**1. Introduction**

Major depressive disorder (MDD) is one of the leading causes of disability worldwide [1], with a high prevalence among individuals of all ages and races [2]. MDD is a chronic condition with a high recurrence rate with a full recovery rate of only 20% and 80% of recovered patients experiencing at least one relapse in their entire life [3]. Antidepressants have long been used in the acute and long-term treatment of MDD, with selective serotonin reuptake inhibitors (SSRIs) being the first-line antidepressants. The process of selecting an antidepressant agent is primarily prescribed based on trial and error. Patients with poor efficacy of the initial course of medication for at least 4–6 weeks require alternative therapeutic strategies, containing changing within and between classes of antidepressants. Unfortunately, the treatment outcomes from antidepressants are discouraging. About 50% of patients enrolled in the Sequenced Treatment Alternative to Relieve Depression (STAR\*D) study failed to respond to standard SSRI treatment, and only about 30% experienced complete remission in response to the first antidepressant used [4]. After unsuccessful treatment for MDD patients with a SSRI, the choice of a second drug is important for remission [5]. Biomarkers for response to antidepressant treatment can reduce the time to symptom relief and costs, minimize unnecessary drug exposure, and improve patient outcomes.

Proteomics, the quantitative analysis of all proteins expressed in samples, is a powerful tool for identifying novel molecular biomarkers and enables the detection of molecular signatures reflecting multiple biological pathways involved in response to treatment in patients with MDD [6]. Proteomic analysis of peripheral body fluids, such as blood plasma and serum, may not only enable prediction of response to treatment in clinical practice, but also assist in monitoring drug activity during early stages of clinical trials. To date, however, there have been few proteomic analyses of peripheral blood samples that can predict response to antidepressant treatment [7,8]. A previous liquid chromatography tandem mass spectrometry (LC-MS/MS) analysis found that several plasma proteins might be potential biomarkers for the prediction of antidepressant response over a 6-week treatment period [7]. A multiplex immunoassay testing of up to 258 blood-based markers related to immune, endocrine, and metabolic mechanisms identified 9 markers as potential pre-treatment biomarkers associated with antidepressants treatment response [8].

Longitudinal data are commonly used in biomedical studies [9,10]. In statistical analyses, mixed-effect models (MEMs) [11] and generalized estimating equations (GEEs) [12] are widely applied. To further elaborate, MEM is a subject-level approach that could employ random effects to acquire a between-subjects variable by considering the correlations with observations from the same subject based on the full-likelihood method. Conversely, GEE is a population-level model that relies on a partial-likelihood function. In this study, repeated drug efficacy measurements (baseline and follow-up visits after treatment) were performed on the surrogate plasma protein over time for each patient, with the subject of interest being some of the time-varying changes. In proteomic studies, the MEM method is more popularly used than the GEE method [13–15]. This is because after fixing the desired effect, it is possible to estimate by measuring a random effect of the technical or biological repeated measurement with actual MS [16,17]. In addition, it is technically easier to reflect the variance of any effect on repeated measurements of the same sample after more than two times of MS. Conversely, GEE is robust to the misspecification of correlation structure using quasi likelihood, and many modified variance estimation methods for small samples have been developed [18]. We identified biological implications primarily with the results of the analysis with linear mixed model (LMM) and compared the results after performing with the same data with GEE.

In this preliminary study, LC–MS/MS profiling was performed to identify candidate blood-based protein biomarkers that could monitor early (0–1 week), mid (1–4 week), or late (4–10 week) response to antidepressants, before and after their administration. This study also assessed whether changes in plasma protein concentrations after antidepressant treatment were associated with changes in the severity of depressive symptoms. Blood samples were collected at four time points during the 10-week treatment of ten depressed patients, five responders, and five non-responders, who were taking escitalopram. Plasma proteins were profiled, as were differences in protein abundance between the two groups. Unlike biomarker studies that don't take into account the time of disease occurrence [19–22], this study attempted to identify more reliable candidate biomarkers by time-dependent longitudinal changes in the plasma proteome of these patients. Furthermore, the identified biomarkers predicting early-drug response were validated in 19 responders and five non-responders by the liquid chromatography-multiple reaction monitoring/mass spectrometry (LC-MRM/MS) technique. In addition, significant markers were identified assessing the correlation between protein concentrations, as determined by molecular diagnostic techniques, and psychological parameters.
