Forty-two native Swiss-German- or German-speaking elderly people (14 men, 26 women) with a mean age of 74.0 ± 7.7 years (range: 63–93 years) and a mean of 12.8 ± 2.3 educational/academic years (range: 9–20 years) were recruited from a behavioral DoG study of a large sample of 120 healthy, cognitively unimpaired elderlies (Forstmeier et al., 2011). A lack of cognitive impairment as indexed by a score of ≥26 on the Mini Mental State Examination (MMSE; Folstein et al., 1975) and age of ≥60 years were the inclusion criteria for the current neuroimaging study. Current mental health disorders, neurological disorders, and history of stroke or dementia were the exclusion criteria. One participant reported a history of apoplexy and one was taking drugs for attention deficit hyperactivity disorder and therefore these two subjects had to be excluded from the current investigation. An additional subject was excluded for the rsfMRI analysis only because the rsfMRI scans did not cover the whole brain. All participants provided written informed consent and were paid CHF 100 for their willingness to participate. The study protocol was conducted in accordance with the Declaration of Helsinki and approved by the ethics committee of the canton of Zurich, Switzerland.

We applied the German version of the DoG test for adults (BAT-E) (Forstmeier et al., 2011), but in addition we also used the German version of the more known (DD) paradigm (Kirby and Marakovic, 1996) that is related to DoG (Forstmeier and Maercker, 2011). The DoG paradigm (BAT-E) has been described in more detail elsewhere (Forstmeier et al., 2011), but is reiterated here in exactly the same wording as formerly already published by our group (Drobetz et al., 2014). “The test requires participants to make 18 choices between immediate smaller and delayed larger primary and secondary reinforcers: snacks, hypothetical money, real money, and magazines. In the snacks subscale, the experimenter asks the participant to decide between one piece of chocolate immediately and two pieces in 2 h. In the real money subscale, the participant must decide between CHF 8 now or CHF 10 in 1 month. In the hypothetical money subscale, the participant must choose between smaller amounts of hypothetical money now and larger amounts of hypothetical money in 1 month. In the magazines subscale, the participant must choose either one real magazine now or two real magazines in 1 month. The snacks and hypothetical money subscales consist of eight items, the real money and magazine subscale consist of two items—in sum, the DoG-A has 18 items. Whereas the snacks and hypothetical money subscores range from 0 to 8, the real money and magazines subscores are dichotomous (immediate real money/ magazine now vs. delayed real money/two magazines). In each of the 18 items, the participant received one point when he/she chose the delayed reward and zero points when he/she took the immediate reward. We used the DoG-A score, the sum of all four subscales, as a composite measure ranging from 0 (nondelayer in all four subscales) to 4 (delayer in all four subscales)” (Drobetz et al., 2014).

Impulsivity has been measured with the German version of Barratt impulsiveness scale (BIS-11; Patton et al., 1995; Preuss et al., 2008) and executive functions were assessed by three different cognitive tests: Trail making test—part B (Reitan, 1958), digit span backward test (Wechsler, 1997), and the Stroop color-word test (Stroop, 1935). For executive functioning, we calculated composite scores (z-scores) for all individuals based on the three executive function tests applied, i.e., the trail making test—part B, digit span backward test and the Stroop color-word test.

We used a 3.0 Tesla Philips Achieva whole-body scanner (Philips Medical Systems, Best, The Netherlands) to acquire the MRI data. The MRI scanner was equipped with a common eight-element head coil array capable of sensitivity encoding (SENSE) and a transmit-receive body coil.

Two T1-weighted images were acquired, in addition to the one resting-state functional MRI (rsfMRI) and one diffusion-weighted imaging sequence that have been applied for each subject under investigation.

For all 40 participants, a volumetric three dimensional T1-weighted gradient echo sequence named fast field echo has been applied twice resulting in two T1-weighted MRI scans. With a spatial resolution of 0.94 × 0.94 × 1.00 mm³ (matrix 256 × 256 pixels, 160 slices), the slices were acquired in the sagittal plane. Other technical parameters of the T1-weighted sequence were: field of view FOV = 240 × 240 mm²; echo time TE = 3.7 ms; repetition time TR = 8.06 ms; flip angle = 8°; and SENSE factor = 2.1. Scan time was about 8 min per scan. To exclude any T2-sensitive tissue anomalies, T2-weighted MRI scans were acquired as well.

Subjects rested quietly with closed eyes in the scanner while rsfMRI time series have been acquired. Participants were instructed to let their mind wander and to think of nothing in particular. Images of rsfMRI were acquired in the transversal plane using a spin-echo echo-planar imaging (EPI) sequence with a measured spatial resolution of 2.5 × 2.5 × 4.0 mm³ (reconstructed 1.72 × 1.72 × 4.0 mm³). Imaging parameters were: TR = 4.0 s (interleaved slice acquisition starting with slice one); TE = 35 ms; FOV = 220 × 220 mm²; matrix 112 × 112 pixels; slice thickness = 4 mm; number of slices = 40; flip angle = 78°; SENSE factor = 1.8. The rsfMRI sequence lasted about 10 min (corresponding to 150 brain volumes).

Subsequently, a diffusion-weighted sequence with a spatial resolution of 2 × 2 × 2 mm³ (matrix: 112 × 112 pixels, 75 slices in transversal plane) was applied. Diffusion was measured along 32 non-collinear directions (b = 1000 s/mm²) preceded by a non-diffusion-weighted volume (reference volume, b = 0 s/mm²). Further imaging parameters were: FOV = 224 × 224 mm²; TE = 55.0 ms; TR = 13.010 s; flip-angle = 90°; SENSE factor = 2.1. Scan time was about 8 min 42 s. T1-weighted, DTI, and resting state fMRI images were acquired in the same session.

The FMRIB software library tools (FSL, version 5.0.7; http://www.fmrib.ox.ac.uk/fsl/; Smith et al., 2004) such as the FDT (FMRIB diffusion toolbox; version 3.0; Behrens et al., 2003) has been used for preprocessing of the diffusion-weighted MRI data. The Diffusion Toolkit (DTK, version 0.6.3) and TrackVis software (version 0.6.0.1; http://trackvis.org/; Park et al., 2009) has been applied to achieve deterministic fiber tractography. The connectivity matrix has been computed in MATLAB software (version 2013b; http://www.mathworks.com/index.html).

To construct the connectivity matrix of the white matter pathways, the following fully automated preprocessing steps were realized: (1) In a first step, a binary brain mask was created using FSL's brain extraction tool. This mask was used in later steps to exclude non-brain tissue. (2) Eddy current distortions and head movements were corrected using the EDDY_CORRECT tool of FDT. (3) Extraction of the mean translation and rotation values according to work done by Yendiki et al. (2014) for later use of these measures as a covariate of no interest in the regressions. This has been achieved by using the TRACULA tool (Yendiki et al., 2011) (https://surfer.nmr.mgh.harvard.edu/fswiki/Tracula) implemented in the FreeSurfer software suite (http://surfer.nmr.mgh.harvard.edu/fswiki). (4) Diffusion gradients were adjusted for rotations introduced by the eddy current and head movement corrections. (5) The preprocessed DTI data were then subjected to the DTK to compute voxel-wise diffusion tensors and to construct the (principal) eigenvector and eigenvalue maps as well as a map of fractional anisotropy (FA). (6) Deterministic tractography was conducted in TrackVis using the “brute force” approach with an interpolated streamline tracking algorithm. Twenty streamlines per voxel were propagated (using the -rseed option in DTK with value 20) and fiber tracking algorithm terminated if the voxel's FA value drop below 0.10 or if the streamline's turning angle between two consecutive voxels increase over 45°. This results in a connectome of the whole-brain comprised by about 2–3 million of streamlines including connections to the cerebellum and subcortical pathways. (7) The individual FA image was mapped onto the FMRIB58-FA reference brain, which is already in the MNI152 standard space, using FSL's linear image registration tool (abbreviated FLIRT). The obtained transformations were stored for later use. (8) We then applied these transformations to the streamlines derived from step 6 to map the streamlines into the MNI152 standard space. (9) To count the number of streamlines between each pair of ROIs we used the automated anatomical labeling (AAL) atlas (Tzourio-Mazoyer et al., 2002), the ROIs of which are already located in MNI152 standard space. Covering the entire neocortex (78 cortical ROIs), the subcortical structures amygdala, hippocampus, thalamus, caudate, putamen, and pallidum (12 subcortical ROIs) as well as 26 cerebellar ROIs, the AAL atlas consists of 116 ROIs in total. However, the 26 cerebellar ROI were excluded because we did not expect any correlates of DoG od DD in the cerebellum, which however was not entirely covered by MRI in all subjects. (10) Through the brainstem running streamlines and that shorter than 5 mm in length were removed. Self-loops, i.e., streamlines that make connections within a ROI were discarded. Using MATLAB scripts written by Zalesky (Zalesky et al., 2010), we counted the number of the remaining streamlines between any pair of ROIs resulting in an undirected, weighted, 90 × 90 connectivity matrix for each individual participant. Hence, the strength of a structural connection between two ROIs was operationalized by the number of reconstructed streamlines. (11) The undirected, weighted 90 × 90-node connectivity matrices were then subjected to a network-based statistical analysis (see below). We also investigated a smaller and more DoG-specific network comprised by only frontostriatal brain regions that were derived from a subcortical brain atlas and from a surface-based morphometric imaging study (Drobetz et al., 2014) that investigated structural gray matter correlates of DoG in the same participants as investigated in the present study.

DPARSFA toolbox (version 3.1) that is part of DPABI (version 1.2, http://rfmri.org/dpabi; Chao-Gan and Yu-Feng, 2010) and depends on functions of statistical parametric mapping software (SPM, version 8; www.fil.ion.ucl.ac.uk/spm/software/spm8) has been used for functional MRI data preprocessing. Following preprocessing steps has been performed: (1) Coregistration of the T1-weighted image onto the functional images. (2) slice timing correction, (3) realignment combined with the extraction of the frame-wise displacement parameters according work by Power and colleagues (Power et al., 2012, 2014) for later use as a covariate of no interest in the regressions, (4) estimation of linear and non-linear normalization of the T1-weighted MRI image using the unified segmentation approach as implemented in SPM8, (5) estimated transformations were then applied to the functional images, (6) voxel re-sampling to 2 × 2 × 2 mm³, (7) smoothing with a Gaussian kernel of 6 mm full width at half maximum, (8) detrending, (9) data filtering in the range 0.01 < f <0.1 Hz, and (10) regressing out the variance of nine nuisance covariates, i.e., the six parameters from head motion correction (three translation and three rotation parameters) as well as the global mean signal, white matter signal, and cerebrospinal fluid signal. Although, there is an ongoing dispute about whether regressing out the global mean signal in rsfMRI data analyses is beneficial or affects data detrimentally (Wong et al., 2012; Qing et al., 2015; Yeh et al., 2015), we applied global mean signal regression because it has also been shown that the global mean signal regression is very effective in controlling movement-related artifacts in functional connectivity measures (Power et al., 2012, 2014; Satterthwaite et al., 2012, 2013; Van Dijk et al., 2012; Yan et al., 2013).

In addition to the 90-node whole-brain network analyses, a 12-node frontostriatal network including DoG-related brain regions has been constructed. This network was comprised by the subcortical structures putamen, caudate nucleus, and nucleus accumbens and the cortical structures OFC as well as DLPFC and VLPFC. The subcortical structures were derived from the Harvard-Oxford subcortical atlas as implemented in the FSL viewer (version 3.0, http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/Atlases), whereas the cortical regions were derived from a former study of our group showing that regional cortical thickness and cortical surface area in the OFC, DLPFC, and VLPFC is related to DoG performance the same participants as the ones investigated in the present study (Drobetz et al., 2014), which also reported associations between DoG and DD performance and the volume of the caudate nucleus and nucleus accumbens. This DoG-specific 12-node network was used for both structural and functional network analyses.

The network-based statistical analyses of the structural (DTI) and functional (rsfMRI) connectivity matrices have been performed using the network-based statistic (NBS) toolbox (Zalesky et al., 2010) (https://www.nitrc.org/projects/nbs/) running in MATLAB (version R2013b). When mass-univariate testing is performed at every connection comprising the graph (network), NBS can be used to control the family-wise error rate. NBS is based on common principles underpinning other traditional cluster-based thresholding procedures applied onto statistical parametric maps and it exploits the extent to which the connections comprising the effect of interest are interconnected (Zalesky et al., 2010).

Based on the general linear model approach, we used the t-test module in NBS to correlate DoG and DD performance with the number of reconstructed streamlines of each connection (structural connectome) or the Fisher's z-transformed correlation coefficient between two brain regions (functional connectome) of the 90- and 12-node networks while simultaneously controlling for the influences of age, sex, and total intracranial volume. In addition to these three covariates, we also used the frame-wise displacement value encompassing translational and rotational motion parameters (Power et al., 2012, 2014) derived from the volume-by-volume realignment as a covariate of no interest in the functional connectivity analyses and the volume-by-volume realignment parameters (Yendiki et al., 2014) in the structural connectivity analyses in order to account for residual head motion-related influences in the connectivity measures. These partial correlations were done with the component extent option in NBS that is suited for detecting an effect of interest that is relatively weak at one connections, but extends to encompass a lot of connections (Zalesky et al., 2010). In all analyses, we simultaneously applied multiple comparisons correction procedures using 5000 permutations at p < 0.05. P-values of the network-based statistical analyses are reported one-tailed due to the directed contrasts.

It is important to note that the main statistical threshold (alpha error probability) was set at p = 0.05 corrected for multiple comparison using 5000 permutations for all network analyses reported in the present manuscript. The t-thresholds reported (see Results Section below) do not represent the alpha error probabilities. These t-thresholds are set (also called sensitivity) thresholds (Zalesky et al., 2010) and are used to determine which edges of the connectivity matrix form the largest subnetwork, which is then subjected to the permutation statistic. These sensitivity thresholds must be determined by exploration and are therefore chosen in an arbitrary way. However, this does not affect the false positive rate of the actual permutation statistic of the alpha error probability.

Two different thresholded subnetworks are reported per behavioral measure (DoG or DD), i.e., one subnetwork with many connections (more liberal set threshold) and one subnetwork with only few connections (more conservative set threshold) by applying two different set (sensitivity) thresholds (operationalized as t-values, called set t-value in the rest of the manuscript; Zalesky et al., 2010). The more conservative set threshold has been chosen to show the specificity of the frontostriatal connections within the subnetworks. However, for both differently thresholded subnetworks alpha error probability was set at p = 0.05 while simultaneously correcting for multiple comparison using 5000 permutations of the behavioral measure across subjects. Positive as well as negative associations have been tested separately. Significant subnetworks are visualized using BrainNet Viewer software (https://www.nitrc.org/projects/bnv/; Xia et al., 2013). Due to reasons of completeness and because of literature cited in the Discussion Section, structural and functional connectivity strength have also been correlated with impulsivity and executive functioning using the same covariates of no interest as used for DoG and DD. However, impulsivity and executive functioning have mainly been investigated to show the specificity of DoG and DD ability with respect to frontostriatal connectivity strength, i.e., these associations represent negative control conditions and hence no hypotheses were posed a priori.

Correlations between behavioral measures and the movement parameters of the functional and structural MRI data have been performed using IBM SPSS Statistics software (version 23.0, http://www-01.ibm.com/software/analytics/spss/). One-tailed p-values are reported for the network analyses because t-contrasts in NBS are directed and two-tailed p-values are reported for all other statistical tests. If exceptions are made, they are explicitly indicated.

The mean values and standard deviations of the demographic, cognitive as well as global brain measures of the 40 subjects under investigation are summarized in Table 1.

As already reported elsewhere (Drobetz et al., 2014), the total DoG score of the BAT-E and the global DD rate (k) of the delay-discounting test are inversely correlated (r = −0.33, p = 0.039). The total DoG score did neither statistically significantly correlate with impulsivity (r = 0.08, p = 0.61) nor with executive functions (r = 0.01, p = 0.97). DD rate (k) did neither statistically significantly correlate with impulsivity (r = 0.15, p = 0.35) nor with executive functions (r = −0.13, p = 0.45). There was also no statistically significant correlation between impulsivity and executive functions (r = −0.16, p = 0.34).

In all the correlations reported below, the behavioral measure (DoG or DD) and the structural connectivity measure (number of streamlines) have been associated by partial correlation analyses while simultaneously controlling for age, sex, and intracranial volume, but also controlling for the mean translation and mean rotation parameters on a volume-by-volume basis derived from the realignment procedure (Yendiki et al., 2014). The alpha error probability (p-value), which is independent and different from the sensitivity/set t-value reported below, was set at p = 0.05 corrected for multiple comparisons using 5000 permutations of the DoG of DD ability across subjects.

In addition, we also checked whether the behavioral measures of interest (DoG and DD) are associated with motion during the acquisition of the DTI data to rule out the possibility that subjects showing enhanced self-control also moved less during scanning. The behavioral measures DoG and DD neither correlated statistically significantly with the mean translation parameters (r = 0.026, p = 0.87 and r = 0.040, p = 0.81, respectively) nor with the mean rotation parameters (r = −0.093, p = 0.57 and r = 0.038, p = 0.81).

In all the correlations reported below, the behavioral (DoG or DD) and the functional connectivity measures have been associated by partial correlation analyses while simultaneously controlling for age, sex, and intracranial volume, but also controlling for the mean frame-wise displacement parameter according Power (Power et al., 2014). The alpha error probability (p-value), which is independent and different from the sensitivity/set t-value reported below, was set at p = 0.05 corrected for multiple comparisons using 5000 permutations of the DoG and DD ability across subjects.

In addition, we also checked whether the behavioral measures of interest (DoG and DD) are associated with motion during the acquisition of the rsfMRI data to rule out the possibility that subjects with better cognitive control also moved less during scanning. Neither DoG nor DD correlated statistically significantly with the mean frame-wise displacement parameter (r = −0.127, p = 0.44 and r = 0.223, p = 0.17, respectively).

We also correlated impulsivity and executive functioning with structural and functional connectivity strength using the same covariates of no interest as used for DoG and DD. The 90-node functional whole-brain connectome analysis and the 12-node functional frontostriatal network analysis did not reveal significant positive or inverse correlations between the functional connectivity strength of any connection of the connectivity matrix and impulsivity or executive functioning. The 90-node structural whole-brain connectome analysis revealed an inverse association between executive functioning and structural connectivity strength (set t-value: 4.50, alpha error p = 0.052) of only two connections involving only three nodes, although only on a trend level toward significance. One connection is located between the right middle occipital gyrus and the left putamen and the other connection is located between right middle occipital gyrus and the right middle cingulate cortex. No further statistical significant associations between structural connectivity and executive functioning have been found.

The 12-node structural frontostriatal network analysis revealed an inverse association between impulsivity and the structural connectivity strength (set t-value: 2.70, alpha error p = 0.018) of also only two connections involving only three nodes. One connection is located between the left nucleus accumbens and the right caudate nucleus and the other connection is located between nucleus accumbens and the orbitofrontal cortex. No further statistical significant associations between structural connectivity and impulsivity have been found.

We used a whole-brain connectome (90-node network, excluding the cerebellum) approach as well as an approach focusing on specific frontostriatal regions (12-node network) in isolation from the rest of the brain. As can be seen in Figure 1 and Table 2, the whole-brain connectome approach highlights the specificity of our predicted findings; at the more liberal set threshold in the DoG structural 90-node network analysis, 15 out of 15 connections showing a positive correlation between structural connectivity strength and DoG performance involved either (pre)frontal or striatal brain regions. Considering the fact that the 90-node structural connectivity matrix theoretically contains 4005 possible connections, finding 15 connections with the effect of interest and all 15 connections are located in the predicted prefrontal and striatal regions makes the specificity of our findings even more remarkable.

As already mentioned, the right putamen serves as a hub region within this subnetwork and the left putamen also appeared in both differently thresholded subnetworks. The caudate nucleus did not show any structural connection with a statistically significant effect of interest, whereas the nucleus accumbens was not modeled as an individual node within the 90-node whole-brain connectome. In contrast, all striatal nodes (both putamina, both nuclei caudati, and both nuclei accumbi) of the 12-node network showed connections with positive associations between DoG performance and structural connectivity strength (see Figure 2, Table 3).

With respect to frontal regions, the medial OFC as well as those of the inferior frontal gyrus, i.e., the pars orbitalis, pars opercularis, and pars triangularis, were all involved in the subnetwork revealed by the structural whole-brain connectome analysis. In this 90-node network, however, the DLPFC and VLPFC were not represented in its full extent and in isolation (as a single node) and can hence not be labeled as such in the 90-node network. But these two regions were represented in the 90-node network in a distributed manner over the superior and middle frontal gyrus nodes with respect to the DLPFC and over the inferior frontal gyrus nodes with respect to the VLPFC meaning that the nodes containing DLPFC and VLPFC also containing other brain regions not part of these two functional areas. In contrast, the 12-node frontostriatal network analysis revealed that both OFC and the left VLPFC and left DLPFC possess connections that run to the putamen and caudate nucleus and show the effects of interest, although the size of the nodes of the 12-node network are smaller compared with the size of the nodes of the 90-node network, but contain only voxels belonging to these two functional regions.

Taken together, our structural network analyses revealed that mainly frontostriatal connections such as those connecting the putamen, caudate nucleus, and nucleus accumbens on the one hand with the DLPFC, VLPFC, and OFC on the other hand are associated with DoG and DD performance in healthy elderly individuals.

In our functional connectivity analyses, we also assessed whole-brain connections (90-node connectome analysis) as well as frontostriatal connections (12-node network analysis) in isolation from the rest of the brain. No functional subnetworks with statistically significant correlations between DoG and functional connection strength in the resting state were found in the 90-node functional whole-brain connectome analysis. However, the 90-node functional whole-brain connectome analysis revealed that the DD ability was statistically inversely correlated with functional connectivity strength in a 29-node subnetwork encompassing 29 mainly striatal and frontal nodes. The right caudate nucleus and the right putamen serve as striatal hub regions within this subnetwork. The right pars opercularis served as frontal hub.

In the 12-node functional frontostriatal network analysis, DoG performance did also not show any statistically significant association with functional connectivity strength. However, the 12-node functional frontostriatal network analysis revealed a subnetwork consistent of 13 connections with inverse correlations between functional connectivity strength and DD. As can be seen in Figure 4, Table 6, the right and left orbitofrontal cortex together with the right caudate nucleus serve as hub regions within this subnetwork.

Taken together, these findings show a large overlap between the structural and functional frontostriatal nodes, which each are positively correlated with DoG and/or inversely correlated with DD. The overlap is not so strong in the explicit connections found between the structural and functional network analyses, but is rather pronounced in the common nodes involved in DoG or DD ability. A recently published study in 120 healthy subjects (20–85 years old) applied whole-brain rsfMRI to measure functional connectivity and DTI to measure structural connectivity using automated fiber tractography of 18 major white matter tracts longitudinally over a time interval of 3.3 years mainly concluded that the anatomical alignment of structural connectivity alterations and functional ones seems restricted to specific networks and tracts and that in general changes in structural connectivity and those in functional connectivity are not necessarily strongly correlated (Fjell et al., 2017).

Here we report further evidence with respect to the network nodes (brain regions) whose structural and functional connectivity is associated with DoG and DD. As previously outlined (Drobetz et al., 2014), the DLPFC in general is a functionally defined brain area that is pivotal in the control of a variety of behavioral aspects, such as bottom-up and top-down control processes for working memory and attentional processes, the feeling of being present in a virtual reality (Jäncke et al., 2009), the regulation of emotions (Davidson et al., 2000), and even risky behaviors (Jäncke et al., 2008). In the context of self-controlled behavior, however, the DLPFC as well as the striatum are the key players in the processing of rewards and in decision making with respect to immediate and delayed rewards (Elliott et al., 2003; Wallis and Miller, 2003; McClure et al., 2004, 2007; Cho et al., 2010; Figner et al., 2010).

Evidence for the involvement of the VLPFC and ACC has been reported in the literature as well. The right VLPFC has been associated with behavioral inhibition (Cools et al., 2002) and bilateral VLPFC activity showed to be related to motor impulsivity as can be measured by go/no-go tasks (Goya-Maldonado et al., 2010), whereas activations in the left ACC have been demonstrated in response inhibition tasks (Horn et al., 2003), another important facet of DoG as shown in a rat experiment (Reynolds et al., 2002).

The left frontal pole/OFC has repeatedly been associated with impulsivity and its related constructs, including DD (McClure et al., 2004; Bjork et al., 2009; Matsuo et al., 2009) and has also been related to the value functions of rewards (Cohen and Ranganath, 2005). The importance of the OFC as a neural substrate in the decision to take delayed over immediate rewards and in subjective valuation has been reported using lesion studies in both animals (Cardinal et al., 2004; Winstanley et al., 2004; Rudebeck et al., 2006) and humans. Lesions to the medial OFC was associated with a significantly higher preference for an immediate over a delayed reward (Sellitto et al., 2010).

It has also been reported that the subjective value and magnitude of rewards are represented in the caudate nucleus (Delgado, 2007; Knutson et al., 2009), a critical component of the dopaminergic reward system (Breiter and Rosen, 1999), while the putamen seems to be activated in immediate and delayed rewards only in older participants (Samanez-Larkin et al., 2011). In a previous study by our group (Drobetz et al., 2014) using the same experimental subjects as investigated here, we found that DoG ability correlated positively with the volume of the left (r = 0.423, p = 0.004, one-tailed) and right (r = 0.300, p = 0.034, one-tailed) caudate nuclei, whereas DD was inversely correlated with the volume of the left (r = −0.273, p = 0.049, one-tailed) and right (r = −0.324, p = 0.024, one-tailed) caudate nucleus, as well as with the volume of the left nucleus accumbens (r = −0.321, p = 0.025, one-tailed; Drobetz et al., 2014).

Recently published neuroimaging-based connectivity studies provide strong support for our findings. With respect to age-related structural connectivity in association with decision making, Samanez-Larkin and colleagues reported that decreased reward learning and decreased white matter integrity in specific pathways running from the thalamus to the medial PFC and from the medial PFC to the ventral striatum are associated with older age. This suggests that the integrity of frontostriatal white matter pathways critically supports reward learning (Samanez-Larkin et al., 2012). Peper et al. (2013) used DTI in young subjects and reported a link between the structural integrity of frontostriatal connections and DD performance. van den Bos et al. (2014) applied structural and functional MRI-based connectivity analyses and reported distinguishable striatal connections differentially related to DD. Structural and functional connectivity between the lateral prefrontal cortex and striatum was related to enhanced patience (corresponding to better DoG), whereas connectivity between the striatum and subcortical regions was related to enhanced impulsivity (corresponding to poorer DoG; van den Bos et al., 2014). The same group also investigated alterations in structural and functional connectivity of different frontostriatal fiber tracts during development. The authors showed that developmental increases in structural connectivity strength for connections reaching or originating from the right DLPFC were significantly associated with enhanced negative functional coupling (increased functional connectivity) with the striatum. An age-related decrease in discount rates was also reported (van den Bos et al., 2015).

In an fMRI study in children using a DoG paradigm, Luerssen and colleagues highlighted that a stronger attentional focus directed away from temptations was associated with enhanced functional coupling (increased functional connectivity) between the nucleus accumbens and several brain regions within the prefrontal and parietal cortices known to support self-control (Luerssen et al., 2015). In another fMRI study applying activity and effective connectivity analyses combined with an intertemporal monetary choice task, Hare and colleagues revealed a region in the left DLPFC (BA46) that showed increase activity in trials where subjects preferred delayed rewards. Furthermore, the strength of the functional connection between this region and a region in the VLPFC, commonly involved in the evaluation of stimulus values, enhanced at the time of the choice, mainly during trials where the subjects chose the delayed rewards (Hare et al., 2014).

A further fMRI study in adolescents and individuals in mid-adulthood used a hypothetical discounting task with monetary rewards delayed in the range of 1 week to 1 year and revealed that reductions in choice impulsivity (corresponding to enhanced DoG or reduced DD) with advancing age were related to alterations in brain activity in the VLPFC, ACC, ventral striatum, insula, inferior temporal gyrus, and in the posterior parietal cortex (Christakou et al., 2011). Even in nonhuman primates, there exists a strong link between frontostriatal connectivity strength and DoG ability. Using DTI, Latzman et al. (2015) investigated whether white matter tracts projecting from the prefrontal cortex to the caudate nucleus and the DoG ability are related in a sample of 49 captive chimpanzees (Pan troglodytes). After accounting for age, sex, and the time interval between the acquisition of the DTI data and the DoG assessment, enhanced white matter connectivity between the right dorsal prefrontal cortex and the caudate nucleus revealed to be significantly related to the acquisition (i.e., training) phase, but not related to the maintenance of the DoG ability. Further support for the association between DoG or DD on the one hand and structural and functional frontostriatal connectivity strength on the other has been reported elsewhere in the human and animal literature (Kim et al., 2012; Jimura et al., 2013; Li et al., 2013; Benningfield et al., 2014; Calluso et al., 2015) and therefore will not be reiterated in detail here.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Especially, we do not have any financial interests in the Delay of Gratification Test for Adults (DoG-A), which is freely available upon request by author SF.