Neurobiological research in autism spectrum disorders (ASD) has largely focused on social communication impairments, with much less attention on brain mechanisms that cause and maintain restricted and repetitive behaviors and interests (RRBIs) [1]. There is accumulating evidence indicating that both symptom clusters might be mediated, in part, by the same mesocorticolimbic system subserving reward-driven, motivational behaviors [2]. But the reward system’s role in both symptom clusters has not been systematically studied in a single sample of youth with ASD to date.

The social motivation hypothesis postulates ASD as a motivation disorder with affected persons preferring to explore and learn from the non-social environment at the expense of the social world [3]. The imbalance between motivation for social versus non-social stimuli is reflected in the responsiveness of the brain’s reward system [4]. Specifically, this hypothesis posits that some RRBIs may originate, in part, from the reward system being hyper-reactive for circumscribed interests (CIs) mostly within the non-social realm (e.g., mechanical and physical aspects of the environment), whereas social impairments may result, in part, from the reward system being hypo-reactive for socially rewarding stimuli and encounters such as social interactions with positive or non-negative affect [3]. Consequently, the developing child with ASD becomes deprived of crucial social learning opportunities, leading to aberrant social skill development and failed specialization of brain regions subserving social information processing [5].

Most recently, functional magnetic resonance imaging (fMRI) studies have started to address the neural system of reward responsiveness as a proxy for motivation functioning in ASD (see for a review, [6]). Although there is accumulating evidence of neural reward processing dysfunction in this population, findings are decidedly mixed, i.e., the direction and specificity of the deviations are inconsistent [2]. Thus, any firm conclusions are premature at this point. Several studies, however, report aberrant blood-oxygen-level-dependent (BOLD) responses of crucial reward regions in individuals with ASD, including ventral and dorsal striatum, ventromedial prefrontal cortex (vmPFC), anterior cingulate cortex (ACC), and orbitofrontal cortex (OFC), in response to both social rewards (e.g., smiling face, approval) and monetary rewards (e.g., gain of 0.50$) [7–11]. Findings of altered reward system responsiveness in ASD have generally been interpreted as potential neural signatures of decreased motivation to seek and appreciate these types of “conventional” desires.

In contrast, the pursuit of RRBIs, particularly CIs, are reported to be a source of pleasure by affected people [12], and the use of RRBIs as reward contingencies in behavioral modification programs has been found to be therapeutically effective [13]. Given the sporadic behavioral studies addressing the rewarding value of RRBIs, it is not surprising that there has only been limited neurobiological research on the reward system’s potential role in RRBIs [1].

Although the reward system may contribute, at least partially, to all forms of RRBIs [4], CIs may make the ideal candidate for investigating neural reward mechanisms of RRBIs—versus social impairments—in ASD: (1) The non-social quality of CIs stand in stark contrast to the diminished social interests of affected people (i.e., CIs usually are non-socially interesting, not shared, and negatively impact interpersonal relations [12]); (2) CIs can be compared to interests of typically developing controls (TDC); TDCs likely exhibit low rates of other RRBIs during the school age years (e.g., stereotypic body movements); and (3) In practical terms, CIs are relatively easier to re-create and measure in the MRI environment than other RRBI symptoms.

To date, two studies have investigated reward system responsiveness for CIs in ASD [14, 15]. Dichter et al. [6] applied a standardized set of stimuli hypothesized to be “autism-specific objects of high interest” (e.g., pictures of trains, computers) as reward outcomes; their presentation was contingent on accurate task performance in a reaction time task. The study revealed diminished ventral striatal activation for monetary reward in adults with ASD, which was accompanied by enhanced activation in vmPFC for autism-specific object rewards. Because Dichter et al. [6] used standardized object images rather than individualized items, it remains unknown whether the participants’ actual interests were covered by the stimulus set used. More importantly, the study did not contrast the interest rewards with social rewards. Thus, the social motivation hypothesis of ASD was tested only indirectly.

Cascio and colleagues [14] used a passive viewing paradigm with personalized CI pictures of each participant with ASD, focusing on youth rather than adults. The authors report heightened BOLD responses in anterior insula as well as mid-dorsal ACC—critical nodes of the “salience network”—when youth with ASD viewed their own versus others’ interests. The use of a passive task that did not require an active response to maximize reward outcome might explain the lack of group differences for CIs in reward circuitries. Reliable between-group activation differences within core reward regions (e.g., striatum) appear to be critically dependent on the requirement for an instrumental response [16].

For the current investigation, we leveraged the strengths of both prior imaging studies. Like Dichter, Felder and colleagues [6], we applied an incentive delay task to assess neural reward responsiveness with fMRI. Here, we chose a blocked design that has previously shown robust reward system activation in youth with vs. without ASD [7]. Existing imaging experiments targeting motivation in ASD relied on static pictures to serve as appetitive stimuli, which are, at best, only weakly rewarding and can fail to elicit motivational processes. Therefore, we employed dynamic stimuli that are perceived as more natural and engaging and, thus, may serve as more vital incentives [17]. To this end, for the current study, we developed a novel set of video clips of social and interest rewards [18]. More specifically, for the CI reward condition, we created individualized video stimuli for each participant based on self- and parent-reported CIs similar to Cascio and colleagues [14].

Thus, the present fMRI study aimed to compare the brain’s reward system responsiveness to individual CI rewards versus social rewards in youth with ASD relative to TDC. We expected enhanced neural signals in participants with ASD in response to their individual CI reward (in particular, in ventral and dorsal striatum as well as in vmPFC, ACC, and OFC), while neural activation would be reduced for social rewards. Additionally, to specifically test predictions derived from the social motivation hypothesis, we explored correlations between differential reward system responsiveness for CI reward versus social reward and ASD symptom severity.

The results of this fMRI study are in line with predictions by the social motivation hypothesis [3, 37]: Compared to TDCs, youth with ASD had stronger reward system responses for CIs mostly within the non-social realm (e.g., video games) than social rewards (e.g., approval). Additionally, aberrant reward system responses—most pronounced within the caudate nucleus-were related to greater social impairment. Behavioral task performance did not differ between the two study groups, emphasizing the power of neurobiological data in revealing atypical motivational mechanisms in ASD and at the same time avoiding a potential confounding factor [38].

The current investigation advanced the strengths of the two preceding imaging studies targeting CIs in ASD [14, 15] by (a) employing a newly created set of video clips as reinforcement in (b) a well-established instrumental reward task to (c) reliably contrast reward system responsiveness for individual CIs versus social approval rewards. Thus, we tested the social motivation hypothesis of ASD more directly, and we applied more ecologically valid reward stimuli. Using a standard set of static images hypothesized to be “autism-specific objects of high interest” versus monetary reward as incentives, Dichter and colleagues [6] previously reported blunted striatal responses to both reward types in adults with ASD, accompanied by enhanced vmPFC activation for object rewards. Standardized objects as a proxy for CIs, however, provide a potentially weaker estimate of the neural reward effects that could be identified with individualized CI items. Moreover, instead of simple static stimuli, we utilized, for the first time in this line of research, dynamic incentives that are commonly experienced as more naturalistic and appealing and, thus, may elicit both stronger and more reliable motivational responses [39].

The aberrant pattern of caudate activation in ASD not only supports prior research but also highlights a potentially fundamental difference in how individuals with ASD prioritize CI stimuli over social stimuli. Our finding of greater BOLD responses for CI rewards than social rewards in youth with ASD compared to a more balanced response pattern in TDCs substantiates the earlier report by Dichter and colleagues [6]. In contrast to our hypothesis, our follow-up analyses did not reveal specifically diminished caudate responses to social reward when directly comparing the two groups, at least with the social incentives used in this study sample. Other recent functional imaging studies, however, did reveal aberrant caudate responses to both social and monetary reward in ASD [7, 11]. Converging evidence from human and nonhuman research links the caudate nucleus directly to decision-making processes, specifically to the selection and initiation of purposeful actions in order to maximize reward outcome [40]. In this regard, we found that greater BOLD responses in the left caudate to CI rewards were associated with faster reaction times by the participants with ASD to obtain this positive outcome (but only at an uncorrected significance level). This suggests that in youth with ASD the caudate nucleus may optimize goal-directed actions for CIs to a greater extent than for social rewards or other types of “conventional” desires (e.g., money). Whether the caudate nucleus is also involved in aberrant instrumental responsiveness to primary rewards, such as food [41], should be investigated in future studies.

Notably, we detected that stronger caudate responsiveness for CIs (versus social reward) was related to greater overall symptom severity among our participants with ASD. Cascio and colleagues [14] recently reported that activation of the insula—as part of the “salience network”—in response to viewing pictures of one’s own vs. others’ interests was positively correlated with CI severity, but not overall clinical symptomatology, in youth with ASD. As noted in the Introduction, an instrumental task with active response-outcome contingencies is critical for stimulating reward pathways [16]. A passive viewing task may be less optimal for observing reward system responsiveness in relation to ASD symptoms. Thus, our findings emphasize the connection between the caudate nucleus as a core reward region involved in goal-directed behavior and the clinical phenotype of ASD (see also [42]).

Although several plausible models have been advanced to explain RRBIs in ASD (e.g., inhibitory control deficit), the relatively overactive caudate nucleus for CIs, as found in the present study, supports the idea that ASD may be in part a motivational disorder of “behavioral dependency” to RRBIs because of the rewarding effects they induce [4]. Different authors have argued that the rewarding effect of RRBIs, including CIs, may originate from the need of people with ASD for predictability, i.e., lawful and deterministic events in their environment, where they can exert more control [43]; rapid and uncertain moment-by-moment changes of a person’s behavior inherent in social encounters are the opposite. The unpredictable nature of social encounters may make them unrewarding or even anxiety-provoking and aversive for individuals with ASD [44]. RRBIs, such as CIs, in turn, may offer a pleasurable compensation for the unpredictable social world.

The exact mechanisms as to how the dominating reward effects of CIs emerge and interfere with the reward value of social engagement in ASD are, however, yet unclear. When CIs are indeed rewarding, their pursuit may be strengthened through self-reinforcement that turns them into rigid habits [45]. The self-reinforcing character of RRBIs, like CIs, may hijack the normal developmental trajectories of entire repertoires of behaviors, including social ones. We suggest that the caudate nucleus—in concert with other frontolimbic structures [46]—may dominate the formation and maintenance of RRBIs. On a daily basis, RRBIs hinder social development and functioning because they absorb resources typically dedicated to social learning opportunities [4]. This view converges with recent evidence demonstrating the role of the caudate nucleus, as part of a cortico-striatal-thalamo-cortical loop [47], in CIs as well as compulsive and ritualistic behaviors in individuals with ASD across different age groups [42, 48–51].

While we could demonstrate robust neural activation differences between ASD and TDC for CI rewards versus social rewards in dorsal striatum, other brain regions that have previously been related to CIs did not emerge in our study (i.e., ventral striatum/Nacc and vmPFC [15] as well as ACC and insula [14]). Despite our specific finding of an imbalance of caudate nucleus responsiveness in ASD, we acknowledge that CIs are not limited to a single anatomical correlate, but are mediated by various distinct yet interacting subcortical and cortical systems [1, 47]. Follow-up studies are warranted to better define how different subcomponents of reward processing—and their neural correlates—contribute to the emergence of and adherence to CIs. Incentive delay tasks, such as the one applied here, can principally be used to test reward learning, reward anticipation and reward valuation as three crucial reward components to consider with regard to CIs and the responsiveness to other “conventional” rewards in ASD [2]. While each reward component has been associated with some distinct (and some interrelated) neural correlates [52], recent human research highlights, however, the difficulty to effectively decompose the neural signals of the various components within a single experimental paradigm (see for a discussion on this issue, [53]). Thus, we foresee that refined inventories of specifically tailored reward measures—preferably grounded in preclinical studies and adequately validated in humans—will benefit this line of research.

Importantly, future fMRI investigations need to control more strictly for multiple comparisons at the whole-brain level. This will help avoiding high degrees of false positive findings and, thus, will ensure that neuroimaging results—particularly between-group findings—are more reliably reproduced. In the present study, we implemented most recent recommendations from the literature using whole-brain cluster thresholding that rigorously controls type I errors [32]; this could also explain why we were not able to replicate some of the previous findings (see Additional file 1: Table S3 for prior analytic approaches in this line of research).

This study had the following limitations: While we presented individually tailored interest clips, we did not utilize personalized social rewards, such as approval clips by caregivers, or preferred peers. Several recent investigations, however, indicate that personally meaningful social incentives hold similar reward value for youth with ASD as unfamiliar social rewards (e.g., [54, 55]); thus, it is unlikely that this potential confound fully accounts for the present findings. An imminent drawback to the use of unique interest items was their diversity among participants, which possibly added noise into the data. We endeavored to strictly equate stimulus properties across participants and conditions. Stimulus diversity (including differential complexity and luminance of video materials), however, may have precluded us from having enough statistical power to reveal additional regions involved in core reward mechanisms of CIs due to BOLD signal heterogeneity (e.g., Nacc, or vmPFC). Moreover, using a blocked design did not allow parsing brain responses for reward learning, anticipation, and consumption. Finally, because we explicitly targeted CIs, our findings are limited to this specific type of RRBI. A logical next step would be to extend this line of research to elucidate reward functions with regard to stereotypies, insistence on sameness, and sensory responses (e.g., [56]) using fMRI as well as other imaging methods, such as event-related brain potentials [57].

In conclusion, the results of this fMRI study add to the emerging clinical and imaging evidence of striatal involvement in the ASD pathophysiology [58]. More precisely, we demonstrated aberrant caudate nucleus responsiveness in ASD, with greater brain responses for CIs relative to social reward that was also associated with social impairment. We did not find, however, specifically diminished reward system responses to social reward when directly comparing youth with ASD to TDC, at least with the social rewards (i.e., approval from unfamiliar adults) used in the present study. We speculate that reward system dysfunction—most pronounced within the caudate nucleus—may contribute to enhanced motivation for RRBIs in ASD, accompanied by diminished motivation for social engagement. If a dysregulated reward system indeed supports the emergence and maintenance of social and non-social symptoms of ASD, then strategically targeting the role of reward mechanisms will allow for the development of more efficacious treatment practices to better support individuals with ASD and their families.