Participants

Twenty-eight (mean age = 22 years, range, 18–25 years; 14 females) healthy, right-handed college students with no history of psychiatric and neurological disorders were recruited as participants. All participants were assessed to ensure normal or corrected to normal vision. Informed consent was obtained from all participants before the experiment according to the requirements of the Ethics Committee of Department of Psychology, South China Normal University.

Stimuli

During the scan, participants viewed two types of stimuli, facial photos for facial beauty and scene drawings for moral beauty. For facial stimuli, 72 photographs of greyscale photographs of non-famous Asian human faces (with only neutral facial expressions) were selected from the photographic databases of South China Normal University and Beijing Normal University, and rated by an independent group of 48 participants on a 7-point scale, forming two equal-sized sets: ‘beautiful’ faces (5.07 ± 0.27) and ‘common’ faces (3.09 ± 0.18) (seeFigure 1). Each set consisted of 18 male and 18 female faces. Facial photographs were digitized at 600 dpi in 8 bit greyscale, spatially downsized and cropped to fit in an oval ‘window’ sized 350 × 470 pixels. The scene drawings were 72 black-and-white drawings, in which the behaviours of cartoon characters in everyday life scenes were described. In the same pilot study, the 48 participants rated 360 candidate drawings based on the degree of moral beauty conveyed in these pictures (e.g. altruistic behaviour), which yielded 36 ‘moral beauty’ scenes (5.67 ± 0.70) and 36 ‘moral neutral’ scenes (4.22 ± 0.42) (seeFigure 1). In both stimulus sets, the main characters in half the drawings were male. Scene drawings were cropped to fit in a square ‘window’ sized 589 × 500 pixels.

Task

During the scan, participants performed the aesthetic judgment task and the gender judgment task on both stimulus types. In the face aesthetic (FA) judgment task, participants were instructed to judge whether the face was beautiful or common. In the scene moral aesthetic (SA) judgment task, participants were instructed to judge whether the scene drawing was morally beautiful or neutral. In the face gender (FG) task, participants were instructed to judge whether the face was male or female. In the scene gender (SG) task, participants were instructed to judge whether there was a male or a female figure in the picture. The aesthetic judgment required participants to make a left-finger response to beautiful faces or moral beauty scenes and a right-finger response to common faces or neutral scenes. The gender judgment task, which served as a controlling baseline, required participants to make a left-finger response to male faces or male main characters and a right-finger response to female faces or female main characters.

Procedure

The combination of the two task types (aestheticvs gender judgment) and the two stimulus types (facevs scene) yielded four conditions. A blocked design with 12 blocks pertaining to each condition was used. Block orders were fixed and counterbalanced across participants. Each block lasted for 15 s and was followed by 10 s of fixation (seeFigure 2). An instruction for task type appeared for 2 s before block onset. Each block included six trials. Within each block, every image was presented for 2 s subsequently after a 500 ms fixation baseline. Participants were instructed to fixate on each image and make judgments, accordingly, using their thumbs on the button box. Each subject had a single functional scan, which lasted ∼25 min.

fMRI data acquisition

Whole-brain fMRI data were acquired on a 3T Siemens Trio scanner with 12-channel phase array head coil. Functional images were obtained in one scanning run with a T2*-weighted gradient echo planner imaging sequence, each comprising 32 contiguous axial slices (3 mm thickness, 1 mm gap). The following parameters were used for scanning: Echo Time (TE) = 35 ms; Repetition Time (TR) = 2000 ms; flip angle = 90°; acquisition matrix size = 64 × 64; field of view = 192 mm. In addition, T1-weighted 3D structural image were acquired using a spoiled gradient-recalled sequence in a steady state (voxel size 1 × 1 × 1 mm³; TR = 19 ms, TE = 5 ms, flip angle = 20°).

fMRI data analysis

Data analysis was performed using statistical parametric mapping (SPM8) (http://www.fil.ion.ucl.ac.uk/spm). For each participant, the first six volumes, during which a fixation cross was presented, were discarded to allow for T1 equilibration. Images were spatially realigned to the first volume to correct for head movements. No individual runs had >3 mm maximum displacement or >1° rotation. The aligned functional images were first coregistered to the high-resolution T1-weighted structural image, normalized to the standard T1 template volume (MNI) and then resampled with voxel size of 2 × 2 × 2 mm³. The data were then smoothed with an isotropic full-width half-maximum 8 mm Gaussian kernel.

At the single participant level, a general linear model was applied to the fMRI time series where stimulus onset was modelled as single impulse response function and convolved with SPM8’s canonical haemodynamic response function (HRF). We modelled four regressors of interest: FA, SA, FG and SG. The four block types were time-locked to the onset of the block with block duration of 15 s. The instructions for task type (seeFigure 2) were modelled as a regressor of no interest. Six motion parameters estimated during the realignment procedure were included in the model as covariates of no interest. A high-pass filter with a cut-off period of 128 s was used to remove any low-frequency noise. At the first level, analyses were performed individually for each participant and contrast images were subsequently entered into a second-level analysis treating participants as a random factor.

At the group level, analyses were conducted at a voxel threshold ofP < 0.05 (FWE corrected) and a cluster threshold ofP < 0.05 (FWE corrected). We first performed the contrasts of ‘FA > FG’ and ‘SA > SG,’ respectively, using the flexible factorial analysis in SPM8. Based on our main interest of the common brain regions involved in the above two contrasts, we then computed a conjunction between the ‘FA > FG’ and ‘SA > SG’ using the minimum statistic approach (Nicholset al., 2005).

The parameter estimates in the regions of interest (ROI) were obtained by an iterative leave-one-out-cross-validation (LOOCV) analysis, in which all but one participant’s data were used to define the ROI for the left-out participant (Kriegeskorteet al., 2009). Each ROI was defined using a sphere with 10 mm radius centred on the peak voxel. This procedure was repeated 28 times by omitting a different participant each time, and the parameter estimates were then averaged across participants and plotted. As a result of the LOOCV analysis, the data used to define an ROI and the data extracted from this ROI (bar graphs inFigures 3 and6) are independent.

Behavioral results

We observed no significant differences in the reaction times (RTs) or accuracy between aesthetic judgment and gender judgment for either faces (RTs:t27 = −1.74, Mean Squared Error (MSE) = 0.03,P > 0.05, Cohen’sd = 0.45; accuracy:t27 = −1.81, MSE = 49.15,P > 0.05, Cohen’sd = 0.34) or scenes (RTs:t27 = 0.08, MSE = 0.02,P > 0.05, Cohen’sd = 0.08; accuracy:t27 = −1.61, MSE = 55.00,P > 0.05, Cohen’sd = 0.34). Thus, the observed brain activation differences could not be attributed to performance variation confounds across the different task types.

Imaging results

For FA judgments, a contrast of ‘FA > FG’ revealed strong activation in the bilateral medial SFG, left ITG and left inferior OFC (seeTable 1 andSupplementary Figure S1). For moral aesthetic judgments, a contrast of ‘SA > SG’ revealed activation in the bilateral inferior OFC, left SFG, bilateral ITG, left rectus gyrus, left middle cingulate, bilateral middle occipital lobe and right precuneus (seeTable 1 andSupplementary Figure S1).

Critically, as we predicted, the conjunction analysis of the two previous contrasts of interest showed that the neural networks involved in moral aesthetic judgments overlapped with those involved in facial aesthetic judgments (seeTable 1 andFigure 3). To further investigate the neural properties of these key regions, we also conducted an ROI analysis by using a LOOCV procedure. We found that the left inferior OFC, left ITG and bilateral medial SFG were more active during the aesthetic judgment task (P < 0.05) (seeFigure 3). Together, these findings suggest a common neural mechanism underlying aesthetic judgments irrespective of beauty forms, involving the left inferior OFC, left ITG and bilateral medial SFG, providing direct support for our hypothesis.

Participants

Twenty-two (mean age = 21, range, 18–25 years; 12 females) healthy right-handed college students who did not participate in Experiment 1 with no history of psychiatric and neurological disorders were recruited as participants. All participants were assessed to ensure normal or corrected to normal vision. Informed consent was obtained from all participants before the experiment according to the requirements of the Ethics Committee of Department of Psychology, South China Normal University.

Stimuli

Facial photographs and scene drawings were used as stimuli. For facial stimuli, we used the same stimulus sets used previously as ‘beautiful’ and ‘common’ faces in Experiment 1 (seeFigure 4). To reduce the visual differences of moral beauty and non-moral beauty scenes, paired scene drawings were created. In each pair, moral beauty and moral neutral scenes had identical backgrounds and characters, and only differed in the character’s actions. Before the experiment, an independent group of 36 participants rated the visual complexity, familiarity, artistry and degree of moral beauty of the candidate stimuli on a 7-point scale. Thirty-six pairs with significant differences in the moral beauty score but not in other indices were thus selected as stimuli for the purpose of achieving a better match of visual processing workload and subjective artistic preference (complexity:t35 = −1.33,P > 0.05; familiarity:t35 = −0.42,P > 0.05; artistryt35 = 1.79,P > 0.05; moral beauty:t35 = 22.73,P < 0.05), forming 36 paired drawings of moral beauty scenes and moral neutral scenes, respectively (seeFigure 4).

Task

The same gender judgment task, the details of which were described previously in Experiment 1, was used as the only task of Experiment 2 to keep the participant’s attention fixated on the presented stimuli in the absence of the explicit aesthetic judgment task.

Procedure

We used a blocked design with 12 blocks pertaining to each of the four experimental conditions (FB and FC, SB and SN). Block orders were fixed and counterbalanced across participants. Each block lasted for 15 s and was followed by 10 s of fixation (seeFigure 5). Each block included six trials. Within each block, every image was presented for 2 s subsequently after a 500 ms fixation baseline. Participants were instructed to fixate on each image and make judgments, accordingly, using their thumbs on a button box. Each participant had a single functional scan, which lasted ∼25 min. After the scan, all participants rated the facial photos based on the degree of facial beauty and the scene drawings based on the degree of moral beauty.

fMRI data acquisition and analysis

The fMRI data acquisition and pre-processing were the same as in Experiment 1. At the single participant level, we modelled four regressors of interest: beautiful face (FB), common face (FC), morally beautiful scene (SB), and neutral scene (SN). The four block types were time-locked to the onset of the block with a block duration of 15 s. Six motion parameters estimated during the realignment procedure were included in the model as covariates of no interest. At the first level, analyses were performed individually for each participant and contrast images were subsequently entered into a second-level analysis treating participants as a random factor.

At the group level, analyses were conducted at a voxel threshold of uncorrectedP < 0.001 and a cluster threshold ofP < 0.05 (FWE corrected). We first performed the contrasts of ‘FB > FC’ and ‘SB > SN’, respectively, using the flexible factorial analysis in SPM8. Based on our main interest on the common brain regions involved in the above two contrasts, we then computed a conjunction between the ‘FB > FC’ and ‘SB > SN’ using the minimum statistic approach advocated byNicholset al. (2005). Given the heavy involvement of OFC in mediating beauty in previous studies, we used small volume correction (SVC) on OFC with 10 mm radius centring on the coordinate of the OFC peak voxel (x = −21,y = 54,z = −9) from a previous study (Winstonet al., 2007). Parameter estimates in each ROI were obtained by a LOOCV analysis (Kriegeskorteet al., 2009). This procedure was repeated 22 times in Experiment 2 by omitting a different participant each time, and the parameter estimates were then averaged across participants and plotted.

Moreover, to further investigate the role of cortical reward region (OFC) and subcortical reward region (putamen) in the cerebral representation of beauty, we used the LOOCV procedure to obtain parameter estimates from OFC and putamen across participants, and then conducted a Pearson correlation analysis to examine the relationship between these parameter estimates and the behavioural ratings of facial and moral beauty obtained at the post-scan session. This procedure ensures that we are not prone to a ‘double dipping’ error in the subsequent correlation analyses (Kriegeskorteet al., 2009). Finally, we examined the difference between correlation coefficients by using the Fisher’s test (Fisher and Bennett, 1973).

Behavioral results

Behavioral results in Experiment 2 did not exhibit significant performance differences between beautiful and common faces (RTs:t21 = 0.11, MSE = 9.18,P > 0.05, Cohen’sd = 0.01; accuracy:t21 = −0.48, MSE = 0.01,P > 0.05, Cohen’sd = −0.11), or moral beauty and moral neutral scenes (RTs:t21 = 0.45, MSE = 10.51,P > 0.05, Cohen’sd = 0.04; accuracy:t21 = −1.79, MSE = 0.01,P > 0.05, Cohen’sd = −0.32). However, for the post-scan rating session, we found significant differences between the ratings of beautiful (5.08 ± 0.46) and common (2.97 ± 0.52) faces (t21 = 11.72, MSE = 0.20,P < 0.05, Cohen’sd = 3.87) as well as significant difference between ratings of moral beauty (4.84 ± 0.64) and neutral (3.58 ± 0.49) scenes (t21 = 12.85, MSE = 0.21,P < 0.05, Cohen’sd = 3.07), indicating that there was no mismatch between our stimuli selection criteria and participants’ perception of facial and moral beauty.

Imaging results

We found that compared with FCs, FBs elicited stronger activation in the left inferior OFC, left putamen, left SFG and right middle occipital cortex (seeTable 2 andSupplementary Figure S2). Stronger brain activation related to scenes conveying moral beauty (contrasted with those related to SNs) was observed in a large-scale brain network, including the left inferior OFC, right medial OFC, left middle temporal gyrus, left middle frontal gyrus, angular gyrus, right paracentral lobule, bilateral middle cingulate, left post cingulate, and left precuneus (seeTable 2 andSupplementary Figure S2).

A conjunction analysis of the above two contrasts showed that the left inferior OFC was activated by both facial beauty and moral beauty (seeTable 2 andFigure 6). Given the heavy involvement of the OFC in mediating beauty in previous studies, we corrected for the multiple comparison error rate by applying an SVC on the statistical results (see ‘Materials and Methods’ section). The ROI was defined as a 20 mm sphere centred on the peak voxel in the OFC (x = −21,y = 54,z = −9), identified from a previous study (Winstonet al., 2007). As a result, we discovered significant activation of the OFC (x = −36,y = 38,z = −14, peakt = 4.58) in the conjunction analysis (SVC,P < 0.05, FWE corrected). So, while the conjunction analysis revealed that there is a common cortical reward region (OFC) involved in the implicit perception of both facial and moral beauty, interestingly, the facial and moral beauty analyses revealed that the implicit perception of facial beauty involves both cortical (OFC) and subcortical (putamen) reward regions, whereas the implicit perception of moral beauty involves only a cortical reward region (OFC).

To further understand the neural properties of the OFC and the putamen in relation to facial and moral beauty, we conducted an ROI-based correlation analysis (see ‘Materials and Methods’ section for details). Both OFC and putamen activity positively correlated with facial beauty ratings (OFC:r₂₁ = 0.49,P < 0.05; putamen:r₂₁ = 0.44,P < 0.05) (seeFigure 7A and B). Fisher’s test (Fisher and Bennett, 1973) showed that there were no significant difference between the above correlation coefficients,z = 0.15,P > 0.05, indicating similar involvement of both cortical and subcortical reward systems in the perception of facial beauty. Additionally, we found a significant positive correlation between OFC activity and moral beauty ratings (r₂₁ = 0.67,P < 0.05) (seeFigure 7C) but no relationship between putamen activity and moral beauty ratings (r₂₁ = 0.21,P > 0.05) (seeFigure 7D). Fisher’s test revealed a significant difference between the above correlation coefficients,z = 1.79,P < 0.05, suggesting a stronger cortical (i.e. OFC) than subcortical involvement in the perception of moral beauty.