There were 68 participants were included in the present study. Thirty-eight MDD patients with sleep disturbances and 30 healthy controls (HCs) were recruited from the Second Affiliated Hospital at Anhui Medical University. All MDD patients met the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-V) criteria (Uher et al., 2014).

The inclusion criteria for patients with MDD encompassed: (1) those experiencing their first episode, whether as outpatients or inpatients; (2) patients who were either naive to pharmacological treatment or had discontinued all antidepressive treatments for a period exceeding four weeks prior to the study’s commencement; (3) patients exhibiting a definitive sleep disorder, as confirmed through sleep assessments; and (4) individuals without a familial history of psychosis. Meanwhile, HCs had no history of DSM-V Axis I disorders, mental health issues, or significant physical ailments, and demonstrated good sleep quality. Furthermore, participants were excluded if they met any of the following criteria: (1) the presence of organic lesions within the central nervous system or neurodegenerative disorders, such as stroke or Parkinson’s disease; (2) secondary mental disorders arising from severe physical conditions; (3) history of alcohol or drug abuse and dependence; (4) significant physical ailments, encompassing endocrine disorders, autoimmune conditions, or impaired liver or kidney function; (4) any form of neoplasia or cerebral trauma; (5) pregnant and breastfeeding women; or (6) history of OSA.

The ethical approval was obtained from the Ethics Committee of the Second Hospital of Anhui Medical University (approval number: SL-YX2024-022). All participants or their legal guardians provided informed consent.

All patients diagnosed with MDD underwent a 14-day antidepressive treatment. During this period, only one type of SSRIs or SNRIs was administered as the primary antidepressant. Because of the primary objective of this study was not to assess the efficacy of antidepressive treatment, the study was designed as an observational follow-up rather than a randomized controlled trial.

The interviews of each participant were completed by a trained psychiatrist. Neuropsychological assessments were conducted for HCs at baseline, as well as for patients with MDD both prior to and following treatment.

In the present study, Pittsburgh Sleep Quality Index (PSQI) (Buysse et al., 1989) was performed to measures sleep quality over the previous month. The questionnaire evaluates seven clinically derived domains of sleep quality, sleep onset latency, sleep duration, sleep efficiency, sleep difficulty, sleep medication, and daytime dysfunction. These domains are collectively scored to determine a single factor representing global sleep quality, with a global score exceeding 11 serving as an indicative marker of significant sleep disturbances.

Furthermore, the 24-item Hamilton Depression Scale (HAMD-24) (Hamilton, 1960) and Self-Rating Depression Scale (SDS) (Zung, 1965) were used to evaluate the depressive symptoms. Meanwhile, each MDD patient also received an individual sleep disturbance score, which was computed by summing the scores of items 4, 5, and 6 from the HAMD-24 (Shi et al., 2020a; Shi et al., 2021a; Shi et al., 2021b).

In the current study, oral biospecimens were collected noninvasively from the tongue dorsum using a sterile cotton swab, placed them in a sterile centrifuge tube with 1.5 mL Tris-EDTA buffer solution (Solarbio) and rapidly transported and stored them at -80°C. All participants completed the oral sample collection process after neuropsychological assessment.

Total microbial genomic DNA was extracted using the Bacterial DNA Extraction Mini Kit (Mabio, Guangzhou, China) according to manufacturer’s instructions. The quality and concentration of DNA were determined by 1.0% agarose gel electrophoresis and a NanoDrop^® ND-2000 spectrophotometer (Thermo Scientific Inc., USA). The hypervariable region V3-V4 of the bacterial 16S rRNA gene were amplified using specific primers (338F and 806R) by an T100 Thermal Cycler (BIO-RAD, USA). Using the NEXTFLEX Rapid DNA-Seq Kit (Bioo Scientific, USA) to generate the sequencing libraries. Purified amplicons were pooled in equimolar amounts and paired-end sequenced on an Illumina NextSeq 2000 PE300 platform (Illumina, San Diego,USA) according to the standard protocols by Majorbio Bio-Pharm Technology Co. Ltd. (Shanghai, China). Raw FASTQ files were de-multiplexed using an in-house perl script, and then quality-filtered by fastp version 0.19.6 and merged by FLASH version 1.2.11. The optimized sequences were clustered into operational taxonomic units (OTUs) using Usearch 11 with 97% sequence similarity level. The metagenomic function was predicted by PICRUSt2 (Phylogenetic Investigation of Communities by Reconstruction of Unobserved States) based on OTU representative sequences. More details could be found in the Supplementary Materials .

Following an overnight fast, peripheral venous blood was collected into a vacutainer tube without anticoagulant. Within 30 minutes after collection, the blood samples were centrifuged at 3500 revolutions per minute at a temperature of 4°C for a duration of 10 minutes, subsequently serum was aspirated and stored at -80°C until needed for analysis. Notably, all MDD patients provided blood samples prior to initiating treatment.

The serum levels of glial fibrillary acidic protein (GFAP) and S100beta protein (S100β) were assayed in triplicate, using the commercial Enzyme-linked Immunosorbent Assay kits (FineTest, Wuhan, China; Catalog Number: EH0410 for GFAP and EH0543 for S100β) in accordance with the manufacturer’s protocols. The protein concentration in each plate was determined based on the standard curves. Both the inter-assay and intra-assay coefficients of variation were less than 5%.

The data were analyzed utilizing SPSS version 22.0 (SPSS, Inc., Chicago, IL, USA) and R software package (version 4.2.1).

Alpha and beta diversity analyses were conducted for the diversity analysis of oral microbiota. In the alpha diversity analysis, Observed species, Chao, and ACE indices were utilized to evaluate the richness of the microbial community, whereas Shannon, Simpson, and Coverage indices were harnessed to assess the diversity of the community. To investigate the variations in the composition of oral microbiota, beta diversity analysis was performed using Partial Least Squares-Discriminant Analysis (PLS-DA). Meanwhile, the microbial dysbiosis index (MDI) acted as an indicator to assess the extent of microbial imbalance within the study population, with a higher MDI value signifying a more pronounced degree of bacterial disturbance (Gunathilake et al., 2020). It was calculated using the formula MDI=log₁₀[(total abundance in genera increased in disease group)/(total abundance in genera decreased in disease group)]. Furthermore, MetagenomeSeq analysis was conducted to compare the relative abundances of oral microbiota between the MDD and HC groups (Barot et al., 2024; He et al., 2024).

To ascertain the normal distribution of the data, the Kolmogorov-Smirnov test was performed. For categorical variables, a chi-squared test was employed for analysis. As for continuous variables, the independent-samples t test was utilized when the data were normally distributed; otherwise, the Mann-Whitney U test was applied. The paired t-test was employed to compare the changes in variables before and after treatment. In MDD patients, partial correlation analysis was conducted to ascertain the associations between the two variables, adjusting for factors such as age, sex, years of education, duration of the disease, and body mass index (BMI). Additionally, mediation analysis in MDD patients, controlling for age, sex, education years, duration of the disease, and BMI, was conducted to determine whether oral bacteria mediated the relationship between neuropsychological assessments and serum molecular indicators, based on a standard three-variable mediation model (Baron and Kenny, 1986; He et al., 2019). A detailed description of the method can be seen in Supplementary Materials . Receiver operating characteristic (ROC) curves were utilized to calculate the area under the curve (AUC), thereby assessing the diagnostic accuracy of oral bacteria in identifying MDD. The Youden index (Youden, 1950) was used to assess optimal values of sensitivity and specificity. Notable, changed rate of scores=(before treatment score - after treatment score)/before treatment score. Statistical significance was established at a p-value of less than 0.05 (two-tailed).

The demographic and clinical characteristics of the participants in the two groups are succinctly summarized in Table 1 . No significant differences were observed between the two groups in terms of age, sex, education years, and body mass index. When compared to HCs, patients with MDD exhibited significantly elevated scores on the PSQI, HAMD-24, SDS, as well as increased serum levels of GFAP and S100β ( Table 1 ). Additionally, there were significant differences in seven factor scores of the PSQI scale and the somnipathy factor scores of HAMD-24 scale between the two groups, as highlighted in Table 1 .

Following a 14-day antidepressive intervention, MDD patients displayed significant reductions in their PSQI, HAMD-24, and SDS scores when compared to their pretreatment assessment scores ( Table 1 ). Additionally, there were notable decreases in seven factor scores of the PSQI scale and the somnipathy factor scores of the HAMD-24 scale ( Table 1 ).

A total 4,024,824 sequences (SRA accession number: PRJNA1279610) were obtained from 68 samples using QIIME software (version 1.91; URL link: https://qiime.org/). The operational taxonomic units were assigned based on a threshold of 97% sequence similarity. The MDD group demonstrated a greater abundance of OTUs compared to the HC group, with a count of 1106 versus 1063, respectively, encompassing 750 shared OTUs ( Supplementary Figure S1 ). The rarefaction curves for the samples reached a saturation plateau at a sequencing depth of 42,952 reads, indicating that the majority of microbial species were adequately covered by the sequencing depths employed, and that the sample size was appropriately sized ( Supplementary Figure S2 ).

In the alpha diversity analysis, six alpha-diversity indices, i.e., Observed species, Chao1, ACE, Shannon, Simpson and Coverage, were analyzed in the present study. The levels of these six indices showed no significant difference between MDD and HC groups ( Supplementary Figure S3A ). Furthermore, in the beta diversity analysis, PLS-DA analysis was conducted for the reduction of the impact of intergroup differences ( Supplementary Figure S3B ).

In the current study, MDD patients exhibited significantly elevated levels of MDI compared to HCs ( Figure 1A ). Furthermore, within the MDD patient group, there were positive correlations observed between MDI levels and scores on the PSQI and HAMD-24 scales, as well as with factor scores pertaining to sleep efficiency and daytime dysfunction, as depicted in Figures 1B-D .

Figure 2A illustrates the top 50 most prevalent bacterial genera, characterized by their highest relative abundances in both groups. These genera belong to ten different phyla, including Firmicutes, Fusobacteriota, Actinobacteriota, Bacteroidota, Proteobacteria, Cyanobacteria, etc. The MDD patients displayed 36 exclusive genera, while 140 identical genera were identified in both the MDD and HC groups ( Figure 2B ). Among the overlapping bacterial genera shared by the two groups, Streptococcus, Prevotella, Neisseria, Veillonella, Leptotrichia, and Haemophilus exhibited the six highest relative abundances ( Figure 2C ).

In addition, the Mann-Whitney U test revealed significant differences in 30 predicted functional pathways between the two groups ( Supplementary Figure S4 ). Among these pathways, “Carbon metabolism”, “Biosynthesis of amino acids”, and “Microbial metabolism in diverse environments” demonstrated the highest relative abundances within the oral microbiota ( Figure 2D ).

Furthermore, the MetagenomeSeq analysis represented 25 genera with significant differences in their relative abundances between the two groups ( Figure 3A ). To ensure the precision of the analysis exploring group disparities, four oral bacteria at the genus level were retained, as the test values for these bacteria could reliably be obtained from each subject ( Figure 3A ). Consequently, MDD patients exhibited a significant decline in the relative abundance of Solobacterium and Granulicatella, whereas Campylobacter and Haemophilus demonstrated a significant increase, in comparison to the HC group ( Figure 3B ). However, only Granulicatella and Haemophilus demonstrated statistically significant differences following multiple comparison correction (adjusted p < 0.0125, Bonferroni method).

To assess the diagnostic efficacy of these four oral bacteria in identifying MDD with sleep disturbances, the ROC curve analysis revealed that Haemophilus showed an optimal AUC value of 0.724, accompanied by a sensitivity of 56.67% and a specificity of 89.47% ( Figure 3C ). Additionally, the remaining three oral bacteria demonstrated tolerable diagnostic performance, each possessing an AUC value exceeding 0.6 ( Figure 3C ).

In Figure 4A , a correlation matrix was constructed utilizing partial correlation analysis, taking into account environmental factors in MDD patients prior to treatment, while adjusting for variables such as age, sex, years of education, duration of the disease, and BMI. This analysis incorporated four oral microbes at the genus level that exhibited specific alterations in MDD patients. The relative abundance of the genus Haemophilus showed significant positive correlations with PSQI and HAMD-24 scores, as well as serum levels of GFAP and S100β ( Figure 4A ). Meanwhile, there were significant correlations between relative abundance of the genus Haemophilus and sleeping quality and sleep onset latency factor scores in MDD patients ( Supplementary Table S1 ). However, no significant correlations were observed between the other three oral microbes and the environmental factors in MDD patients.

After undergoing a 14-day antidepressive treatment, MDD patients exhibited significant reductions in PSQI, HAMD-24, and SDS scores compared to their pretreatment levels ( Table 1 and Supplementary Figure S5 ). Additionally, correlation analyses were performed for the environmental factors in MDD patients after treatment, with controlling age, sex, years of education, duration of the disease, BMI, and antidepressant medicines. In the MDD group, there were significant negative correlations between the relative abundance of the genus Haemophilus and the changed rates of PSQI and HAMD-24 scores ( Figures 4B, C ).

Further mediation analyses conducted in MDD patients revealed that environmental factors played a significant role in affecting the relationship between the relative abundance of oral Haemophilus and the severity of sleep disturbances and depression. These analyses accounted for covariates such as age, sex, years of education, duration of the disease, and BMI. Additionally, the serum levels of GFAP were found to influence the relative abundance of Haemophilus genera in relation to PSQI and HAMD-24 scores ( Figure 4D ). Notably, serum GFAP exhibited a direct mediating effect on the association between the relative abundance of Haemophilus and PSQI scores, as depicted in Figure 4D .

The sample size calculation was performed using an online calculator available at https://sample-size.net/. Based on the relative abundance of Haemophilus, with parameters set at α = 0.05 and β = 0.2, the analysis indicated that 27 subjects per group would achieve a statistical power of 0.8082. Consequently, the current sample sizes of 38 and 30 subjects in the respective groups are adequate, as they provide a power value exceeding the conventional threshold of 0.8. Nevertheless, a larger sample size remains necessary to validate the current findings and improve their reliability.