Diminished neural responses predict enhanced intrinsic motivation and sensitivity to external incentive

The duration and quality of human performance depend on both intrinsic motivation and external incentives. However, little is known about the neuroscientific basis of this interplay between internal and external motivators. Here, we used functional magnetic resonance imaging to examine the neural substrates of intrinsic motivation, operationalized as the free-choice time spent on a task when this was not required, and tested the neural and behavioral effects of external reward on intrinsic motivation. We found that increased duration of free-choice time was predicted by generally diminished neural responses in regions associated with cognitive and affective regulation. By comparison, the possibility of additional reward improved task accuracy, and specifically increased neural and behavioral responses following errors. Those individuals with the smallest neural responses associated with intrinsic motivation exhibited the greatest error-related neural enhancement under the external contingency of possible reward. Together, these data suggest that human performance is guided by a “tonic” and “phasic” relationship between the neural substrates of intrinsic motivation (tonic) and the impact of external incentives (phasic). Electronic supplementary material The online version of this article (doi:10.3758/s13415-014-0324-5) contains supplementary material, which is available to authorized users.

Supplemental Results includes exploratory analyses further parsing the relationship between external incentives and intrinsic motivation.

Supplemental Data
• Figure S1 shows that duration of free choice time is not related to behavioral accuracy.
• Figure S2 shows that external incentive does not affect neural responses following correct responses.
• Figure S3 illustrates post-hoc analyses suggesting that the increased error reactivity conferred by external reward was evident in the High Intrinsic group and not in the Low Intrinsic group.
• Table S1 shows that increased duration of free-choice time is predicted by diminished neural response in regions associated with cognitive/affective regulation.

Remote associates word problems
Remote-associates-type word problems (Mednick and Mednick, 1967;Bowden and Jung-Beeman, 2003) were used in each phase of this study. Participants were presented with two words separated by a blank and were asked to produce a word that makes a common phrase or compound word both when it follows the first word and when it precedes the second word. For example, a solution to the problem "SUN _____ HOUSE" would be "LIGHT" because "sunlight" and "lighthouse" are both common words. Subjects were instructed to try to produce the most common solution. A behavioral pilot was conducted with 24 subjects to select the remote associates problems to be included in the study. 150 items were tested, and 126 singlesolution problems were chosen to be included. The final problem set consisted of 126 problems, and the mean accuracy of the pilot participants on the final chosen problems was 75.1%. The mean interest rating by the pilot participants of the included problems was 5.26 (Likert scale from 1 to 7; not at all, to very interesting).
During the scanning session, participants were presented with one remote associates problem at a time and entered their response using an on-screen keyboard (Figure 1b). A four button box was used in the scanner to navigate the on-screen keyboard, and all subjects practiced "typing" on the keyboard before the practice period until they were comfortable with the controls. Each remote associates problem was displayed until one of three criteria was met: the participant entered and submitted an answer, the participant chose to pass on the trial, or the time limit for the trial was reached. From the time that a problem was presented, participants had 60 seconds to provide and Marsden et al. Supplemental Material 3 submit an answer. If the subject was in the process of typing when the 60 seconds ran out, the program allowed them to continue until either they ended the trial with either a submission or pass, or no buttons were pressed for five seconds. At the end of each trial, the most common solution was displayed (e.g., " The most common answer is 'light'."). Feedback display time was jittered between 4 to 6 seconds. The items were presented in a random sequence, and no participant ever saw the same item more than once throughout the entire laboratory visit.

Procedure
A schematic diagram of the experimental procedure is displayed in Figure 1. When participants arrived, informed consent was obtained, and the experimenter provided instructions about the remote associates problems as well as scanner safety instructions. Upon entering the scanner, subjects completed several sample problems to verify that they understood the instructions and could easily navigate the keyboard. Throughout the task, instructions were both displayed on screen and read to the subjects. Before the practice period, all subjects received the same instructions and were told that they would have first have the opportunity to practice and "get used to the puzzles". After the practice period, individuals in the No Reward group were instructed that the main problem-solving period would begin and "Try to do your best and name as many of the common solutions as you can." Those in the Reward group were further instructed they could receive a bonus of "up to $20" depending on performance, and the experimenter would "let you know how much you have earned at the end." At the end of the performance period, the Reward group was told that they performed "well" on the puzzles and "earned an additional $15." The No Reward group was also told that they performed "well," but there was no mention of a reward. All participants then received the same instructions for the

Marsden et al. Supplemental Material 4
remainder of the experimental session. The study concluded with the free-choice period with participants remaining in the scanner. Subjects were instructed that an anatomical scan was being collected and "While you're waiting, you can do some more word problems, read recent articles from the New York Times, or just wait, and you can switch freely between these choices." Subjects used the button boxes to alternate freely among these options. During this period, we recorded the time subjects spent completing remote associates items, reading the newspaper, or resting. All participants received $35 dollars for their participation, regardless of group membership.

Behavioral analyses
The behavioral analyses focused on intrinsic motivation and task performance under the external contingency of possible reward (or not). As described above, intrinsic motivation was operationalized as the percentage of time each participant spent completing remote associates problems during the free-choice period, and the incentive treatment was categorical as described above, with a Reward group and a No-Reward group. Measures of performance included cumulative accuracy and response time, and accuracy and response time after incorrect and correct responses, respectively. Accuracy was computed as the number of remote associates problems that a participant answered correctly in the performance period divided by the total number of problems that the subject completed during the performance period. Response time was defined as the amount of time between when a problem was first displayed to when the participant either submitted an answer, pressed the pass button, or timed out, thus representing the amount of time that the subject spent on a problem. For accuracy following incorrect and

Marsden et al. Supplemental Material 6
Although the present paradigm was fully powered to detect significant neural and behavioral effects of intrinsic motivation (and did so), some of the follow up analyses may not have been sufficiently powered to detect significance at traditional alpha levels for interactions between intrinsic motivation and external rewards. For example, the data presented in Figure S3, while showing significant effects of Reward within the High and Low Intrinsic motivation groups, fell just short of traditional alpha values at the interaction level when two between group factors were included (p = .12). The effect size of the results shown in Figure S3 fell in the small to medium range (partial η 2 = 0.07), suggesting a small interaction effect that would require a larger sample size to reach alpha = .05. Thus, in this case, power may be limiting our ability to detect significant interactions, and interpretation of the associated results should be made with caution.
Future studies aimed to examine the effects suggested by the post-hoc analyses will be sufficiently powered.

Exploratory analyses
In this manuscript, we focus primarily on the regions involved in intrinsic motivation and how external incentives affected processing in these regions. Nonetheless, for the interested reader we have conducted several additional exploratory analyses to further parse the relationship between external incentives and intrinsic motivation.
These analyses revealed greater activity specifically in right inferior frontal gyrus (IFG) in the No Reward > Reward comparison at trial onsets (p < .001, uncorrected). This pattern suggested the hypothesis that individuals who showed greatest neural responses associated with both general intrinsic motivation and external incentives would show optimal performance on both free-choice time (associated with intrinsic motivation) and accuracy after incorrect responses (associated with the Reward condition). To test this possibility, we divided participants into (i) those who ranked above the median on both intrinsic motivation and Reward group neural responses (first component of the intrinsic motivation PCA and beta coefficients extracted from the Reward-related IFG ROI, respectively) and (ii) those who ranked below the median on both neural measures. This classification yielded 14 participants below the median on both external reward and intrinsic motivation neural responses and 13 participants above the median on both responses (recall that the total N = 40; 13 participants were discordant on the intrinsic and extrinsic motivation neural measures). Univariate ANOVAs revealed that the High and Low "Neural Intrinsic Motivation response + Neural Reward response" groups differed significantly on both free choice time (F(1,25) = 4.84, p = .037, partial η 2 = 0.31) and behavioral accuracy following incorrect responses (F(1,25) = 6.25, p = .019, partial η 2 = 0.2). For both comparisons, the 'Low' neural response group showed optimal performance (greater free choice time, and greater behavioral accuracy following incorrect responses. As detailed in the main text, diminished neural responses during task performance predicts increased intrinsic motivation; the additional No Reward > Reward comparison identified here shows that diminished activity in IFG (typically activated during response inhibition; Aron et al., 2014) distinguishes the Reward group. Together, these data suggest that during task performance, reduced neural responses in regions associated with neurocognitive/affective regulation may be a biomarker (e.g., the 'brakes are off') for those who are both intrinsically motivated and sensitive to the performance boosts that external incentives confer.

Marsden et al. Supplemental Material 8
To address this possibility, we generated a composite metric of general neural responses to intrinsic and extrinsic motivation for each participant (the sum of the beta coefficient from the Reward IFG ROI and the first component of the intrinsic motivation PCA); this metric showed a significant negative correlation with neural error reactivity (r = -.61, p = .0001). The negative relationship supports both the possibility of a 'composite biomarker' and our hypothesis (discussed in the manuscript) that diminished general neural activation facilitates enhanced neural and behavioral responses during critical task periods (here, post-error responding).
Although the present paradigm was not designed to evaluate the neural effects of obtained rewards (our Reward treatment simply informed participants they would be paid based on