A multidimensional astrocytic framework links serum biomarkers to depression subtypes cognition and sleep EEG Authors: Zi-Jie Ma, Tian-Tian Zou, Yu-Shun Zhou, Xue-Yan Li, Zhen-Yu Hu, Gao-Xia Liu, Lan Xia, Gui-Hai Chen Published: 2026 ABSTRACT AbstractObjectiveTo investigate changed levels of serum astrocytic biomarkers, including glial cell line-derived neurotrophic factor (GDNF), aldehyde dehydrogenase (ALDH) and connexin-26, and the correlations with depression subtypes, cognitive dysfunction and electroencephalogram measures in patients with depression disorder (DD).MethodsSeventy-three patients and 46 healthy subjects (HC) were enrolled. Questionnaires, Nine Box Maze Test (NBMT) were used to assess sleep and neuropsychology. Serum biomarkers were evaluated with enzyme-linked immunosorbent assay. Partial subjects (45 DD and 10 HC) completed polysomnography.ResultsCompared to the HC, the patients had lower levels of GDNF, ALDH and connexin-26, and increased errors in NBMT tasks, including spatial/object working memory (SWM/OWM) and spatial/object reference memory (SRM/ORM). The connexin-26 concentration was higher in recurrent depression relative to dysthymia and bipolar depression; and GDNF concentrations was higher in the first episode depression compared to the recurrent depression and bipolar depression, with the lowest concentration in dysthymia. Partial correlation analysis for patients showed that the GDNF positively associated with N3% and power of delta wave; ALDH positively associated with MMSE score and OWM error; and connexin-26 negatively correlated with ORM and SRM errors.ConclusionDepressive patients had decreased serum levels of GDNF, ALDH and CX26, with subtype-specific difference in GDNF and CX26, which were associated with impaired cognition and slow wave power of electroencephalogram.Keywords: Acetaldehyde dehydrogenase, Connexin-26, Depression, Electroencephalography, Glial cell line-derived neurotrophic factor Introduction Depression disorder (DD) represents a leading global cause of disability, accounting for a significant portion of the mental disease burden worldwide [1]. Its clinical presentation is highly heterogeneous, encompassing not only core affective symptoms such as persistent low mood and anhedonia but also frequent comorbid cognitive dysfunction, sleep disturbances (e.g., insomnia), and vegetative symptoms [2]. This symptomatic diversity is mirrored by a lack of consistency in treatment response and course of illness, suggesting the existence of distinct neurobiological subtypes beneath the common diagnostic umbrella. Critically, the current diagnosis of DD relies primarily on subjective clinical assessment, lacking objective, biologically-based tools to parse its heterogeneity, predict outcomes, or guide targeted therapies. This gap underscores an urgent need to identify reliable biomarkers rooted in the disorder's pathophysiology. Regarding the pathogenesis and progression of DD, accumulating evidence underscores the critical involvement of astrocytic pathology and functional impairment [3]. Neuropathological findings from postmortem studies of DD patients have demonstrated a significant reduction in astrocyte density across multiple brain regions [4], accompanied by decreased immunoreactivity of glial fibrillary acidic protein (GFAP), a principal component of astrocytic intermediate filaments [5]. Functionally, astrocytes in DD patients exhibit impaired glutamate homeostasis, leading to dysregulated glutamate signaling [6]. This pathological alteration can induce excessive neuronal glutamate accumulation and subsequent excitotoxicity, a mechanism implicated in depressive pathophysiology [7]. Preclinical studies using chronic mild stress (CMS) rodent models have further revealed that depressive-like behaviors are associated with astrocytic pyroptosis, an inflammatory form of programmed cell death. Notably, pharmacological inhibition of pyroptosis via selective serotonin reuptake inhibitor (SSRI) administration attenuated both astrocyte loss and depressive phenotypes in CMS-exposed animals [8]. These findings collectively highlight astrocytic dysfunction as a pivotal component in the pathogenesis of depression, offering a novel perspective beyond neuron-centric views for identifying biomarkers. However, astrocytic functions are highly diverse, and a single biomarker is insufficient to capture their multifaceted roles. A systematic assessment of their distinct functional dimensions is crucial for understanding disease heterogeneity. This study focuses on three pathway-specific, functionally complementary key astrocytic proteins to construct a multidimensional evaluation framework: ① Glial cell line-derived neurotrophic factor (GDNF): As an intracellular neurotrophic factor and cytoskeleton-associated protein, it maintains the structural integrity of astrocytes while mediating critical neuroprotective and plasticity signals [9–11], representing the cells' trophic support and structural maintenance function. ② Aldehyde dehydrogenase 1 family member L1 (ALDH1L1): As a highly specific cytosolic metabolic marker enzyme regulating folate metabolism, it reflects the core state of cellular metabolism and antioxidant defense [12, 13]. ③ Connexin-26 (CX26): As a transmembrane channel protein forming gap junctions, it directly regulates direct communication between astrocytes and ionic homeostasis [14, 115], representing the cells' network connectivity and microenvironment regulation function. Currently, no research has simultaneously and systematically measured the levels of these three biomarkers in the peripheral serum of patients with depression—each representing a distinct core functional dimension of astrocytes (trophic/structural, metabolic, and connective). More importantly, whether this multidimensional dysfunction pattern can objectively reflect core brain phenotype characteristics across diagnostic dimensions (such as domain-specific cognitive impairments and sleep EEG slow-wave activity) remains entirely unknown. Building upon this multidimensional framework, we further sought to ground these peripheral biomarkers in a key objective brain phenotype. Among the various central nervous system functions disrupted in depression, sleep–wake regulation—quantifiable through electroencephalography—offers a critical and mechanistically relevant window into neuro-glial pathophysiology. Astrocytes are integral to sleep–wake regulation, particularly in orchestrating slow-wave sleep (SWS) through calcium-dependent signaling and neuro-glial communication [16, 17]. Importantly, disruptions in astrocytic function—including aberrant calcium dynamics and network connectivity—are known to directly cause sleep architecture disturbances, such as SWS fragmentation and abnormal brain rhythms [18–20]. In depression, where both astrocytic impairment and pronounced sleep abnormalities (e.g., reduced SWS, delta power) are core pathophysiological features, a critical translational question arises: can the peripheral serum signatures of astrocytic dysfunction (such as GDNF, ALDH, and CX26) serve as accessible biomarkers that reflect these central sleep-regulatory deficits? While preclinical evidence strongly links astrocytic mechanisms to sleep physiology, and clinical studies confirm sleep alterations in depression, a significant gap exists in directly connecting specific, measurable serum astrocytic markers with objective sleep EEG measures (e.g., N3%, delta power) in patients. This study aims to bridge this gap by investigating whether the levels of these multidimensional astrocytic proteins correlate with polysomnographic parameters, thereby offering a novel, integrative biomarker approach to understanding sleep disturbances in depression. The clinical classification of DD remains a subject of ongoing debate and refinement. As defined by the DSM-5 diagnostic framework, DD primarily encompass disruptive mood dysregulation disorder, major depressive episode (MDE), persistent depressive disorder (dysthymia), and premenstrual dysphoric disorder [21]. Notably, MDE manifest in heterogeneous forms, including distinctions between bipolar and unipolar depression [22]. Current clinical research has proposed various supplementary classification approaches to enhance diagnostic precision and treatment guidance. For instance, temporal characteristics of symptom duration have been investigated as potential diagnostic markers to optimize depression classification criteria [23]. Similarly, cognitive dysfunction profiling has emerged as a stratification method to inform pharmacological interventions [24]. However, these phenomenological classification systems lack robust neurobiological validation, particularly regarding distinct neural substrates underlying different depressive subtypes. This critical gap in neurophysiological evidence underscores the imperative for multidisciplinary investigations integrating clinical psychiatry with neuroscientific methodologies. Further research should prioritize elucidating subtype-specific neural circuitry abnormalities, neurochemical alterations, and functional connectivity patterns. Such mechanistic insights could ultimately facilitate the establishment of a more precise, biologically grounded taxonomy for DD, with direct implications for personalized therapeutic strategies. Based on the aforementioned gaps, the primary objectives of this study are to systematically investigate the neurophysiological and clinical manifestations of DD through four integrated approaches: firstly, to characterize alterations in sleep architecture, affective symptomatology, and cognitive performance patterns in DD patients; secondly, to conduct comparative analyses of astrocyte-specific serum biomarkers and quantitative EEG spectral parameters between DD cohorts and matched healthy controls (HCs); subsequently, to establish potential correlations between peripheral concentrations of astrocytic markers and central nervous system indicators including oscillatory brain activity, sleep disturbance metrics, mood severity scores, and neuropsychological test outcomes; and ultimately, to identify biomarker profile variations across clinically defined DD subtypes, thereby advancing the pathophysiological understanding and potential stratification strategies for DD. Methods Subjects This study was conducted between February 2020 and March 2021. A total of 119 subjects aged 18—65 years were included in this study, including 73 DD patients and 46 healthy people. Among them, patients with depression were classified into four types, including first episode of major depression (FED), dysthymia, recurrent depressive disorder and bipolar disorder (bipolar disorder included in the group is characterized by the current depressive phase), and the duration of the disease was counted. DD patients were from the Department of Sleep Disorders in the Fourth Affiliated Hospital Anhui Medical University and met the diagnostic criteria of the DSM-5. The main inclusion criteria are as follows: 1. Aged 18—65 years old. 2. The main complaint is depression, which is subdivided by group into a. FED depression (n = 17): There is no previous history of depression or other mental illness attacks. This visit is the first episode that lasts for more than 2 weeks and the onset time is no more than 2 years. b. Persistent mood disorder (dysthymia; n = 11): Persistent or recurring depressive mood for at least 2 years, with normal mood rarely lasting for more than a few weeks without a hypomanic episode. c. Recurrent depressive disorder (RDD; n = 18): There has been one or more depressive episodes in the past, which may be mild, moderate, or severe depression, lasting at least 2 weeks, and the current episode is at least 2 months apart from the last episode. d. Bipolar disorder (BD; n = 27): After a depressive episode, there needs to be a reverse or mixed phase attack. 3. Regarding medication status, patients in the recurrent depressive disorder subgroup had received pharmacotherapy during their first episode but underwent a washout period exceeding three months prior to enrollment. All other participants were medication-free at the time of the study. Thus, the potential confounding effect of recent medication use was minimized. 4. The total score of the 17-item Hamilton Depression Rating Scale (HAMD-17) is > 7. The HCs collected volunteers and physical examination personnel from the same hospital who were in good mood, slept well, and had no other organic diseases, and were roughly matched according to gender, age, education level, and body mass index (BMI) of the experimental group. Regarding polysomnography (PSG) monitoring, we excluded those who did not cooperate and those who were involved in the recording process from the above experimenters. Baseline data collection Collect baseline data at the clinic, including general information (i.e. age, gender, BMI and educational information) and medical history. Neuropsychological evaluation The HAMD-17 is used to assess the severity of depression. The HAMD-17 score ranges from 0 to 52, with higher scores indicating more severe depression [25]. The severity of anxiety symptoms was assessed using the 14 item-Hamilton Anxiety Scale (HAMA-14), which includes a set of symptoms [26]. Beck-Raffinson Mania Scale (BRMS) has a total of 13 items, of which 0 to 5 are rated as no manic symptoms; 6 to 10 are definitely manic symptoms; and 22 or more are rated as severe manic symptoms. The higher the total score, the more severe the manic symptoms are [27]. Evaluation of cognitive function The modified version of the Nine Box Maze Test (NBMT) includes five tasks, i.e.,spatial/object working memory (SWM/ORM), spatial/object reference memory (SRM/ORM), and object recognition memory (ORcM) [28]. Its content refers to previous research [29]. The Mini Mental State Examination (MMSE) is a 30 point questionnaire widely used in clinical research environments to assess cognitive impairment, including time and place tests, repetitive word lists, arithmetic, language use and comprehension, and basic motor skills. According to Chinese standards, illiteracy is ≤ 13, education for 1–6 years is ≤ 19, and education for 7 years or more is ≤ 24 [30]. Serum sample collection In the morning, blood was extracted from all subjects and serum was obtained from blood samples. It was immediately stored in a -80° C freezer until analysis. According to the manufacturer's instructions (biotechnology company), measure the serum concentrations of GDNF, ALDH, and CX-26 using quantitative sandwich enzyme-linked immunosorbent assay. Sleep assessment The Pittsburgh Sleep Quality Index (PSQI) is used to evaluate the subjective sleep quality of participants in the past month (including subjective sleep quality, sleep time, sleep efficiency, sleep disorders, daytime damage, and sleep medication use). The total score is 21 points, with higher scores indicating poorer sleep quality [31]. All depression patients and control subjects filled out this questionnaire as a self-report measure. According to the screening scale and clinical interviews, unqualified participants were excluded. In this study, 45 DD patients and 10 healthy sleepers participated in completing PSG monitoring. PSG was primarily employed for sleep and dream research, as well as for the diagnosis of depression and sleep apnea syndrome. Participants refrained from medication for two weeks prior to the study commencement. They were instructed to maintain their regular routines the day before the sleep monitoring session and to abstain from consuming stimulants such as tea, coffee, tobacco, and alcohol. On the monitoring night, participants cleansed the skin areas for electrode placement and removed any surrounding metallic objects. The sleep laboratory was maintained at a constant temperature, and lights were turned off according to each participant's habitual bedtime. All recordings utilized the Compumedics Grael v2 polysomnography system (Compumedics Ltd., Australia). Electrodes were placed according to the international 10 ~ 20 system, including left frontal (F1), right frontal (F2), left central (C3), right central (C4), left occipital (O1), and right occipital (O2) placements. All electrodes were referenced to the contralateral mastoid processes (M1, M2). To mitigate the first-night effect, all participants underwent one adaptation night prior to the formal recording. Sleep stages were manually scored in 30-s epochs according to the american academy of sleep medicine (AASM) manual for the Scoring of sleep and associated events. Recorded parameters included total sleep time (TST), sleep efficiency (SE), sleep latency (SL), number of awakenings (NAWK), and the percentage of time spent in each sleep stage. Sleep EEG analysis All EEG analyses were conducted on leads C3 and C4 using MATLAB 2014b (Mathworks, Natick, USA). The EEG signal is sampled at a rate of 1024 Hz. Use the Welch method to calculate the power spectral density between 1 ~ 30 Hz at 2 ~ second intervals and 1 ~ second overlap. Classify the power spectral density and calculate the power spectral density of delta (1 ~ 4 Hz), theta (4 ~ 8 Hz), alpha (8 ~ 12 Hz), sigma (12 ~ 15 Hz), and beta (15 ~ 20 Hz) for each frequency band. Multiply by the sampling rate to obtain the absolute power. To control the power difference caused by sleep time between individuals, we used relative power for comparison (i.e. relative power = [absolute power/average power of total sleep duration] × 100%). Before automatic analysis, experienced raters manually exclude EEG artifacts during sleep. Statistical analysis All statistical analyses were conducted using SPSS 22.0 for Windows. Use Student's t-test or non parametric test to analyze and compare the clinical characteristics and background information of each population group. Partial correlation analysis was used to determine serum sample concentration and relative spectral EEG power. Kruskal Wallis H test was used to compare the relationship between serum sample biomarker concentrations among the four groups. Bonferroni corrected P-value < 0.05 is considered to indicate statistical significance. Results General data, neuropsychological evaluation and Nine Box Maze Test There was no significant differences in gender, age, BMI and educational level between the DD patients and the HCs. The DD patients had higher scores of PSQI, HAMD, HAMA and BRMS, and lower MMSE scores than the HCs (Ps < 0.05); In the NBMT, the DD patients had increased numbers of errors in all five tasks, i.e., OWM, SWM, SRM, ORM and ORcM (Ps < 0.05). The detailed information is shown in Table 1. Table 1Background data, evaluation of cognition, neuropsychology, the objective sleep quality and relative power in different frequency bands in the patients and control subjectsCategoriesTermsDepression disorderHealthy controlStatisticsp-valueTotal sample (DD: n = 73, HC: n = 46)DemographicAge (yr)52.5 ( 42.25, 58.75)49.5 (39.0, 54.0)Z = − 1.53 > 0.05Sex (male/female)20 / 5320 / 26X2 = 3.71 > 0.05Years of education (yr)9.0 (6.0, 9.0)7.0 (3.75, 11)Z = − 1.38 > 0.05BMI (kg/m2)23.44 (21.53, 24.55)22.53 (21.01, 25.36)Z = − 1.38 > 0.05Neuropsychological evaluationHAMD (score)16 (11, 23)3 (1.75, 5)Z = − 8.96 < 0.01MAMA (score)14 (12, 19)3 ( 1, 5)Z = − 8.91 < 0.01PSQI (score)14 (12, 16)3 ( 1, 5)Z = − 9.02 < 0.01BRMS (score)4 (3, 5)1 ( 0, 2)Z = − 7.19 < 0.01BRMS (score)27 (24, 28)29 (29, 30)Z = − 6.86 < 0.01Cognitive performancesOWM (number of errors)2 (0, 3.75)0 (0,0)Z = − 3.21 < 0.01SWM (number of errors)1 (0, 3)0 (0, 0)Z = − 2.61 < 0.01SRM (number of errors)3 (1.25, 5.75)0 (0, 1)Z = − 3.40 < 0.01ORM (number of errors)2 (0, 3)2 (0, 3)Z = − 2.04 < 0.01ORcM (number of errors)0 (0, 1)0 (0, 0)Z = − 2.95 < 0.01SUM (Total number of errors)8.75 (1.25, 16.25)1 (0, 3)Z = − 3.03 < 0.01PSG subsample (DD: n = 45, HC: n = 10)DemographicAge (yr)48.78 (± 14.27)37.40 (± 11.70)F = 0.30 > 0.05Sex (male/female)9/364/6X2 = 1.81 > 0.05Years of education (yr)9.00 (5.00, 12.00)9.00 (8.25, 13.00)Z = − 0.99 > 0.05BMI (kg/m2)22.66 (20.51, 24.61)23.97 (22.06, 25.73)Z = 1.41 > 0.05Neuropsychological evaluationHAMD (score)18.00 (12.00, 23.00)0.00 (0.00, 2.00)Z = − 4.94 > 0.05MAMA (score)14.00 (11.00, 20.50)1.00 (0.00, 1.25)Z = − 4.94 < 0.01PSQI (score)15.00 (13.00, 17.00)0.00 (0.00, 1.00)Z = − 4.95 < 0.01BRMS (score)4.00 (4.00, 5.00)0.00 (0.00, 1.00)Z = − 4.74 < 0.01MMSE (score)27.00 (25.00, 29.00)30.00 (29.75, 30.00)Z = − 4.01 < 0.01Polysomnographic parametersTST (min)440.06 (± 58.79)431.20 (± 43.55)F = 1.040.31SE (%)90.8 (82.10, 94.15)89.15 (86.73, 93.00)Z = − 1.420.89SL (min)10.50 (2.75, 21.75)5.50 (2.25, 11.00)Z = − 1.170.24N1%13.20 (9.05, 18.90)9.65 (8.23, 11.25)Z = − 2.21 < 0.01N2%55.40 (46.45, 70.25)53.60 (44.90, 56.25)Z = − 1.330.18N3%12.30 (6.25, 21.05)23.40 (20.08, 25.38)Z = − 3.06 < 0.01REM%11.90 (7.80, 21.85)15.15 (10.95, 22.43)Z = − 1.190.24NAWK (time)16.00 (10.00, 24.50)12.50 (10.75, 18.25)Z = 3.00 < 0.01Spectral bandsDelta (DB)35.12 (± 5.89)45,86 (± 9.38)F = 2.48 < 0.01Theta (DB)5.66 (± 2.00)6.39 (± 3.46)F = 7.00 < 0.01Alpha (DB)4.46 (2.67, 6.46)3.36 (2.52, 6.84)Z = − 0.430.67Beta (DB)1.65 (1.04, 2.35)1.12 (0.80, 2.00)Z = − 1.390.17Normal distribution variables (Mean ± SD); non-normal distribution variables (P50 [P25, P75])DD Depression disorder, HC healthy controls, BMI Body Mass Index, HAMD Hamilton Depression Rating Scale (17-items), MAMA Hamilton Anxiety Scale (14-items), PSQI The Pittsburgh Sleep Quality Index, BRMS Beck-Raffinson Mania Scale, MMSE The Concise Mental State Examination, ORcM object recognition memory, ORM object reference memory, OWM object working memory, SRM spatial reference memory, SWM spatial working memory, SUM Total number of errors, TST total sleep time, SE sleep efficiency, SL sleep latency, N3% percentage of the sleep stage 3, REM% percentage of the rapid eye movement sleep time, NAWK number of awakenings Background data, evaluation of cognition, neuropsychology, the objective sleep quality and relative power in different frequency bands in the patients and control subjects Normal distribution variables (Mean ± SD); non-normal distribution variables (P50 [P25, P75]) DD Depression disorder, HC healthy controls, BMI Body Mass Index, HAMD Hamilton Depression Rating Scale (17-items), MAMA Hamilton Anxiety Scale (14-items), PSQI The Pittsburgh Sleep Quality Index, BRMS Beck-Raffinson Mania Scale, MMSE The Concise Mental State Examination, ORcM object recognition memory, ORM object reference memory, OWM object working memory, SRM spatial reference memory, SWM spatial working memory, SUM Total number of errors, TST total sleep time, SE sleep efficiency, SL sleep latency, N3% percentage of the sleep stage 3, REM% percentage of the rapid eye movement sleep time, NAWK number of awakenings Comparison of serum biomarkers Figure 1 Comparison of serum levels between depression disorder patients (DD) and healthy controls (HC). There were statistically significant differences in serum levels of GDNF (Z = − 3.33, P < 0.05), ALDH (Z = − 6.62, P < 0.05), and CX26 (Z = − 7.60, P < 0.05) between the HCs and DD group with lower GDNF, ADLH, and CX26 in the DD group. Comparison of serum levels between depression disorder patients (DD) and healthy controls (HC). Notes: t-test for independent samples; **P < 0.01. *P < 0.05. Abbreviations: ALDH, aldehyde dehydrogenase; GDNF, glial cell line-derived neurotrophic factor; CX26, connexin-26 Among the DD patient subgroups, the serum concentrations of GDNF differed significantly (H = 21.00, P < 0.01), and pairwise comparisons revealed significantly lower GDNF levels in the BD patients compared to the both FED and DM patients (P < 0.01). Additionally, GDNF concentrations in the FED patients were significantly higher than that in the RDD patients (P < 0.01), the remaining pairwise comparisons for GDNF were not statistically significant (P > 0.05). The serum CX26 concentrations also differed significantly across subgroups (H = 13.30, P < 0.01), and pairwise comparisons showed significantly higher CX26 levels in the RDD patients compared to both BD and DM patients (P < 0.01), and the remaining pairwise comparisons for CX26 were not statistically significant (P > 0.05). Interestingly, no statistically significant difference was found in ALDH concentrations among the patient’s subgroups (H = 2.41, P > 0.05). See Table 2 and Fig. 2. Table 2The comparision of serum astrocytic markers among four subgroups of patientsVariablesFEDDMRDDBDStatisticsp-vauleGDNF(ng/mL)0.56 (0.46, 0.84) d0.46 (0.42, 0.84)0.35 (0.23, 0.49) a0.22 (0.19, 0.44) bH = 21.00 < 0.01ALDH(ng/mL)1.17 (1.13, 1.29)1.16 (1.08, 1.27)1.16 (1.07, 1.21)1.20 (1.10, 1.33)H = 2.410.42CX26(ng/mL)3.47 (2.28, 5.44)2.34 (1.97, 6.13)6.63 (4.50, 15.64) b2.50 (1.97, 6.22) cH = 13.30 < 0.01ALDH, aldehyde dehydrogenase; GDNF, glial cell line-derived neurotrophic factor; CX26, connexin-26In the table, a−d represents statistical differences: a vs. the first episode of depression; b vs. dysthymia, c vs. recurrent depression, and d vs. bipolar disorder. All comparisons were corrected by Bonferroni. 2. The four groups were non normally distributed variables (P50 [P25, P75]), and Kruskal Wallis H test was performed. 3. Abbreviations: FED, first episode of depression; DM, dysthymia; RDD, recurrent depression disorder; BD, bipolar disorder; GDNF, glial cell line-derived neurotrophic factor; ALDH, aldehyde dehydrogenase; CX26, connexin-26 The comparision of serum astrocytic markers among four subgroups of patients ALDH, aldehyde dehydrogenase; GDNF, glial cell line-derived neurotrophic factor; CX26, connexin-26 In the table, a−d represents statistical differences: a vs. the first episode of depression; b vs. dysthymia, c vs. recurrent depression, and d vs. bipolar disorder. All comparisons were corrected by Bonferroni. 2. The four groups were non normally distributed variables (P50 [P25, P75]), and Kruskal Wallis H test was performed. 3. Abbreviations: FED, first episode of depression; DM, dysthymia; RDD, recurrent depression disorder; BD, bipolar disorder; GDNF, glial cell line-derived neurotrophic factor; ALDH, aldehyde dehydrogenase; CX26, connexin-26 Box scatter plot: The comparision of serum astrocytic markers among four subgroups of patients. Abbreviations: ALDH, aldehyde dehydrogenase; GDNF, glial cell line-derived neurotrophic factor; CX26, connexin-26; FED, first episode of depression; DM, dysthymia; RDD, recurrent depression disorder; BD, bipolar disorder; GDNF, glial cell line-derived neurotrophic factor; ALDH, aldehyde dehydrogenase; CX26, connexin-26. Notes: In the figure,the levels of three biomarkers were compared among the four groups. **P < 0.01. *P < 0.05 General data analysis and analysis of power spectral density of EEG There was no significant differences in gender, age, BMI and educational level between the DD patients and the HCs. The DD patients had higher scores of PSQI, HAMD, HAMA and BRMS, and lower MMSE scores than the HCs (Ps < 0.01). Although the PSG sample is limited, the PSG results show that the N3% of DD patients is significantly lower than that of HC patients (P < 0.01), N1% and NAWK of DD patients is significantly higher than that of HC patients (P < 0.01, Table 1). TST, SE, SL, N2%, and REM% showed no statistical significance (P > 0.05). Relative power in different frequency bands show that the delta and theta of DD patients is significantly lower than that of HC patients (P < 0.01), alpha and beta showed no statistical significance (P > 0.05, Table 1). Correlations between biomarkers, sleep parameters, and cognitive performance in DD After adjusting for age, sex, PSQI, BMI, and education, the partial correlation analysis revealed some significant associations between serum biomarker levels and cognitive/performance measures. Specifically, the serum ALDH level correlated positively with MMSE score (r = 0.274, P < 0.05) and negatively with OWM errors (r = − 0.248, P < 0.05) and SUM errors (r = − 0.234, P < 0.05). Serum CX26 level was negatively correlated with ORM errors (r = − 0.243, P < 0.05) and SRM errors (r = − 0.251, P < 0.05) (see Table 3 & Fig. 3A-E). Additionally, partial correlations (adjusted for age, sex, education, and BMI) between serum biomarkers and objective sleep parameters (Table 4 & Fig. 3F) indicated that serum GDNF concentration was positively correlated with N3% (r = 0.101, P < 0.05). In contrast, no significant correlations were observed in the healthy control (HC) group. Table 3Partial correlation analysis between GDNF/CX26/ALDH and cognitive performanceTermsMMSEORMSRMOWMSWMORcMSUMTotal sample (DD: n = 73)GDNF(ng/mL)0.0820.0720.0390.0250.017 − 0.0480.057CX26(ng/mL)0.090 − 0.243* − 0.251* − 0.0330.0940.107 − 0.067ALDH(ng/mL)0.274*0.076 − 0.101 − 0.248* − 0.116 − 0.157 − 0.234*Total sample (HC: n = 46)GDNF(ng/mL)0.1310.0920.076 − 0.035 − 0.156 − 0.163 − 0.073CX26(ng/mL) − 0.039 − 0.237 − 0.1000.045 − 0.113 − 0.181 − 0.061ALDH(ng/mL) − 0.0170.1840.170 − 0.0050.021 − 0.1430.067ORcM object recognition memory, ORM object reference memory, OWM object working memory, SRM spatial reference memory, SUM Total number of errors, GDNF glial cell line-derived neurotrophic factor, ALDH aldehyde dehydrogenase, CX26 connexin-26*Controlling for age, sex, education, PSQI, BMI*P < 0.05 Partial correlation analysis between GDNF/CX26/ALDH and cognitive performance ORcM object recognition memory, ORM object reference memory, OWM object working memory, SRM spatial reference memory, SUM Total number of errors, GDNF glial cell line-derived neurotrophic factor, ALDH aldehyde dehydrogenase, CX26 connexin-26 *Controlling for age, sex, education, PSQI, BMI *P < 0.05 (A–G). Scatter plots: Correlation between CX26 and ORM, SRM in the CID; Correlation between ALDH and MMSE, OWM, SUM in the CID; Correlation between GDNF and N3%, Delta in the CID. Abbreviations: ALDH, aldehyde dehydrogenase; GDNF, glial cell line-derived neurotrophic factor; CX26, connexin-26; ORM, object reference memory; SRM, spatial reference memory; SUM, Total number of errors; N3%, percentage of the sleep stage 3 Table 4Partial correlation analysis between GDNF/CX26/ALDH and objective sleep qualityTermsTST(min)SE(%)SL(min)N1%N2%N3%REM%AF(time)Total sample (DD:n = 45)GDNF (ng/mL) − 0.0280.0640.181 − 0.1720.0960.351*0.258 − 0.111CX26 (ng/mL) − 0.0340.020.0360.0550.0390.1510.1130.074ALDH (ng/mL) − 0.010 − 0.1210.0220.085 − 0.2740.0240.2280.116Total sample (HC: n = 10)GDNF (ng/mL)0.091 − 0.3210.1880.5560.195 − 0.176 − 0.2250.347CX26 (ng/mL)0.115 − 0.2120.4260.4670.164 − 0.091 − 0.1880.213ALDH (ng/mL)0.285 − 0.1270.5530.96 − 0.0850.1030.0490.097Controlling for age, sex, education, PSQI, BMI; *P < 0.05TST total sleep time, SE sleep efficiency, SL sleep latency, N1% percentage of the sleep stage 1, N2% percentage of the sleep stage 2, N3% percentage of the sleep stage 3, REM% percentage of the rapid eye movement sleep time, AF awakening times, GDNF glial cell line-derived neurotrophic factor, ALDH aldehyde dehydrogenase, CX26 connexin-26 Partial correlation analysis between GDNF/CX26/ALDH and objective sleep quality Controlling for age, sex, education, PSQI, BMI; *P < 0.05 TST total sleep time, SE sleep efficiency, SL sleep latency, N1% percentage of the sleep stage 1, N2% percentage of the sleep stage 2, N3% percentage of the sleep stage 3, REM% percentage of the rapid eye movement sleep time, AF awakening times, GDNF glial cell line-derived neurotrophic factor, ALDH aldehyde dehydrogenase, CX26 connexin-26 Correlations between biomarkers and relative power of each frequency band in DD Table 5 lists the partial correlations (adjusted for age, sex, education, and BMI) between serum biomarkers and relative band power. Serum GDNF concentrations showed a significant positive correlation with delta power (r = 0.338, P < 0.05, Fig. 3G). In contrast, no significant correlations were observed between serum ALDH concentrations and theta, alpha, or beta power (all P > 0.05). Similarly, serum CX26 concentrations showed no significant correlations with any of the analyzed frequency bands (all P > 0.05). Table 5Correlation between serum biomarkers and relative power of each frequency bandVariablesDeltaThetaAlphaBetaTotal sample (DD: n = 45)GDNF (ng/mL)0.338*0.085 − 0.1210.043ALDH (ng/mL)0.0640.0750.1310.304CX26 (ng/mL) − 0.074 − 0.025 − 0.1040.078Total sample (HC: n = 10)GDNF (ng/mL)0.3210.1880.2580.006ALDH (ng/mL) − 0.2000.5030.2480.139CX26 (ng/mL) − 0.1880.3700.406 − 0.103GDNF glial cell line-derived neurotrophic factor, ALDH aldehyde dehydrogenase, CX26 connexin-26Controlling for age, sex, education, PSQI, BMI; *P < 0.05 Correlation between serum biomarkers and relative power of each frequency band GDNF glial cell line-derived neurotrophic factor, ALDH aldehyde dehydrogenase, CX26 connexin-26 Controlling for age, sex, education, PSQI, BMI; *P < 0.05 Discussion Glial biomarker alterations in depression: implications for astrocyte-mediated pathophysiology The observed reductions in serum GDNF, ALDH, and CX26 levels in depressive patients align with emerging evidence of astrocytic pathology in mood disorders. GDNF, a key neurotrophic factor secreted predominantly by astrocytes, supports neuronal survival, synaptic plasticity, and dopamine system integrity [32]. The downregulation of GDNF observed in depression is corroborated by postmortem brain tissue analyses, which show reduced GDNF expression in key regions like the prefrontal cortex and hippocampus of patients with depressive disorder (DD) [33]. Preclinical studies suggest GDNF deficiency exacerbates stress-induced dendritic atrophy and serotonin depletion, potentially explaining the hippocampal volume loss observed in neuroimaging studies of depression [34]. Similarly, ALDH—a critical enzyme in alcohol metabolism and retinoic acid synthesis—was significantly lower in depressive patients. ALDH1A1, an isoform highly expressed in astrocytes, protects neurons from toxic aldehydes (e.g., 4-hydroxynonenal) generated during oxidative stress [35]. Chronic stress models demonstrate that ALDH inhibition induces depressive-like behaviors and dendritic simplification in rodents [36]. Our results extend these findings to humans, implicating ALDH depletion as a mediator of oxidative damage in depression. This is further supported by the positive correlation between ALDH levels and MMSE scores, suggesting that ALDH-driven detoxification may preserve global cognitive function. CX26, a gap junction protein facilitating intercellular communication between astrocytes and neurons, showed the most pronounced subtype-specific differences. Gap junctions enable synchronized astrocyte networks to regulate ion homeostasis, neurotransmitter clearance, and metabolic support [37]. CX26 downregulation may disrupt these processes, contributing to glutamate excitotoxicity and impaired neurovascular coupling—both implicated in depression [38]. Intriguingly, animal studies show that chronic stress reduces connexin expression in the prefrontal cortex, while antidepressants like fluoxetine restore gap junction coupling [39]. CX26 and GDNF as complementary biomarkers for depression subclassification The differential CX26 and GDNF levels across depression subtypes challenge the traditional symptom-based classification of mood disorders and point towards distinct underlying neurobiological mechanisms. Specifically, the elevated CX26 concentration in recurrent depression (compared to dysthymia/bipolar) suggests a state of heightened astrocytic activity, potentially acting as a compensatory mechanism. Characterized by episodic relapses, recurrent depression may involve transient astrocytic activation aimed at repairing stress-induced neural damage. This elevated CX26 could reflect a compensatory upregulation of gap junctions to enhance intercellular nutrient transfer or buffer glutamate surges during acute episodes [40]. Conversely, dysthymia and bipolar depression—marked by chronicity or cyclical mood instability—may exhibit progressive astrocyte dysfunction or loss, diminishing CX26 reserves. This parallels neuroimaging studies showing distinct gray matter trajectories: recurrent depression associates with reversible hippocampal atrophy [41], while bipolar disorder shows persistent volume reductions in the anterior cingulate cortex [42]. Complementing the CX26 findings, the observed GDNF pattern (first episode > recurrent/bipolar; bipolar < dysthymia) further underscores subtype heterogeneity and likely reflects differential neurotrophic support dynamics. The significantly higher GDNF level in first-episode depression could represent an initial, robust neuroprotective response to the onset of pathological stress, attempting to preserve neuronal integrity and plasticity. In stark contrast, the lower GDNF level in recurrent depression suggest a potential exhaustion or failure of this neurotrophic response over repeated depressive episodes, contributing to the cumulative neurobiological burden associated with recurrence. The particularly low GDNF level in bipolar depression, even compared to dysthymia, strongly align with evidence of more pronounced and persistent neuroprogression, synaptic deficits, and reduced neuroplasticity in this subtype, potentially reflecting a core deficit in neurotrophic resilience crucial for mood stability. These combined subtype differences in CX26 (astrocytic connectivity/compensation) and GDNF (neurotrophic support) highlight their significant potential as complementary biomarkers for differential diagnosis between major depressive disorder subtypes and bipolar depression. Future work should explore the interplay between these markers and their correlation with clinical features (e.g., episode frequency, chronicity, cognitive symptoms) and, crucially, treatment response. For example, interventions like electroconvulsive therapy, which enhances astrocytic connectivity and potentially GDNF signaling, may preferentially benefit recurrent depression patients with low baseline CX26 and/or GDNF [43]. Conversely, treatments specifically targeting neurotrophic pathways might be particularly relevant for subtypes like bipolar depression exhibiting profound GDNF deficits. Cognitive dysfunction in depression: insights into astrocyte-neuron network dysregulation The inverse relationships between glial biomarkers and memory errors provide mechanistic clues to depression-related cognitive decline. ALDH positive correlation with object working memory aligns with its role in detoxifying reactive aldehydes that impair prefrontal synaptic plasticity [44]. ALDH1A1 knockout mice exhibit working memory deficits and reduced dendritic spine density in the medial prefrontal cortex—phenotypes rescued by ALDH activation [45]. In humans, ALDH2 polymorphisms are linked to executive dysfunction in alcoholism and Alzheimer’s disease [46]. Our results extend these associations to depression, proposing ALDH as a therapeutic target for cognitive remediation. Similarly, CX26’s negative correlation with spatial reference memory errors implicates disrupted astrocyte networks in hippocampal dysfunction. Astrocytic gap junctions synchronize theta oscillations between the hippocampus and entorhinal cortex, a process critical for spatial memory [47]. CX26 knockdown in rodents disrupts place cell firing and Morris water maze performance [48]. The current findings suggest that similar mechanisms may underlie spatial memory deficits in depression. Notably, the lack of correlation between GDNF and cognitive measures contrasts with its established role in hippocampal neurogenesis [49]. This discrepancy may reflect GDNF’s preferential influence on emotional memory circuits (e.g., amygdala-prefrontal) rather than spatial navigation networks [50]. In the present study, partial correlation analyses revealed distinct patterns of associations when comparing DD patients with healthy controls. Multiple significant correlations were identified in the DD group, particularly between GDNF and cognitive performance, as well as between CX26 and objective sleep parameters. In contrast, no significant correlations were observed in the healthy control group across any domain. In healthy individuals, homeostatic mechanisms likely maintain these molecular systems within a narrow physiological range, preventing variations from translating into linear correlations with functional outcomes. Conversely, in major depressive disorder, dysregulation of neurotrophic support (GDNF), cochlin (CX26), and oxidative stress-related enzymes (ALDH) may become pathophysiologically coupled with clinical manifestations, unmasking associations that are otherwise latent under normal conditions. This pattern aligns with the concept of “state-dependent” biomarker–phenotype relationships. Furthermore, the absence of significant findings in the HC group mitigates concerns about spurious correlations and supports the specificity of the associations observed in the DD group. It should be noted that the relatively smaller sample size of the HC group may have limited statistical power to detect modest correlations. Sleep and EEG abnormalities: exploring slow-wave activity in the context of neuroplasticity The association between GDNF and slow-wave sleep bridges glial biology with sleep neurophysiology. Slow-wave activity during NREM sleep stage 3 (N3) facilitates synaptic downscaling and memory consolidation—processes regulated by astrocyte-derived neurotrophins like GDNF [51]. GDNF enhances slow oscillations in cortical slices by modulating astrocytic ATP release, which synchronizes neuronal networks via purinergic receptors [52]. Depressive patients’ reduced GDNF levels might thus contribute to slow-wave activity deficits, perpetuating cognitive and emotional dysfunction. Supporting this, delta power—a surrogate for slow-wave activity—positively correlated with GDNF, suggesting that GDNF replacement therapies could restore sleep-dependent plasticity. Conversely, the lack of ALDH/CX26 correlations with sleep parameters may reflect their primary roles in wake-related processes. ALDH’s detoxification function peaks during active wakefulness when oxidative stress is highest [53], while CX26-mediated metabolic support is critical for sustained neuronal firing during cognitive tasks [54]. These temporal specializations highlight the need for circadian assessments of glial biomarkers to capture dynamic astrocyte-neuron interactions. Astrocytes at the crossroads of sleep, mood, and cognition: implications for insomnia and depression comorbidity Chronic insomnia frequently co-occurs with depression. While depression was historically considered a primary cause of insomnia, accumulating epidemiological evidence now suggests that insomnia itself may constitute a significant risk factor for developing depression. Extensive research into the relationship between sleep and depression has established that the two conditions exert a bidirectional influence [55]; Nevertheless, the precise nature of this relationship remains under investigation—specifically, whether insomnia acts as a precursor to depression, represents a pre-morbid characteristic, or functions as an independent risk factor [56]. Crucially, emotional disorders, including BD, MDD, and anxiety-depressive states, are intimately linked to astrocyte function [57]. Furthermore, disruptions in light exposure and seasonal rhythms can impact mood regulation by altering normal sleep–wake patterns. Importantly, evidence indicates that the recovery and repair of glial cells, particularly astrocytes, can lead to improvements in both sleep quality and emotional states. Consequently, comprehensive clinical assessment necessitates a thorough understanding and precise evaluation of the patient's emotional and cognitive functioning. Additionally, the utilization of astrocyte-derived serum biomarkers alongside EEG proves highly valuable in elucidating the complex interrelationships between insomnia, mood disturbances, and cognitive function. This integrated approach is essential for developing optimal treatment strategies and achieving satisfactory therapeutic outcomes for patients. Limitations and future directions While this study offers significant advances, several limitations must be addressed: Cross-sectional Design: The temporal sequence between glial dysfunction and depression onset remains unclear. Longitudinal studies tracking biomarker changes during remission/relapse could establish causality. Cross-sectional Design: The temporal sequence between glial dysfunction and depression onset remains unclear. Longitudinal studies tracking biomarker changes during remission/relapse could establish causality. Sample Heterogeneity: The inclusion of both unipolar and bipolar depression, despite distinct pathophysiologies, may confound subtype comparisons. Future work should stratify analyses by diagnostic criteria (e.g., DSM-5 vs. ICD-11). Sample Heterogeneity: The inclusion of both unipolar and bipolar depression, despite distinct pathophysiologies, may confound subtype comparisons. Future work should stratify analyses by diagnostic criteria (e.g., DSM-5 vs. ICD-11). Small Healthy Control Sample: The healthy control group comprised only 10 participants, which limits statistical power and may increase the risk of sampling bias or imprecise effect estimates. This relatively small control sample also restricts the robustness of subgroup analyses and multivariate adjustments. Small Healthy Control Sample: The healthy control group comprised only 10 participants, which limits statistical power and may increase the risk of sampling bias or imprecise effect estimates. This relatively small control sample also restricts the robustness of subgroup analyses and multivariate adjustments. Future studies should: explore multi-omics approaches (transcriptomics, proteomics) to identify upstream regulators of GDNF/ALDH/CX26; test glial-targeted interventions (e.g., ALDH activators, connexin modulators) in animal models of depression; integrate neuroimaging (fMRI, DTI) to map biomarker correlations with network connectivity. Additionally, replication in larger, independent cohorts with adequately sized control groups is necessary to validate the present findings. Future studies should: explore multi-omics approaches (transcriptomics, proteomics) to identify upstream regulators of GDNF/ALDH/CX26; test glial-targeted interventions (e.g., ALDH activators, connexin modulators) in animal models of depression; integrate neuroimaging (fMRI, DTI) to map biomarker correlations with network connectivity. Additionally, replication in larger, independent cohorts with adequately sized control groups is necessary to validate the present findings. Conclusion This study positions serum GDNF, ALDH, and CX26 as clinically relevant biomarkers of astrocyte dysfunction in depression. Their subtype-specific profiles and associations with cognitive/EEG parameters underscore the need for a gliocentric framework in depression research. By bridging molecular psychiatry with systems neuroscience, these findings pave the way for precision therapies targeting astrocytic resilience—whether through GDNF enhancers to restore slow-wave plasticity, ALDH activators to mitigate oxidative cognitive decline, or connexin modulators to repair neural networks. Validating these biomarkers in larger, longitudinal cohorts could revolutionize depression subtyping and treatment personalization. Acknowledgements We would like to express our sincere gratitude to all the participants for their support and cooperation in this study. We also thank all the members of our laboratory for their assistance in data collection and analysis. Finally, we extend our thanks to all our peers and reviewers who provided valuable suggestions during the manuscript preparation and revision process. Author contributions Zi-Jie Ma,Gui-Hai Chen: wrote the main manuscript text; Zi-Jie Ma, Yu-Shun Zhou, Zhen-Yu. Hu, Gao-Xia Liu: data collection; Zi-Jie Ma,Gui-Hai Chen,Lan Xia,Tian-Tian Zou,Xue-Yan Li: methodology. Funding No funding was received for this research. Data availability The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request. Declarations Ethics approval This study was conducted in accordance with the Declaration of Helsinki, and approved by the Ethics Committees of the Fourth Affiliated Hospital of Anhui Medical University (Number KYXM-202108-005). Consent to participate In this multicenter biomarker study for depression, all participants were informed in detail about the study purpose, procedures (including blood draws, cognitive and sleep tests, and adverse event reporting), potential risks and benefits. Participation was voluntary, confidentiality was guaranteed, and aggregated results will be fed back upon study completion. Informed consent was obtained from all participants prior to their inclusion in the study. Consent to publish Not applicable. Footnotes Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Lan Xia, Email: xialanxl1983@163.com. Gui-Hai Chen, Email: doctorcgh@163.com. References Associated Data Data Availability Statement The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.