This article investigates the functional connectivity patterns of the nucleus accumbens (NAcc) in 18 healthy participants using a resting state functional connectivity (rsFC) protocol. Also, a meta-analytic connectivity modeling (MACM) was used to characterize patterns of functional coactivations involving NAcc: The results of a structure-based meta-analyses of 57 fMRI and PET studies were submitted to activation likelihood estimation analysis to estimate consistent activation patterns across the different imaging studies. The results of the combined rsFC and MACM analyses show that spontaneous activity in NAcc predicts activity in regions implicated in reward circuitries, including orbitomedial prefrontal cortex, globus pallidus, thalamus, midbrain, amygdala, and insula. This confirms the key role of NAcc in the mesocorticolimbic system, which integrates inputs from limbic and cortical regions. We also detected activity in brain regions having few or no direct anatomical connections with NAcc, such as sensorimotor cortex, cerebellum, medial and posterior parietal cortex, and medial/inferior temporal cortex, supporting the view that not all functional connections can be explained by anatomical connections but can also result from connections mediated by third areas. Our rsFC findings are in line with the results of the structure-based meta-analysis: MACM maps are superimposable with NAcc rsFC results, and the reward paradigm class is the one that most frequently generates activation in NAcc. Our results overlap considerably with recently proposed schemata of the main neuron systems in the limbic forebrain and in the anterior part of the limbic midbrain in rodents and nonhuman primates.
The nucleus accumbens (NAcc) is increasingly recognized as a pivotal center within brain systems regulating motivation and reward. NAcc is located at the conjunction between the head of the caudate and the anterior portion of the putamen, laterally to the septum pellucidum (Groenewegen, Wright, Beijer, & Voorn, 1999; Groenewegen, Wright, & Beijer, 1996; Heimer, Zahm, & Alheid, 1995); together with the olfactory tubercle, NAcc forms the ventral striatum, a critical element of the mesocorticolimbic system (Heimer & Wilson, 1995).
Although NAcc is traditionally described as a collection of neuronal ensembles with different functional and behavioral connotations within the BG (Pennartz, Groenewegen, & Lopes da Silva, 1994), it is widely accepted that there are two major functional components, the core and the shell, which are characterized by specific input and output channels (Surmeier, Ding, Day, Wang, & Shen, 2007).
Some of the main afferent projections to NAcc are from the cortex, more specifically from the orbitomedial prefrontal cortex (OMPFC), that is, Brodmann's areas 11, 13, 24, 25, and 32 (Haber & McFarland, 1999). OMPFC inputs to NAcc are massive and represent the defining feature of separate functional circuits (Goldman-Rakic & Selemon, 1986) involving OFC, ACC (Parkinson, Willoughby, Robbins, & Everitt, 2000), and medial prefrontal cortex (MPFC) (Carmichael & Price, 1994). It has been proposed that OMPFC–NAcc pathways play a key role in the development of reward-guided behaviors by linking reward experiences with their motivational and emotional features (Cummings, 1995).
NAcc is also interconnected with several subcortical structures, namely hippocampal region, midline, medial parafascicular, medial dorsal (MDN) and intralaminar thalamic nuclei, ventral pallidum, dopaminergic ventral tegmental and retrorubral cell groups, basal amygdaloid complex, dorsal and medial raphe nucleus, and noradrenergic cell group in the nucleus of the solitary tract (Morgane, Galler, & Mokler, 2005; Groenewegen et al., 1996, 1999; Brog, Salyapongse, Deutch, & Zahm, 1993). Of note, the projections of these structures are not restricted to NAcc but extend to the olfactory tubercle, caudate, putamen, and more caudal ventral striatal areas. Therefore, NAcc integrates the limbic and cortical inputs and projects to other BG nuclei, which send feedback projections into the prefrontal cortex via the MDN (Zahm, 1998; O'Donnell, Lavin, Enquist, Grace, & Card, 1997).
The connectivity pattern of NAcc provides support to the traditional hypothesis that NAcc constitutes the neural substrate for limbic–motor interactions (Mogenson, Jones, & Yim, 1980). More recently, neurochemical studies have suggested that NAcc plays a prominent role in the reward and motivation process via dopaminergic innervation (Knutson & Cooper, 2005; Glimcher & Rustichini, 2004; McClure, York, & Montague, 2004; Schultz, 2004). Current research on the behavioral correlates of NAcc focuses on the role of the mesolimbic dopamine system in reward and reinforcement (Wise, Bauco, Carlezon, & Trojniar, 1992) and, to a lesser extent, in response selection and intensification processes (Redgrave, Prescott, & Gurney, 1999; Salamone, Cousins, & Snyder, 1997; Robbins & Everitt, 1992) or arousing and preparatory effects of reinforcers (Robbins & Everitt, 1992; Taylor & Robbins, 1984; Koob, Riley, Smith, & Robbins, 1978).
NAcc has been found to be recruited in multiple forms of positive (Burgdorf & Panksepp, 2006) and negative (Carretie, Albert, Lopez-Martin, & Tapia, 2009) affective states; likewise, functional neuroimaging studies have found activity in the human accumbens following both rewarding (Mobbs, Greicius, Abdel-Azim, Menon, & Reiss, 2003; Aharon et al., 2001) and aversive stimuli (Gottfried, O'Doherty, & Dolan, 2002; Becerra, Breiter, Wise, Gonzalez, & Borsook, 2001). Moreover, several differential functions within each component of NAcc have been reported in animal studies (McFarland, Lapish, & Kalivas, 2003; Grill & Coghill, 2002; Zahm, 1999), albeit not in humans. Finally, disturbances at the level of NAcc have been implicated in drug abuse and schizophrenia, in addition to affective disorders (Kienast & Heinz, 2006; Totterdell, 2006).
Despite clear-cut clinical relevance, our knowledge of NAcc connectivity is mainly based on experimental studies using tractographic techniques in animals (Haber, Kim, Mailly, & Calzavara, 2006; Middleton & Strick, 1994, 2002; Ferry, Ongur, An, & Price, 2000; Haber, Fudge, & McFarland, 2000; Cavada & Goldman-Rakic, 1991; Selemon & Goldman-Rakic, 1985). The literature in humans has flourished only recently, with 2 diffusion tensor imaging (DTI) studies (Leh, Ptito, Chakravarty, & Strafella, 2007; Lehericy et al., 2004), one on fMRI resting state functional connectivity (rsFC) (Di Martino et al., 2008) and one on meta-analysis of coactivation patterns, from 126 fMRI and PET studies (Postuma & Dagher, 2006). However, the primary goal of these studies was the investigation of BG functional subdivisions and connectivity patterns; none of them was specifically targeted at NAcc. Thus, we set out to investigate the connectivity pattern of NAcc in 18 healthy volunteers by using an rsFC protocol and comparing our findings with a structure-based meta-analysis (Laird, Eickhoff, Kurth, et al., 2009) of 59 fMRI and PET studies.
rsFC (Fox & Raichle, 2007; Margulies et al., 2007; Vincent et al., 2007; Damoiseaux et al., 2006; Fox et al., 2005; Greicius, Krasnow, Reiss, & Menon, 2003; Hampson, Peterson, Skudlarski, Gatenby, & Gore, 2002) is a recently developed technique that allows in vivo assessment of brain networks by detecting coherent patterns of spontaneous activity in the resting brain. It has been shown that correlations in slowly fluctuating spontaneous brain activity tend to reflect intrinsic functional networks. Resting state networks (RSNs) are localized in the gray matter and are likely related to ongoing neuronal activity, as demonstrated by aliasing of cardiac and respiratory cycles (De Luca, Beckmann, De Stefano, Matthews, & Smith, 2006). Moreover, RSNs are characterized by changes in BOLD signals that are comparable with task-related changes (i.e., up to 3%) and are consistent across individuals and stable across repeated measurements (Damoiseaux et al., 2006).
Structure-based meta-analyses focus on specific anatomical regions and address global coactivation patterns across a diverse range of tasks and experimental designs. The consistency of coactivation patterns across experiments is interpreted as a piece of evidence for the functional connection of groups of regions; this type of functional connectivity has also been indicated as meta-analytic connectivity modeling (MACM) (Robinson, Laird, Glahn, Lovallo, & Fox, 2009). In fact, two recent meta-analysis studies (Smith et al., 2009; Toro, Fox, & Paus, 2008) demonstrated that the set of major covarying activation networks identified from large-scale meta-analyses overlaps almost completely with the set of networks that are visualized in the resting brain. These results provide strong evidence that RSNs reflect functional neural networks and that these dynamic networks are engaged even at rest (Fox & Raichle, 2007). Therefore, similar to previous studies (Toro et al., 2008; Postuma & Dagher, 2006; Lancaster, Laird, Fox, Glahn, & Fox, 2005; Koski & Paus, 2000), we used MACM (Laird, Eickhoff, Kurth, et al., 2009) to characterize patterns of functional coactivations in the human brain and compared the emerging patterns with the results of our rsFC analysis to validate each other.
Eighteen right-handed healthy volunteers (nine women; mean age = 51.2 years, SD = 19.2 years), free of neurological or psychiatric disorders, not taking medications known to alter brain activity, and with no history of drug or alcohol abuse, participated in the study. Written informed consent was obtained from each subject, in accordance with the Declaration of Helsinki; the study was approved by our institutional committee on ethical use of human subjects. The fMRI study was performed at the Ospedale Koelliker in Turin.
Task and Image Acquisition
Subjects were instructed to simply keep their eyes closed, to think of nothing in particular, and not to fall asleep. After the scanning session, participants were asked if they had fallen asleep during the scan, and data from subjects with positive or doubtful answers were excluded from the study.
Images were gathered on a 1.5-Tesla INTERA scanner (Philips Medical Systems, Andover, MA) with a SENSE high-field, high-resolution (MRIDC) head coil optimized for functional imaging. Resting state functional T2*-weighted images were acquired using EPI sequences, with a repetition time of 2000 msec, an echo time of 50 msec, and a 90° flip angle. The acquisition matrix was 64 × 64, with a 200-mm field of view. A total of 200 volumes were acquired, with each volume consisting of 19 axial slices parallel to the anterior–posterior (AC–PC) commissure; slice thickness was 4.5 mm with a 0.5-mm gap. To reach a steady-state magnetization before acquiring the experimental data, two scans were added at the beginning of functional scanning: The data from these scans were discarded.
Within a single session for each participant, a set of three-dimensional high-resolution T1-weighted structural images was acquired, using a fast field echo sequence, with a repetition time of 25 msec, an ultrashort echo time, and a 30° flip angle. The acquisition matrix was 256 × 256; the field of view was 256 mm. The set consisted of 160 contiguous sagittal images covering the whole brain. In-plane resolution was 1 × 1 mm, and slice thickness was 1 mm (1 × 1 × 1 mm3 voxels).
Selection of NAcc ROIs
For rsFC analysis, two bilateral anatomical ROIs were drawn according to the AFNI brain structure atlas (afni.nimh.nih.gov/afni/doc/misc/afni_ttatlas/). The mean ROI volume was 143 mm3 (see Figure 1).
For MACM analysis, we used rectangular ROIs, because BrainMap does not currently allow to draw anatomical ROIs. Therefore, two rectangular ROIs were drawn bilaterally according to the Talairach Daemon Database (Lancaster et al., 2000) around the following coordinates: left, X = −10 Y = 8 Z = −8; right, X = 12 Y = 7 Z = −8 (see Figures 1 and 2). ROIs for activation likelihood estimation (ALE) meta-analysis were bigger than the rsFC ROIs to account for the nonanatomical ROI shape and the mean “between-template variance” introduced by different normalization strategies between different studies (Eickhoff et al., 2009). The boundaries for the right ROI were (8, 4, −12) to (16, 14, −5); the boundaries for the left ROI were (−15, 5, −12) to (−7, 13, −5).
BOLD imaging data were analyzed using BrainVoyager QX software (Brain Innovation, Maastricht, Holland); an in-house developed Matlab Script has been used to create masks (see Supplementary Data for a description). Functional images were preprocessed as follows to reduce artifacts (Miezin, Maccotta, Ollinger, Petersen, & Buckner, 2000): (i) slice scan time correction was performed using a sinc interpolation algorithm; (ii) 3-D motion correction was applied: using a trilinear interpolation algorithm, all volumes were spatially aligned to the first volume by rigid body transformations and the rototranslation information was saved for subsequent elaborations; (iii) spatial smoothing was performed using a Gaussian kernel of 8 mm FWHM; (iv) temporal filtering (linear trend removals) and a band pass filter of 0.01–0.08 Hz was used (as described in Greicius et al., 2003; Biswal, Yetkin, Haughton, & Hyde, 1995), as it has been shown that the 0.01- to 0.08-Hz frequency range had the greatest power in revealing the underlying connectivity (Greicius, Supekar, Menon, & Dougherty, 2009; Hagmann et al., 2008; Vincent et al., 2007; Achard, Salvador, Whitcher, Suckling, & Bullmore, 2006; Fransson, 2006; Biswal et al., 1995).
Preprocessing was followed by a series of steps to facilitate accurate anatomical localization of brain activity and intersubject averaging. For each subject, the slice-based functional scan was coregistered with the 3-D high-resolution structural scan, and the 4-D data set was transformed into Talairach space (Talairach & Tournoux, 1988).
The first step to perform all FC analyses was to extract BOLD time courses from each ROI (defined as described above) by averaging over voxels within each region. Several nuisance covariates were included in our analyses to control for the effects of physiological processes (such as fluctuations related to cardiac and respiratory cycles) (Bandettini & Bullmore, 2008; Birn, Murphy, & Bandettini, 2008; Napadow et al., 2008) and motion. Specifically, we included nine additional covariates that modeled nuisance signals from white matter, global signal (Fox, Zhang, Snyder, & Raichle, 2009), and CSF, as well as six motion parameters.
All seed-based predictors were z-normalized; temporal autocorrelation correction (Woolrich, Ripley, Brady, & Smith, 2001) was used. Seed ROI-driven FC maps were computed on a voxel-wise basis for each previously selected region. The individual participant multiple regression analysis was carried out using the general linear model (Friston, Ashburner, Kiebel, Nichols, & Penny, 2007) and resulted in a t-based map (SPMt) FDR corrected for multiple comparisons (Genovese, Lazar, & Nichols, 2002; Benjamini & Hochberg, 1995) (q < 0.05, cluster threshold K > 10 voxels in the native resolution).
Group Statistical Map
Random effect group level analyses, controlling for age and gender effects, were conducted using the BrainVoyager QX ANCOVA module [p < .05 FDR corrected (Genovese et al., 2002; Benjamini & Hochberg, 1995), cluster threshold K > 10 voxels in the native resolution].
To evaluate the spatial consistency of functional connectivity patterns across subjects, we computed spatial probabilistic maps. The probability map describes the relative frequency (expressed in percentage) wherein the same network is represented over different brain areas.
To achieve maximum precision in the location description, we developed a Matlab script (see Supplementary Data) that, in case of large blobs, splits the activation in different BAs and Gyri that compose the original blob. When the script finds a voxel that does not match the BA mask or the Gyral mask (e.g., white matter voxels), this voxel is labeled as “Out of Gyrus.”
To evaluate the spatial consistency of functional connectivity patterns across subjects, we computed the split half reliability index: We calculated the reliability coefficient with the Spearman–Brown (Brown, 1910; Spearman, 1910) formula, , wherein the term rh, in our case, is the spatial similarity of the maps obtained by two random selected, equally numerous subgroups. The rh term is a measure of the intersection of two fuzzy sets, the Sørensen index (Sorensen, 1948), defined as: , wherein A and B are the elements in sample A and B, respectively, and C is the number of elements shared by the two samples (this is equivalent to Dice metric).
We followed the workflow indicated by Laird, Eickhoff, Kurth, et al. (2009) for the structure-based meta-analyses, also indicated as MACM (Laird, Eickhoff, Li, et al., 2009; Robinson et al., 2009): (i) we extracted from the BrainMap database (Laird, Lancaster, & Fox, 2005) all the studies involving only normal subjects that reported an activation in at least one of the two NAcc ROIs previously described, irrespective of class or behavioral paradigm that had generated that activation. The specific query was [Diagnosis = Normals] AND [ROIs = “left accumben”, (−15, 5, −12) to (−7, 13, −5) OR “right accumbens”, (8, 4, −12) to (16, 14, −5)] and yielded a total of 42 articles (see Supplementary Table 1) corresponding to 57 experiments, leading to a total of 762 foci (see Figure 2).
BrainMap (Laird et al., 2005) is a database of published functional neuroimaging studies (mainly PET and fMRI) that contains both metadata descriptions of experimental design and activation locations in the form of stereotactic coordinates. BrainMap contains 1843 neuroimaging publications that describe analyses from 8618 experimental contrasts using 81 unique paradigm classes, yielding 69,681 locations (November 22, 2009).
Activation Likelihood Estimation
An ALE meta-analysis (Laird et al., 2005; Turkeltaub, Eden, Jones, & Zeffiro, 2002) was performed on the sets of coordinates identified as coactivated during each ROI activation. Regions of convergence were calculated using GingerAle 2.0.
The ALE analysis is a quantitative method that can be used to estimate consistent activation across different imaging studies (Laird, Eickhoff, Kurth, et al., 2009). ALE maps of coactivations are derived based on patterns of foci of interest, where multiple studies have reported statistically significant peak activation. To limit the intersubject and interlaboratory variability, we used an algorithm that estimates the spatial uncertainty of each focus, taking into account the possible differences among the neuroimaging studies (Eickhoff et al., 2009). This algorithm was preferred to a prespecified FWHM as in the original ALE approach. The advantage of such an algorithm is that it limits the meta-analysis to an anatomically constrained space specified by a gray matter task. Furthermore, it comprises a method to calculate the above-chance clustering between experiments (i.e., random effects analysis) rather than between foci (fixed effects analysis; Eickhoff et al., 2009).
The original studies contributing these foci for each domain are presented in Supplementary Table 5.
Paradigm Class Profiles
Aside from MACM, we were interested in examining what mental processes are underpinned by the activation of our ROI. In BrainMap, metadata are organized under three experiment level fields: context, paradigm class, and behavioral domain. The “context” represents the purpose for which an experiment was designed. Possible contexts include normal mapping, age effects, disease effects, etc. The “paradigm class” is the experimental task isolated by the contrast. For a given experiment, multiple paradigm classes may apply. Paradigm classes include, among others, action observation, episodic recall, task switching, etc. A complete list of BrainMap's paradigm domains can be accessed at brainmap.org/scribe/.
Both rsFC and MACM techniques revealed a pool of areas that were connected to NAcc: these included putamen, caudate head, anterior and posterior cingulate, subcallosal gyrus, thalamus (MDN nucleus), medial frontal gyrus, amygdala, inferior parietal lobule, insula, caudate body, globus pallidus, parahippocampal gyrus, inferior frontal gyrus, culmen, precuneus, and cerebellum (Figures 3,45–6).
In addition, the rsFC analysis showed functional connections with the middle temporal gyrus, superior frontal gyrus, superior temporal gyrus, middle frontal gyrus, inferior temporal gyrus, supramarginal gyrus, fusiform gyrus, hippocampus, precentral gyrus, angular gyrus, hypotalamus, and pulvinar. Finally, MACM identified connections with the postcentral gyrus and the inferior occipital gyrus, which were missed by the rsFC technique (Figures 3,45–6).
Overall, the two techniques generally converge, but rsFC seems to have a better sensibility showing a richer pool of connected areas. It has to be considered that these two connectivity techniques are based on very different types of data: rsFC uses resting state scans, whereas MACM uses activation paradigms (see Figures 3,4–5, Supplementary Figures 1 and 2, and Supplementary Tables 1 and 2). A summary image can be seen in Supplementary Figure 8.
Paradigm Class Profiles
The paradigm class profiles, which more frequently lead to an activation in NAcc, are the reward tasks. Error detection, lexical decisions, and rest were all associated with weaker activations (see Supplementary Figure 6).
We repeated the FC analysis by using monolateral ROIs and by comparing the maps generated by right ROIs with the maps generated by left ROIs for both rsFC and MACM results (two-sample t test, p < .05 FDR corrected, K > 10 voxels in the native resolution). Lateralization analysis of the rsFC findings evidenced that the right NAcc is more connected with uncus, subcallosal gyri, insulae, parahippocampal gyri, and cerebellum whereas the left NAcc is more connected with OMPFC, subgenual, temporal, medial prefrontal, and posteromedial cortices (Figures 7 and 8, Supplementary Figures 3 and 4, and Supplementary Tables 3 and 4). Lateralization analysis of the MACM findings showed a similar pattern of lateralization: The insular, thalamic, anterior cingulated, pontine, and cerebellar areas are more right-lateralized, whereas the subgenual, motor/premotor, prefrontal, and occipital cortices are more left-lateralized (Figures 7 and 8, Supplementary Figures 3 and 4, and Supplementary Tables 3 and 4).
We assessed possible differences in the paradigm class profiles activating the left and right NAcc: The right NAcc is more activated by emotional paradigms, whereas the left NAcc is more activated by pain paradigms (see Supplementary Figure 7).
The split-half test performed with the Spearman–Brown method between each ROI in the two split groups shows our results: rsb = 0.73.
Probability maps computed for assessing the spatial consistency and reproducibility of seed-generated maps yielded a high level of overlap among specific ROI-related rsFC maps for each subject (Figure 8 and Supplementary Figure S4).
To the best of our knowledge, this is the first study combining rsFC with MACM to investigate NAcc connectivity. We evaluated the functional network associated with NAcc activity by examining temporally correlated patterns of low-frequency spontaneous activity during rest in a group of 18 right-handed healthy volunteers and MACM derived from the representative sample of the fMRI and PET literature present in the BrainMap database. Our findings are consistent with the results of both animal models (Postuma & Dagher, 2006; Morgane et al., 2005; Haber & McFarland, 1999) and DTI and fMRI investigations in humans (Stoeckel et al., 2009; Di Martino et al., 2008; Postuma & Dagher, 2006; Knutson & Cooper, 2005; Becerra et al., 2001; Breiter & Rosen, 1999). Spontaneous activity in NAcc predicted activity in regions implicated in reward circuitries, including OMPFC, globus pallidus, thalamus, midbrain, amygdala, and insula. Furthermore, we detected activity in brain regions that are described to have few or no direct connection with NAcc, such as sensorimotor cortex, cerebellum, medial and posterior parietal cortex, and medial/inferior temporal cortex (Haber & McFarland, 1999). It has been shown that functional connectivity patterns result not only from direct connections but also from connections mediated by third areas, thus suggesting that not all functional connections can be explained by anatomical connections, although FC overlaps considerably with tract tracing analysis (Damoiseaux & Greicius, 2009). Consequently, our FC patterns give a representation of the brain regions working together with NAcc as a coordinate network and can be, in great part, explained by indirect connections (Damoiseaux & Greicius, 2009). For example, several areas directly connected with NAcc, such as insula (Cauda et al., 2011), dorsal striatum, and thalamus, are characterized by strong connections with sensorimotor as well as parietal and cerebellar cortices. These observations are further validated by our MACM results, which essentially replicate the rsFC findings with the exception of the wider connectivity pattern with the fronto-temporal neocortex.
NAcc is a key element of the mesocorticolimbic system, which integrates inputs from limbic and cortical regions, linking motivation with action (Mogenson et al., 1980). Specifically, NAcc has a well-established role in mediating the rewarding effects of drug abuse and fundamental rewards such as food and sexual behavior (Carlezon & Thomas, 2009). Interestingly, our results overlap considerably with recently proposed schemata of the principal neuron systems in the limbic forebrain and in the anterior part of the limbic midbrain in rodents and nonhuman primates (Figure 4) (Morgane et al., 2005; de Olmos & Heimer, 1999; Haber & McFarland, 1999). We found that NAcc positively correlated with OMPFC, insulae, MDN, amygdalae, and hippocampi. Of these, OMPFC, insulae, and MDN are known to be active during reward (Breiter & Rosen, 1999), whereas the strong interconnections between NAcc, amygdala (Cardinal, Parkinson, Hall, & Everitt, 2002), and hippocampus (O'Donnell & Grace, 1995) are well described in the modified model of Lawrence, Sahakianb, and Robbins (1998) on corticostriatal circuits.
Our rsFC findings are further confirmed by the results of the structure-based meta-analysis: the reward paradigm class most frequently generates activation in NAcc. The ALE maps of the meta-analytical connectivity modeling are similar to the rsFC maps (Supplementary Table 5), thus confirming the validity of the rsFC results. The fact that in the midbrain we only found substantia nigra to be connected with NAcc can be explained by taking into account the relative low resolution of our method: Relatively small structures like ventral tegmental or dorsal raphe nucleus can be too small to be detected. Likewise, the small midline thalamic nuclei that are described as interconnected with NAcc (Haber & McFarland, 1999) might be under the threshold of our resolution power; in fact, we detected only a significant cluster in the MDN nucleus, one of the largest thalamic nuclei known to have strong connections with several regions in the prefrontal and limbic regions (Cauda et al., 2009; Zhang et al., 2008; Morgane et al., 2005; McFarland & Haber, 2002), including pallido-thalamic fibers belonging to the ventral striatal or limbic loop (Alexander, DeLong, & Strick, 1986; Haber, Groenewegen, Grove, & Nauta, 1985).
Our results are in agreement with most DTI, fMRI, and meta-analytic (Di Martino et al., 2008; Postuma & Dagher, 2006; Breiter & Rosen, 1999) studies in the literature. There are, however, a few differences. For example, a meta-analysis of 126 PET and fMRI studies by Postuma and Dagher (2006) failed to demonstrate connections with OFC and left insula, contrary to the present study and a previous work (Di Martino et al., 2008). Although this meta-analytic study is of great importance because it first introduced the structure-based meta-analysis technique, the use of different ROIs and statistical methods makes its findings only partially comparable with ours.
There are also notable differences between our study and Di Martino et al.'s (2008) analysis of the rsFC of the striatum. Although most of the findings show considerable overlap, Di Martino et al. described the insulae to be more strongly connected with dorsal striatal ROIs and a number of deactivations that we failed to find; interestingly, some of those deactivations are in areas where we found small but significant positive activations, such as the sensorimotor cortex. The IC has been described by Chikama, McFarland, Amaral, and Haber (1997) to have somatotopic anatomic connections with the striatum, wherein the dorsal posterior insula projects to the dorsal putamen whereas the more anterior and ventral insula projects to the caudate nucleus and ventral striatum. These differences are likely to be related to ROIs placed in slightly different positions and multiple regression analyses performed with all the predictors orthogonalized.
The striatum is known to have a ventral-to-dorsal gradient via circuits that spiral from emotional/motivational to decision making and executive motor control (Haber, 2003; Haber et al., 2000). Because of partial volume effects and smoothing, our ROI time courses may be affected by a small but significant residual sampling of more dorsal (i.e., executive/motor) striatal region time courses. However, this bias should also affect MACM, because the results of the former are superimposable to rsFC results. This is remarkable, given the independent nature of these two analyses on fundamentally different types of data as well as the heterogeneity of data contained in BrainMap due to differences in subjects, scanners, analyses, and paradigms. Therefore, the overall convergence of rsFC and MACM results supports the validity of our functional connectivity map.
Interestingly, in our analysis, the right ROI and the left ROI generated a different pattern of connectivity, suggesting that the right NAcc is more connected with uncus, subcallosal gyri, insulae, and parahippocampal gyri, whereas the left NAcc is more connected with OFC, temporal gyri, and PMC. This difference in functional connectivity is consistent with the lateralization pattern emerging from studies with subjects suffering from affective disorders. Converging evidence from neuroimaging, neuropathological and lesion analysis studies revealed predominant involvement of the left MPFC, subgenual ACC, and related limbic and striato-pallido-thalamic structures in regulating emotional expression in cases with recurrent depressive episodes (Drevets, Price, & Furey, 2008; Drevets, Savitz, & Trimble, 2008).
Resting state functional connectivity analysis has a number of limitations that merit consideration. First, it has been shown that functional connectivity can change during task performance (Fransson, 2006; De Luca, Smith, De Stefano, Federico, & Matthews, 2005). Second, resting state analysis faces the same potential limitations as task-related fMRI studies with regard to interindividual variability in ventral striatum organization and connectivity. However, the patterns of functional differentiation observed using our seeding approach were reliable and detectable at the individual participant level. Third, our subjects were distributed over a wide age span; however, we attempted to take into account the intersubject variability using a random effect analysis and to reduce the variability induced by age and gender differences by controlling these factors, inserting age and gender as covariates in the statistics. Furthermore, the interpretation of our findings needs caution, because the networks described here are detected in the absence of specific functional activity. We are inferring functional roles for the ventral striatal areas on the basis of their belonging to intrinsic connectivity networks, whose functional relevance is reasonably well established in the literature. Although the exact functional significance of temporal correlations in very low frequency neural fluctuations remains largely unclear, it has been argued that this basal, task-independent, intrinsic connectivity is important to avoid disuse-related pruning of critical synapses (Luo & O'Leary, 2005) and/or to maintain networks in a primed state, thus improving response efficiency (Fox & Raichle, 2007). MACM also has limitations. A major limitation of MACM is that this technique identifies regions that tend to be coactivated when NAcc is activated in the absence of a control group. This means that some coactivations might be attributable to common task requirements or mental states beyond reward processing rather than functional connectivity per se.
A recent fMRI study investigated both regional and interregional functional connectivity patterns while subjects performed a gambling task featuring unexpectedly high monetary gains and losses (Camara, Rodriguez-Fornells, & Munte, 2008). The authors found that monetary gains and losses activated a similar fronto-striato-limbic network, in which main activation peaks were observed bilaterally in the ventral striatum. Our resting state connectivity findings provide confirmation to the role of NAcc as a seat for major “hedonic hotspots” in the widespread pleasure-activated brain networks in humans (Kringelbach & Berridge, 2009). Such hedonic hotspots are anatomically distributed (NAcc shell and ventral pallidum, other forebrain and limbic cortical regions, and deep brainstem regions including the parabrachial nucleus in the pons) but interact to form a functional integrated circuit, which is sensitive to stimulation with opioids, endocannabinoids, and other neurochemical modulators. From a clinical perspective, the identification of this network is essential to the understanding of the brain mechanisms underlying reward experiences and neuropsychiatric conditions such as addiction, impulse control disorders, and obsessive compulsive spectrum disorders, in addition to disorders of affect (Camara, Rodriguez-Fornells, Ye, & Münte, 2009). Therefore, our results stress the importance of studying functional connectivity in addition to standard fMRI analysis in reward-related studies in a wide range of neuropsychiatric conditions. Finally, because an assortment of human behaviors is thought to be driven by reward-based processes, including novelty seeking, decision making, economic choice, reinforcement learning, and incentive motivation, future research will be able to show the exact contribution of the NAcc system in the different behavioral contexts to describe both physiological and pathological reward mechanisms more appropriately.
Movement was assessed by summing the deviations used to compensate for head motion within the fMRI scanner. The overall quantity of movement was very mild; furthermore, we failed to find a correlation between movement and the age of the subject. Hence, we excluded any influence by these confounds.
In the meta-analysis, a possible confound is generated from the fact that different groups used slightly different templates to transform each subject's MRI or PET image into stereotaxic space. The use of different template brains would result in slight variation in localization of peaks. In particular, depending on the applied normalization procedure, the between-subject variances ranged from 11.0 to 12.1 mm (Eickhoff et al., 2009), hence the need of a slightly bigger ROIs for the meta-analysis with the respect to the rsFC ROIs.
Our results show good reproducibility: The split half reliability (rsb = 0.73) and the probabilistic maps (Figure 8) together with the good correspondence observed between the resting state and meta-analytic results lead us to rule out the possibility that the patterns result merely from random fluctuations or that they are due to unintentional tasks by the single subject. In addition, our results are in agreement with anatomical and functional data obtained in previous studies on nonhumans as well as human primates.
We thank all the subjects who participated in this study. The research has been founded by Piedmont Region, Human and Social Science 2008, GIRS project “The invisible grammar of social relations.”
Reprint requests should be sent to Franco Cauda, Dipartimento di Psicologia, Via Po 14, 10123 Turin, Italy, or via e-mail: email@example.com.