Patients were recruited from the Movement Disorders Outpatient Clinic at the Department of Neurology and control participants were recruited in Aachen and the surrounding area. Before inclusion in the study, patients and volunteers were screened for mental disorders using a short German version of a DSM-IV interview [SKID-PIT light (38)]. Only volunteers with no self-reported history of mental illness were included in the study. 14 patients with Parkinson’ disease (PD) and 13 healthy controls (HC) underwent the study protocol. Data of one patient had to be excluded from analysis due to technical problems. Data of 13 PD (5f, 8m) and 13 age-matched HC with no self-reported history of neurological diseases (6f, 7m) were analyzed. All participants were right handed (39) and had normal or corrected-to-normal vision. PD were medicated with antiparkinsonian drugs according to their symptoms and measured in the “ON” state. The mean levodopa equivalent daily dose (LEDD) [calculated according to Ref. (40)] across all PD was 586.52 (SD = 494.88). The averaged disease duration for PD was 5.94 years (SD = 4.39, range 0.9–15.4 years). Patients with PD underwent the Mini-Mental-Status-Test (41). The averaged sum score was 28.31 (SD = 1.55; Min = 25). An experienced movement disorder specialist assessed motor dysfunction in PD using the motor subscale of the Unified Parkinson’s Disease Rating Scale [(UPDRS-III) (42) M = 24.21 (N = 12, SD = 9.60)].

Groups did not differ with regard to age [Median (Mdn)_HC = 65, Mdn_PD = 68; U = −0.13 p = 0.898], years of school education [M_HC = 10.92, SD_HC = 1.32, M_PD = 10.15, SD_PD = 1.68; t(24) = 1.3, p = 0.206] or crystallized intelligence, assessed by a German neuropsychological test [Mehrfachwahl-Wortschatztest version B MWT-B (43) M_HC = 118.62, SD_HC = 17.15, M_PD = 113.54, SD_PD = 20.17; t(24) = 0.69, p = 0.496].

Participants underwent neuropsychological assessment and a functional magnetic resonance imaging (fMRI) measurement, completed a post-scanning questionnaire (PSQ) and a computer-based emotion recognition test.

We used the same experimental setup as in our study examining subclinical carriers of a single mutant Parkin allele (10) that was based on an earlier study on healthy participants (34). To assess emotion recognition, we used the Facial Expressions of Emotions—Stimuli and Test battery [FEEST (44)]. The FEEST contains two tests comprising photographs and morphed photographs of emotional facial expressions from the Pictures of Facial Affect series (45). These photographs are commonly used to estimate emotion recognition deficits in PD (46). In our study, participants underwent the Emotion Hexagon test. In this test, photographs of two emotions are blended, whereby the proportion of each emotion differs. Since sensing and expressing mixtures of emotions is very common (47), we assume a high ecological validity of the Emotion Hexagon test. The fMRI experiment enabled us to examine disturbances of the involvement of human homologs of mirror neuron areas in PD.

Participants were asked to judge emotions in a six-alternative forced choice paradigm from morphed emotional facial expressions of different levels of difficulty (44). The FEEST consists of blended continua of two different emotional facial gestures, which are presented on a computer screen (45). Similar emotional expressions (happiness—surprise, surprise—fear, fear—sadness, sadness—disgust, disgust—anger, anger—happiness) are blended with different proportions of each emotion (90–10, 70–30, 50–50, 30–70, 10–90%) resulting in a total number of 30 morphs. The test contains 150 test trials split into five runs, in which, the 30 morphs were presented once in a randomized order. In total, each emotion occurred ten times in each degree of difficulty. The labels of the six basic emotions were provided at the bottom of the computer screen as alternative choice. Participants were asked to name the emotion that best described the facial expression displayed on the computer screen. Participants were familiarized to the task by 30 test trials prior to the actual test. Responses were entered into the computer by the experimenter. Responses to the 90–10 and the 70–30% morphs were defined correct if the predominant emotion was chosen by the participant. Responses to the 50–50% morphs were considered correct if one of the two emotions included in the morph was chosen by the participant.

The stimulus material consisted of video clips, which had previously been used by us and our collaboration partners (10, 34, 48, 49). These clips consisted of emotional (smile), non-emotional (lip protrusion), and neutral stimuli (relaxed face without motion). The latter was used as a high level baseline. Each of the three facial expressions was shown by 24 actors (12 men), giving a total of 72 videos. Each video started with the actor displaying a neutral face for 1 s than expressing the facial gesture for 3 s and finally returning to display a neutral face for 1 s. Each video clip lasted 5 s in total. In addition, pixelated videos were created (Photoshop CS3 v.10.0^® and Adobe Premiere Pro CS3^®). In these pixelated videos, the actors’ faces were scrambled into randomly moving squares, whereas the background remained original. A fixation cross either highlighted in red, blue, or green was presented on the scramble for the mid 3 s of the video to cue participants for execution of one of the three facial expressions (see below). The stimuli were presented with MR-compatible goggles (Resonance Technology, Inc., Northridge, CA, USA) using the Presentation© software package v.11.0 (Neurobehavioral Systems, Inc., Albany, CA, USA).

The fMRI experiment was planned as a 3 × 2 factorial design with factors “facial expression” (emotional, non-emotional, neutral face) and “task” (execution, observation). Participants were instructed to (i) attentively observe the videos displaying actors or (ii) in case of pixelated video clips to execute a facial expression. The kind of facial expression (emotional/smile, non-emotional/lip protrusion, neutral/relaxed face) was designated by the color of the fixation cross. The participants were to execute the facial expression as long as the fixation cross showed up. Participants’ facial expressions were filmed with a scanner-compatible camera and monitored online. The measurement was interjected in case of repeated errors (e.g., participant executed a wrong facial expression or mimicked the actor’s face during observation). The participant was then reinstructed and the run was restarted (this happened in case of two patients and one healthy control).

Immediately after the fMRI measurement, participants were asked to rate their subjective feeling of happiness during each condition (PSQ).

We analyzed behavioral data using IBM SPSS Statistics version 21. The proportional FEEST data were arcsine root square transformed before comparison of mean values by calculating a repeated measures ANOVA (50). Normal distribution of demographic and PSQ data was checked by evaluating skewness and kurtosis of these data in the two groups (51). A variable was assumed normally distributed when skewness [S = skewness/SE (skewness)] and kurtosis [K = kurtosis/SE (kurtosis)] were not greater than ±2. Because normality was assumed for the PSQ and most demographic variables, repeated measures ANOVAs and/or (post hoc) independent t-tests (p < 0.05, two-tailed) were calculated (see also Section “Results” and the Section “Procedure and Participants”). The assumption of normal distribution was violated concerning the age of participants. Therefore, a Mann–Whitney U-test was calculated to compare the age of the two groups (52). Post hoc pairwise comparisons were Bonferroni-corrected (p = 0.05/number of tests).

All fMRI data analyses were performed with SPM8 (Wellcome Department of Imaging Neuroscience, London, UK) implemented in Matlab 8.1 (Mathworks Inc., Sherborn, MA, USA).

We obtained anatomical and functional images with a Siemens 3T Trio MR-scanner. Functional T2* weighted echo-planar images (EPIs) were obtained with the following parameters: TR = 2,400 ms, TE = 30 ms, flip angle = 90°, FoV = 240 mm, matrix size = 64 × 64, in-plane resolution = 3.8 mm × 3.8 mm, 36 slices with slice thickness 3.0 mm, and distance factor 3%. A high-resolution T1-weighted anatomical 3-D magnetization prepared rapid gradient echo image (TR = 1,900 ms, TE = 2.52 ms, TI = 900 ms, flip angle = 9°, FoV = 250 mm, 256 × 256 matrix, 176 slices per slab) was recorded.

The first five EPI volumes were discarded to allow for T1 equilibration effects. The remaining functional images were realigned to the first image to correct for head motion (53). Participants, except for one, moved less than 4.1 mm (translation) and 4.6° (rotation). One patient had a rotation value of 14°. Therefore, movement parameters were included as six additional regressors into the general linear model (GLM) as covariates of no interest to model variance related to absolute head motion. Prior to that, for each participant, the T1 image was co-registered to the mean image of the realigned functional images. The mean functional image was normalized to the MNI template [Montreal Neurological Institute (54, 55)], using a segmentation algorithm (56). Normalization parameters were applied to all EPI images and the T1 image. The images were resampled to 1.5 mm × 1.5 mm × 1.5 mm voxel size and spatially smoothed with an 8 mm full width half maximum isotropic Gaussian kernel.

Data were subsequently analyzed by a two-level approach. Using a GLM, each experimental condition (emotional observation, non-emotional observation, neutral observation, emotional execution, non-emotional execution, neutral execution) was modeled on the single-subject level with a separate regressor convolved with a canonical hemodynamic response function and its first temporal derivative. The parameter estimates for each voxel were calculated using maximum likelihood estimation and corrected for non-sphericity.

We calculated two second-level models, one to replicate previous findings on shared representations of facial expressions and one to identify brain activation differences of PD and HC. As a proof of concept, we identified shared representations for the emotional and the non-emotional facial expressions the current sample first. Following the analyses of our previous study in healthy participants (34), the neutral/static facial expression was conceptualized as high-level baseline. Activation during observation and execution of neutral facial expressions was subtracted on the single subject level: Observation: emotional minus neutral facial expression (E_OBS minus N_OBS) and non-emotional minus neutral facial expression (NE_OBS minus N_OBS); execution: emotional minus neutral facial expression (E_EXE minus N_EXE) and non-emotional minus neutral facial expression (NE_EXE minus N_EXE). Resulting four contrast images for each group (HC, PD) were fed into a flexible factorial second-level analysis using a one-way four-level ANOVA (factor: condition; blocking factor: subject). To identify shared representations for execution and observation, we calculated four conjoint conjunction analyses across the corresponding execution and observation contrast, separately for HC and PD and for the emotional and the non-emotional facial expression. In our previous study examining shared representations for happy and non-emotional facial expressions in healthy participants, we found significant differences between happy and non-emotional facial gestures, which we attributed to the enhanced communicative information of the emotional facial expression. We expected to replicate this previous finding here.

The second model was computed to assess differences between HC and PD. Here, we were especially interested in the communicative emotional facial gesture. All six first-level contrasts for each group (E_OBS, NE_OBS, N_OBS, E_EXE, NE_EXE, N_EXE) were fed into a flexible factorial analysis using a one-way 12-level ANOVA (factor: condition; blocking factor: subject). Four contrasts were computed: E_EXE: HC minus PD, HC minus PD; E_OBS: HC minus, HC minus PD. All contrasts were masked inclusive with OBS and EXE of the HC (t > 3.14) to restrict the analysis to brain regions activated during execution and observation, respectively, of emotional facial expressions in HC. Group comparisons were restricted to the emotional facial expression, but effects of interest plots were generated to display activation of areas of interest during all conditions. We obtained a mask, encompassing the right IFGop from the AAL atlas embedded in the WFU PickAtlas [Wake Forest University, Winsotn-Salem, NC, USA (57)] as IFGop was the region of most interest derived from the literature. Results of regions of interest analyses are reported FWE-corrected at a threshold of p < 0.05. Further exploratory whole brain analyses are reported at a threshold of p < 0.001, uncorrected, with an extent threshold of k > 10 voxel, to consider the small sample size. Brain structures were labeled using the Anatomy Toolbox v 1.6 (58, 59).

We extracted brain data from all clusters in the IFGop or IPL that were identified in the group comparison, because these two regions were the homologs of monkey mirror areas. We selected the eigenvariate option (adjusted for the effects of interest). Pearson’s correlation coefficient was used to assess the relation of brain data and emotion recognition in the FEEST as well as medication [LEDD (40)] and disease duration in PD, with results being reported significant at a threshold of p < 0.05 (two-tailed).

Means and SDs of the PSQ and the emotion recognition accuracies in the FEEST are presented in Table 1.

A 2 × 2 × 6 repeated measures ANOVA with within-subject factors “difficulty” (90–70 and 70–30%) and “emotion” (happiness, surprise, fear, sadness, disgust, anger) and between-subject factor “group” revealed a significant main effect of difficulty (F_{1, 24} = 20.30, p < 0.001) and a significant main effect of emotion (F_{5, 120} = 12.45, ε = 0.57, p < 0.001). There was neither a significant main effect of group (F_{1, 24} = 1.10, p = 0.306), nor any significant interaction effect (all p ≥ 0.328). A separate analysis was run for the 50–50% morphs because responses to these morphs were counted as correct if the participant recognized either of the two emotions contained in the morph. For the 50–50% morphs, PD performed significantly worse than HC (t₂₄ = 2.43, p = 0.023).

Post-scanning ratings were analyzed with a repeated measures ANOVA with “facial expression” (emotional, non-emotional, neutral) and “task” (observe, execute) as within-subject factors and “group” as a between-subject factor. There was a significant main effect of facial expression (F_2,46 = 12.79, p < 0.001) and a trend for the main effect task (F_1,23 = 3.42, p = 0.077), but no significant main effect of group (p = 0.637) and no significant two or three-way interaction effect (all p > 0.32). Three post hoc tests were calculated resulting in a Bonferroni-corrected threshold p = 0.017. Participants rated their happiness higher after production and observation (values of both tasks were averaged) of emotional compared to non-emotional (t₂₅ = 2.86, p = 0.008) or neutral facial expressions (t₂₅ = 5.77, p < 0.001). The difference of happiness ratings between the non-emotional and neutral facial expression did not survive the Bonferroni corrected threshold of p = 0.017 (t₂₅ = 2.38, p = 0.025).

In the fMRI results section, brain regions are labeled according to Anatomy Toolbox (58, 59) with cytoarchitectonic areas listed in brackets.

Both groups activated the right IFGop during execution and observation of emotional faces, but the response of the dorsal section of the IFGop and adjacent rostral part of the dorsal premotor cortex was reduced during observation of the emotional facial expression in PD relative to HC. Supposed to be the human homolog of monkey premotor mirror neuron area F5 (23), the IFGop has consistently been shown to be activated during observation and imitation of goal-directed movements (61) and processing of facial expressions (62). In our previous studies in healthy participants (34) and in Parkin mutation carriers (10), using a similar experimental design (but another scanning protocol and sample) conjoint activation for execution and observation of emotional facial expression was found inferior to the weaker than normal IFGop activation in PD in the current study. However, other studies found dorsal IFGop activated during observation (33, 63, 64) as well as during execution of emotional facial expressions (33) remarkably close to the peak in the present study. Activation of dorsal IFGop during observation of facial expressions was thought to represent activation of the mouth motor area (32) due to preparation of an automatic motor response, “facial mimicry” or a resonant activation of a dorsal sector of the mirror neuron system (33, 65, 66). Thus, reduced muscular mimicry accompanied by reduced resonance in the IFGop might account for emotion recognition deficits in PD. Besides this, lowered right dorsal IFGop activation was related with self-reported difficulties in identifying one’s own feelings (Alexithymia) in a recent study (67), and PD have been shown to suffer from Alexithymia double as often as control subjects (68). Thus, this region might also be important for the understanding of one’s own feelings. Interestingly, in contrast to Parkin mutation carriers (10), emotion recognition accuracy in PD with manifest Parkinson’s disease was not related to IFGop activity in the current study. Due to the small sample size, we had not enough power to detect small to medium size relations. Moreover, the FEEST might have been too easy to uncover emotion recognition deficits in PD. Future studies might use a more difficult emotion recognition test to enhance variance of emotion recognition accuracy and increase the sensitivity to detect relations between IFGop activation and behavioral performance.

We found widespread weaker than normal activation in patients with PD in the right IPL including supramarginal gyrus (PFt), somatosensory cortices, and the anterior IPS and a more localized decrease of activation in the left supramarginal gyrus (PFt) and somatosensory cortex.

Decreased activation of the supramarginal area PFt is in line with our hypothesis of a disturbed resonance in homologs of monkey mirror areas in PD, in contrast to stronger involvement of supramarginal gyri as shared representation of execution and observation of facial affect in HC [concerning HC, see also (69)]. Furthermore, somatosensory cortices [bilateral SI and SII (22), left SII (32), right SII (33)] have previously been shown to be activated during observation and execution/imitation of emotional facial expressions (but see two studies that did not find overlapping activity in the inferior parietal lobules 31, 34). In accordance with the positive relation of right IPL activation and emotion recognition accuracy in patients with PD in the current study, lesions of right IPL encompassing somatosensory areas and supramarginal gyrus have previously been reported to cause emotion recognition deficits (70). Furthermore, abnormal activation and connectivity patterns of the right IPL in patients with PD were shown in previous resting state studies (71, 72). Decreased connectivity of IPL and primary motor cortex and SMA was supposed to reflect a disturbance in networks linked to motor preparation and initiation in PD (73).

The IPS was assumed to be part of a frontoparietal network involved in adaptive online control of actions (74–76) and coding of action goals (75). Self- and other-generated actions are commonly represented in the IPS (75). Right IPS involvement was reported to be selective for face compared to object processing (62), shown during short presentation of facial expressions (77) and during emotion differentiation (78).

Analogously to our antecedent study (34), HC activated the right (pre-)SMA during observation and execution of emotional facial expressions. Interestingly, activation of this region was significantly decreased in PD patients. The Pre-SMA is involved in motor preparation (79). Activation during observation might, therefore, represent a starting contagious motor response of the observers face. Deviant activation in PD might result from functional changes in the basal ganglia-cortical motor loops, as pre-SMA is interconnected with the striatum (80) and the subthalamic nucleus (81), which in turn is connected to globus pallidus. The weaker activity in patients with PD compared to HC in our study is in accordance with a previously reported decreased brain activation (indexed through the amplitude of low frequency fluctuations) in the SMA during resting state (82) and an increase of activation during mental simulation of actions when PD patients were on compared to off dopaminergic medication (1).

We examined emotion recognition and processing of emotional facial expressions in a small sample of medicated patients with PD. Given the small sample size, the power of our study is appropriate to detect large correlations, but error probability β is high concerning medium and small effects (83). Furthermore, we measured medicated patients. And although our data are in accordance with and complement our findings in Parkin mutation carriers (10), and emotion recognition deficits in PD have been shown to be unrelated to dopaminergic medication (46), the results of this study have to be interpreted with caution until replication in larger samples and with unmedicated patients. Moreover, although similar brain activation patterns were shown during observation and execution of facial affect irrespective of which facial emotions were included (22, 32, 33), we cannot generalize our brain activation results to other emotions. Future studies could test the relation of emotion recognition and brain activation during observation of diverse facial emotions.