Searching for biomarkers in schizophrenia and psychosis: Case‐control study using capillary electrophoresis and liquid chromatography time‐of‐flight mass spectrometry and systematic review for biofluid metabolites

Abstract Metabolomics has been attracting attention in recent years as an objective method for diagnosing schizophrenia. In this study, we analyzed 378 metabolites in the serum of schizophrenia patients using capillary electrophoresis‐ and liquid chromatography‐time‐of‐flight mass spectrometry. Using multivariate analysis with the orthogonal partial least squares method, we observed significantly higher levels of alanine, glutamate, lactic acid, ornithine, and serine and significantly lower levels of urea, in patients with chronic schizophrenia compared to healthy controls. Additionally, levels of fatty acids (15:0), (17:0), and (19:1), cis‐11‐eicosenoic acid, and thyroxine were significantly higher in patients with acute psychosis than in those in remission. Moreover, we conducted a systematic review of comprehensive metabolomics studies on schizophrenia over the last 20 years and observed consistent trends of increase in some metabolites such as glutamate and glucose, and decrease in citrate in schizophrenia patients across several studies. Hence, we provide substantial evidence for metabolic biomarkers in schizophrenia patients through our metabolomics study.


| INTRODUC TI ON
Schizophrenia is a chronic and disabling psychiatric illness that affects around 1% of the general population. 1 Previous studies suggest that neuronal dysfunction in dopaminergic and glutaminergic systems, 2 metabolic abnormalities in glycolysis and oxidative stress responses, and hyperactive immune systems could be involved in the pathogenesis of schizophrenia. 3 Neurochemical imaging tools, such as magnetic resonance spectroscopy, positron emission tomography, and magnetic resonance imaging, have long been used along with salivary cortisol and bloodbased biomarkers (inflammatory, oxidative stress, immune analytes) 4 in the pathological study of schizophrenia. Recently, metabolic biomarkers have attracted attention. The number of studies trying to identify biomarkers of schizophrenia and other psychotic disorders in biofluids (blood, urea, cerebrospinal fluid), the post-mortem brain, and through magnetic resonance spectroscopy has rapidly grown in recent years. 3 According to previous systematic reviews, 3,5 glutamate, N-acetyl aspartate, and some lipids and lipid-like molecules are candidates for biomarkers that may be involved in the pathogenesis of schizophrenia. However, no robust biomarkers have been identified thus far.
Mass spectrometry (MS) coupled with gas chromatography, liquid chromatography (LC) or capillary electrophoresis (CE), and nuclear magnetic resonance (NMR) spectroscopy has been commonly used for biofluid analysis. 3,6 As there is no single analytical platform capable of detecting all metabolites, an integrated approach using more than one platform is often adopted in recent studies to provide the most sensitive and reliable measurements. 6 Here, we analyzed 378 metabolites in the serum, including fatty acids, amino acids, and other organic acids in patients with schizophrenia and healthy controls, by combining CE-time-of-flight mass spectrometry(TOFMS) and LC-TOFMS. We subjected these data to multivariate analysis and tried to identify the metabolites that can differentiate patients with schizophrenia from healthy control subjects to build a biomarker that can assess the disease trait of patients. We also compared the acute and remission state of the same patients and tried to identify a biomarkers corresponding to the difference between these two states.
In biomarker research, univariate analysis using Mann-Whitney or predictive models is more commonly used, and there are not so many reports of studies using multivariate analysis. Multivariate analysis has been attracting attention in recent years and is now being used in the field of metabolomics. 7,8 The metabolome is actually a result of the intersection of a large number of factors, and multivariate analysis allows us to snapshot the obtained results from multiple directions, thereby enabling us to have more comprehensive view than does univariate analysis.
Additionally, we conducted a systematic review of studies from the last 20 years investigating metabolite differences between schizophrenia or psychotic disorder subjects and control subjects.

| Subjects
This study was approved by the Ethical Committee of Kobe University Graduate School of Medicine. All the participants were of Japanese descent and were recruited in the Hyogo prefecture, Japan. We obtained written informed consent after describing the study to the subjects. The psychiatric assessment of each participant was performed as previously described. 9,10 As recommended by the Diagnostic and Statistical Manual of Mental Disorders, 4th Edition (DSM-IV) criteria, at least two psychiatrists diagnosed the patients with schizophrenia. The control group consisted of healthy volunteers. None of the control subjects had any present, past, or family (first-degree relatives) histories of psychiatric disorders or substance abuse excluding nicotine dependence. We conducted interviews and screening for psychiatric disorders on all the control subjects.
Demographic and clinical characteristics are shown in Table 1.
The subjects consisted of 20 unrelated patients with chronic schizophrenia (SCZc) and 20 unrelated healthy controls (CTL). Additionally, we took blood samples from 20 patients who were admitted due to acute psychosis, once at the time of admission (SCZa-Acute) and once just before being discharged (SCZa-Remission).

| Statistics
The collected data were imported to SIMCA software v. 16  S-plots of this model were used to see which metabolites help in strongly differentiating these two groups. We checked the significance of the metabolites selected by OPLS-DA using the non- Finally, we used the OPLS regression model to compare the metabolites of the paired samples of SCZa-Acute and SCZa-Remission. This is called OPLS-effect projection (OPLS-EP) and was used to test which metabolites contribute most to differentiating the acute and remission states of the same patients. For metabolites selected by OPLS-EP, we performed the paired non-parametric Wilcoxon signed rank test (two-tailed) in R to assess the changes between two states of the same patients.
Principal component analysis and OPLS-DA were performed with pareto-scaled datasets and OPLS-EP was performed with unit variance (UV) scaled datasets. We performed this analysis similar to previous studies. [12][13][14] We considered P < .05 as statistically significant for all statistical analyses. However, since we have 378 comparisons, we used Bonferroni correction to obtain the adjusted (α = 0.05/378 ≒ 0.000132). All the univariate analyses of our data were performed using R Version 4.0.2 (R Core Team, Vienna, Austria; https://www.R-proje ct.org).

| Systematic review for possible biomarkers
We conducted a systematic search according to PRISMA guidelines. 15 We used the following search terms by referring to a previous review 3 : "(schizophreni? OR psychosis OR "at risk mental state" OR "at-risk mental state" OR ARMS OR "ultra-high risk" OR UHR) AND We summarized only the metabolites that showed significant changes and did not include those without significant differences.

| Principal component analysis
A total of 378 metabolites were detected in the serum of the participants. The PCA score plot of PC1 and PC2 showed reasonably clear independent groupings (Figure 2). The scores of PC1, PC2, and the residuals were 25.7%, 18.8%, and 55.5%, respectively.
The CTL and SCZc groups were clearly separated from each other, while SCZa-Acute and SCZa-Remission groups were distributed between these two groups. The distribution of SCZa-Remission was similar to, but narrower than, that of SCZa-Acute.

| Comparison of control subjects and chronic schizophrenia patients
In the score plot of OPLS-DA with two components, both R2 and Q2 were statistically significant (R2 = 0.79, Q2 = 0.76) and the two groups SCZc and CTL were clearly separated ( Figure 3A).
The resulting S-plot is shown in Figure 3B. The horizontal axis (ie, the loading) describes the magnitude of each variable, and the vertical axis represents the reliability of each variable in the OPLS-DA data. Metabolites that exceeded the threshold of 0.70 for p(corr) [1] values and 0.1 for p [1] values are labeled in Figure 3B, and considered as significantly different between the two classes. Full data sets are shown in Table S1.
In total, six unique metabolites (alanine, glutamate, lactic acid, ornithine, serine, and urea) were identified to be responsible for class separation. We performed the non-parametric Mann-Whitney U-test (two-tailed) on these six selected metabolites and observed significant differences in all these metabolites (P < .000132) ( Figure 3C).
We also performed a multivariate logistic regression analysis to determine whether these six metabolites, sex, and age are biomarkers for schizophrenia. We first tested the six metabolites individually and found that the AUCs ranged from 0.807 to 1.0 ( Figure S1A-F).
In addition, the ROC analysis including sex and age as fixed factors We also performed a regression analysis using a generalized linear model with gamma distribution and log link, in which serum metabolite level was the response variable, and age, sex, and antipsychotic dose were the explanatory variables (Table S2) Table S3.
Levels of these five metabolites were significantly higher in SCZa-Acute than SCZa-Remission according to Wilcoxon signed rank test (P < .05) ( Figure 4B). In addition, Spearman's correlation analysis was performed to determine the effect of antipsychotic

| Systematic review on metabolite biomarkers
We reviewed and summarized 33 reports including our data (Tables S4-S6). The metabolites reported to be significant in four or more reports were summarized ( Figure 5). We divided results into sections based on metabolite classes: (1) lipids and lipid-like molecules (including fatty acids, steroids, and other lipid-like molecules), (2) carbohydrate metabolites, organic acids, and derivatives, and (3) other metabolites.

| Lipids and lipid-like molecules
Significant changes in fatty acids, such as oleic acid, 16 were reported to be significantly different in schizophrenia patients in four or more reports. Increase in glucose was reported in several papers and was not necessarily accompanied with antipsychotic use as four of these studies were conducted on drug-free patients. 18,19,22,23 We also observed a trend that citrate decreased 18,19,22,24 and glutamate increased 19,21,26,27 in psychotic patients. However, the differences in aspartate, taurine, and lactic acid were inconsistent across studies.

| DISCUSS ION
We observed significant differences in the levels of several amino acids, lactic acid, and urea between CTL and SCZc groups in our case-controlled experiments. Moreover, we found significant differences in several fatty acids and thyroxine between the same individuals during the acute and remission phases (SCZa-Acute and SCZa-Remission groups). As these substances were not correlated with the amount of antipsychotic medications, we concluded that the metabolites identified in the first set may be biomarkers of schizophrenia while the metabolites in the second set may be state markers of acute psychosis.
Glutamate is an amino acid known to be a major excitatory neurotransmitter in the brain. 27 Several studies report dysregulation of the glutamatergic system in schizophrenia. 34,35 As summarized in Section 3.2.2, recent studies showed trends toward increased plasma glutamate levels in patients with psychotic disorders consistent with our data. However, glutamate levels could also increase due to improvement of symptoms or use of antipsychotic medicines 34,36,37 and benzodiazepines. 35 Hence, more studies are needed to fully reveal the relationship between medication and peripheral glutamate concentration.
Serine and alanine are categorized as non-essential amino acids that can be synthesized in the body. D-serine is an intrinsic D-amino acid synthesized in the central nervous system from L-serine by serine racemase, whereas D-alanine is an extrinsic D-amino acid found in the gut microbiome. 38 Both of these D-amino acids are known to Lactic acid is an organic acid used as a marker for mitochondrial diseases. In such disorders, metabolism relies on extra-mitochondrial glycolysis and lactic acids accumulate since they are not metabolized. 44 Mitochondrial dysfunction in schizophrenia has been previously reported 45 and could be related to lactic acid accumulation.
Both increased and decreased lactic acid levels were reported in studies of patients with schizophrenia 18,19,22,23 and our data were consistent with some reports. 18,19,22 However, the potential antipsy- Cis-11-Eicosenoic acid (gondoic acid) is a monounsaturated omega-9 fatty acid found in a variety of plant oils and nuts. While not many studies have examined this fatty acid, our data were consistent with Yang et al. 19 Our findings could suggest systematic alternations in glycogenolysis and lipid metabolism in patients with acute psychosis, similar to suggestions by previous studies.
Thyroxine, the hormone produced and released by the thyroid gland, has also recently attracted attention in relation to schizophrenia. Although we did not find any consistent reports of thyroxine in our systematic review, our metabolomics study indicates dysregulation of the pituitary-thyroid axis, which could be related to major neurosignaling systems in psychosis as suggested by Santos et al. 49 Following a large-scale systematic review in 2018, 3  would be able to collect more reliable data to identify possible biofluid markers in schizophrenia.
One of the biggest challenges in this field is the difficulty in excluding biases due to effects of antipsychotic medication, diet, activities, and hospitalization. Although some studies were con- The other big challenge we face is that schizophrenia may not be a single disease, but rather a syndrome, which is affected by a wide variety of genetical and biological variables. 53 Inconsistent results of metabolites in schizophrenia could be due to these diverge pathologies. Furthermore, if we could identify metabolomic effects of psychotic disorders, medication, and acute or remission state of psychosis more correctly, they might also reveal pathologies of schizophrenia, which is still far from being fully understood.
As this research field is rapidly expanding, more studies can shed further light on these questions.
One of the major limitations of our study was that the sample size was not large with 20 participants for each group. Moreover, the participants in our study were all Japanese, which limits the extrapolation of our findings to a more general population. Additionally, the scores of PC1 and PC2 were 25.7 and 18.8, which may not be considered robustly significant. We suggest that it is difficult to degrade all the variables into two components since there were too many variables and various directions.
Furthermore, there could be biases due to hospitalization between each group. We could not exclude these biases because all the patients included in the study were severely ill and needed to be treated in a hospitalized setting.
Nevertheless, we believe our study is an important primer for further studies that could substantiate and expand our findings.

ACK N OWLED G M ENT
We thank Ms Y. Nagashima for her technical assistance.

CO N FLI C T O F I NTE R E S T
The authors declare that they have no conflict of interest.

A PPROVA L O F TH E R E S E A RCH PROTO CO L BY A N I N S TITUTI O N A L R E V I E WER B OA R D
The protocol for this research project has been approved by a suitably constituted Ethics Committee of the institution and it conforms to the provisions of the Declaration of Helsinki.

I N FO R M E D CO N S E NT
All informed consent was obtained from the subjects or guardians.

R EG I S TRY A N D TH E R EG I S TR ATI O N N O. O F TH E S T U DY
n/a

A N I M A L S TU D I E S
n/a

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that supports the findings of this study are available in the Supporting Information of this article.