Multi-block data integration analysis for identifying and validating targeted N-glycans as biomarkers for type II diabetes mellitus

Abstract

Plasma N-glycan profiles have been shown to be defective in type II diabetes Mellitus (T2DM) and holds a promise to discovering biomarkers. The study comprised 232 T2DM patients and 219 healthy individuals. N-glycans were analysed by high-performance liquid chromatography. The multivariate integrative framework, DIABLO was employed for the statistical analysis. N-glycan groups (GPs 34, 32, 26, 31, 36 and 30) were significantly expressed in T2DM in component 1 and GPs 38 and 20 were related to T2DM in component 2. Four clusters were observed based on the correlation of the expressive signatures of the 39 N-glycans across T2DM and controls. Cluster A, B, C and D had 16, 16, 4 and 3 N-glycans respectively, of which 11, 8, 1 and 1 were found to express differently between controls and T2DM in a univariate analysis [Formula: see text]. Multi-block analysis revealed that trigalactosylated (G3), triantennary (TRIA), high branching (HB) and trisialylated (S3) expressed significantly highly in T2DM than healthy controls. A bipartite relevance network revealed that HB, monogalactosylated (G1) and G3 were central in the network and observed more connections, highlighting their importance in discriminating between T2DM and healthy controls. Investigation of these N-glycans can enhance the understanding of T2DM.

Figure 1

Flow chart of N-glycan data processing. Participants with no prior history of T2DM were recruited from the Kumasi metropolis. Ethics was approved and each participant was asked to complete a questionnaire. After this, demographic and anthropometric data were obtained, and fasting blood samples were collected for biochemical and N-glycan analysis. Statistical analyses were performed in SPSS and R.

Figure 2

Workflow of N-glycan analysis with UPLC-FLR. Plasma samples were aliquoted into 96 well plates and denatured with sodium dodecyl sulphate (SDS). The plate was sealed and incubated at 65 °C for 10 min. IGEPAL CA-630 was added and sample mixed by pipetting up and down. This was then followed by incubation at room temperature. Glycans were freed from their bound glycoproteins by adding peptide N-glycosidase F (PN-Gase F) and incubation at 37 °C for 18 h. glycans were then fluorescently labelled with 2-aminobenzamide and incubated for 2 h at 65 °C. This was followed by four-step washing procedure with acetonitrile and 2AB glycans were eluted using ultra-pure water. Samples were injected into the UPLC and analysed under the following conditions: solvent A = 100 Mm ammonium formate, solvent B = acetonitrile, flow rate 0.1 ml/min, pH = 4.4. Structural assignments and normalisation of glycan peaks were then performed.

Figure 3

(A) Glycans correlation analysis for healthy controls and T2DM cases. The matrix presented are hierarchically clustered to highlight the signature of glycans expression in healthy controls and T2DM cases. (B) Expression of glycans in healthy controls and T2DM, ranked in terms of significant differing expression.

Figure 4

Principal component and discrimination analysis of top expressive glycans in T2DM and healthy controls. Feature selections are important in the refinement of biological and biochemical hypotheses. We identified a combination of discriminative features from a disparate block of N-glycan data set. N-glycan peaks loaded differently along the two principal components (PC), with estimates of positive and negative weights. A large absolute value indicates the importance of the variable to the PC and the colour codes indicate how prominent the biomarker expressed in T2DM and healthy controls. Selected variables were ranked from the least important to the most important. The classification accuracy the training and testing of the discriminant function formulated using the 10 glycans associated with component 1 reveals a model with good learning rate.

Figure 5

Hierarchical clustering of derived N-glycans traits in cases and controls. Hierarchical clustering of the cases and control samples using the measurement of sugars and lipids from block sPLSDA-reg network. Agglomerative hierarchical clustering was derived using the Euclidean distance as the similarity measure and Ward methodology. The red colour indicates that the row-column clusters are positively correlated, and the light blue colour indicates a negative correlation, whereas yellow indicate weaker correlation values.

Figure 6

Correlation and relevance N-integrative supervised analysis with DIABLO. (A) Circos plots showing inter-block correlations. Spearman rank correlations were calculated for each pairwise comparison of variables. Variables with r = 0.5 between-block correlation were presented, (B) relevance network visualisation of the selected features.