A total of 108 people, including 76 subjects with CID and 32 HCs, were involved in the study. The flowchart of the study participants is presented in Figure 1. All subjects were recruited from the Clinic of Sleep Disorders in the Affiliated Chaohu Hospital of Anhui Medical University. The patients (30 males and 46 females between 18 and 60 years old) were diagnosed with CID following the criteria in the third edition of the International Classification of Sleep Disorders (33). Furthermore, they should meet the following criteria: (1) age is 18–60 years old; (2) normal ability to complete the assessment scales and the memory test without problem in hearing and vision; (3) not experiencing any somatic disease (including immunologic, endocrine, cardiovascular, neurologic, liver, or kidney disease or organic brain disease) or any other psychiatric disorders; (4) not taking sedatives, antidepressants, antipsychotics, or any other drugs within 4 weeks; (5) non-pregnant or non-lactating females.

Thirty-two HCs were chosen as the control group using the following criteria: (a) no complaints or history of insomnia or mood disorders, (b) total score <7 on the Pittsburgh Sleep Quality Index (PSQI), (c) total score <7 on the Hamilton Depression Rating Scale-17 (HAMD), (d) a total score <7 on the Hamilton Anxiety Scale-14 (HAMA), and (e) Montreal Cognitive Assessment, Chinese-Beijing Version (MoCA-C) score >26 (34–37). Control subjects were recruited from the physical examination centers at the same hospitals. They received a comprehensive medical involving a medical history review, electrocardiogram, blood testing (hematology and serology), and physical examination. The control subjects were also matched to cases by age, gender, geographic region, and ethnicity. As pre-established criteria, any somatic disease (including immunologic, endocrine, cardiovascular, neurologic, liver, or kidney disease or organic brain disease) or any other psychiatric disorders were excluded.

Information on demographic characteristics, medical history, and family history were collected using a questionnaire developed by our group.

Depression and anxiety severity were separately assessed with the HAMD and HAMA, respectively (35, 36). The maximum score on the HAMD is 52, and the highest score on the HAMA is 56. A higher total score indicates greater depression or anxiety symptom severity. In China, the HAMD score cut-offs are as follows: ≤ 7, no depression; ≥8, mild depression; ≥18, moderate depression; and ≥24, severe depression. HAMA ≥7 indicates that the patient may have anxiety, and HAMA <7 indicates that the patient has no anxiety.

The PSQI is designed to measure subjective quality of sleep over the past month. It measures seven components of sleep: latency, quality, duration, disturbances, efficiency, the use of sleep medications, and daytime dysfunction. A higher total score indicates worse sleep quality (34).

All participants underwent one night of polysomnography (PSG) to acclimatize to the novel environment followed by 2 nights of PSG to collect study data. The recording assessed sleep continuity and structure. The former was measured as total sleep time (TST), sleep efficiency (SE), and sleep onset latency (SOL); the latter was evaluated for the following parameters: percentages of sleep stage 1 (N1%) and slow wave sleep (SWS, N3%), time awake after sleep onset, and time (REM) and percentage (REM%) of rapid eye movement sleep, apnea hypopnea index (AHI), oxygen desaturation index (ODI), SpO2 time <90% (TS90). Sleep parameters were staged according to the 2012 American Academy of Sleep Medicine Manual for the Scoring of Sleep and Associated Events recommendations (37).

Cognitive function was evaluated by experienced clinicians using MoCA-C and a modified version of the Nine Box Maze Test (NBMT). The MoCA-C scale is used for a screening of general cognitive abilities. The maximum score for the MoCA-C is thirty, however, the score <26 indicates the cognitive impairment in China (38). A modified version of the Nine Box Maze Test (NBMT) was used to evaluate memory, including spatial/object working memories (SWM/OWM), spatial/object reference memories (SRM/ORM), and object recognition memory (ORcM) (39). The testing procedures are described in detail elsewhere (40). The numbers of errors were counted to examine SWM, OWM, SRM, ORM, and ORcM performance.

Serum from blood samples was obtained the morning following PSG and immediately stored at −80°C until analysis. Serum concentrations of AQP4, CX43, and CX30 were measured by quantitative sandwich enzyme-linked immunosorbent assays according to the manufacturer's instructions (Wuhan colorfulGene biological technology, Wuhan, China).

All data were analyzed using SPSS^® 20.0 for Windows (IBM Corp., Armonk, NY, USA). Normally distributed data are expressed as mean ± standard deviation (SD). Mann-Whitney U-tests were performed to analyze non-parametric data expressed as the 25th, 50th, and 75th percentiles (P25, P50, and P75, respectively). Partial correlation analyses were used to analyze correlations between serum biomarkers and sleep parameters and PSQI (adjusted for sex, age, education, HAMD, and HAMA) and between serum biomarkers and general cognitive function and multi-dimensional memory (adjusted for sex, age, education, HAMD, HAMA, and PSQI). The diagnostic accuracies of biomarkers were evaluated by constructing receiver operating characteristic (ROC) curves and calculating the areas under the curve (AUCs). Differences were considered significant at P < 0.05 for all analyses.

There were no significant differences in sex, age, or education between the two groups (Ps > 0.05, Table 1). Compared to the HC group, patients with CID had significantly higher HAMD and HAMA scores (Ps < 0.001, Table 1). Meanwhile, the patients with CID had significantly longer SOL and N1, and significantly lower SE, TST, and N3 than the HC (Ps < 0.001, Table 2). There were no significant differences in AHI, ODI, and TS90 between the two groups (Ps > 0.05, Table 2). These results indicated that the patients who were recruited into the study met our inclusion criteria. Sleep parameters are detailed in Table 2.

Patients with CID had significantly lower MoCA-C scores (P < 0.001) and more SWM (P < 0.001) and ORcM (P < 0.01) errors than HC (Table 3). CID patients had significantly lower serum levels of AQP4, CX43, and CX30 (Ps < 0.001) compared to the HC group (Figure 2).

Partial correlations (adjusted for age, sex, education HAMD, and HAMA) between serum biomarkers and objective sleep parameters are listed in Table 4. The serum concentrations of AQP4, CX43, and CX30 were positively correlated with N3 (r = 0.250, P < 0.05; r = 0.323, P < 0.01; r = 0.334, P < 0.01). The same results were also observed in N3% (r = 0.233, P < 0.05; r = 0.302, P < 0.05; r = 0.335, P < 0.01). There were no significant correlations between serum astrocyte-related marker levels and PSQI scores, AHI, ODI, and TS90 (Ps >0.05). Partial correlation analysis (adjusted for age, sex, PSQI, HAMD, HAMA, and education) indicated that serum AQP4 level was positively correlated with MoCA-C score (r = 0.357, P < 0.01) and negatively correlated with SWM errors (r = −0.435, P < 0.01; Table 5). CX30 level was negatively correlated with MoCA-C score (r = 0.2.71, P < 0.01; Table 5).

In present study, we found that serum AQP4, CX43, and CX30 levels were significantly lower in the CID group compared to healthy individuals. These results indicate a certain extent of astrocytes dysfunction in patients with CID.

Our group has spent several years exploring the correlation between insomnia and astrocytes (16, 17). We found altered serum levels of astrocyte-related biomarkers in CID patients, with increases in S100B and glial fibrillary acidic protein and decreases in BDNF and GDNF, which led us to hypothesize a role of astrocyte dysfunction in CID (14). Interestingly, serum S100B levels did not normalize after effective therapy, indicating that astrocyte dysfunction may be an inherent trait in subjects who develop CID (13). However, this evidence is still preliminary (16). To obtain more results to support this hypothesis, we evaluated another group of astrocyte-related markers: AQP4, CX43, and CX30.

AQP4 and CXs are mainly expressed in astrocytes in the brain. We are aware of only one study describing the levels of AQP4 in the blood in neurological disorders, suggesting the lower the serum AQP4 levels, the higher the number of cerebral microbleed evaluated by MRI in the patients with Intracerebral Hemorrhage (32). This indicated an association between serum AQP4 levels and brain structure damage to some extent. Moreover, astrocytes dysfunction is found in most of brain pathologies, and are associated with changes in CX expression and functions. The basal swelling of astrocytic end-foot processes is accompanied by decreased expression of CX at the membrane (31). Obstructive sleep apnea (OSA) commonly co-occurred with insomnia and related to cognitive problems (41, 42). In addition, a recent study showed an increase in astrocytic AQP4 levels in mice, following exposure to chronic intermittent hypoxia (one of the main features of OSA) (43). In this study, none of the patients had OSA based on the respiratory PSG parameters. Thus, we have reason to believe that patients with CID have some astrocytes dysfunction according to the current study.

Astrocytes are able to detect and respond to neuronal activity by increasing local calcium (Ca²⁺) transients. These astrocytes communicate through CXs and release gliotransmitters to modulate neuronal activity at multiple locations (44). Therefore, astrocytes are ideally positioned to promote synchronization of large-scale neuronal networks, such as slow wave sleep (SWS), which is a reliable index of sleep homeostasis (45). The role of astrocytes in sleep homeostasis was previously investigated in the dnSNARE (dominant-negative soluble N-ethylmaleimide-sensitive fusion protein attachment protein receptor) mouse model. The authors concluded that defective astrocyte transmission could reduce SWS but did not influence REM sleep or awakening (13). Others reported that blockade of astrocytic CX43 also reduced slow wave amplitude in mice (46). Moreover, the study found that AQP4 was also linked with SWS (22). However, previous reports were mainly based on animal experiments, and this association has not yet been established in CID patients. In the current study, AQP4, CX43, and CX30 levels in subjects with CID all positively correlated with N3 sleep, also known as SWS. Our results support the idea that astrocytes may play an important role in sleep homeostasis. Interestingly, there were no significant correlations between serum astrocyte-related marker levels (AQP4, CX43, and CX30) and subjective sleep parameters (PSQI scores). This suggests that subjective and objective sleep appear to reflect two distinct neurophysiological processes.

Astrocytes play a significant role in cognitive abilities. They are capable of (1) regulating extracellular glutamate and potassium ion uptake as well as Ca²⁺ concentration, (2) releasing and taking up neurotransmitters, and (3) controlling energy supply to neurons to facilitate cognition (47). Indeed, a recent study find that AQP4 knockout mice exhibited impaired memory consolidation in the Morris Water Maze used to test learning and memory function (48). In addition, microinjection of TAT-Gap19 (a Cx43-specific hemichannel inhibitor) into the ventricle of mice impaired short-term spatial memory in a delayed spontaneous alternation Y-maze task (49). One study also showed that mice lacking CX30 have impaired memory (50). However, it is not clear whether astrocyte dysfunction is associated with cognitive dysfunction in patients with CID. Our findings showed that serum CX30 and AQP4 levels positively correlated with cognition assessed by MoCA-C and NBMT in patients with CID. Contrary to expectations, this study did not reveal a correlation between CX43 and cognitive impairment. This may be due to the high molecular weight of CX43, which cannot easily pass the blood-brain barrier (19). It is unfortunate that we were only able to assess blood samples; future investigations can expand our findings by measuring AQP4 and CX levels in cerebrospinal fluid (CSF). It is notable that lower AQP4 expression was also observed in CSF from patients with Alzheimer's disease and Optic Neuromyelitis (51, 52). In addition, recent studies also demonstrated that the distribution of cerebrospinal fluid appears to be under a circadian control and that AQP4 seems to support this rhythm (53).

ROC analysis was conducted to assess these biomarkers' discriminatory power for CID. The results showed that three proteins had AUC values of 0.726–0.823. The optimum cut-off values for serum AQP4, CX43, and CX30 for identifying subjects with CID were 2109.7, 2542.7, and 6348.5 pg/mL, respectively (Table 6). The use of biomarkers as an objective index may help support making a diagnosis of CID. Combining these three indicators may provide a sensitive and specific approach to distinguish between the CID and healthy people.