Study design and participants

We collected blood samples from a cohort of 30 MDD patients. For every participant, we collected samples both before and after ketamine treatment, which allowed us to examine the proteomic changes at different time points. At the end of the treatment, 20 responders (also referred to as R) and 10 non-responders (also referred to as NR) were identified (Fig. 1). We compared demographic differences between the responders and non-responders. Except for body mass index (BMI) and family history, there was no significant difference between the two sub-populations (Table 1). We then performed MS-based proteome profiling on each sample. The MS-based method implemented in this study has high robustness in terms of reproducibility, with a Pearson correlation coefficient of nearly 1 and p < 2.2e-16 (Supplementary Fig. 1). Afterwards, downstream analyses of the obtained proteomics data were carried out.

Fig. 1

Overview of study population and schematic proteomic workflow. MS in DIA mode was used to profile plasma samples of 30 MDD patients before (pre) and after (post) ketamine treatment. Pre- and post-treatment proteomics data sets were also referred to as DIA-pre and DIA-post, respectively. Merged data, DIA-merge, was generated by merging DIA-pre and DIA-post. Afterwards, (1) proteins as indicators of patients’ response to ketamine were evaluated and selected by receiver operating characteristic (ROC) curve analysis (using DIA-pre), (2) proteomic dynamics reflecting temporal proteome changes caused by ketamine from pre- to post-treatment was discerned (using DIA-merge), and (3) proteome alterations in responders versus non-responders were investigated to capture response-relevant protein signatures in the bloodstream to reveal antidepressive mechanisms of ketamine (using DIA-post)

Table 1 Demographics of the MDD patients involved in this studyPre-treatment proteomic profiles

The in-depth proteomic landscape of 30 MDD patients before ketamine treatment was resolved using a high-resolution DIA MS-based method, and the proteome data, namely DIA-pre, was acquired.

From the cohort, 8034 peptides in total were obtained, with the number ranging between 3669 and 6280 in individual samples, and with 2881 peptides repeatedly detected across all samples (Fig. 2A). After mapping these peptides to the human reference proteome, a total of 562 proteins were quantified in at least one sample, with the number ranging between 320 and 530 proteins in individual samples, and with 264 proteins commonly quantified across all samples (Fig. 2B). These identified proteins were supported by an average of 14 peptides, and 493 of these proteins (87.7%) could be traced and supported by at least 2 peptides (Fig. 2C). The number of samples where a protein was observed was counted. A small number of proteins were found to be restricted to a few samples while most of the proteins were quantified in the majority of the samples analyzed. The distribution of proteins identified in different samples was analyzed, and we found that only 21 proteins were detected in a single sample, that 541 proteins were quantified in at least 2 samples, that up to 264 (47%) proteins were simultaneously quantified in all 30 samples, and that, on average, proteins could be detected across 22 samples (Fig. 2D).

Fig. 2

Profiling of pre-treatment samples. A Distribution of the number of quantified peptides across samples; the dash-dotted line shows the number of common peptides in all samples. B Distribution of the number of quantified proteins across samples; the dash-dotted line shows the number of common proteins in all samples. C Distribution of peptide numbers of quantified proteins. D Distribution of protein numbers in samples. E Assessment of individual sample quality with respect to three contamination indices, i.e., coagulation, erythrocyte, and platelet, respectively; the dashed magenta line indicates two standard deviations from the average value. F Comparison of proteins identified in pre-treatment samples with Human Plasma Proteome Project (HPPP). G Location distribution of proteins identified in pre-treatment samples. H Existence distribution of proteins identified in pre-treatment samples; PE1-PE5, evidence levels in the neXtProt database; NA, no information available in the neXtProt database

To further evaluate the quality of DIA-pre, we compared it against public data sets. The sample contamination analysis revealed that no sample was contaminated by erythrocytes, platelets, or coagulation factors (Fig. 2E). Notably, the majority (522/562) of the proteins we identified were also in HPPP, constituting 12% of the HPPP data set (Fig. 2F). In line with the nature of our plasma samples, the identified proteins were mainly secreted into the blood (Fig. 2G). Nearly all of the identified proteins presented PE1 and PE2 levels of existence, and the majority had a level of PE1 (Fig. 2H). Therefore, these findings indicate high reliability of the MS-based analysis and high quality of the DIA-pre data set.

Post-treatment proteomic profiles

The proteome of MDD patients treated with ketamine holds great potential for advancing biomedical research and personalized medicine, but it has not been fully explored. In this study, we present an in-depth proteomic analysis of 30 MDD patients following ketamine treatment using a high-resolution, DIA-MS-based strategy. This resulted in the generation of a comprehensive proteome dataset, referred to as DIA-post.

From the cohort, 8118 peptides in total were obtained, with the number ranging between 4876 and 6231 in individual samples, and with 3606 peptides repeatedly detected across all samples (Fig. 3A). After mapping these peptides to the human reference proteome, a total of 528 proteins were quantified in at least one sample, with the number ranging between 401 and 490 proteins in individual samples, and with 315 proteins commonly quantified across all samples (Fig. 3B). These identified proteins were supported by an average of 15 peptides, and 457 of these proteins (86.5%) could be traced and supported by at least 2 peptides (Fig. 3C). The number of samples where a protein was observed was counted. A small number of proteins were found to be restricted to a few samples while most of the proteins were quantified in the majority of the samples analyzed. The distribution of proteins identified in different samples was analyzed, and we found that only 5 proteins were detected in a single sample, that 523 proteins were quantified in at least 2 samples, that up to 315 (59.7%) proteins were simultaneously quantified in all 30 samples, and that, on average, proteins could be detected across 24 samples (Fig. 3D).

Fig. 3

Profiling of post-treatment samples. A Distribution of the number of quantified peptides across samples; the dash-dotted line shows the number of common peptides in all samples. B Distribution of the number of quantified proteins across samples; the dash-dotted line shows the number of common proteins in all samples. C Distribution of peptide numbers of quantified proteins. D Distribution of protein numbers in samples. E Assessment of individual sample quality with respect to three contamination indices, i.e., coagulation, erythrocyte, and platelet, respectively; the dashed magenta line indicates two standard deviations from the average value. F Comparison of proteins identified in post-treatment samples with Human Plasma Proteome Project (HPPP). G Location distribution of proteins identified in post-treatment samples. H Existence distribution of proteins identified in post-treatment samples; PE1-PE5, evidence levels in the neXtProt database; NA, no information available in the neXtProt database

We compared DIA-post with public data to further evaluate its quality. The sample contamination analysis revealed that no sample was contaminated by erythrocytes, platelets, or coagulation factors (Fig. 3E). Notably, the majority (489/528) of the proteins we identified were also in HPPP, constituting 11% of the HPPP data set (Fig. 3F). In line with the nature of our plasma samples, the identified proteins were mainly secreted into the blood (Fig. 3G). Nearly all of the identified proteins presented PE1 and PE2 levels of existence, and the majority had a level of PE1 (Fig. 3H). Therefore, these discoveries indicate high reliability of the MS-based analysis and high quality of the DIA-post data set.

Proteome differences between ketamine responders and non-responders

Hypothesizing that the differences in patients’ response to ketamine are merely superficial reflections of underlying molecular differences deeply rooted in the proteome, we analyzed altered proteins between ketamine responders and non-responders using the DIA-post data set.

Differential abundance analysis revealed 45 DAPs, accounting for 11.3% of the total quantified post-treatment proteome. Compared to non-responders, 38 out of the 45 DAPs were up-regulated in responders, while the remained were down-regulated (Fig. 4A and B). We then performed unsupervised principal component analysis (PCA) and hierarchical clustering of the samples on the basis of the DAPs. The PCA result revealed a clear separation between responders and non-responders in the first two components (Fig. 4C). In agreement with the PCA result, hierarchical clustering clearly separated the two groups of the samples (Fig. 4D).

Fig. 4

Proteomic alterations in post-tretment samples between responders (R) and non-responders (NR). A Expression levels of DAPs in responders versus non-responders. Data points indicate the data of individual patients and are presented as medians with interquartile ranges; the center line within each box shows the median, and the top and bottom of each box represent the 75th and 25th percentile values, respectively; the upper and lower whiskers extend from the hinge to the largest and smallest values no further than 1.5 times the distance between the first and third quartiles, respectively; and outliers are presented as circles. B Differential protein abundance profile. C Separation of responders and non-responders by DAPs using principal component analysis. D Hierarchical clustering of responders and non-responders by DAPs

Correlations between protein abundance and HAMD scores measured post ketamine treatment were analyzed. As shown in Fig. 5A and B, there were 31 proteins significantly correlated with post-treatment HAMD scores (p < 0.05), and the correlations were predominantly negative. Among the 31 proteins, there were 9 DAPs, and they were all negatively correlated with post-treatment HAMD scores, with the exception of IGLV2-23.

Fig. 5

Protein correlation with HAMD scores. A Correlation of proteins with HAMD scores measured post-treatment across samples. B Proteins in A significantly correlated with post-treatment HAMD scores. C Correlation of proteins with decreases in HAMD scores from pre-treatment to post-treatment across samples. D Proteins in C significantly correlated with HAMD score decreases. E-J Correlations of individual proteins with HAMD score decreases. In A and C, magenta dashed lines denote the p value (i.e., 0.05) threshold for a significant correlation; Sig represents significantly correlated; Sig DAP denotes significantly correlated DAP

Correlations between protein abundance and decreases in HAMD scores from pre-treatment to post-treatment were analyzed as well. As shown in Fig. 5C and D, 24 proteins were significantly correlated with HAMD score decreases (p < 0.05), and most of the correlations were positive. Among them, six proteins were also DAPs, namely IGKV2D-30, COMP, IGHV3-64, IGHV3-15, ADAMTSL4, and IGLV2-23. Intriguingly, these six proteins were also significantly correlated with post-treatment HAMD scores. Upon analyzing the correlations of the six proteins, we found a striking pattern: proteins negatively correlated with post-treatment HAMD scores were positively correlated with the reduction in HAMD scores, and vice versa (Supplementary Fig. 2 and Fig. 5E-J). These proteins may constitute a core set of plasma proteins associated with the mechanism by which ketamine alleviates depression.

To further elucidate the function of the six proteins, we annotated them with Gene Ontology terms. The main biological process they are involved in is related to immune response, and they mainly possess the molecular function of binding to other biomolecules (Supplementary Table 1). We then examined their differential expression profiles and found that all but IGLV2-23 were up-regulated in ketamine responders (Fig. 6A-F).

Fig. 6

Protein function. A-F Expression differences of individual proteins between responders and non-responders. * p < 0.05. (G) Global correlation map showing co-regulation of proteins and clinical parameters; proteins are shown in red font color while clinical parameters are shown in blue font color; protein accession numbers are shown on the x axis and the corresponding genes are shown on the y axis; clinical parameters grp and v14hamdsum denote group label and post-treatment HAMD score, respectively

Proteins frequently collaborate with each other to exert their functions. We used 398 proteins in the DIA-post data set and incorporated 2 clinical parameters (group label and post-treatment HAMD score). The proteins, along with the clinical parameters, comprised a vector of 400 elements. We cross-correlated them to generate a matrix of 160,000 correlation coefficients. We then developed a global correlation map to capture the coordination between the proteins and clinical metrics, aiming to uncover the antidepressant mechanisms of ketamine. As can be seen in Fig. 6G, five of the six proteins form two big cluster areas with other proteins and the group clinical parameter. On the contrary, IGLV2-23, the only protein up-regulated in non-responders and positively correlated with post-treatment HAMD scores, clusters closely with the corresponding clinical parameter and maintains a distinct distance from the other two areas. We further carried out enrichment analysis of proteins belonging to the two clusters, and the enriched pathways are related to immune response. These findings indicate that the antidepressive effect of ketamine results from co-regulation of proteins that perform similar functions in the body.

Bearing these results in mind, we propose that the differences in patients’ response to ketamine is manifested by the six-protein panel and that the antidepressive mechanisms of ketamine is associated mainly with immune response. Compared with the other five proteins, IGLV2-23 functions in an opposite way. It is down-regulated in responders, negatively correlated with HAMD score decreases, and positively correlated with post-treatment HAMD scores. Ketamine is likely to take antidepressive effect by down-regulating IGLV2-23 and up-regulating the other five proteins.

Dynamic proteome rearrangements reveal the effect of ketamine over time

Drug response is reflected in the change that occurs between two states (e.g., pre- and post-treatment), rather than a snapshot of one of them. To identify how ketamine affects the proteome over time, we used matched samples from the same donors in our cohort and performed a paired Student’s t test between matched pre- and post-treatment samples for responders and non-responders, respectively. For easy writing and reading, the comparison in responders is named R_pre_post, and that in non-responders is named NR_pre_post.

Here the differential abundance analysis was performed via paired Student’s t-test using R software. P values were adjusted using the Benjamini–Hochberg method via the p.adjust function in R. A more stringent significance cutoff was used to select DAPs. We chose adjusted p < 0.01 instead of p < 0.05 to reduce the proportion of common DAPs between R_pre_post and NR_pre_post (Supplementary Fig. 3). The change in the cutoff mainly affected the number of DAPs in NR_pre_post, whereas the impact was subtle in R_pre_post. The number of DAPs in NR_pre_post dramatically decreased from 83 to 11. This stringent approach is designed to minimize false positives and uncover accurate results.

We identified 81 significantly altered proteins in responders over the entire time course, accounting for 22.3% of all proteins; however, 97% of the proteins remained unaltered after treatment in non-responders (Fig. 7A and Supplementary Fig. 4). Subsequent PCA and hierarchical clustering revealed that pre- and post-treatment samples of responders were clearly separated by the altered proteins (Fig. 7B and C). Enrichment analysis of up-regulated proteins upon treatment in responders revealed a significant enrichment of adaptive immune response, humoral immune response, plasma lipoprotein remodeling, hydrogen peroxide catabolic process, amyloid fiber formation, response to glucocorticoid, NABA ECM regulators, extracellular matrix organization, hemostasis, and axon development (Fig. 7D). Enrichment analysis of down-regulated proteins upon treatment in responders revealed a significant enrichment of adaptive immune response, blood vessel morphogenesis, response to wounding, and negative regulation of cell differentiation (Fig. 7E). The results indicate ketamine triggers changes in proteins belonging to different functional groups to execute its effect over time post administration.

Fig. 7

Proteomic rearrangements in ketamine responders over time. A Differential abundance profile. B PCA with DAPs in A. C Hierarchical clustering with DAPs in A, with up- and down-regulated proteins in all samples from pre- to post-treatment shown in insets on the left. D-E Functional enrichment of up- and down-regulated proteins, respectively

Proteins capable of differentiating ketamine responders from non-responders

Identifying a metric to distinguish responders from non-responders before treatment is beneficial to patients and would ultimately facilitate personalized medicine in MDD treatment. Here, we used proteome data from pre-treatment samples to identify proteins that can predict the response of MDD patients to ketamine before treatment.

We performed PCA before differential abundance analysis and found that data points of two samples were located far from other data points (Supplementary Fig. 5). They seemed to be outliers and were removed from analysis to reduce distortion caused by them.

We found 14 DAPs in responders versus non-responders (Supplementary Fig. 6A). Next, PCA revealed that the two groups can be separated by these DAPs to a certain extent (Supplementary Fig. 6B). Even though there were two samples incorrectly clustered into the NR group by hierarchical clustering, the divergence of the two clusters was obvious (Supplementary Fig. 6C).

Next, we performed correlation analysis of protein abundance with HAMD scores measured before treatment. Twenty proteins significantly correlated with pre-treatment HAMD scores were discovered (Fig. 8A and B). Among them, three proteins were also DAPs (Supplementary Fig. 6C) and were negatively correlated with pre-treatment HAMD scores (Fig. 8C-E). These three proteins were also up-regulated in responders (Supplementary Fig. 6C). To determine whether a higher expression level of these proteins indicates a better clinical outcome, we performed receiver operating characteristic (ROC) analysis. Indeed, all three proteins achieved good performance in predicting ketamine response, using the pre-treatment proteomics data only (Fig. 8F-H). Therefore, it is reasonable to apply these proteins as treatment response biomarkers in clinic to identify MDD patients suitable for ketamine treatment at an early stage.

Fig. 8

Three proteins with predictive value in the response of MDD patients to ketamine before treatment. A-B Protein correlation with pre-treatment HAMD scores. C-E Correlations of pre-treatment HAMD scores with PI16, NEO1, and TNXB, respectively. F–H Predictive performance of PI16, NEO1, and TNXB, with 95% confidence intervals shown as shadows