Blood metabolomic and transcriptomic signatures stratify patient subgroups in multiple sclerosis according to disease severity

Abstract

There are no blood-based biomarkers distinguishing patients with relapsing-remitting (RRMS) from secondary progressive multiple sclerosis (SPMS) although evidence supports metabolomic changes according to MS disease severity. Here machine learning analysis of serum metabolomic data stratified patients with RRMS from SPMS with high accuracy and a putative score was developed that stratified MS patient subsets. The top differentially expressed metabolites between SPMS versus patients with RRMS included lipids and fatty acids, metabolites enriched in pathways related to cellular respiration, notably, elevated lactate and glutamine (gluconeogenesis-related) and acetoacetate and bOHbutyrate (ketone bodies), and reduced alanine and pyruvate (glycolysis-related). Serum metabolomic changes were recapitulated in the whole blood transcriptome, whereby differentially expressed genes were also enriched in cellular respiration pathways in patients with SPMS. The final gene-metabolite interaction network demonstrated a potential metabolic shift from glycolysis toward increased gluconeogenesis and ketogenesis in SPMS, indicating metabolic stress which may trigger stress response pathways and subsequent neurodegeneration.

Keywords: Classification Description; Machine learning; Metabolomics; Molecular network; Transcriptomics.

Graphical abstract

Figure 1

Serum metabolomics can stratify patients with SPMS vs. RRMS Related to Table 2, Figure S2 and Tables S5–S8. NMR serum metabolomics analysis in patients with SPMS (n = 29) and RRMS (n = 52). Metabolomic signatures associated with SPMS vs. RRMS determined using machine learning. (A) Confusion matrices showing number of correct (blue squares) and incorrect (green squares) classifications for each model. The sum (Σ) of each row and column is given. The models were: logistic regression (LR) with/without interactions (I) including/excluding age and EDSS (expanded disability status score), bagged and boosted LR, support vector machine (SVM), random forest (RF), neural network (NN) and sparse partial least square discriminant analysis (sPLS-DA). See Figures S2A and S2B. (B) Comparison of metabolites selected by each machine learning model (black squares). Metabolites selected by > 6 models highlighted in red bold; metabolites selected by ≥ 3 models in black bold. (C) sPLS-DA plot (sparse partial least squares-discriminant analysis) to validate metabolomic signature in (B) in metabolites identified by ≥ 3 models. See Figure S2D for features in components 1 and 2. (D) Area under the curve-Receiver operator characteristic (AUC-ROC) of top 14 metabolites and/or clinical features identified by ≥ 6 models and the best performing machine learning model, boosted LR. See Table S2 for metabolite abbreviations. (E and F) Metabolite features in (B) analyzed using the Youden Index to identify features that best stratify patients with SPMS vs. RRMS. Features with Youden J statistic >0.5 were analyzed using the Autoscore model. (E) Importance ranking of top variables (See Tables S7 and S8). (F) AUC-ROC plot showing performance of autoscore test set.

Figure 2

Metabolite set enrichment analysis (MSEA) on SPMS vs. RRMS metabolomic signature Related to Table S5. Metabolites identified as associated with SPMS by > 1 model (Table S5) were analyzed by MSEA. (A) Enrichment analysis using KEGG database. Pathways with a significant enrichment p < 0.05 are shown (–log10(p value)). Enrichment ratio (ER) is indicated along y axis. (B) Network of metabolic pathways derived from SMPDB database. Nodes are sized using the -log10(p value) and colored on a red to yellow gradient. The larger and redder the node, the greater the significance of the p value. (C–G) Bar charts showing the relative expression of metabolites (normalised values) in serum from patients with RRMS n = 52 (blue) vs. SPMS n = 29 (orange) (C) lactate and glutamine, representative metabolites of gluconeogenesis pathway; (D) acetoacetate and bOHbutyrate (ketogenesis pathway) (E) citrate (TCA); (F) alanine and pyruvate (glycolysis-related) and (G) total triglycerides, cholines and total fatty acids associated with lipid metabolism. T-tests were performed to identify statistically significant differences between patients with RRMS (blue) and patients with SPMS (red). p value, mean and +/− SD shown.

Figure 3

Metabolism of RNA, cellular responses to stress, immune effector response, cellular respiration, metabolism of lipids and tRNA processing pathways are dysregulated between patients with SPMS and RRMS Related to Figure S3 and Tables S9 and S10. RNA-sequencing was performed on n = 8 patients with SPMS and n = 5 patients with RRMS followed by differential gene expression and pathway enrichment analysis. Refer to Figures S3A–S3F. (A) Principal component analysis on all 1052 DEGs clustering patients with SPMS (orange) from patients with RRMS (blue). (B) Volcano plot showing differentially expressed genes (DEGs) Log2 fold change (>1.5 or < -1.5) and FDR-adjusted p value<0.05. Colored points represent significantly up- (red) and down- (blue) regulated genes in SPMS compared to RRMS. (C and D) Pathway enrichment analysis of DEGs analyzed by Metascape to identify regulated pathways. (C) Bar chart of top 20 significantly enriched pathways upregulated in patients with SPMS (Gene Ontology (GO), Reactome, Hallmark, Wikipathways). Pathways are ranked by p value. (D) Bar chart of top 14 significantly enriched pathways downregulated in patients with SPMS (Gene Ontology (GO), Reactome, Hallmark, Wikipathways). Pathways are ranked by p value. (E) Validation of gene expression pathways using an independent dataset. Gene lists were taken from. This was performed on multiple cell types including, myeloid cells, lymphocytes, whole blood (myeloid and lymphocytes combined) and oligodendrocyte precursor cells. For consistency only genes with a log2FC of 0.585 and FDR<0.05 was considered.

Figure 4

Gene-metabolite interaction network Related to Table S12. Gene-metabolite interaction network was constructed using the network analysis tool in MetaboAnalyst. This was performed by combining all 1052 DEGs and any metabolite that was identified by the machine learning models. The gene-metabolite interaction network illustrates the crosstalk between 27 upregulated (red circles) and 4 downregulated genes (blue circles) with 7 metabolites from the metabolomic signature (black squares). Genes are also colored based on the strength of their fold-change. Whilst phenylalanine was not identified as part of the metabolomic signature, it has been found to be lowered in MS, in previous publications. These genes and metabolites are associated with the metabolic pathways: aminoacyl-tRNA biosynthesis, glycolysis/gluconeogenesis, pyruvate metabolism and the TCA cycle.