brain template using Freesurfer (http://surfer.nmr.mgh.harvard.edu/).
It has recently been argued that whole-head group-level fMRI approaches are problematic for the investigation of some midbrain areas (D'Ardenne et al., 2008; but see Düzel et al., 2009). Specifically, D'Ardenne and coworkers (2008) concluded that standard spatial resolutions and normalization strategies are insufficient to characterize hemodynamic activity in the VTA. To ameliorate such problems, we optimized our data acquisition and analysis approach to investigating midbrain activity (including small voxel sizes/high-resolution matrices and small smoothing kernels); results were overlaid on the average of the spatially normalized proton-density weighted structural scans from the same subjects, which provided excellent anatomical contrast for those midbrain areas. More importantly, however, the present study did not attempt to separate activations in the VTA from activations in the SN. In humans, these two structures build a continuous dorsal tier, which is much larger than the VTA alone (Ahsan et al., 2007), that projects to the striatum and cortical areas (Haber and Knutson, 2010). Moreover, such a distinction may not be crucial here, because in humans (in contrast to rodents), most of the dopaminergic cells are located within the SN rather than the VTA (Björklund and Dunnett, 2007).
Pupil dilation experiment.
Pupil dilation has been found to be a reliable indicator of task demands (Cabestrero et al., 2009) and can thus provide an additional window into the perceived level of task demands. Because such an analysis was not possible based on the eye-tracking data that was acquired during fMRI scanning, we ran an independent sample of subjects while recording their pupil sizes outside of the scanner (Eyelink 1000; SR-Reseach). Ten subjects were run (average age, 22.6 years; six females), using the identical experimental setup, and again counterbalancing cue colors between subjects. The resulting data were averaged across trials and baseline corrected with respect to a 200 ms precue baseline. Results for the low-demand and high-demand conditions were then compared using a sliding one-sample t test (p < 0.05, two-tailed).
Results
Behavioral results
The present experiment entailed two levels of task-demand that were precued by a colored arrow (see Materials and Methods, above) (Fig. 1). Following a short interval, two symbols were simultaneously flashed and subjects had to either compare the shapes (high task demands) or detect a deviant color (low task demands). No explicit performance feedback was provided to avoid indirect trial-based motivational factors (in addition to avoiding direct reward), and subjects were told in advance that their performance would have no influence on their compensation. Targets were rare, and the more frequent nontargets were identical under both conditions. This experimental set-up served two important functions: the target trials provided a measure of the task-demand manipulation, which was indeed very effective [high/low, 71% vs 98% correct responses; p < 0.001; response time (RT), 747 vs 560 ms; p < 0.001] and, even more importantly, it allowed us to focus our fMRI analysis entirely on nontarget trials (additionally excluding false-alarm trials, <3% of all trials) that did not entail motor responses. This approach excluded processes related to response execution that could also influence midbrain activity. Furthermore, the nontarget stimuli were identical under both instructions, so that fMRI effects also cannot be accounted for by any difference in visual stimulation. Hence, the fMRI results reported below are only related to the different levels of cognitive demand and effort during nontarget trials.
Pupil dilation results
To further corroborate our notion that the behavioral differences between our main conditions indeed reflect differences in perceived task demands, we sought to provide an additional independent measure. One such measure is pupil dilation, which is found to increase with increasing task demands (Cabestrero et al., 2009). We recorded such data from an independent set of subjects (n = 10) outside of the scanner. The behavioral data of this group was similar to that of the main experiment (accuracy, 61% vs 95%; RT, 614 vs 478 ms; both p < 0.001). Figure 2 illustrates the grand-average pupil dilation results separated into the low-demand and the high-demand condition. It is obvious from this plot that pupil dilation was relatively more pronounced for the high-demand task, and that this effect was biphasic, showing clear differences both in response to the cue and to the targets. On an absolute level, pupil size appeared to shrink starting ∼400 ms after both stimuli (presumably in response to the bright visual input), but at the same time, the relative effects between high- and low-demand trials are very pronounced and significantly different over a wide time range. Note that high- and low-demand trials give rise to differential effects somewhat earlier than one would expect based on the known slow dynamics of the pupil response. It is conceivable that this is partly due to overlap from the preceding stimulus because the timing of the different events was optimized for the analysis of fMRI data, which is not optimal for the analysis of pupil responses.
fMRI results
When comparing high minus low task-demand trials, our group-level results revealed enhanced bilateral cortical fMRI activity in the anterior cingulate cortex, the pre-supplementary motor area, the anterior insula extending into the frontal operculum, the dorsolateral prefrontal cortex, and in parietal and lateral occipitotemporal structures (corrected cluster size p value < 0.05; with an auxiliary voxel-level threshold of t > 4.5; p < 0.0005) (Fig. 3, red–yellow scale; Table 1). Additionally, large parts of the default-mode network (Raichle et al., 2001; Greicius et al., 2003), including the posterior cingulate cortex and the angular gyrus bilaterally, were deactivated more strongly for high than for low task demands (Fig. 3, blue–cyan scale; Table 2).
Most importantly, we observed enhanced fMRI signals in the high-demand task minus low-demand task bilaterally in the SN [local maxima at MNI (x/y/z): 8/−20/−14 (t = 12.55) and −6/−18/−16 (t = 6.55)] (Fig. 4), in line with our prediction. Further subcortical modulations were observed in the left ventral striatum and different parts of the thalamus. These results corroborate our hypothesis that dopaminergic midbrain structures are involved in the recruitment of processing resources even in the absence of any rewards or other salient extrinsic factors.
Discussion
Here we demonstrate, for the first time, that activity in dopaminergic midbrain areas can occur in the complete absence of extrinsic factors such as reward or salient perceptual stimulus properties. Specifically, we found enhanced hemodynamic activity in the SN for nontarget trials under conditions of high compared with low task demands. The importance of this finding is underscored by current models on dopamine function that assign dopaminergic inputs a pivotal role in driving activity in frontal control regions and corticobasal ganglia processing loops related to a wide variety of executive functions (Seamans and Yang, 2004; Frank and O'Reilly, 2006; Braver et al., 2007).
In the present paradigm, the neural modulation by task demands in the SN clearly indicates that this region can be recruited volitionally in the absence of reward or highly salient stimuli, presumably via projections from prefrontal cortex (Moore et al., 1999). This finding suggests that motivational systems that are driven by extrinsic and intrinsic motivating factors overlap on the level of dopaminergic midbrain areas and functionally associated structures (most prominently the striatum, plus medial and lateral prefrontal regions). Despite this substantial overlap of neural networks, the mechanisms by which the dopaminergic midbrain areas are recruited may be distinct. Responses of dopaminergic neurons to extrinsic factors such as reward or salience are likely mediated by the superior colliculi (Dommett et al., 2005) or limbic areas, in particular the orbitofrontal cortex and the amygdala (Mohanty et al., 2008; Croxson et al., 2009; Schoenbaum et al., 2009). In contrast, volitional recruitment of the dopaminergic areas is more likely driven by higher-level prefrontal control areas (Cho and Strafella, 2009).
Although the present work focuses on dopaminergic midbrain areas, the activations observed here arise from the interplay with a broad network of cortical and subcortical regions. Some of these areas are presumably instrumental in operations that are specific to the two tasks that were compared (comparing the shape of two stimuli vs detecting a deviant color). Such specific operations, however, are unlikely to underlie the activity in high-level control areas and even less likely in the SN. Moreover, many of the identified frontal regions (including anterior–insular together with lateral and medial frontal areas) have been suggested to be part of a system that generally responds to situations of increased cognitive demand (Duncan and Owen, 2000; Dosenbach et al., 2006). Based on resting-state connectivity combined with personality-trait data, a further subdivision of this demand system into an executive-control network and a salience network has been recently suggested (Seeley et al., 2007). The latter is implicated with detecting relevant stimuli, and the SN is assumed to be a part of this network that is processing reward-related and other salient stimuli. In the context of this distinction, the present data indicate that the role of the SN goes beyond being part of a salience network.
The above findings of enhanced hemodynamic activity during high task demands were complemented by substantially stronger deactivations in large parts of the default-mode network (Raichle et al., 2001; Greicius et al., 2003), which is in accord with suggestions that dopamine is involved in task-related deactivations of this network (Argyelan et al., 2008; Engelmann et al., 2009; Tomasi et al., 2009). Together, the dopaminergic system appears to impact multiple neural targets, suggesting that its widespread innervations of many crucial control areas (Björklund and Dunnett, 2007) can be used to promote a state of enhanced availability of cognitive and behavioral resources. From a computational perspective, in particular with respect to the frontal cortex, it has been suggested that dopamine modulates the response to incoming information (Servan-Schreiber et al., 1990; Braver and Cohen, 2000; Seamans and Yang, 2004). Such a mechanism is assumed to effectively enhance the signal-to-noise ratio, but may come at the expense of higher energy demands and less influence of unattended inputs (Servan-Schreiber et al., 1990) and is therefore likely to be used in an on-demand fashion only (Robbins and Arnsten, 2009). This putative gating mechanism might be of particular relevance in situations where proactive preparation is possible (Braver et al., 2007), as in the present experiment.
On a conceptual level, our results imply that neural activity in mesolimbic reward regions observed in reward-prediction experiments, which are typically attributed directly to the reward-predicting properties of stimuli, could in fact be (at least partly) related to the cognitive effort that is exerted to achieve the reward—two factors that are inherently hard to dissociate (Maunsell, 2004). In this context, it is important to note that, because of the total absence of any direct reward or punishment, which are integral to most animal studies, the present task could only be performed by human participants. Moreover, although a direct paradigmatical combination of task difficulty and reward is appealing (Engelmann et al., 2009), it was mandatory for the purpose of the current study to avoid a reward context altogether, since neutral trials in the context of rewarded trials can be perceived as disappointing or might be avoided in a choice situation (Croxson et al., 2009; McGuire and Botvinick, 2010).
In accord with our findings, pharmacological and lesion studies in animals have indicated a central role of dopamine in the propensity to exert (physical) effort to achieve rewards (for review, see Salamone et al., 2005). Moreover, fMRI experiments have found that modulations of reward-related activity within one of the key dopaminergic target structures, the striatum, strongly depend on whether subjects have to perform a simple detection task to obtain a reward (Zink et al., 2004; Bjork and Hommer, 2007). These observations imply a more general role of this SN signal, possibly related to the amount of effort exerted (Wise, 2004; Salamone et al., 2005). Importantly, effort in animal experiments is usually operationalized in choice settings that require physical labor, thus putting it at some distance to the present study. However, more closely related studies in humans have indicated a similar role for the involved structures in tasks that invoke a form of effort that is rather cognitive in nature (Zink et al., 2004; Bjork and Hommer, 2007).
Evidently, modulations of fMRI signals in dopaminergic midbrain areas should not be equated directly with dopamine release in their target structures. Nonetheless, several observations corroborate the notion that the observed SN activity is indeed linked to enhanced dopaminergic neurotransmission: attentional deficits (especially concerning top-down, endogenous attention) have been reported for PD patients, along with deficits in other executive functions (Brown and Marsden, 1988; Yamaguchi and Kobayashi, 1998; Nieoullon, 2002; Cools, 2006). In accord with these findings, pharmacological and lesion studies in animals reported influences of the dopaminergic system on functions related to the recruitment of attentional resources (for review, see Robbins, 2005), and pharmacological investigations in humans have also yielded corroborating results (Clark et al., 1987; Servan-Schreiber et al., 1998; Bullmore et al., 2003). Moreover, demonstrations of a strong correlation between reward-related mesolimbic fMRI responses and reward-related striatal dopamine release in the same human subjects ([11C]raclopride displacement PET) suggests a strong relationship between the two measures (Schott et al., 2008).
The current findings extend our knowledge about the processes that underlie the recruitment of neural processing resources and thereby have implications on the understanding of major neurological and psychiatric disorders. Specifically, deficits in top-down resource recruitment have not only been described in PD patients, but also in attention deficit hyperactivity disorder, schizophrenia, autism, and drug abuse—conditions that have all been related to disturbed dopaminergic neurotransmission (Nieoullon, 2002; Everitt and Robbins, 2005). Additionally, cognitive deficits in normal aging have been linked to a decline of dopaminergic cells (Arnsten et al., 1995).
In summary, we conclude that dopaminergic midbrain areas play a central role in the control of neural processing resources beyond the activation by extrinsic motivating factors. Specifically, it appears that these areas can be recruited in a top-down fashion to adjust the amount of available cognitive and motor resources to be able to meet changing situational demands and opportunities without wasting energy when such resources are not required.
Footnotes
- Received September 14, 2010.
- Revision received February 8, 2011.
- Accepted February 10, 2011.
This work was supported by grants from German Research Council (BO 3345/1-1 to C.N.B., SFB-TR31/TPA8 to T.N., and SFB 779 to M.A.S, J.M.H., and H.J.H.), and the German Ministry for Education and Research (Contract 01GO0202) to the Center for Advanced Imaging, Magdeburg. We thank Dr. Michael Scholz for technical support, Johanna Starke for assistance with data collection, and Joseph A. King for valuable comments.
- Correspondence should be addressed to either Dr. Carsten Nicolas Boehler, Center for Cognitive Neuroscience, Box 90999, Duke University, LSRC Building, Room B203, Durham, NC 27708-0999, c.n.boehler{at}gmail.com; or Dr. Tömme Noesselt, Institute of Psychology, Otto-von-Guericke-Universität, PO Box 4120, 39116 Magdeburg, Germany, E-mail: toemme{at}med.ovgu.de
- Copyright © 2011 the authors 0270-6474/11/314955-07$15.00/0