Setting and participants
This study was conducted at Tongde Hospital of Zhejiang Province in Hangzhou, China between April 2015 and March 2017. The study protocol was approved by the Medical Ethical Committee [equivalent to the Institutional Review Board (IRB)] of Tongde Hospital of Zhejiang Province on March 11, 2014 (The approved document is attached in Additional file 1) and registered at http://www.clinicaltrials.gov (NCT02346682) on January 27, 2015 (https://clinicaltrials.gov/ct2/show/NCT02346682). All participants gave voluntary, written, informed consent before entering the study. We reported this study according to the Minimum Standards of Reporting Checklist (see Additional file 2).
Participants were sought from out-inpatients. Subjects were eligible for this study if they: (a) were aged 18–65 years; (b) were currently experiencing a recurrent, moderate or severe depressive episode according to the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5), as evidenced by a score of at least 21 on the 24-item Hamilton Rating Scale for Depression (HAMD-24) [21]; (c) met the diagnostic criteria of the LQS or HSD subtype as defined in Table 1 with representative tongue manifestations as shown in Fig. 1; and (d) had received no treatment with antidepressants and other psychotropic drugs in the previous 3 months. Tongue diagnosis is a critical diagnostic approach in discriminating different TCM patterns according to manifestations of the tongue coating and body [22].
Patients were excluded from this study if they: (a) were experiencing their first episode of depression; (b) had serious comorbid cardiac, hepatic, or renal conditions; (c) had a history of brain injury or surgery; (d) had a history of manic, hypomanic, or mixed episodes; (e) had received investigational drug treatment within the previous 6 months; (f) had experienced alcohol or drug abuse within the previous 12 months; or (g) were pregnant or breastfeeding.
A group of healthy volunteers who had no personal or family history of significant mental and physical illness was recruited from Tongde Hospital and local community to serve as controls.
Screening, assessment, and TCM diagnosis
The screening of patients and healthy volunteers was done by a psychiatrist and a TCM practitioner. The severity of the depression symptoms of the patients and healthy controls was assessed using HAMD-24 [21]. The assessment was conducted by a trained rater (L.Y.L., C.Q.Z.). TCM subtypes diagnoses were made by at least two senior TCM practitioners (Y.P.L., C.Q.Z.) and a third TCM practitioner (L.Y.L.) was involved in the diagnosis process if the first two practitioners could not reach an agreement.
To ensure consistency of assessment and diagnosis, a manual was provided and a training workshop was carried out on videotaped patients with different TCM subtypes before the study was initiated. An inter-rater reliability coefficient of > 0.85 was achieved after the completion of the training workshop.
Collection and preparation of serum and urine samples
Blood and urine sample for each were collected within 2 days after the completion of clinical assessment. Following the sample collection, patients immediately received antidepressant and other psychotropic treatment at psychiatrists’ discretion based on their clinical condition, with routine subsequent monitoring.
In the morning (09:00–10:00) after an overnight fasting, 10 ml of midstream urine were collected; 15 ml of blood were drawn and sera were immediately separated. Samples were further prepared for metabolomic measurement as described previously [17, 23]. Briefly, 500 μl of urine or sera was mixed with 50 μl heptanoic acid and 10 mg/ml para-chlorophenylalanine as the internal standard solution. 500 μl methanol was added to precipitate protein, mixed for 5 min, and then centrifuged at 12,000 rpm/min for 15 min at 4 °C. A 500-μl aliquot of supernatant was transferred to a clean Eppendorf tube and dried under a low temperature vacuum drier. The residue was derivatized by adding 100 μl methoxyamine hydrochloride (15 mg/ml in pyridine) at 30 °C for 90 min. For each sample, 100 μl BSTFA (1% TMCS) was added and the mixture was heated at 70 °C for 60 min. The derivative was cooled and filtered in 0.45 μm membrane prior to GC–MS analysis. Quality control samples and the reference standard were processed as done for experimental samples.
GC–MS acquisition
One μl aliquot of derivatized sample was injected into a Varian 450-GC/240-MS equipped with 19091 N-113 capillary column (30 m × 0.32 mm × 0.25 μm, Agilent J & W Scientific, USA) at a split ratio of 10:1. Helium was used as the carrier gas with a constant flow rate of 1 ml/min. The initial temperature was set at 70 °C for 4 min, elevated to 300 °C at a rate of 8°C/min, and then maintained for 3 min. Temperature for the injector, transfer line, and ion source was set at 280 °C, 250 °C, and 220 °C, respectively. The mass range (50–800 m/z) in a full-scan mode for electron impact ionization (1.0 kV) was applied. The solvent delay time was set to 6 min.
Statistical analysis
There were no studies detecting metabolomic effects in TCM subtypes of MDD. Sample size estimation was based on one previous study that has revealed differential metabolomic profiles between MDD patients with and without early life stress [17]. As the LQS and HSD subtypes are appear to be differentially associated with stress-related (reactive) and endogenous (melancholic) depression [2, 9], we assumed that metabolomic differences between the two TCM subtypes was similar to that of MDD patients with and without early life stress [17]. The study has shown an averaged 36% difference in plasma level of major metabolites between MDD patients with and without early life stress with an averaged standard deviation of 41% [17]. A sample size of 22 each group would be sufficient to yield an 80% power at a statistical level of 0.05.
For baseline data, one-way analysis of variance (ANOVA) was used to detect differences in continuous variables among healthy controls and the two TCM subtypes. Student’s t-test was used to detect differences in continuous variables between the two TCM subtypes. Categorical baseline variables were analyzed using Chi square (χ2) test.
For metabolomic data, the pretreatment process was performed, including novel nonlinear retention time alignment, baseline filtration, peak identification, matching, and integration. The resulting data matrix consisting of variables, sample code, and peak area was further processed using Microsoft Excel program. The original spectral data obtained from GC-MS spectroscopy were scaled to unit variance (z), which was calculated from the formula z = (x – y)/s, where x, y, and s represent the level of the particular metabolite in one subject, the mean level and the standard deviation of this metabolite across all subjects, respectively.
Nonparametric Mann–Whitney U test was used to detect significantly differential metabolites. All metabolites were determined by standard samples and/or a similarity of > 70% that was obtained by comparing with the mass spectral database of the US National Institute for Standards and Technology (NIST).
Metabolites were determined using variable importance in the projection (VIP) which value was defined as > 1 and t-test was set at a level of 95%. The principal component analysis (PCA) was used to discriminate metabolic patterns among the three groups. The orthogonal projection to latent structures-discriminant analysis (OPLS-DA) model was further constructed to identify meaningful metabolites that could differentiate between the three groups using SIMCA-P 13.0 program. A regression method was applied to establish the optimal discriminant model with controlling gender as a confounding factor. The quality of the model was tested with cross-validation and R2X, R2Y, and Q2 values were obtained. OPLS-DA models which all R2X, R2Y, and Q2 values were ≥ 0.5 were acceptable. Hierarchical clustering patterns were further established with heatmap analysis using R i386 3.3.0 software for visual verification of OPLS-DA models. The impact pathway was determined using MetaboAnalyst 3.0 (http://www.metaboanalyst.ca), a web-based tool for pathway analysis, with the criteria as p-value < 0.05, false discovery rate (FDR) < 0.05, and impact value > 0. Visualization of metabolomic correlation network was drawn with CytoScape 3.3.0. The receiver operating characteristic (ROC) curve analysis was conducted to determine the optimal metabolite combination patterns that could well dichotomize the subtypes and healthy controls at acceptable sensitivity and specificity (defined as greater than 80% for both).
Statistical analysis was conducted using SPSS 18.0 (SPSS Inc., Chicago, IL, USA) and, unless otherwise indicated, significance level was set at a two-tailed P < 0.05.