Spatial migration of human reward processing with functional development: Evidence from quantitative meta‐analyses

Abstract Functional magnetic resonance imaging (fMRI) studies have shown notable age‐dependent differences in reward processing. We analyzed data from a total of 554 children, 1,059 adolescents, and 1,831 adults from 70 articles. Quantitative meta‐analyses results show that adults engage an extended set of regions that include anterior and posterior cingulate gyri, insula, basal ganglia, and thalamus. Adolescents engage the posterior cingulate and middle frontal gyri as well as the insula and amygdala, whereas children show concordance in right insula and striatal regions almost exclusively. Our data support the notion of reorganization of function over childhood and adolescence and may inform current hypotheses relating to decision‐making across age.

A common issue with empirical findings is that studies comparing brain responses across different age groups are oftentimes inconsistent (Richards, Plate, & Ernst, 2013). For example, some show suprathreshold activity in the basal ganglia for children, and no suprathreshold activity in superior/medial frontal gyri for adults when they process rewards (Kappel et al., 2013), whereas, others have demonstrated less basal ganglia activity and more medial frontal cortex activity in late compared with early adolescence (e.g., Forbes et al., 2010). Heterogeneity of task design and complexity of the behaviors being studied have been identified as explanations for such inconsistencies (Richards et al., 2013 for review). Richards et al. (2013) also emphasize that the developing brain is a moving target, meaning that different systems of regions may develop with different trajectories (Giedd, 2004). For instance, the amygdala, hippocampus and insula are implicated in aversive behaviors and appear to follow a quadratic developmental trajectory (Arnett, 1999;Larson, Moneta, Richards, & Wilson, 2002;Silk, Steinberg, & Morris, 2003;Weinstein, Mermelstein, Hankin, Hedeker, & Flay, 2007), whereas executive regions such as the anterior cingulate and prefrontal cortex operate as an executive regulation subsystem (Haber & Knutson, 2010) that develops linearly with age (Casey, Jones, & Hare, 2008;Li, 2017;Marsh et al., 2006;Rubia et al., 2006;Rubia, Smith, Taylor, & Brammer, 2007). In summary, different developmental trajectories may explain variability among empirical findings.
Reward-related studies have proposed top-down modulation of subcortical regions via direct corticostriatal projections (Haber & Knutson, 2010), whereas others have demonstrated indirect modulation of the ventromedial prefrontal cortical regions (coding for valuation) by the executive subsystem within the dorsolateral prefrontal cortex (coding for self-control; Hare et al., 2009). This framework is consistent with studies revealing an increase in reward sensitivity and risky decision making during adolescence (Schneider et al., 2012;van Duijvenvoorde et al., 2014;. Perhaps reward-processing during adolescence may be the result of an executive system that is still under development (Prencipe et al., 2011;Steinbeis, Haushofer, Fehr, & Singer, 2014;Steinberg et al., 2009); an executive system that fails to maintain the balance between an overcompensating striatum and a diminished insula (Ernst & Fudge, 2009). Meta-analytic approaches on age offer a quantitative approach for addressing these types of examinations (e.g., Yaple and Yu, 2020).
Activation likelihood estimation (ALE), for example, evaluates concordance of brain coordinates reported across functional magnetic resonance imaging (fMRI) studies. Past ALE meta-analyses show that adolescents compared with adults exhibit greater concordance within the insula, ventral and dorsal striatum, amygdala and anterior and posterior and anterior cingulate cortex (Silverman et al., 2015; also see Bartra, McGuire, & Kable, 2013). The authors attributed their results to specific cognitive mechanisms associated with higher reward seeking behaviors in adolescents, since adults showed no activation greater than adolescents.
However, many studies included participants that were younger than 13 years in the adolescent group Schlund et al., 2010;van Leijenhorst, Crone, & Bunge, 2006;van Leijenhorst, Moor, et al., 2010;van Leijenhorst, Zanolie, et al., 2010) or included both children and adolescents (Christakou et al., 2011;Ernst et al., 2005;Jarcho et al., 2012;May et al., 2004). A comprehensive meta-analysis of functional brain correlates of children performing reward tasks has not yet been conducted and estimates of conjunction and disjunction of brain responses to rewards among children, adolescents, and adults are lacking.
In order to investigate concordance of brain correlates across studies and find overarching patterns in the literature we perform a series of quantitative ALE meta-analyses across data derived from children, adolescents and adults. We first examine data associated with general reward processing, to identify regions that engage in all reward-related functions. Based on the notion that the executive control system is still under development during adolescence, we expected our fMRI metaanalyses to reveal greater prefrontal and cingulate activity across studies for adults compared to adolescents and children, and no intact executive system in children. To further explore the role of each of the regions in the reward network, we also performed supplementary analyses on

| Literature search and article selection
Eligible studies were recovered from past meta-analyses of adults (e.g., Sescousse et al., 2013; n = 22 eligible) and adolescents (Silverman et al., 2015; n = 6 eligible studies). Literature from subsequent years was searched using the Web of Science database (http:// www.webofknowledge.com). Due to the vast number of rewardrelated studies in adults, we performed three independent searches using keywords: (a) "reward" AND "youth", (b) "reward" AND "children," and (c) "reward" AND "adolescents." These searches were Eligible articles included reward-related contrasts (e.g., reward anticipation, reward outcome, positive vs. negative feedback, etc.) to correspond with previous fMRI meta-analyses on reward processing (Diekhof et al., 2008;Mohr et al., 2010;Liu et al., 2011;Diekhof et al., 2012;Sacchet and Knutson, 2013;Sescousse et al., 2013;Montoya et al., 2014;Morellia et al., 2015;Wesley & Bickel, 2014;Silverman et al., 2015;Oldham et al., 2018). Exclusion criteria include articles that did not report whole brain fMRI coordinates in MNI or Talairach space, articles that did not report reward-related contrast associated with risky decision-making, delay discounting or feedback learning, and articles that did not report healthy human volunteers within specified mean ages for the following age groups: children (between 6 and 12.9 years), adolescents (13 and 17.9 years), and young adults (18 and 35 years). See Supplemental Tabes S1-S3 for the list of eligible articles included in the meta-analyses. Figures S1 and S2 for flowcharts showing the yield of the searches and the steps taken to screen and identify eligible articles for children/adolescents and adults.
Two authors independently selected articles meeting these criteria, and final decisions were made in agreement. The final dataset contained data from 18 eligible articles (28 experiments) for children, 29 articles (46 experiments) for adolescents, and 70 articles (90 experiments) for adults. Because our main between-group variable was age, we excluded studies that tested groups with large age-ranges (e.g., 18-70 years). Participant groups and foci included in the three meta-analyses were exclusive.
Several articles reported more than one relevant experiment, all of which were included in the analyses to improve statistical power, as the latest and currently recommended ALE analysis algorithm accounts for within-group effects (Turkeltaub et al., 2002 (Eickhoff et al., 2017). Reward subcategories associated specifically with reward anticipation did not fulfill the criterion of a minimum of 17 experiments for all age groups, therefore related results are reported in supplementary material. We also performed contrast analyses and computed conjunctions and differences among age groups.

| Software and analysis
We analyzed data coordinates using GingerALE, which is a freely available, quantitative meta-analysis method. This method was first proposed by Turkeltaub et al. (2002), with the latest version described by Eickhoff et al. (2009Eickhoff et al. ( , 2017. GingerALE (version 2.3.6) was used, which relies on ALE (http://brainmap.org/ale/). ALE compares foci from multiple articles and estimates the magnitude of overlap between foci, yielding clusters most likely to become active across studies. The most recent algorithm minimizes within-group effects and provides increased power by allowing for inclusion across all possible contrasts (Eickhoff et al., 2017;Turkeltaub et al., 2012). All coordinates were transformed into a common atlas space (Talairach) using the Lohrenz, McCabe, Camerer, and Montague (2007) transformation algorithm. Resulting statistical maps were thresholded at p < .05 using a cluster-level correction for multiple comparisons and a cluster forming threshold at p < .001 (Eickhoff et al., 2017). Contrast and conjunction analyses were also performed to compare differences and overlap across age groups, respectively. The threshold for groupcontrasts was set to p < .01 uncorrected for multiple comparisons (5,000 permutations, 50 mm 3 minimum cluster-size; e.g., Arsalidou et al., 2018), because group-contrast analyses use cluster-level thresholded ALE maps for each group, which have already been controlled for multiple comparisons.

| RESULTS
Data from a total of 3,444 participants were used for this study. Participant sample size and mean ages (± SD) in our resulting groups were 554 children with a mean age of 10.80 ± 1.48 (range: 6.9-12.5) years, 1,059 adolescent participants with a mean age of 14.82 ± 0.96 (range: 13.39-17.1), and 1,831 young adults with a mean age of 24.38 ± 2.52 (range: 19.6-29.9) years. Participants for each meta-analysis were 44.18, 59.53, and 55.45% male for children, adolescents and young adults, respectively.

| ALE maps
Tables 1-3 shows a complete list of concordant brain regions associated with reward processing with stereotaxic coordinates in Talairach space identified by all ALE meta-analyses by age group, conjunction and contrast analyses, respectively. Significant results were separated by age group and illustrated on Figure 1. Supplementary analyses on T A B L E 1 Concordant brain regions related to reward processing  Figure S3). Note that the latter two meta-analyses were performed by combining all age groups for the purpose of exploring brain maps associated with these events. Supplementary analyses revealed concordance patterns similar to the main meta-analyses: with exception of the insula, which shows no significant concordances during reward anticipation tasks.

| Post hoc analysis
To assess any systematic activity across different age groups we tested the frequency of foci reported with multiple bins associated with age for three key regions: the right dorsolateral prefrontal cortex, posterior cingulate cortex, and anterior cingulate cortex. We explored this relation by extracting foci from the raw data which fell within a 10 mm 3 radius of the peak cluster from the main meta-analysis. These values were then plotted in a histogram and viewed for changes across age

| DISCUSSION
In a series of quantitative meta-analyses we investigate concordance in brain responses to reward processing in children, adolescents, and adults. Specifically, we examine common and distinct executive and subcortical brain regions across different age groups. From these meta-analyses we reveal that: (a) children show concordance in subcortical regions, yet lack implication of brain regions associated with In general, these findings support the notion that all age groups recruit the subcortical system, yet differences by age group rely on brain areas associated with the executive system. This is the first study that examined concordant brain areas among children, adolescents and adults, which allowed us to assess the pivotal moments of implication of certain executive regions (dorsolateral prefrontal cortex, anterior cingulate cortex and the posterior cingulate cortex) by plotting the frequency of these regions across age. Specifically, development of the executive component of reward processing seems to involve two dependent components maturing at different developmental stages: a logical-reasoning component and a psychosocial/motivational component (Steinberg, 2007). Whereas adolescents are thought to attain adult-like reasoning by age 15, psychosocial abilities are thought to follow a more protracted linear development Luciana, Wahlstrom, Porter, & Collins, 2012;Steinberg, 2007;Steinberg et al., 2009). This mechanism has been illustrated in Figure 2. Consequentially, an executive system that is still developing, along with psychosocial factors may be hindered in adolescents who often make risky decisions in social settings, and thus brain responses of adolescents may be associated with more salient experiences of reward anticipation and the reception of reward outcomes (Chein et al., 2011). Throughout the following, we discuss the brains regions involved in reward processing in the attempt to emphasize their functional role in children, adolescents, and adults. The anterior cingulate cortex is a functionally heterogeneous region that is anatomically connected to various anterior and posterior regions (Vogt, Finch, & Olson, 1992) including the prefrontal cortex (Barbas, 2015;Ray & Zald, 2012;Yeterian, Pandya, Tomaiuolo, & Petrides, 2012), but also subsections of the cingulate including the subgenuate, pregenuate, postgenuate, dorsal anterior cingulate areas (Mao et al., 2017;Palomero-Gallagher et al., 2019;Stevens, Hurley, & Taber, 2011). The anterior cingulate cortex may be related to detection of prediction errors in monetary (Brown & Braver, 2005;Hauser et al., 2014;Holroyd & Coles, 2002;Garrison et al., 2013) and social contexts (Eisenberger & Lieberman, 2004;Lockwood et al., 2015;  . Moreover, the anterior cingulate cortex may play a crucial role in motivated social cognition (Apps, Rushworth, & Chang, 2016;Eisenberger & Lieberman, 2004;Hughes & Beer, 2012;Park et al., 2016;van der Molen et al., 2017;Wittmann, Lockwood, & Rushworth, 2018) perhaps by estimating the motivation of others and updating this information based on erroneous predictions (Apps et al., 2016).

| Dorsal anterior and posterior cingulate: Adolescents and adults
A specific methodological consideration is that adult task protocols may reflect higher demands compared with those used in younger children and should be regarded in the interpretation of the results. An alternative interpretation to the lack of cingulate activity in children may be that adolescents and adults may monitor performance and thereby experience error-related processing differently than children, such that children rely on model-free decision-making processing each trial more independently (Decker et al., 2016). This would support the notion that children lack specific cognitive abilities that would allow one to regulate decision-making (e.g., Arsalidou & Pascual-Leone, 2016).
In general, the anterior and posterior regions of the cingulate cortex are associated with the detection and monitoring of change or unexpected stimuli (Pearson et al., 2009;Pearson et al., 2011;Apps et al., 2012). Within the context of reward, while the anterior cingulate cortex is involved in the experience of pleasure or happiness (Lindgren et al., 2012;Matsunaga et al., 2016;Rolls et al., 2003;Suardi, Sotgiu, Costa, Cauda, & Rusconi, 2016), and value-guided decision-making (Holroyd & Coles, 2002;Kolling et al., 2016;Shenhav, Cohen, & Botvinick, 2016), the posterior cingulate cortex involves the monitoring of action-reward outcome associations (Hayden, Nair, McCoy, & Platt, 2008;Tabuchi et al., 2005). Together, anterior and posterior cingulate cortices have been associated with different aspects of motivation; the anterior cingulate processes motivational choices for complex cognitive tasks (i.e., decision-making) while the posterior cingulate processes self-referential motivational choices. Neurologically, the relative increase in cingulate foci reported in studies may be explained by cerebral developments at the onset of puberty such as pruning or redundant synaptic connectivity and myelination, which continue to develop into early adulthood (Giedd et al., 1999;Kelly et al., 2008;Rakic, Bourgeois, & Goldman-Rakic, 1994).
The dorsal striatum has been suggested to be involved in encoding of habitual learning (Patterson & Knowlton, 2018) and with learning new stimulus-reward contingencies (Knutson & Cooper, 2005;Rogers et al., 2004). Since dorsal parts of the basal ganglia have been implicated in processing rewards in children, adolescents, and adults we propose that these subcortical regions develop early with respect to cortical regions. This is consistent with the theory of constructive operators, which suggest that fundamental aspects of the Affective (A)-operator, housed in the limbic system, are ontologically and phylogenetically the first to develop (e.g., Arsalidou & Pascual-Leone, 2016;Pascual-Leone & Johnson, 2005).

| Amygdala: Adolescents, and adults
Large clusters that peaked over the caudate extended to the amygdala for both the adolescent and adult groups. We also find amygdala to be significantly concordant in the conjunction of these two groups. The amygdala is traditionally associated with emotional learning (Huff, Miller, Deisseroth, Moorman, & LaLumiere, 2013;Nieh, Kim, Namburi, & Tye, 2013) and processing of fear conditioning and anxiety (LaLumiere, 2014;LeDoux, 2000;Maren & Quirk, 2004;Nieh et al., 2013;Pape & Pare, 2010); however, it is also a key area of the mesolimbic dopamine reward system which projects to the nucleus accumbens during rewarding events (Nieh et al., 2013).

| Insula and claustrum: Children, adolescents, and adults
Laterally adjacent to the dorsal striatum are the claustrum and insular cortex, which were also found to be concordant across studies in all three age groups. Along with the anterior cingulate cortex, the insula is another region that activates to an array of cognitive, emotional and interoceptive events, to which some have suggested that these regions are key nodes in a salience network associated with responding to stimuli deserving of attention (Calder, Keane, Manes, Antoun, & Young, 2000;Calder, Lawrence, & Young, 2001;Menon & Uddin, 2010). In a coordinate-based meta-analysis it was revealed that the insula assumes multiple functions, anatomically portrayed as a topographic map (Kurth, Zilles, Fox, Laird, & Eickhoff, 2010). Specifically, the anterior-dorsal part of the insula was found to be associated with executive/cognitive functions, while the anterior-ventral part corresponds with social-emotional functions such as emotional processing and empathy. The idea that the insula may be related to motivated cognitive behavior has been proposed in earlier developmental studies of working memory (Yaple & Arsalidou, 2018) and mathematical cognition .
Some reward-related studies suggest that the insula is primarily involved in the processing of negative events (Phillips et al., 1998 (Camara, Rodriguez-Fornells, & Münte, 2009;Choi, Padmala, Spechler, & Pessoa, 2014). Systematic reviews on reward processing have suggested that the insula responds to expectation of rewards (Knutson & Bossaerts, 2007;Knutson & Greer, 2008;Liu et al., 2011;Moreira, Pinto, Almeida, Barros, & Barbosa, 2016), yet other studies have found that the insula responds to reward anticipation as well as reward delivery (Boecker et al., 2014;Liu et al., 2011;Padmala & Pessoa, 2011;Samanez-Larkin et al., 2007). To address this inference, we emphasize the results of the supplementary meta-analyses on reward anticipation and reward outcome across age groups, revealing concordant activity within the insula for reward anticipation, but not reward outcome. This supports the notion that insula may not necessarily be functionally associated with observing reward outcomes (See Table S4; Figure S3).
Interestingly, previous meta-analyses on cognitive abilities in children revealed concordant right-lateralized insula cortex activity, suggesting that right insula cortex activation is essential for problem solving (Yaple & Arsalidou, 2018;Arsalidou et al., 2018). In the current research, we found this region to be highly significant, especially within the meta-analysis on reward anticipation (See Table S5; Figure S3). Because the insulae is implicated in different constructs related to rewards as well as other qualitative different tasks

| Limitations
Our meta-analyses evaluated coordinates from fMRI studies that examined reward processes in children, adolescents, and adults. To achieve sufficient power for the analyses we cotableh study heterogeneity. To this regard, we omitted contrasts that included monetary losses to specifically focus on reward processing. In addition, we separately performed secondary analyses on reward outcomes, reward anticipation and a task-relevant dataset across all age groups (see Supplementary Materials section). We had initially considered performing separate meta-analyses on losses, risk taking and delay dis- counting; yet the number or reported articles were insufficient. When further data becomes available, future meta-analyses can address specific questions related to these processes across age.
Further, the number of studies considered for each age group was different with the least number of studies in the children group; albeit all age-related analyses reported in the main text adhere to minimum experiment requirements for sufficient statistical power (Eickhoff et al., 2017). These are a main disadvantage of performing meta-analyses across age groups, however, as this is the nature of tasks variability in reward processes in the literature, it is the state of the art. Optimally, future developmental studies should consider parametric tasks with a common goal but variable levels of difficulty to ensure that individuals with variable performance levels can complete the task (e.g., Arsalidou & Im-Bolter, 2017). Finally, many studies were not included in our meta-analyses due to the relatively wide in range in age. We encourage future research in this field to focus on discrete or narrower age ranges, as opposed to studies using a wider age range to allow for improved option for determining the relative shifts in brain activity throughout development.

| CONCLUSIONS
In these large-scale meta-analyses with a total of 554 children, 1,059 adolescents, and 1,831 adults, we showed that all age groups yield consistent activity in the striatum and the insula. Children lack concordant activation of regions implicated in associative "higher-order" regions.
Across studies, adolescents engage the right dorsolateral prefrontal cortex, a key region involved in executive control, whereas adults show concordance in anterior cingulate cortex but no concordant activity within the dorsolateral prefrontal cortex. Our findings suggest that these executive regions undergo dramatic changes across adolescence through to adulthood. These findings coincide with the notion that these executive regions may develop twofold: distinguished by dorsolateral prefrontal cortex concordance in adolescents representing the development of executive control processing at around 15 years of age, and anterior cingulate cortex concordance signifying later development of psychosocial abilities in early adulthood.

ACKNOWLEDGMENTS
We gratefully acknowledge support in part from the Russian Science

DATA AVAILABILITY STATEMENT
Development of reward processing: Over-arching brain clusters in children, adolescents, and adults. The data that support the findings of this study are openly available in OSF at DOI 10.17605/OSF.IO/5XUQW Data sharing is not applicable to this article as no new data were created or analyzed in this study.