Zinc-ion binding and cytokine activity regulation pathways predicts outcome in relapsing-remitting multiple sclerosis

Abstract

Multiple sclerosis (MS) is a demyelinating disease characterized by an unpredictable clinical course with intermittent relapses that lead over time to significant neurological disability. Clinical and radiological variables are limited in the ability to predict disease course. Peripheral blood genome scale analyses were used to characterize MS patients with different disease types, but not for prediction of outcome. Using complementary-DNA microarrays we studied peripheral-blood gene expression patterns in 53 relapsing-remitting MS patients. Patients were classified into good, intermediate and poor clinical outcome established after 2-year follow-up. A training set of 26 samples was used to identify clinical outcome differentiating gene-expression signature. Supervised learning and feature selection algorithms were applied to identify a predictive signature that was validated in an independent group of 27 patients. Key genes within the predictive signature were confirmed by quantitative reverse transcription-polymerase chain reaction in an additional 10 patients. The analysis identified 431 differentiating genes between patients with good and poor clinical outcome (change in neurological disability by the expanded disability status scale was -0.33 +/- 0.24 and 1.6 +/- 0.35, P = 0.0002, total number of relapses were 0 and 1.80 +/- 0.35, P = 0.00009, respectively). An optimal set of 29 genes was depicted as a clinical outcome predictive gene expression signature and classified appropriately 88.9% of patients. This predictive signature was enriched by genes related biologically to zinc-ion binding and cytokine activity regulation pathways involved in inflammation and apoptosis. Our findings provide a basis for monitoring patients by prediction of disease outcome and can be incorporated into clinical decision-making in relapsing-remitting MS.

Figures

Fig. 1

Flowchart of the study design. Overview of the strategy used for the identification and validation of predictive clinical outcome gene-expression signature in relapsing–remitting multiple sclerosis.

Fig. 2

(a) Heatmap of clinical outcome differentiating genes. Heatmap of the 431 differentiating genes that distinguishes between patients with good and poor clinical outcome. Each row of the heatmap represents a gene and each column represents a patient sample. Genes with increased expression are shown in progressively brighter shades of red, and genes with decreased expression are shown in progressively darker shades of green. The bottom matrix shows corresponding clinical outcome attributes marked in black when positive. (b) Functional annotation histogram. Distribution of differentiating gene expression signature according to biologically relevant functional groups. (c) Overabundance analysis. Actual number of genes (blue line) is significantly more abundant than expected (red line) for threshold number of misclassifications (TNoM) statistical test. x-Axis denotes P-value; y-axis denotes number of genes. (d) Leave-one-out cross-validation (LOOCV) classification. Division of errors between patients with good and poor clinical outcome using TNoM, Info and t-test demonstrated high classification rate of 90% at P < 0·0001. x-Axis denotes P-value; y-axis denotes error rate in percentage.

Fig. 3

(a) Predictive classification chart. The classification rate of 29 predictive genes is demonstrated. Highest classification rate is achieved using only seven genes, yet according to the feature selection algorithm, genes are added to the subset as long as the classification rate is not decreased. y-Axis denotes classification rate; x-axis denotes the number of genes. (b) Gene enrichment. Direction of an over-expressed (1) or down-expressed (− 1) gene is demonstrated in the enriched groups within the poor versus good outcome signature. (c) Heatmap of module analysis using differentiating clinical outcome genes. Enrichment of zinc-ion binding gene set for patients with relapses and cytokine activity gene set for patients with stable disease [no change in neurological disability, Expanded Disability Status Scale (EDSS) = 0] are demonstrated. The upper left graphical panel is a matrix of gene sets versus arrays, where a coloured entry indicates that the genes in the gene set had changed significantly in a co-ordinated fashion in the respective array (red, increased; green, decreased). The centre graphical panel shows individual clinical outcome attributes to which each array belongs. The bottom graphical panel demonstrates overall clinical outcome attributes in which gene sets were significantly enriched. (d) Reconstructed zinc-ion binding pathway. Pathway analysis performed using genes from the predictive signature (yellow circles) and genes brought into the pathway based on literature-known relationships according to PathwayArchitect software (green circles). Arrows indicate regulatory interactions confirmed by literature database, dashed arrows indicate suggested gene interactions. (e) Reconstructed cytokine activity pathway. Pathway analysis performed using genes from the predictive signature (grey circles) and genes brought into the pathway based on literature-known relationships according to PathwayArchitect software (blue circles). Arrows indicate regulatory interactions confirmed by literature database, dashed arrows indicate suggested gene interactions. (f) Gene expression regulatory network module. The single gene expression module from the gene expression regulatory network of 431 differentiating genes is demonstrated. Each node in the regulation tree represents a regulating gene. The expression of the regulating genes themselves is shown below their node. Cluster of gene expression profiles (rows represent genes, columns represent patients' arrays) arranged according to the regulation tree. Note that zinc-ion binding related genes KLF4 (regulating gene) and S100B (regulated gene) belong to the same regulatory module (black asterisks).