A total of 37 Chinese students from East China Normal University (6 males; mean age = 20.6 years, SD = 1.8) voluntarily attended the current study. Data from 8 participants attending the pilot study was not analyzed, and the remaining 29 participants were used for further analysis (4 males; 14 in Red Group and 15 in Green Group; mean age = 20.5 years, SD = 1.7). All were right-handed with normal or corrected-to-normal vision and normal color perception, and reported no psychiatric or neurological history. After completing all tasks, participants were debriefed and paid as compensation for their time. Written informed consents were obtained from all subjects, and the protocol was approved by the University Committee on Human Research Protection (UCHRP) at East China Normal University.

The scanning was conducted on a 3-Tesla Siemens Trio MR scanner, including four functional runs and one anatomical run in total. For functional images, 35 axial slices (FOV = 240 mm × 240 mm, matrix = 64 × 64, in-plane resolution = 3.75 mm × 3.75 mm, thickness = 4 mm, without gap) covering the whole brain were obtained using a T2^∗-weighted echo planar imaging (EPI) sequence (TR = 2000 ms, TE = 30 ms, flip angle = 90°). A high-resolution structural image was also acquired using 3D MRI sequences for anatomical co-registration and normalization (TR = 1900 ms, TE = 3.43 ms, flip angle = 7°, matrix = 256 × 256, FOV = 240 mm × 240 mm, slice thickness = 1 mm).

Dispositional measures were listed as mean (SD). The mean score of IRI was 69.5 (9.3) with the mean of the empathic concern subscale was 19.0 (4.6). The mean score of MEIM was 56.7 (8.5). For the measurement of HVIC, no significant difference was found between the scores of Collectivism subscale and Individualism subscale [6.8 (1.0) and 6.5 (0.8), respectively; t₂₈ = 1.4, p = 0.18]. All these scores were within the range of the norm values.

For the implicit racial attitude, the results of the IAT showed that D-index was significantly greater than zero (t₂₈ = 4.6, p < 0.001), indicating an implicit preference for own-race to other-race group members. For the explicit racial attitude, participants preferred Chinese people to Western people [3.5 (0.8); t₂₈ = 3.1, p = 0.005] and showed more positive attitudes toward own-race people than other-race people [Chinese: 7.6 (1.1); Western: 6.4 (1.2); t₂₈ = 3.7, p = 0.001].

Neither main effect nor interaction on response accuracy toward in- or outgroup Chinese/Western faces (Fs_1,28 < 1.2, ps > 0.1) was found between pre- and post-scanning categorization tests. The average response accuracy of 29 participants in categorization task was shown in Table 1.

In the fMRI session, faces penetrated by a needle were rated more painful than those touched by a Q-tip (F_1,27= 167, p < 0.001). Furthermore, participants displayed higher rating scores toward Chinese faces in comparison with Western faces (F_1,27= 49.0, p < 0.001) both in the nopain and pain condition (ts > 5.1, ps < 0.001). Regarding the reaction time, both main race effect and race × condition interaction were found (Fs > 25.9, ps < 0.001), and these differences came from the Q-tip condition of the Western faces, which was much faster than any other conditions (ts > 7.0, ps < 0.001). No other main effect or interaction was found (Fs_1,27< 3.9, ps > 0.06). Please see Table 2 for details.

The main effects of empathic responses for pain (i.e., contrast of pain vs. nopain) were found in the ACC and bilateral AIs, where participants showed stronger activity for the pain condition than the nopain condition, in line with previous findings (Lamm et al., 2011). In these brain regions, the main effect of race was found in the ACC (Figure 2A), where participants displayed increased activity to Chinese members than Western members (Fs_1,28 > 7.9, ps < 0.01) both in the nopain condition and in the pain condition (ts₂₈ > 2.2, ps < 0.05, Figure 2B). The main effect of group was found in the right AI (Figure 3A), where participants showed increased responses toward ingroup members than outgroup members (Fs_1,28 > 7.9, ps < 0.01) both in the nopain condition and in the pain condition (ts₂₈ > 1.7, ps ≤ 0.05, Figure 3B). Participants also displayed increased responses toward ingroup members than outgroup members in the orbitofrontal cortex (Small Volume Correction), whereas increased responses toward Western members than Chinese were observed in the visual cortices (FWE corrected). No significant interaction was found.

We calculated correlations between contrast values of nopain/pain vs. rest and dispositional measures, and found a negative correlation between D-index and racial bias (i.e., contrast of own-race vs. other-race) of ACC of non-painful stimuli vs. rest (r = -0.47, adjusted p = 0.022 after Bonferroni correction, Figure 4A) but a significant positive correlation between D-index and racial bias of ACC of painful stimuli vs. rest (r = 0.46, adjusted p = 0.025, Figure 4B). Post hoc T-test further showed a significant difference between the correlation coefficients in the pain condition and that in the nopain condition (t₂₆ = 4.54, p < 0.001).

We further assessed the relationship between the BOLD signals of the ACC and that of the rAI. In the nopain condition, we found a significant positive correlation between the contrasts of own-race vs. other-race (i.e., racial bias; r₂₈ = 0.49, adjusted p = 0.028 after Bonferroni correction) but not between the contrasts of ingroup vs. outgroup (i.e., ingroup bias; r₂₈ = 0.35, p = 0.07,) in these two regions (Figure 5A), which was stood for both ingroup (r₂₈ = 0.67, p < 0.001) and outgroup members (r₂₈ = 0.38, p = 0.04, Figure 5B). In the pain condition, on the contrary, we found a significant correlation between the contrasts of ingroup vs. outgroup (i.e., ingroup bias; r₂₈ = 0.57, adjusted p = 0.004) but not between the contrasts of own-race vs. other-race (i.e., racial bias; r₂₈ = 0.17, p = 0.39) in these two regions (Figure 6A), and the correlations stood both for Chinese members (r₂₈ = 0.53, p = 0.003) and Western members (r₂₈ = 0.76, p < 0.001; Figure 6B).

We found that Chinese models elicited stronger activation than Western models did in the ACC, both for nopain and for pain conditions. Consistently, Xu et al. (2009) reported that the empathic neural response in the ACC decreased significantly when participants viewed faces of other races. Stanley et al. (2008) also suggest that the ACC is involved in the detection and regulation, respectively, of implicit attitudes. Our results confirmed Stanley’s suggestion in that activity in the ACC was modulated by IAT. Moreover, we found that the activity of the ACC was modulated by implicit racial bias differently in the pain condition and nopain condition. To be specific, we found a positive correlation in the pain condition, but a negative correlation in the nopain condition, between racial contrasts of the ACC and D-index. Combined, these results suggest that the ACC is involved in empathic racial bias.

On the other hand, we found neural responses in the AI, but not the ACC, were modulated by minimal group, in line with a previous study (Lamm et al., 2007), as participants showed stronger BOLD signals to ingroup members than to outgroup members in the rAI. We assume that the difference between the BOLD signals is due to neither stimulus features nor affect, because all faces were neutral and were counterbalanced between participants. In addition, there was no explicit between-group affective or competitive difference during minimal group assignment. The difference between BOLD signals comes from minimal group assignment that is just a simple separation between ‘us’ and ‘them.’ Our results significantly extend previous studies about ingroup bias in face recognition/evaluation (Kurzban et al., 2001; Lebrecht et al., 2009; Hehman et al., 2010) and affective evaluation (Van Bavel and Cunningham, 2009).

To further investigate whether racial bias in neural response for pain is modulated by minimal group, we calculated the correlations between neural responses of the rAI (which is related to ingroup bias as we have discussed) and the ACC (which is related to racial bias as we have discussed) in the pain condition. We found a positive correlation in the contrasts of ingroup vs. outgroup, but not in the contrasts of Chinese vs. Western, between the ACC and rAI in the pain condition, indicating that racial bias can be modulated by ingroup bias in the pain condition. Our results are in line with a previous ERP study that racial empathic bias can be affected by minimal group (Sheng and Han, 2012).

We also calculated the correlations between neural responses of the rAI (which is related to ingroup bias as we discussed) and the ACC (which is related to racial bias) in the nopain condition. Interestingly, we found a positive correlation in the contrasts of Chinese vs. Western, but not in the contrasts of ingroup vs. outgroup, between the ACC and rAI in the nopain condition, indicating that ingroup bias modulates racial bias in the nopain condition. We also note that neural responses of the ACC were modulated by implicit racial bias differently between nopain and pain conditions, i.e., a positive correlation between racial contrasts of the ACC and D-index in the nopain condition, but a negative correlation in the pain condition. It is high unlikely that the increased activity in the nopain condition reflects higher empathy in participants with higher racial bias. Instead, these results suggest that different mechanisms are underlying nopain and pain conditions. Further detailed study may tackle this question. Nevertheless, these differences could lead to larger individual differences and make it harder to detect possible effects of ingroup bias on racial bias in previous studies (Contreras-Huerta et al., 2013, 2014).

Our study bears several limitations. First, there was a large discrepancy in the gender of our participants (only four males). Stereotypically, females are portrayed as more nurturing and empathetic (Christov-Moore et al., 2014) and self-report data consistently indicates greater empathy in women (Baez et al., 2017). Despite experimental and neuropsychological measures show no consistent sex effect on empathy in most studies (Baez et al., 2017) but not all (Yang et al., 2009), it is possible that our results may be affected by gender discrepancy and further study is needed to address this question. The second issue is that no significant interaction was found in the voxel-wise analysis of imaging results. We note that the effects of minimal grouping on racial bias are very weak. Thus, this study may lack the power to detect such small difference. Further study using more participants is desired to address this issue. The third one is related to frequently reporting extremely high correlations between measures of emotion and BOLD signals of brain, or “Voodoo correlations” (Vul et al., 2009). Note that in this kind of study, typically, a voxel-wise approach was used and voxels were selected because they correlated highly with the behavioral measure of interest. However, Vul et al. (2009) also argued that the “Voodoo correlations” can be avoided by an ROI approach in which ROI is selected “blind” to the correlations of BOLD signals with the behavioral measure, as in the present study. Moreover, a sample correlation based on 25 subjects has an approximate 95% confidence interval of ± 0.4 (Nichols and Poline, 2009). In our case, data from 29 participants were used for the correlation analysis. Thus, we believe that our results are robust and the possibility of “Voodoo correlation” is very limited.