A group of 22 young (11 males, mean age = 25.5 years, SD = 3.1 age-range 20–32) and 32 healthy elderly participants were tested. One elderly subject had to be excluded due to brain anomalies; the remaining 31 participants consisted of 14 males (mean age = 67.3 years, SD = 6.2 age-range 56–78). All (young and elderly) were right-handed, had normal or corrected-to-normal vision (including color-vision) and none of the participants reported any history of neurologic or psychiatric disorders, or current medical problems (excluding blood pressure).

The elderly successfully completed the Geriatric Depression Scale (GDS; mean GDS = 1.4, SD = 1.9, GDS ≤ 5 for all participants; GDS ranges from 0 to 15, scores higher than 10 indicate depression; Sheikh et al., 1991) and the neuropsychological battery of the “Consortium to Establish a Registry for Alzheimer Disease” (CERAD; Welsh et al., 1991) including the Mini Mental State Examination (MMSE; Folstein et al., 1975; mean MMSE = 29.5, SD = 0.77, MMSE ≥ 27 for all participants; MMSE ranges from 0 to 30, scores smaller than 25 indicate pathologies) to access mental well-being and cognitive integrity. Data from this group of elderly participants have already been published (Steiger et al., 2016). This study was carried out in accordance with the recommendations of the local ethics committee (Medical Association Hamburg). All subjects gave written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by the local ethics committee (Medical Association Hamburg).

The experimental task was divided into three phases: association, encoding and recognition (Figure 1). In the first phase (association) participants were informed about the color-reward association and subsequently had to press two different buttons (indicating high vs. low reward) in response to a blue or green frame (10 frames; in this phase, no reward feedback was given). The color-reward association was counterbalanced across participants, and the association phase was followed by two short training blocks that were identical to the subsequent encoding and recognition phase. Therefore, encoding during the actual experiment was not incidental.

During the encoding phase, 30 indoor and 30 outdoor scene images (gray scaled) were presented for 1.5 s each in random order, with the prompt “Indoor/Outdoor” below the image. Accordingly, participants indicated the indoor/outdoor status of the stimulus by button press using their index and middle finger of the right hand. Importantly, during picture presentation, either a green or a blue frame surrounded the image and predicted a high or low reward for a correct indoor/outdoor classification. For each participant, the color reward magnitude association was identical to the association phase, and the images were randomly assigned to a frame. The direct feedback (“+1€” or “+0.1€” for correct classification, “X” for incorrect classification) was presented for 1.5 s after a jittered inter-stimulus interval (ISI) of 1.75 ± 0.25 s in which a fixation cross was presented at the center of the screen. The contingency between cue and reward outcome was 100% and participants were informed before the experiment that they will be paid part of their actual earnings at the end. During a 10 min break after encoding, participants had to answer geography questions in order to avoid memory retrieval from working memory in the subsequent recognition test.

During the recognition phase, the previously encoded 60 images were randomly presented and intermixed with 30 new images together with the prompt “Old/New/Unsure” below each image. In order to prevent guessing, participants were instructed to only press “old” or “new” if they were confident. The decision had to be indicated via button press (index, middle and ring finger of the right hand) during the presentation of the image (3.5 s), which was followed by a central fixation cross in a jittered ISI of 2.25 ± 0.25 s before the next image appeared.

All pictures were presented at the center of the screen with a size of 7.6° × 4.5° (visual angle). The surrounding frame had an additional size of 0.4°.

The encoding and recognition phase were repeated for a second time after a short break with new images, resulting in a total of 120 encoded images (60 high and 60 low rewarded) that were intermixed with 60 new images during retrieval. Participants were paid 12% of their total earnings in addition to their hourly reimbursement rate at the end of the experiment.

Importantly, during all tasks, subjects were instructed to respond as quickly and accurately as possible.

For the encoding phase, the median RTs of each participant for the indoor/outdoor classification of high and low reward images were calculated using MATLAB (The Math-Works Inc., version 2014b). The median was chosen because it is a measure that is less susceptible to outlying responses. The hit rate (percentage of correct indoor/outdoor classification) was taken as a measure of accuracy.

For the recognition phase, we analyzed median RTs for correctly remembered images and corrected hit rates (CHR) separately for images that were either high or low reward predicting during encoding. CHR were calculated by subtracting the false alarm rate (i.e., new images that were classified as old) from the hit rate (correct old classification) for each participant. Items classified as “unsure” were not included in the data analysis. All data were analyzed using IBM SPSS Statistics (version 21).

During encoding and recognition, electroencephalographic (EEG) activity was acquired with a 60-channel active electrode system positioned according to the 10-20 system using acticap (Brain Products GmbH, Munich, Germany) and Brain Vision Recorder (version 1.03.0003). FCz was used as a reference and the right mastoid as a ground electrode. In accordance with the manual and previous studies using active electrodes (Eckart et al., 2014, 2016) impedances were kept below 20 kΩ. To control for horizontal and vertical eye movements, two pairs of additional electrodes were used. Data were recorded with a sampling rate of 500 Hz and a high-pass (0.1 Hz) and low-pass (1000 Hz) filter.

Preprocessing of the EEG data was conducted using EEGLAB (version 13_4_4b; Delorme and Makeig, 2004) and customized MATLAB tools. First, the continuous data were high-pass (1 Hz) and low-pass (60 Hz) filtered; subsequently, all trials of the encoding phase were epoched from 600 ms before to 1400 ms after onset of the image- (or feedback-) presentation and downsampled to 250 Hz. This resulted in epochs for the following four conditions: high reward image, low reward image, high reward feedback and low reward feedback. Note that four of the young participants did not have epochs for the feedback-stimuli due to technical issues.

Subsequently, all trials were visually inspected to identify major atypical artifacts (mostly movements or muscle artifacts) and bad channels. Moreover, artifacts due to blinks or eye movements were removed using independent component analysis (ICA; Delorme and Makeig, 2004). Bad channels had to be interpolated in five participants (only one single channel for each subject). After this procedure, all epochs were visually inspected again and rejected when still containing artifacts. Finally, epochs were re-referenced to the average reference. After preprocessing, an average number of 54 trials per participant and condition remained (55 for high reward images, 55 for low reward images, 54 for high reward feedback, 53 for low reward feedback).

The analysis of the EEG data was conducted using Fieldtrip (version 2015-11-12; Oostenveld et al., 2010) with customized MATLAB scripts. Time-frequency decompositions were conducted from 4 Hz to 30 Hz using convolution on the single-trial time series with complex Morlet wavelets (four cycles) in steps of 0.5 Hz in the frequency- and 8 ms in the time-domain. Power was averaged across trials for each condition of interest. Subsequently, a baseline correction was applied using the condition-specific relative baseline (100–90 ms prestimulus). Note that this relatively short baseline comprises data points from a 570–1000 ms time window for the frequencies of interest (4–7 Hz) due to the temporal smoothing introduced by the 4-cycle wavelet transformation.

For analyzing power differences between conditions in the theta-band (4–7 Hz), non-parametric cluster-based permutation tests (Maris and Oostenveld, 2007) were conducted in a time window from 0 ms to 900 ms after stimulus onset to avoid edge effects. T tests were conducted for all contrasts on each individual sample and adjacent significant samples (p < 0.05) were clustered with the restriction of considering only effects significant on three or more neighboring channels. A Monte Carlo estimate of the permutation p-value was calculated to control for multiple comparisons. Condition labels were randomly permuted (n = 1000) and test statistics were computed again for every of these random partitions. The proportion of randomly drawn partitions resulting in larger test statistics than in the real data gave the p-value. Clusters with p < 0.05 were considered significant.

We took advantage of structural MRI data for all elderly (but not young) participants acquired in a separately conducted study using a 3T MR system (Siemens Trio) with a standard 32-channel head coil as described in Steiger et al. (2016). Briefly, whole-brain multiparameter mapping (MPM; Draganski et al., 2011) was conducted on the basis of multi-echo 3D FLASH (fast low angle shot) images at 1 mm isotropic resolution with predominantly proton density (PD), magnetization transfer (MT) or T₁ weighting. The total scanning time of MPM protocol was approximately 20 min.

The Statistical Parametric Mapping (SPM8) framework (Wellcome Trust Center for Neuroimaging, London) and customized MATLAB tools were used for data processing. The semi-quantitative parameter map of MT represents the percentage loss of magnetization induced by the MT saturation pulse and was calculated as described in Helms et al. (2008).

The MT maps were segmented into gray matter (GM), white matter (WM) and cerebrospinal fluid (CSF) for voxel-based morphometry (VBM) within the unified segmentation approach (Ashburner and Friston, 2000, 2005). MT maps were used for segmentation to help separating the effects of iron concentration from atrophy (Helms et al., 2009; Lorio et al., 2014). The GM images were non-linear transformed to standard Montreal Neurological Institute (MNI) space using the diffeomorphic registration algorithm (DARTEL) implemented in SPM8 (Ashburner, 2007), scaled by the Jacobian determinants of the deformation field and smoothed with an isotropic Gaussian Kernel of 6 mm full width at half maximum (FWHM). Due to their a priori rather low specificity, WM volume maps were excluded.

A voxel-based quantification (VBQ) analysis was used (Draganski et al., 2011) on the MT images to preserve quantitative parameter values, reducing effects of residual misregistration and partial volume effects and enhance tissue-specificity, allowing us to conduct analyses in WM and GM separately. The MT images were normalized into MNI space using the participant-specific deformation fields from the DARTEL procedure without modulating by the Jacobian determinant. Instead, a combined tissue-specific weighting/smoothing procedure (3 mm FWHM isotropic Gaussian smoothing kernel) was used (Draganski et al., 2011).

Whole-brain linear regression models as implemented in SPM8 were used to investigate the relationship between the elderlies’ GM or MT maps, respectively, and their individual reward benefit on RT during encoding (i.e., RT low reward divided by RT high reward for each participant). Clusters with more than 25 voxel and a p < 0.05 after family-wise error correction at cluster-level were regarded significant.

Furthermore, on the basis of our a priori hypotheses (see “Introduction” Section), the SN/VTA and the NAcc were defined as regions of interest (ROIs). The mask for the NAcc (Figure 2A) was taken from the Harvard-Oxford-Atlas (50% probability mask), implemented in the FMRIB Software Library (FSL; Jenkinson et al., 2012). Since there was no SN/VTA-mask available in the Harvard-Oxford-Atlas, it was manually drawn in MRIcron (Rorden and Brett, 2000) on the basis of the mean MT-weighted (MTw)-images of the participants (Figure 2B). Mean GM and MT values were extracted from each participant’s ROI. Values from left and right hemisphere were averaged, since we had no hypothesis about laterality. Using IBM SPSS Statistics, the resulting values were entered into partial correlation analyses with reward benefit on RT during encoding, controlled for age. P-values were corrected for multiple comparisons (four comparisons: GM and MT in both ROIs) resulting in a significance threshold of p = 0.05/4 = 0.0125.

EEG power differences between high and low reward during encoding were calculated for each participant using the statistical mask derived from the non-parametric permutation test (see “Result: Effect of Reward Anticipation on EEG Spectral Power” Section). Since there was no effect across all elderlies, the significant cluster from the younger subjects (see Figure 4A) served as mask to extract the mean power for the elderly. As described above, the mean power differences were used as regressors on the elderlies’ GM and MT maps, respectively, in a whole brain regression analyses and also in the subsequent ROI analyses. Finally, the derived power values were also used for correlations with individual reward related RT differences during encoding.

2 × 2 analysis of variance (ANOVAs) were calculated to investigate the effects of reward (high vs. low) and age group (young vs. elderly) on the behavioral data. In the encoding phase, both young and elderly participants performed with high accuracy in the indoor/outdoor classification (mean hit rate = 97.77%, see Table 1), with no main effect of reward (F_(1,51) = 0.091, p = 0.76) or age group (F_(1,51) = 0.008, p = 0.928) and no interaction of both factors (F_(1,51) = 0.091, p = 0.76). For RT, there was a significant main effect of reward (F_(1,51) = 5.130, p = 0.028), a marginally significant main effect of age group (F_(1,51) = 3.914, p = 0.053) and a trend for an interaction between reward and age group (F_(1,51) = 2.860, p = 0.097). To further examine this trend, two-tailed t tests comparing RTs for high vs. low reward were performed within the two groups separately. It revealed a significant effect only in the young (t₍₂₁₎ = −2.08; p = 0.050) but not in the elderly (t₍₃₀₎ = −0.57; p = 0.575, Figure 3A). Finally, a one-tailed t test was conducted between both age groups on the difference between RT for high vs. low reward. It revealed a significant difference between both groups (t₍₅₁₎ = −1.69; p = 0.048, Figure 3B), which is in line with our a priori hypotheses of an effect of reward anticipation only in the young but not elderly subjects.

For the recognition memory task, there was no main effect of reward or age group on CHR and RTs, and no significant interaction between both factors (p > 0.05, see Table 1).

A first Monte Carlo cluster-based permutation test revealed a significant difference (p = 0.038) between spectral powers of high vs. low reward predicting images in the theta-band (4–7 Hz) of the young participants. We found that theta power was increased in the high compared to the low reward condition. The effect lasted from 56 ms until 552 ms after stimulus onset and the cluster was widely distributed across occipito-parietal and fronto-central electrodes (17 electrodes in total; Figure 4A). In contrast, for the elderly subjects, neither a positive nor a negative effect was found in the theta-band (Figure 4B). To further quantify this potential age effect, the spectral difference was calculated for every participant by subtracting the power for low reward from high reward predicting images. Importantly, a subsequent Monte Carlo cluster-based permutation test revealed a significant difference (p = 0.026) between both groups for reward anticipation in the theta-band in the time range from 40 ms to 624 ms after stimulus onset at central and occipital-parietal electrodes (14 electrodes in total; Figure 4C). This pattern mirrors the RT effect (Figure 3).

For reward feedback, no statistically significant difference between high and low reward was observed, neither in the young nor in the elderly participants. Similarly, there were no significant correlations between reward related behavioral differences in RT and changes in theta power.

To further investigate the absence of a reward anticipation effect in the elderly, MRI data of this group were analyzed (see “Material and Methods: Relationship Between Brain Structure and Behavior” Section). Specifically, the ratio between RTs of the two reward magnitude conditions during encoding (RT low reward divided by RT high reward) was taken as a measure of RT benefit from reward. It has been used as regressor in a whole brain analyses and was subsequently correlated with the structural markers (GM and MT) of the NAcc and SN/VTA, controlling for age (partial correlation). There was no significant correlation between RT reward benefit and MT or GM on a whole brain level, or with values extracted from the NAcc ROI. However, there was a significant positive correlation (r = 0.479; p = 0.007) between RT reward benefit and MT (but not GM) values of the SN/VTA WM (Figure 5).

The analyses of correlations between reward-related differences in the elderlies’ EEG signal and structural markers revealed no significant effect (p > 0.05).