Serum seleno-proteins status for colorectal cancer screening explored by data mining techniques - a multidisciplinary pilot study

Abstract

In this case–control pilot study a recent method based on HPLC hyphenated to ICP-MS was employed for the quantification of serum glutathione peroxidase type 3 (GPx3), seleno-protein P (SelP) and seleno-albumin (SeAlb) in 42 patients with colorectal cancer (CRC) and 20 controls. Patients with early cancer stage (TNM I) showed a significantly higher level of SeAlb (19 ± 3 ng/mL) in respect to both metastatic CRC patients (TNM IV, 16 ± 4 ng/mL) and healthy controls (16 ± 3 ng/mL). Classification models based on logistic regression analysis, classification trees and artificial neural networks were constructed using seleno-proteins concentrations as predictors. Neural networks lead to the best performances, up to 95% of corrected predictions in TNM I vs. controls discrimination. These results suggest a potential association between individual seleno-proteins and CRC progression. Age and radiochemoteraphy were assessed as confounding factors, showing no significant effects. Still, SeAlb level tended to reduce with the age in healthy persons, but did not in CRC patients. Seleno-proteins concentration was also compared with a number of clinical parameters considered as prognostic factors in CRC. Significant Spearman's correlations were revealed between SelP and SeAlb, and presence of peritumoural lymphocytic infiltration (ρ = − 0.57 and ρ = − 0.37, respectively); and SeAlb and degree of cellular differentiation (grading, ρ = − 0.37). This study marks the importance to systematically introduce speciation analysis and multidisciplinary approaches in the investigation of the role of seleno-proteins as a potential combined biomarker for CRC.

Introduction

Selenium (Se) is an essential trace element whose important function in human health has attracted great attention in the last decade. Its role in oxidative stress regulation makes it potentially involved in the development, progression and prognosis of several diseases, in particular cancers [1]. Colorectal cancer (CRC) is the third most commonly diagnosed form of cancer in women and fourth in men worldwide [2], and its association with Se has been investigated in a number of studies. Some of them observed a significantly lower level of Se in serum of CRC patients than in healthy subjects, or a significantly higher risk to develop CRC and lower cumulated cancer-related survival rates for subject presenting low serum Se levels [3], [4], [5]. A similar behaviour was noticed in patients affected by colorectal adenoma, the precursor lesion in most CRC cases [6], [7], [8], [9]. Selenium levels have also been associated with the CRC stage (from adenomatous polyps to local and metastatic cancer, respectively) [10]. However, other studies regarding Se and colorectal adenoma did not observe significant correlations [11], [12], [13]. These contradictions reflect a more general picture emerging from a wide number of trials and cross-sectional studies on the cancer preventive action of supplemented Se, that observed both beneficial than irrelevant effects, suggesting a possible pathology-specificity [14], [15], [16], [17]. Hence, the association between Se status and cancer is probably more complex than initially expected.

A critical aspect is that different biological functions of Se are carried out by individual proteins (Se-proteins), where it participates to the active site. Se is incorporated into Se-proteins as the amino acid Se-cysteine (SeCys) through a genetically encoded pathway. Complex regulatory mechanisms showed to be responsible for the preferential incorporation of Se into certain Se-proteins, so that in marginal element deficiency the metabolism of these compounds is unaffected whereas the content of other species might be decreased [18], [19]. Se-proteins have to be also distinguished from the Se-containing proteins, that are biologically inactive (from the point of view of Se) [1], [20], [21], where Se is unspecifically incorporated as Se-methionine (SeMet) by random replacement of (sulphur) methionine (Met). Thus, the individual Se-proteins level rather than total Se has been suggested as alternative biomarker for Se status assessment, including for cancer studies, which could reflect more accurately the real dynamics of the element [1], [22]. In this context, the human plasma/serum Se-proteins glutathione peroxidase type 3 (GPx3) and Se-protein P (SelP), as well as the Se-containing protein Se-albumin (SeAlb, for simplicity also refereed below as a Se-protein) are of high interest. An inverse correlation has been observed between various types of cancer (including CRC) and serum GPx3 or plasma SelP [4], [23], [24], but other works did not confirm this data [12], [25]. To our knowledge only a single study by Early et al. [12] took into account the complete plasma Se-proteins status in CRC. In such work GPx3, SelP and total Se were determined in distinct steps recurring to enzymatic assay, radioimmunoassay and fluorimetry, respectively, while the SeAlb level was calculated by difference, and only basic data analysis was provided. Thus, new investigations are needed to elucidate the role of serum Se-proteins in CRC on the bases of both robust analytical methods and comprehensive data elaboration.

For this type of applications, predictive data mining techniques has been increasingly used in recent years [26]. These approaches are able to combine a wide range of patients-specific different information to obtain descriptive models that can turn in decision systems effectively supporting clinical processes such as diagnosis, prognosis and treatment. Data mining for classification proposals manipulates data sets to elucidate hidden patterns by a variety of methods. Among them artificial neural networks [27], [28], k-nearest neighbours [29], hierarchical clustering [30] and principal component analysis [31] have been already explored to discover unknown data structures in proteomic and genomic profiling of CRC, allowing not only to discriminate CRC from normal tissues but also to properly classify tumor subgroups. Support vector machine has been also applied to serum proteomic profiling to predict the early-stage response of CRC to radiochemotherapy [32]. Other data mining methods among the most used for clinical tasks include decision trees and rules, logistic regression, and the more classical chemometric techniques [26], [33].

The aim of our study was to combine a newly developed analytical methodology for the simple, fast and accurate simultaneous determination of GPx3, SelP and SeAlb in human serum [34] with clinical information and data mining tools. Samples from 42 patients affected by CRC and 20 healthy control subjects were analyzed in this pilot work, in order to assess the possible association between Se-proteins pattern and the presence of cancer, progression stage and also prognostic criteria. A comprehensive post-hoc statistical approach based on data mining techniques including logistic regression analysis, classification trees and artificial neural networks was addressed in order to explore the potentialities of Se-proteins level as an integrated biomarker for CRC screening.