Two hundred twenty female college students in Beijing University of Chinese Medicine were recruited by questionnaires (including PMS scale, BAI and BDI-II), 196 students fully completed all questionnaires. In accordance with the diagnostic criteria of PMS and health, 57 students took part in the study and were divided into PMS group (n = 33, age = 23.36 ± 3.42 years) and healthy group (n = 24, age = 24.43 ± 3.06 years). There was no significant difference between the groups (Z = 1.042, p = 0.297).

According to John Bancroft’s PMS Scale (31) and the American College of Obstetricians and Gynecologists (ACOG), PMS was defined as:(1) scoring≥6 on PMS scale; (2) showing any one symptom of 10 somatic symptoms (breast distending pain, gastrointestinal symptoms, fatigue, headache, abdominal pain, general malaise, diet and taste changes, weight gain, sweating, skin problems) or 11 emotional symptoms (anxiety, depression, irascibility, sensitivity, easy to cry, attention deficit, reduced sociability, confusion, insomnia, increased naps, fluctuations in sexual desire) within the period of 5 days before to 3 days after the menstruation. The symptoms disappeared 4 days after the menstruation and continued to occur for 3 months.

According to World Health Organization, health was defined as: (1) scoring<45 on Beck Anxiety Inventory (BAI); (2) scoring<20 on Beck Depression Inventory (BDI); (3) qualified by physical examination, without major diseases and chronic diseases.

Participants were included into PMS group when: (a) meeting the diagnostic criteria of PMS, and (b) having regular menstrual cycles (21–35 days); and were included into healthy group when: (a) meeting the criteria of healthy, (b) having regular menstrual cycles (21–35 days), and (c) reporting no major adverse life events in the past 2 months. Individuals with psychiatric disorders were excluded.

We conducted the experiment in the middle of second semester since students reported lower academic pressure in the second semester in our previous study (33). The experiment was scheduled to carry out 1–7 days before the menstruation (luteal phase) according to subjective report in the last 3 months, combined with the premenstrual discomfort of each participant. All experiments were completed between 9 and 11 a.m. Subjects were told to clean their heads before the experiment to maintain the sensitivity of electrodes. They took a break after arriving at the laboratory, wore electrode caps and fixed each electrode. After resting for 3 min, they watched angry, neutral and sad videos (the videos were presented randomly), and had a 10–15 min rest between videos. The emotional rating scales were filled out after resting and each video. The videos in our study are from the emotion video library of the State Key Laboratory of Cognitive Neuroscience and Learning at Beijing Normal University. Each video is about 3 min, scoring 5 points in emotional stimulation, with Asian leads and Chinese dubbing or subtitles.

The physiological data was recorded by MP150 physiological multi-channel recorder of Biopac. Indicators include skin conductivity level (SCL) and heart rate variability (HRV). For HRV, we selected both time domain indices: Standard deviation of NN intervals (SDNN), root mean square of successive NN intervals differences (RMSSD), numbers of pairs of successive NN intervals that differ by more than 50 ms (NN50), and frequency domain indices: very low frequency (VLF), low frequency (LF), high frequency (HF), total power (TP) of HRV signal, LF/HF. SDNN, RMSSD, and NN50 was related to parasympathetic activity (34). TP is the total power of spectrums, reflecting the activity of ANS. VLF was reported to be closely associated to parasympathetic activity (Taylor, Carr, Myers, & Eckberg, 1998). LF was correlated with sympathetic nervous system (SNS) and part of parasympathetic nervous system (PNS). HF was linked to PNS, and LF/HF was used to describe the balance of SNS and PNS, lower LF/HF usually indicates lower sympathetic activity and higher parasympathetic activity (35). The sampling rate of physiological data was 1,000 samples/s. The sensors were attached to the pulps of left hand’s index and middle fingers, and the subjects were asked to keep their left hand as still as possible during the experiment to reduce the interference of noise on the signal. The physiological data were extracted and analyzed by AcqKnowledge 4.2, in which the standardized index of SCL (SCL = (raw-min)/(max–min)) was used because of the large basic difference between subjects.

In addition, we collected saliva samples to measure the level of salivary cortisol (SCort), and the subjects were informed not to eat irritating foods before the experiment. About 1.5 mL saliva was collected at the resting stage and 10–15 min (36) after each video, and was stored at −20°C within 2 h. The SCort were detected by Beijing Protein Innovation using competitive enzyme-linked immunosorbent assay (Salivary (Human) Cortisol ELISA SLV-2930, range: 0.09–30 ng/mL, sensitivity: 0.09 ng/mL). Meanwhile, we measured systolic blood pressure (SBP) and diastolic blood pressure (DBP) using electronic blood pressure monitor (Omron model HEM-7052) at resting stage and immediately after each video.

We recorded EEG data by NuAmps 40 Channel Digital DC EEG Amplifier from Neuroscan (sampling frequency: 1000 Hz, filter bandwidth: 0-100 Hz), using bilateral mastoid as reference electrodes (A1 and A2), and 0.5 cm below the hairline of the forehead as GND. Thirty electrodes in the experiment were placed according to guidelines for standard electrode positions. Any power of electrode exceeding 100 μv was regarded as an artifact, and corrected data were exported using scan4. The EEG data was processed by MATLB with Fourier transform, filtering the power of δ band (0.15–4 Hz), θ band (4–8 Hz), α band (8–13 Hz), β band (13–25 Hz), and γ band (25–100 Hz).

The experiment data were analyzed using a 2 group (PMS, healthy group) × 4 condition (resting, angry, neutral, sad) repeated-measures ANOVA, and corrected according to Greenhouse–Geisser correction. The analysis mainly focused on the differences between groups. In correlation analysis, we relied on the Spearman correlation (0.1 < | r | < 0.3 as weak correlation, 0.3 < | r | < 0.5 as moderate correlation, | r | >0.5 as strong correlation). The threshold level for significance was p < 0.05. All statistical analyses were carried out in SPSS 26.0.

This work was approved by the Ethics Committee of Beijing University of Chinese Medicine, the ethics review number is 2017BZHYLL0313. All participants take part in this study voluntarily and signed informed consent.

The emotional results showed significant condition main effects (F_{pleasure (3,153)} = 70.885, p < 0.001, η² = 0.582; F_{arousal (3,153)} = 48.786, p < 0.001, η² = 0.489; F_{dominance (3,153)} = 24.524, p < 0.001, η² = 0.325; F_{serenity (3,153)} = 45.770, p < 0.001, η² = 0.473; F_{delight (3,153)} = 55.825, p < 0.001, η² = 0.523; F_{anger (3,153)} = 120.388, p < 0.001, η² = 0.702; F_{fear (3,153)} = 3.603, p = 0.026, η² = 0.066; F_{sadness (3,153)} = 79.618, p < 0.001, η² = 0.610; F_{disgust (3,153)} = 62.668, p < 0.001, η² = 0.551; F_{surprise (3,153)} = 3.925, p = 0.019, η² = 0.071). No obvious group main effect and group × condition interaction effect was found. The results are shown in Table 1.

After watching angry video and sad video, scores in pleasure, serenity and delight significantly declined and surged in arousal, dominance, anger, sadness and disgust, among which participants scored higher in anger after angry video and higher in sadness after sad video. These results showed that the angry and sad videos successfully induced anger and sadness, respectively.

Statistical results showed that there were significant group main effects on VLF (F_(1,43) = 4.183, p = 0.047, η² = 0.089) and SCort (F_(1,48) = 4.709, p = 0.035, η² = 0.089), differences were found in resting state. Significant condition main effects were observed on DBP (F_(3,150) = 8.744, p < 0.001, η² = 0.149), SCL (F_(3,138) = 3.615, p = 0.015, η² = 0.073), SDNN (F_(3,141) = 1435.085, p < 0.001, η² = 0.968), RMSSD (F_(3,144) = 201.978, p < 0.001, η² = 0.808), NN50 (F_(3,135) = 1050.835, p < 0.001, η² = 0.959), VLF (F_(3,129) = 70.414, p < 0.001, η² = 0.621), LF (F_(3,138) = 107.523, p < 0.001, η² = 0.700), HF (F_(3,129) = 8.276, p < 0.001, η² = 0.161), LF/HF (F_(3,117) = 70.205, p < 0.001, η² = 0.643), TP (F_(3,132) = 53.039, p < 0.001, η² = 0.547), SCort (F_(3,144) = 22.467, p < 0.001, η² = 0.319).

To be specific, in PMS group, DBP significantly raised during sad video, SCL raised during both angry and sad video, and HF raised during neutral video. In both groups, SDNN, RMSSD and LF/HF dropped dramatically during sad video, VLF surged during sad video, and NN50 raised during angry video and dropped continuously during neutral and sad videos. LF raised during neutral video and declined during sad video in both groups and declined during angry video in PMS group. TP surged during sad video in both groups and declined during angry video in PMS group. SCort declined after neutral video in PMS group and declined continuously in all three video conditions in healthy group.

No significant interaction effect was found. The VLF and SCort results are shown in Figure 1, and the mean value of each parameter in different conditions and groups are shown in Table 2.

There was significant condition main effect on grand mean of θ band (F_(3,144) = 4.776, p = 0.003, η² = 0.090), α band (F_(3,144) = 36.851, p < 0.001, η² = 0.434), γ band (F_(3,144) = 9.279, p < 0.001, η² = 0.162), and interaction effect on α band (F_(3,72) = 3.382, p = 0.040, η² = 0.066) and γ band (F_(3,72) = 3.330, p = 0.021, η² = 0.065). No group main effect was found. Simple effect analysis showed that the α band power of PMS group (F_(3,72) = 40.04, p < 0.001, η² = 0.440) and healthy group (F_(3,72) = 10.00, p < 0.001, η² = 0.164) during sad video were lower than that during the other three conditions. For γ band, the power of PMS group (F_(3,72) = 6.61, p < 0.001, η² = 0.115) during angry video was higher than that during neutral and sad video, and the power of healthy group (F_(3,72) = 4.85, p = 0.003, η² = 0.087) during sad video was lower than that during resting state.

Among specific electrodes, there was significant group main effect on β band power at T₆ (F_(1,51) = 4.235, p = 0.045, η² = 0.077), and interaction effects on δ band power at F₇ (F_(3,78) = 5.268, p = 0.004, η² = 0.094), F₈ (F_(3,78) = 4.404, p = 0.017, η² = 0.079), α band power at CP₃ (F_(3,78) = 4.177, p = 0.032 η² = 0.076), CP₄ (F_(3,78) = 3.696, p = 0.041, η² = 0.068), TP₈ (F_(3,78) = 2.725, p = 0.046, η² = 0.051), T₅ (F_(3,78) = 3.527, p = 0.016, η² = 0.065), P₃ (F_(3,78) = 4.533, p = 0.023, η² = 0.082), P₄ (F_(3,78) = 3.890, p = 0.033, η² = 0.071), and γ band power at F₃ (F_(3,78) = 3.358, p = 0.020, η² = 0.062), F_Z (F_(3,78) = 2.721, p = 0.046, η² = 0.051) FC₃ (F_(3,78) = 3.092, p = 0.029, η² = 0.057).

Simple effect analysis showed that the δ band power at F₇ and F₈ of PMS group was lower than that of healthy group during neutral video (F_{delta F7 (1,51)} = 5.261, p = 0.026, η² = 0.094; F_{delta F8(1,51)} = 4.076, p = 0.049, η² = 0.074), but higher than that of healthy group during sad video (F_{delta F7 (1,51)} = 4.968, p = 0.030, η² = 0.089; F_{delta F8(1,51)} = 5.112, p = 0.028, η² = 0.091). The α band power at CP₃, CP₄, P₃ and P₄ of PMS group was higher than that of healthy group at rest (F_{alpha CP3 (1,51)} = 6.119, p = 0.017, η² = 0.107; F_{alpha CP4(1,51)} = 4.347, p = 0.042, η² = 0.079; F_{alpha P3 (1,51)} = 9.288, p = 0.004, η² = 0.154; F_{alpha P4(1,51)} = 5.399, p = 0.024, η² = 0.096). The γ band power at F₃, F_Z, and FC₃ of PMS group was higher than that of healthy group during angry video (F_{gamma F3 (1,51)} = 5.338, p = 0.025, η² = 0.095; F_{gamma FZ(1,51)} = 5.022, p = 0.029, η² = 0.090; F_{gamma FC3 (1,51)} = 5.016, p = 0.029, η² = 0.090). The main results are shown in Table 3, and outcomes of β band power at T₆, δ band power at F₇, α band power at CP₃ and γ band power at F₃ are shown in Figure 2.

We used spearman correlation to analyze the relations between the score of PMS scale and the variations of indices in different video stages (compared with the resting stage, denoted by Δ below). The results showed that, among PMS participants, the score of PMS scale was positively correlated with Δθ band power (r = 0.434, p = 0.015) during sad video, negatively correlated with ΔSBP (r = −0.362, p = 0.039) and ΔScort (r = −0.529, p = 0.002) during angry video and ΔScort (r = −0.379, p = 0.036) during neutral video. The correlations between the score of PMS scale and the EEG data are shown in Table 4.

Among the power at specific electrodes, the fluctuations of θ band and α band were typically associated with PMS. Their variations are shown in Figure 3.

PMS participants reported lower scores on serenity for neutral video. This result suggested that for students with PMS, neutral stressors also stirred their emotions. Another explanation would be the persistent effect of other negative emotional videos. Previous studies implied that women with PMS might suffer from a weakening of positive experience (37), and even had trouble disengaging attention from negative emotions in severe cases (38), which is consistent with our results. However, since there was no significant group main effect, study with a larger sample size is necessary to get a definite conclusion.

Salivary cortisol, as an indicator of HPA axis’ activity, is an important regulatory hormone of stress reaction (39). Both acute and chronic stress reactions activate the HPA axis, leading to an increase in cortisol (40–42). The cortisol content reaches its peak after 10 min in acute stress (36). The low level and slow response of cortisol weakened the reaction capacity to psychological stress (43). The results that the salivary cortisol content of PMS group at rest was significantly lower than that of the healthy group, might indicate a weaker activity of the HPA axis for college students with PMS, which impaired psychological resilience and brought about more negative feelings (44). Other studies reported blunted cortisol reactivity to Trier Social Stress Test in women with PMS (45) and lower cortisol levels at rest and under mental stress in women with premenstrual dysphoric disorder (PMDD) (46). In our study, no statistic difference between the groups could prove a distinct response pattern of HPA axis in PMS participants to emotional videos, but the correlation analysis indicated an inverse relationship between the activity of HPA axis and PMS symptoms in coping with angry and neutral stressors.

It is notable that cortisol decreased significantly in video conditions compared to resting state, which appears to contradict general cognition and assumptions (47), but is consistent to some earlier research (48, 49). As participants seemed relatively stable at rest according to post hoc multiple comparison of mood scores, we excluded the stress caused by experiment itself, which might give rise to a high level of cortisol at baseline, and speculated that our emotion induction inhibited the HPA axis’ activity.

Furthermore, post hoc multiple comparison showed that the cortisol level after viewing angry video was higher than that after sad video, and was remarkably higher than that after neutral video. No statistical difference between the latter two. This indicated a higher activity of HPA axis in anger compared with sadness.

HRV is a non-invasive indicator for ANS activity. Our study found that compared with healthy participants, PMS participants showed higher VLF at rest. Considered all indicators, our findings more likely indicated stronger parasympathetic activities among PMS students in luteal phase, which was inconsistent with previous research (50, 51). Since there is no accordant conclusion about the relationship between VLF and PNS, the explanation needs to be taken with caution.

Under the stimulation of sadness, among all the participants, the VLF and TP increased significantly, and no difference was observed in HF compared with the baseline. The other indexes showed a downward trend. It is interesting that LF and LF/HF declined significantly, which mainly represented the weakening of sympathetic activity or the strengthening of parasympathetic activity, yet SDNN, RMSSD and NN50 also drastically reduced, suggesting a decreased activity of PNS. Contrary to common believes, researches questioned the utility of LF, HF and LF/HF as indicators in measuring sympathetic and parasympathetic activities because the physiological basis of them are hard to be confirmed (52, 53). These contradictory outcomes might implicitly approve the inaccuracy of LF and LF/HF. Together, we prefer to believe that the sympathetic activity strengthened and parasympathetic activity got weakened, and the overall activity of ANS, which was less sensitive to anger, was notably enhanced after the sad video. Similar result has been reported (54). The study claimed that lower parasympathetic activity may interfere self-regulation and behavioral flexibility. We therefore believe that the sad video in this experiment triggered stronger negative affectivity.

Additionally, the changes of SBP after inducing anger, compared to the baseline, correlated negatively and moderately with the degree of PMS. SBP is associated with the sympathetic activity (55). Research claimed that, compared with control subjects, women with PMS showed higher SBP and DBP at baseline and after cold pressor test (56), which seems to contradict our findings. However, reduced SBP and DBP were reported to be related to depression, anxiety and stress (57, 58). The difference in results may partly come from the different stressors.

Taken together with the results above, angry video did not seem to result in detectable stress response, and even inhibited the activity of HPA axis and SNS. Another explanation is that long-term pressure has changed the response pattern of the HPA axis, since chronic psychological stress substantially works on the HPA axis (59), and the sustained high activation may wear out the resilience, thus impairs acute stress response (60). In contrast, sadness elicited more intense response in ANS, which means stronger emotional perception. For ANS, previous research reported lower specificity in response to stress stimuli and poor correlation with PMS, compared with HPA axis (45). However, our results suggested that the activities of both HPA axis and ANS had certain bearing on PMS, especially in response to anger.

PMS participants presented different patterns of CNS activities under all four conditions in different bands, prominently in δ, α and γ bands.

In the resting state, higher power of β band in the right posterior temporal lobe and α band in the parietal lobe were observed in PMS group. α frequency band mainly suppresses the cortex, manifesting as enhanced power and increased inhibition under the stimulation of negative emotions, which thereupon narrows attention and range of cognitive (61). Liu et al. (10) also reported higher α activity in women with PMS at rest. β band is associated with maintaining the current sensorimotor or cognitive state (62) and is enhanced to keep attention when in high arousal of negative emotions (63). The result indicated that, for PMS students in resting state, heightened parietal inhibition and right posterior temporal arousal might be allied to the low activity of HPA axis and the high activity of PNS. However, previous studies mainly reported heightened activities in the left temporal lobe among PMS participants (64, 65). We suspected that the different recording time could be responsible for part of the inconformity.

During the angry video, PMS group manifested higher power of γ band in the left frontal lobe, and the variation of γ band power in frontal midline was positively related to the severity of PMS. γ band is involved in visual processing and sensitive to emotional stimuli (66, 67). Its enhancement is positively correlated with subjective emotional arousal (68), and is linked to sustained emotional elaboration (69). In addition, the orbitofrontal cortex (OFC) is essential in emotion processing, in which γ band will be activated when dealing with negative emotions and feedback (70).

Besides, in angry phase, α band power mainly heightened in healthy group, which is accord with the conclusion that negative emotions raise α band power, while that in PMS group mainly declined. However, in left anterior temporal lobe and frontal midline, the variation of α band power was positively correlated with the score of PMS, that is, the manifestation of PMS might get worse with higher inhibitory effect of α band in mentioned lobe when dealing with anger.

During neutral video, the δ band power of the anterior temporal lobe in PMS group was lower than that in healthy group. The results during the sad video were in contrast. δ band plays a major role in the cognitive process. It was reported to be more responsive to emotional stimulation, especially to negative emotions, and sensitive participants showed larger increments in δ band (63). Our results suggested that the δ band of students with PMS had relatively poor response to neutral stimulation but significantly stronger response to sadness, indicating a susceptibility to sad emotion in the anterior temporal lobe. Previous study also claimed that people with PMDD had higher activation in the temporal lobe under emotional stimulation (19).

We found it interesting that the increased γ band power in frontal lobe during neutral video was also in line with a higher PMS score. Research reported more intensive high γ band (50–80 Hz) to negative stimuli (71). Our results suggested that, among PMS participants, γ band correlated moderately with anger as well as neutral stressors in frontal lobe, especially in frontal midline, which might have some relation to reduced SCort. Moreover, declined power of α band in right posterior temporal and left occipital lobes, and β band in left middle temporal lobe was moderately associated with worse PMS manifestation.

During sad video, the PMS score was positively related to the variation of θ band power, prominently in the right frontal lobe, right central sulcus and parietal midline. Similarly, the increased power of θ band in right frontal lobe, right central sulcus and left parietal lobe during angry video, in the right central sulcus and parietal midline during neutral video was also positively correlated with PMS scores. θ activity was reported to be associated with cognitive demand (72). Higher θ activity in central and parietal lobe, similar regions in our correlation results, was observed among major depressive disorder (73). The results possibly suggested that PMS participants with more severe symptoms engaged more attention on videos.

In conclusion, our study implied that subjects with PMS had lower SCort and higher VLF at baseline, possibly indicating their lower activity of HPA axis and higher activity of PNS, which might be associated with the heightened inhibition of α band on the parietal lobe and the increased arousal of β band on the right posterior temporal lobe. In emotional regulation, the severity of PMS was moderately related to the decrease of Scort and the enhancement of γ in frontal midline during angry and neutral videos. The enhanced θ in right central sulcus and parietal lobe during all three conditions, the increased inhibition of α in left anterior temporal lobe and frontal midline under anger, the weakened inhibition of α in right posterior temporal and left occipital lobes as well as weakened β activity in left middle temporal lobe under neutral video also had medium correlations with more severe PMS. We speculate that γ activity in frontal lobe may play a role in lower SCort in dealing with angry and neutral stressors, but in the case of dealing with sadness, no distinct response pattern had been found in the HPA axis and ANS, different from the CNS results. It cannot be dismissed that, no intergroup differences were observed in subjective emotion scores for either sadness or anger, and the result did not correspond to physiological manifestation. We suspect that the conflict might stem from the suppression use that contributes to hide feelings (32, 74), yet more research is needed to be sure.