Pathobiochemistry
Comparative study on serum levels of macro and trace elements in schizophrenia based on supervised learning methods

https://doi.org/10.1016/j.jtemb.2017.03.010Get rights and content

Abstract

The etiology and pathophysiology of schizophrenia (SCZ) remain obscure. This study explored the associations between SCZ risk and serum levels of 39 macro and trace elements (MTE). A 1:1 matched case-control study was conducted among 114 schizophrenia patients and 114 healthy controls matched by age, sex and region. Blood samples were collected to determine the concentrations of 39 MTE by ICP-AES and ICP-MS. Both supervised learning methods and classical statistical testing were used to uncover the difference of MTE levels between cases and controls. The best prediction accuracies were 99.21% achieved by support vector machines in the original feature space (without dimensionality reduction), and 98.82% achieved by Naive Bayes with dimensionality reduction. More than half of MTE were found to be significantly different between SCZ patients and the controls. The presented investigation showed that there existed remarkable differences in concentrations of MTE between SCZ patients and healthy controls. The results of this study might be useful to diagnosis and prognosis of SCZ; they also indicated other promising applications in pharmacy and nutrition. However, the results should be interpreted with caution due to limited sample size and the lack of potential confounding factors, such as alcohol, smoking, body mass index (BMI), use of antipsychotics and dietary intakes. In the future the application of the analyses will be useful in designs that have larger sample sizes.

Introduction

Schizophrenia is a severe and costly mental disorder that affects 0.5–1% of the population worldwide [1]. It is common for people with SCZ to suffer from problems such as long-term unemployment, poverty, homelessness [2] and a higher suicide rate [3]. The average life expectancy of people with this disorder is ten to twenty five years less than that of people without this disorder [4].

The real cause of schizophrenia is mysterious and mostly believed to be a combination of environmental and genetic factors [5]. In recent years, many molecular signatures were discovered as significant biomarkers in schizophrenia patients’ serum. In 2011 Schwarz et al. claimed that his study was the first to identify a biological signature for schizophrenia in blood serum [6]. In the same year, Yang et al. discovered that an array of serum markers, including glycerate, eicosenoic acid, β-hydroxybutyrate, pyruvate, cysteine and urine β-hydroxybutyrate, could achieve an area under the receiver operating characteristic curve (AUC) of 1 in both the training set and the test set [7]. In 2014, Tregellas et al. showed that intrinsic hippocampal activity could serve as a biomarker for cognition and symptoms in schizophrenia [8]. Neuroimaging biomarkers were also introduced by Tregellas for early drug development of schizophrenia in the same year [9]. In 2015, Chiappelli et al. introduced myo-inositol as a potential biomarker for depression in schizophrenia [10]. However, the mechanisms of these biomarkers are still unknown.

MTE play essential roles in the biological processes [11], [12]. A number of studies have shown that changes of MTE levels might be linked to the etiology and pathophysiology of some psychiatry diseases, including schizophrenia [13], [14]. However, there have existed numerous debates regarding the link between MTE levels and schizophrenia risk [15], [16], [17]. In addition, a number of MTE, such as molybdenum, neodymium, nickel, praseodymium, rubidium, antimony, stannum, strontium, thorium, titanium, thallium, uranium and vanadium, have not been included or completely discussed. The underlying interactions among these dozens of elements can be complex, traditional single variable analysis or correlation analysis may not be applicable to discover accurate predictions. Recently machine learning techniques, such as support vector machines (SVM) and feature selection methods, are gaining popularity in this field for handling high-dimensional input features and yielding better diagnostic capability. In this study, we conduct a 1:1 matched case-control study to probe the differences of 39 MTE levels between SCZ patients and healthy controls using supervised learning methods and classical statistical testing approaches.

Section snippets

Materials and methods

All experimental procedures were performed in accordance with the principles of the Declaration of Helsinki and later amendments. The study protocol was approved by the Ethics Review Committee of Health Science Center, Peking University (IRB00001052-12065). Each of the study participants was informed about the objectives of the study, and a written consent was obtained. If the independent capacity of any subject was doubted, the written consent of his or her primary caregiver was obtained

Results

A 228 × 39 matrix consisting of 114 patients and 114 healthy subjects with 39 feature macro and trace elements from the schizophrenia patient data is used. We test the aforementioned six classifiers on the original feature space as well as on the embedding spaces by using six dimensionality reduction methods. The dimension of the most embedding spaces is set as 10 for a trade-off between accuracy and complexity, except that the FDA algorithm can only project into 1D because of its inherent

Discussion

In this article, we present a study of macro and trace elements (MTE) in serum for SCZ patients based on supervised learning methods. In order to make a comparison to machine learning methods, the traditional hypothesis testing approaches, including T-test and rank sum test, are applied. As shown in Table 4, at the level of 5%, more than half of MTE are shown to be statistically significantly different between cases and controls. The means and standard deviations differ enormously between the

Conclusion

This study might be useful for diagnosis and prognosis of SCZ disorder. It also indicated other potential applications in pharmacy and nutrition. The results of this study provide useful information for designing future studies with larger sample sizes. However, the results should be interpreted with caution due to the limited sample size and the lack of controlling for the potential confounding factors, such as the use of antipsychotics, BMI, alcohol, dietary intakes and smoking.

Funding

This work was supported by the National Basic Research 973 Program of China under Grant No. 2011CB302202, the National Natural Science Foundation of China under Grant Nos. 61375051 and 61075119, and the Seeding Grant for Medicine and Information Sciences of Peking University under Grant No. 2014-MI-21.

Author contributions

J.W. and T.B.L. proposed and supervised the project. T.L. and Y.L. designed and carried out the experiments and analyzed the data. L.Y. and T.B.L. contributed to the sample collection, macro and trace elements measurement, quality control, and finding of related literature. Z.C. participated in the discussion and helped to improve the proposed methods. T.L., Y.L. and T.B.L. wrote the manuscript.

Competing financial interests

The authors declare no competing financial interests.

Acknowledgements

We thank Jing Guo, Fangbo Feng, Jinyun Qiu for screening participants, performing cognitive and symptom assessments, and collecting the samples and information of study subjects. We also thank Yaqiong Liu for the technical support of the determination of macro and trace elements.

References (32)

  • E. Schwarz et al.

    Identification of a biological signature for schizophrenia in serum

    Mol. Psychiatry

    (2012)
  • J. Yang et al.

    Potential metabolite markers of schizophrenia

    Mol. Psychiatry

    (2013)
  • J.R. Tregellas et al.

    Intrinsic hippocampal activity as a biomarker for cognition and symptoms in schizophrenia

    The Am. J. Psychiatry

    (2014)
  • J. Chiappelli et al.

    Evaluation of myo-Inositol as a potential biomarker for depression in schizophrenia

    Neuropsychopharmacology

    (2015)
  • M. Nechifor et al.

    The influence of some antipsychotics on erythrocyte magnesium and plasma magnesium calcium, copper and zinc in patients with paranoid schizophrenia

    J. Am. Coll. Nutr.

    (2004)
  • A. Rahman et al.

    Zinc, manganese, calcium, copper, and cadmium level in scalp hair samples of schizophrenic patients

    Biol. Trace Elem. Res.

    (2009)
  • Cited by (13)

    • Reprint of: Elemental dysregulation in psychotic spectrum disorders: A review and research synthesis

      2022, Schizophrenia Research
      Citation Excerpt :

      Nevertheless, five studies found increased levels of Mg in the sera and scalp hair of adults with schizophrenia compared to controls (Cao et al., 2020; Chen et al., 2018; de Souza Pessoa et al., 2020; Kaya et al., 2012; Z. Li et al., 2018). Eight studies found no difference in serum or plasma (Arinola et al., 2010; Cai et al., 2015; Cruz et al., 2020b; T. Lin et al., 2017; Mazhari et al., 2020; Nechifor et al., 2004; Ruljancic et al., 2013; Devi et al., 2008). However, of these studies, one found lower Mg in erythrocytes of patients (Nechifor et al., 2004) and another found higher platelet Mg specifically among suicidal schizophrenia patients, but not among non-suicidal patients (Ruljancic et al., 2013).

    • Elemental dysregulation in psychotic spectrum disorders: A review and research synthesis

      2021, Schizophrenia Research
      Citation Excerpt :

      Nevertheless, five studies found increased levels of Mg in the sera and scalp hair of adults with schizophrenia compared to controls (Cao et al., 2020; Chen et al., 2018; de Souza Pessoa et al., 2020; Kaya et al., 2012; Z. Li et al., 2018). Eight studies found no difference in serum or plasma (Arinola et al., 2010; Cai et al., 2015; Cruz et al., 2020b; T. Lin et al., 2017; Mazhari et al., 2020; Nechifor et al., 2004; Ruljancic et al., 2013; Devi et al., 2008). However, of these studies, one found lower Mg in erythrocytes of patients (Nechifor et al., 2004) and another found higher platelet Mg specifically among suicidal schizophrenia patients, but not among non-suicidal patients (Ruljancic et al., 2013).

    • Association of alkali metals and Alkaline-earth metals with the risk of schizophrenia in a Chinese population: A Case-Control study

      2020, Journal of Trace Elements in Medicine and Biology
      Citation Excerpt :

      As for Ca, limited studies indicated that Ca signaling dysfunction may be associated with SCZ [41,42] and the amount of peripheral Ca could affect neurons and hence cognitive functions [15]. While previous studies reported inconsistent results [43,44], our analysis found no significant differences of serum Ca concentration between participants with SCZ and HCs. In Spearman correlation analysis, we found that the concentrations of Ba was negatively related to the positive symptoms of PANSS.

    • Association between trace elements in serum from bipolar disorder and schizophrenia patients considering treatment effects

      2020, Journal of Trace Elements in Medicine and Biology
      Citation Excerpt :

      In the present study, the focus is the analysis of blood serum samples from healthy controls and treated BD and SCZ patients by determining Se, Zn, Fe, K, Ca, Mg, P, Al, Cu, Mn, and Ni using ICP-MS, in order to measure the alterations in their concentrations and point out differences between the disorders. The set of macro/trace elements was selected based on previous studies related with BD and SCZ [8,9,11–14]. The results herein presented are expected to complement metabolomics and proteomics data, thus offering a complete potential biomarker panel that can be employed for the diagnosis of these disorders in the future.

    • Discrete sampling based-flow injection as an introduction system in ICP-MS for the direct analysis of low volume human serum samples

      2019, Talanta
      Citation Excerpt :

      Applicability and the potential use of ICP-MS for routine analysis have been therefore increased in several clinical laboratories. Recent applications of ICP-MS for multielement determination in human serum require, however, long pre-treatments based on acid digestion procedures [9,11,12,14,17–21]. Despite the fact that organic matter amount is reduced after applying these procedures, the remaining solutions contain high acid concentrations, which could be a source of interferences and matrix effect.

    View all citing articles on Scopus
    1

    These two authors contributed equally to this work.

    View full text