*3.3. Time-Dependent Changes in Plasma Proteins in Responders and Non-Responders*

Statistical comparisons of paired plasma protein abundances at baseline and after 1, 4, and 10 weeks of treatment showed that seven, four, and six proteins, respectively, were upregulated in responders and 16, 17, and 10 proteins, respectively, were upregulated in non-responders using the Mann–Whitney test without correction (*p*-value < 0.05; Figure 1A). The Venn diagram of the three time points, T1, T4, and T10 is shown in Figure 1B. Proteins upregulated in non-responders were associated with responses to wounding and stimuli, responses to wounding, and tube morphogenesis in the gene ontology (GO) biological process (Figure 1C). These findings may reflect the greater number of active inflammatory pathways with neural circuits of the brain in non-responders [45]. Proteins that fit the GO terms extracellular structure organization, regulation of complement activation, and triglyceride-rich lipoprotein particle remodeling were enriched in both groups.

**Figure 1.** Plasma proteomic analyses and functional annotations identifying changes in differentially abundant proteins over the first week of drug administration. (**A**) Time-dependent up- and downregulation of differentially abundant proteins compared with the start of drug administration; T0. The number of proteins altered at each time point is shown above each time point. (**B**) Venn diagram of proteins differentially abundant at T1, T4, and T10 vs. T0. (**C**) Gene ontology terms of proteins differentially up- and downregulated at T1, T4, and T10 vs. T0.

LMM is appropriate for identifying differentially abundant plasma proteins based on longitudinal proteome data. The response/time interaction term is important in measuring inter-group differences in time-dependent responsiveness to SSRIs. Through the LMM multiple comparison analysis, we identified 37 significant proteins which were corrected by a SGoF method [37] (adjusted *p*-value < 0.05; response/time interaction term). These proteins over time, as well as the between-group differences, are shown as lowest adjusted *p*-values in Table 2.


**Table 2.** 37 differentially abundant proteins corresponding to response/time interaction.

<sup>a</sup> Spearman's correlation coefficient of protein abundance and Montgomery and Asberg Depression Rating Scale (MADRS) for each protein. <sup>b</sup> Adjusted *p*-values < 0.05 on a permutated correlation test based on Spearman's coefficient analysis.

To better understand the abundance patterns and to cluster proteins with similar patterns, protein abundance at three different times (T1, T4, and T10) was subtracted from that at baseline (T0), followed by t-SNE and affinity propagation (Figure 2A). Six clusters of unique patterns were obtained (Figure 2B). The three and five proteins in clusters 1 and 4, respectively, decreased over time in responders and increased over time in non-responders. Cluster 2, which included five proteins, showed little change over time in responders but decreased over time in non-responders. In cluster 3, three proteins showed increase from week 1 and appeared flat week 4 onward in responders; conversely, the proteins showed a sharp decrease at week 4 and then flattened at week 10 in non-responders. In cluster 5, 10 proteins showed little change over time in responders but increased over time in non-responders. In cluster 6, 11 proteins showed decreases at 4 weeks and increases at 10 weeks in responders but little change over time in non-responders. The individual abundance profiles of the 37 proteins are shown in Supplementary Figure S2.

**Figure 2.** Affinity propagation clustering, profile analysis, and public database search of the 37 proteins found to differ significantly in the response/time interaction of linear mixed model (LMM). (**A**) Identification of seven protein clusters by t-SNE-based affinity propagation clustering. (**B**) Change over time in protein amount in responders and non-responders. (**C**) Association of 14 proteins found on PsyGeNet with psychiatric diseases. (**D**) Association of ten proteins found in the DrugMatrix category of Enrichr with responses of rat tissues and cells to selective serotonin reuptake inhibitors (SSRIs).

To assess whether the functional roles of these proteins were associated with antidepressant response and psychiatric disorders, we searched for the 37 proteins in the PsyGeNet (Figure 2C) [36]. APOD, APOE, BCHE, DBH, GGH, GSN, ITIH3, LCN2, MMP2, PHOX2B, PON1, TNXB, VWF, YWHAE, 14 of these proteins were found to be associated with psychiatric symptoms, such as schizophrenia, bipolar disorder, cocaine use disorders, substance-induced psychosis, alcohol use disorders, and depression. In addition, we assessed whether these 37 proteins were associated with citalopram, an analog of escitalopram, by searching responses of rat tissues and cells to SSRIs in the DrugMatrix category of Enrichr [46], a web-based gene enrichment analysis tool. We found that expression of nine proteins, ITIH3, PON1, MMP2, MYH9, APOE, GC, CD14, LCN2, and CTSD, differed significantly in SSRI-treated and control, corn oil-treated rat liver; the expression of seven proteins, ITIH3, PON1, LCN2, APOE, GC, CLU, and CTSD, differed significantly in SSRI- and corn oil-treated rat hepatocytes; and three proteins, PLXNB1, MMP2, and CTSD, differed significantly in SSRI- and corn oil-treated rat hearts (Figure 2D). It indicated that the drug reaction of these proteins causes quantitative changes not only in the blood but also in the organs of the liver and heart.
