We recruited 28 right-handed Chinese University students (14 males and 14 females). They ranged from 18 to 25 years old (four participants did not report their age; the mean age of 24 participants = 21.21 years ± 1.82). All participants had finished at least high school (i.e., they each had at least 12 years of formal schooling). They each had their own microblog (Weibo) account and had experience reading and reposting information through social media. All of them had normal or corrected-to-normal vision. None had a history of traumatic brain injury, medical conditions, or any psychiatric disorder that could affect neural activity and brain functioning.

This study was approved by the Human Research Ethics Committee for Non-Clinical Faculties in South China Normal University. All participants signed on a consent form which exhibited the study purpose, the experimental procedure and the amount of payment in the beginning of the study. They were allowed to quit at any stage of the experiment without any punishment.

Across three categories, 120 pieces of microblog messages were extracted from Sina Weibo microblog system (Weibo²). The microblog messages in the positive category reported positive events in social life (e.g., someone was praised by the public for his/her contribution), whereas the microblog messages in the negative category reported negative events (e.g., a terrible accident happened, killing dozens of students). The microblog messages with no emotional significance were categorized as neutral and were included to provide a more realistic simulation of microblog messages. Each piece of microblog message was revised or modified to be a standard length (90–140 Chinese characters), and was written in the third person (as is commonly used in the newspaper). In order to control for possible bias from the original poster, the microblog messages used in this study were posted by four different publishers: Tencent, Sina, Sohu, and NetEase (all common, official publishers in Sina Weibo microblog).

Thirty participants were recruited to rate the microblog messages on (1) the emotional valence shown by each piece of microblog message, (2) the emotional arousal elicited by each piece, and (3) their familiarity with the microblog message provided in each piece (all on 7-point scales). Using these ratings, 90 out of the initial 120 pieces of microblog messages were selected (30 in each category). The three categories of microblog messages differed significantly in valence [F(2,87) = 380.52, p < 0.001] and arousal [F(2,87) = 124.07, p < 0.001], but not in familiarity (p > 0.1, see Table 1). Post hoc analyses showed that the valence of all three categories differed significantly from each other (ps < 0.001). In addition, neutral microblog messages elicited less arousal than both positive microblog messages [t(58) = 12.15, p < 0.001] and negative microblog messages [t(58) = 14.13, p < 0.001], whereas no significant difference in arousal was observed between positive and negative microblog messages [t(58) < 1]. In order to exclude a familiarity effect, pieces of microblog messages rated higher than 4.5 (out of 7) on the familiarity scale were excluded. The three categories of microblog messages did not differ in length [F(2,87) = 1.37, p = 0.258].

Participants read all microblog messages after resting state fmri scanning in the scanner. Each trial began with a fixation cross in the center of the screen for 1 s. A piece of microblog message was then presented on the screen for 15 s. This duration was based on the average reaction time from participants in the pilot study. The screen layout mimicked the interface of the Sina Weibo microblog. In a subsequent response stage, the participants’ task was to make a prompt decision (within 3 s) to repost or not repost this piece of microblog message. The decisions “repost” and “not-repost” corresponded to two buttons both on response pad and on screen (one on the left and one on the right, counterbalanced). After the decision was made, feedback was given in the form of the words “Reposted” or “Not Reposted” appearing on the screen for 1 s. The inter-stimulus interval was jittered between 3 and 6 s.

In total, each participant read 90 pieces of microblog messages (all microblog messages were listed in Supplementary Materials). All microblog messages were pseudorandomized within each participant such that no more than three pieces of microblog messages with the same valence were presented successively. The total duration of the task was 36 min.

Magnetic resonance imaging (MRI) data were acquired using a Siemens Tim Trio 3T MRI scanner at South China Normal University. The scanner was equipped with a standard eight-channel head coil. Functional images were acquired based on a T2^∗-weighted gradient-echo echo-planar imaging pulse sequence parallel to the AC-PC plane with following parameters: 160 volumes, 36 slices in each volume, TR = 3000 ms, TE = 30 ms, slice thickness = 3.5 mm, flip angle = 90°, matrix = 64 × 64, and a FOV = 224 mm × 224 mm. Anatomical images were obtained based on a T1-weighted gradient-echo pulse sequence (176 slices, TR = 1900 ms, TE = 2.52 ms) with a spatial resolution of 1 mm × 1 mm × 1 mm.

Preprocessing of the imaging was carried out by SPM8 software (Wellcome Department of Cognitive Neurology, London, UK), DPARSFA toolbox Version 3.0 (Yan and Zang, 2010), and DPABI toolbox (Yan et al., 2016) implemented in Matlab R2012a (Version 7.14.0.739, MathWorks Inc., Sherborn, MA, USA). The first 10 volumes of rs-fMRI data were discarded and the preserved functional images were corrected to the middle slice in each volume in order to adjust the differences in acquisition times of multiple slices. They were realigned to the first volume in the resting state scanning session for head motion correction and resliced to form a mean image. They were then co-registered to high-resolution T1 images at an individual level. After that, six rigid body head motions parameters, white matter, CSF and global signal of the whole brain were regressed out (Yan et al., 2013). Functional images were normalized to the Montreal Neurological Institute (MNI) template (resolution = 3 mm × 3 mm × 3 mm) using unified segmentation T1 images and smoothed with a Gaussian kernel (FWHM = 8 mm) to decrease the individual differences and increase the normality of signals. The smoothed signals were detrended and band-pass filtered (0.1–0.01 Hz) to remove the cardiac and respiratory signals before they were ready for further analysis.

Priori regions of interest (ROIs, 8 mm spheres) including bilateral TPJ (MNI coordinates of center: -48, -51, 15; and 51, -60, 18), DMPFC (MNI coordinates of center: -12, 42, 37; and 11, 39, 43) and MPFC (MNI coordinates of center: -3, 51, -3) from Falk et al. (2012, 2013) were engaged as seeds separately for functional connectivity analyses. The mean neural activities of all voxels within the corresponding seed ROIs for each participant were extracted and correlated (using Pearson’s correlation, r) with all other brain voxels for every time point within the time course of the preprocessed rs-fMRI data. R-values were then transformed to approximately normally distributed Fisher’s Z scores to produce the standardized functional connectivity (z-fc) maps at individual level. The z-fc maps for all participants were entered into second-level, voxel-wise, multiple regression models for identifying the regions where we could observe significant correlations between z-fc and reposting rate of each type. Results were corrected for multiple comparisons using Gaussian Random Field theory (GRF: voxel threshold: |Z| > 2.32, cluster-level threshold: p < 0.05). The mean functional connectivity of all voxels within each activated cluster was further extracted to calculate the magnitude of correlation between rs-fMRI and reposting rate of corresponding type of microblog messages.

A one-way ANOVA was conducted to explore the difference in reposting rate associated with emotional valence. A significant difference was observed across three valences of microblog messages, F(2,54) = 18.19, η² = 0.40, p < 0.001. Microbologs with positive valence (M = 0.48, SD = 0.24) were less reposted than those with negative valence (M = 0.64, SD = 0.21, Mean Difference = -0.16, Bonferroni corrected p = 0.012) but more reposted than neutral ones (M = 0.30, SD = 0.28; Mean Difference = 0.18, Bonferroni corrected p = 0.005).

Functional connectivity was computed based on seeds at bilateral TPJ, DMPFC, and MPFC. The correlations between average functional connectivity with reposting rate of each type of microblog messages were compared with zero, respectively.

Significant correlation with reposting rate of positive microblog messages was observed for the functional connectivity between right TPJ (ROI center at 51, -60, 18, MNI coordinates) and right superior temporal lobe (BA 48, peak at 48, -15, 3, Z = 3.87), R = 0.70, p < 0.001 [see Figure 1A], indicating participants with larger connectivity between these two regions would repost more positive microblog messages. Significant positive correlations with reposting rate of negative microblog messages were observed for the functional connectivity between left TPJ (ROI center at -48, -51, 15) and left middle frontal lobe (BA 9, peak at -30, 30, 39, Z = 4.33), R = 0.75, p < 0.001, and right insula (BA 48, peak at 39, 3, -9, Z = 3.43), R = 0.56, p < 0.001, as well as the functional connectivity between left DMPFC (ROI center at -12, 42, 37) and MOFC (BA 11, peak at -6, 33, -15, Z = 4.01), R = 0.59, p < 0.001 [see Figure 1B]. Significant positive correlations with reposting rate of neutral microblog messages were observed for the functional connectivity between left TPJ (ROI center at -48, -51, 15) and left VLOFC (BA 11, peak at -30, 39, -9, Z = 4.02), R = 0.72, p < 0.001, as well as the functional connectivity between bilateral DMPFC (BA 32, peak at -12, 33, 33, Z = 3.93), R = 0.68, p < 0.001 [see Figure 1C]. All significantly correlated regions are shown in Table 2 and Figure 1.

Sharing information is a productive activity in the social network. Some previous studies show people are thought to have higher intention to share the “good news” with peers (Gable and Reis, 2010; Choi and Toma, 2014) because it improves the individual’s well-being and the atmosphere in a group (Otto et al., 2015). However, our behavioral findings indicate that positive and neutral news tend to be ignored by microblog users, while negative microblog messages are more frequently reposted. Negative stimuli exert a larger effect in individuals’ cognitive performance and attention (Ito et al., 1998; Ohira et al., 1998), helping them avoid dangers (Neuberg et al., 2011), and thereafter robustly attract participants’ interest and are more likely to be propagated than other emotion in SNS. These findings are comparable to those in recent studies on the prevalence of “bad news.” People are more interested in negative gossip about celebrities (e.g., the president did the wrong thing) and positive gossip about themselves. In previous research, hearing negative gossip about others induce higher emotion ratings, as well as greater neural activity in the reward system (Takahashi et al., 2009). Moreover, our findings further extend the boundary of existing knowledge of emotion contagion through SNS. People are not only attracted by negative messages but also were eager to distribute them, making the bad news travel fast.

As for rs-fMRI, TPJ is one of key node of mentalizing network (Mars et al., 2012). It plays a key role in social communication (O’Donnell et al., 2015; Tang et al., 2016), even in the implicit condition (Zerubavel et al., 2015; Welborn et al., 2016). In the conditions without face-to face communication, activity at TPJ was highly increased when participant engaged in more consideration to other’s perspective (Falk et al., 2013). This hyperactivity helps participant successfully spread his/her own idea to others. In another study, the amplified activity at this system can predict participant’s usage of social language to expose one’ own thought (O’Donnell et al., 2015). Our result further extends these results and demonstrates that TPJ also engaged in mentalizing, an essential process of information propagation through on-line SNS. The increased reposting rate of negative messages is parallel with increased rs-functional connectivity of TPJ with insula and with middle frontal lobe.

Insula is the key node of empathy, especially the feeling of other’s pain (meta analysis from Fan et al., 2011; Lamm et al., 2011). Kanske et al. (2015) propose the distinct functions of insula and TPJ: the former underlie empathy, while the latter is involved in mentalizing or theory of mind. Since our messages in negative category report negative events in social life, the functional connectivity between these two regions might be associated with individual selective ability to feel other’s pain emotionally and cognitively. The increased functional connectivity reflects participants’ higher sensitivity to the pain of character described in negative messages, leading to their higher tendency to reposting such type of messages. Another possibility of positive correlation between the reposting rate of negative messages and rs-functional connectivity of TPJ and insula reflects participants’ sensitivity to disgust stimuli exposed by negative messages. Participants who exhibit a higher propensity of disgust both from their own and from recipient’s perspectives (Phillips et al., 1997) would be more likely to repost this type of messages due to its social significance.

It is worth noticing that increased rs-functional connectivity between TPJ with middle frontal lobe was also observed. More than evaluating the emotional significance of microblog messages from self and other’s perspectives, our brains are also preparing for future successful social interactions by considering the social significance of message. Thus, the cognitive control network is engaged in reposting negative messages due to its significant effect on the survival of human beings. This regulation conflicts with an intuitive process that represents self-interest and motivation (Feng et al., 2015), in which making decisions needs extra cognitive resources to resolve the ambivalence, inducing higher activity at bilateral DLPFC (Cikara et al., 2010; Kennerley and Walton, 2011). Moreover, it is demonstrated that activity at TPJ not only reflects participant’s consideration to other’s benefit, it is also related to the conflict resolution of benefits between self and others (Strombach et al., 2015). Our findings indicate participants are involved in a similar regulating process when participants are sharing on the web rather than face-to-face. The microblog messages that make us to undergo a feeling of pain disgust or pain might be those with significant social effects need to be reposted to the public. Participants with higher sensitivity to negative stimuli would have a higher tendency to intuitively avoid them but with have higher consciousness to its significance, so the simultaneous hyperactivity at these two regions leads to the increased rs-functional connectivity between TPJ with middle frontal lobe which compensates the increased rs-functional connectivity between TPJ with insula, as well as the increased reposting rate. This shows the reason why negative messages are given priority in the information sharing on the microblog.

Firstly, the relationships between neural activity and emotional valence have eluded consensus (Lamm et al., 2015; Wagner et al., 2015). In this study, we only measured rs-fMRI and reposting rate of each emotional type in lab, exploring the association between rs-functional connectivity and reposting rate. It is far from drawing a conclusion on the neural basis of reposting behavior in real-life SNS. Task-related fMRI study would be further conducted to measure the neural activity induced by microblog messages with different valence and related to information propagation. On the other hand, the study investigating the association between neural activity and behavioral performance in real-life by excluding confounding factors is also needed.

Second, microblog messages are restricted to one period of time, which might induce different emotional arousal and familiarity to the participants who participate in earlier stage and the participants who participate in later stage. Thus in this study, there were only 28 participants recruited within 1 month. The number is relatively small in imaging study exploring functional connectivity. Further studies with larger sample size may be conducted by better selection and controlling the experimental materials that are not time-restricted.

Third, we explored the functional connectivity of self-reference processing system and mentalizing system but we observed no significant correlation between the functional connectivity of self-reference processing regions and reposing rate. The discrepancy between our result and previous study might be due to the difference in measures and materials. Meshi et al. (2016) calculated the self-reported sharing score by participant’s self-reported updating frequency on Facebook, which might be more related to participant’s self-construct. We calculated the reposing rate of microblog messages, which described almost social events. This selection of our materials decreased its relatedness to self. Further analysis would be conducted for comparing the distinct neural associates with self and social sharing.