Liquid Biopsy in Type 2 Diabetes Mellitus Management: Building Specific Biosignatures via Machine Learning

Abstract

Background: The need for minimally invasive biomarkers for the early diagnosis of type 2 diabetes (T2DM) prior to the clinical onset and monitoring of β-pancreatic cell loss is emerging. Here, we focused on studying circulating cell-free DNA (ccfDNA) as a liquid biopsy biomaterial for accurate diagnosis/monitoring of T2DM.

Methods: ccfDNA levels were directly quantified in sera from 96 T2DM patients and 71 healthy individuals via fluorometry, and then fragment DNA size profiling was performed by capillary electrophoresis. Following this, ccfDNA methylation levels of five β-cell-related genes were measured via qPCR. Data were analyzed by automated machine learning to build classifying predictive models.

Results: ccfDNA levels were found to be similar between groups but indicative of apoptosis in T2DM. INS (Insulin), IAPP (Islet Amyloid Polypeptide-Amylin), GCK (Glucokinase), and KCNJ11 (Potassium Inwardly Rectifying Channel Subfamily J member 11) levels differed significantly between groups. AutoML analysis delivered biosignatures including GCK, IAPP and KCNJ11 methylation, with the highest ever reported discriminating performance of T2DM from healthy individuals (AUC 0.927).

Conclusions: Our data unravel the value of ccfDNA as a minimally invasive biomaterial carrying important clinical information for T2DM. Upon prospective clinical evaluation, the built biosignature can be disruptive for T2DM clinical management.

Keywords: DNA methylation; circulating cell free DNA; machine learning; type 2 diabetes.

Figure 1

Workflow of our study. Abbreviations: TD2M: Type 2 diabetes mellitus, ccfDNA: circulating cell free DNA, qMSP: quantitative Methylation Specific PCR, AutoML: Automated Machine Learning.

Figure 2

(A) Box plot of serum ccfDNA levels between healthy individuals (n = 71) and T2DM patients (n = 96) (p = 0.552). (B) Box plot of INS unmethylated alleles between healthy individuals and T2DM patients (p = 0.001). (C) Box plot of IAPP unmethylated alleles between healthy individuals and T2DM patients (p < 0.001). (D). Box plot of GCK unmethylated alleles between healthy individuals and T2DM patients (p < 0.001). (E) Box plot of KCNJ11 unmethylated alleles between healthy individuals and T2DM patients (p < 0.001). (F) Box plot of ABCC8 unmethylated alleles between healthy individuals and T2DM patients (p = 0.534). Small circles (°) correspond to “outlier” values and stars (*) to the “extreme” values of the dataset. Abbreviations: T2DM: Type 2 diabetes mellitus, ccfDNA: circulating cell free DNA.

Figure 3

Representative capillary electropherograms showing DNA fragment size distribution in ccfDNA isolated from sera of two T2DM patients and one healthy control. (A) A T2DM patient ccfDNA sample showing a peak at ~160 bp indicative of apoptosis. (B) Another T2DM patient ccfDNA sample showing multiple peaks at ~160 bp, ~300 bp and ~500 bp indicative of apoptosis and an additional wide peak at ~2000–3000 bp indicative of active release. (C). A healthy volunteer ccfDNA sample with a wide peak at ~2000–3000 bp indicative of active release. Peaks at 35 bp and 10,380 bp in all electropherograms represent high and low ladders, respectively. (D) Distribution of ccfDNA fragment analysis in the patient’s group and the healthy volunteer’s group, respectively. Abbreviations: T2DM: Type 2 diabetes mellitus, ccfDNA: circulating cell free DNA, bp: base pairs.

Figure 4

ROC curve analysis results. (A) ROC curve of age reaching an AUC of 0.566 (95% CI 0.468–0.664). (B) ROC curve of BMI reaching an AUC of 0.658 (95% CI 0.573–0.743). (C) ROC curve of ccfDNA levels reaching an AUC of 0.527 (95% CI 0.438–0.616). (D) ROC curve of INS gene reaching an AUC of 0.650 (95% CI 0.562–0.737). (E) ROC curve of IAPP gene reaching an AUC of 0.727 (95% CI 0.649–0.805). (F) ROC curve of GCK gene reaching an AUC of 0.848 (95% CI 0.787–0.910). (G) ROC curve of KCNJ11 gene reaching an AUC of 0.712 (95% CI 0.619–0.806). (H) ROC curve of ABCC8 gene reaching an AUC of 0.528 (95% CI 0.439–0.617). Abbreviations: ROC curve: receiver operating characteristic curve, BMI: Body Mass Index, AUC: area under the curve, CI: confidence interval, ccfDNA: circulating cell free DNA.

Figure 5

Predictive modelling results. (A) ROC curve of whole dataset reaching an AUC of 0.927 (95% CI 0.874–0.967). (B) Supervised Principal Component Analysis (PCA) plot of whole dataset depicting discrimination between T2DM patients and healthy individuals. (C) Feature importance plot of the features of the best-performing model for the whole dataset. Feature importance is defined as the percentage drop in predictive performance when the feature is removed from the model. (D) ROC curve of train sub-dataset (blue line) reaching an AUC of 0.898 (95% CI 0.845–0.944) and test sub-dataset (green line) showing an AUC of 0.923 for the best-performing model. (E) PCA plot of test sub-dataset depicting discrimination between T2DM patients and healthy individuals. (F) Feature Importance plot of the features of the best-performing model in the train/test 70/30 split sub-datasets. Abbreviations: ROC curve: receiver operating characteristic curve, AUC: area under the curve, CI: confidence interval, T2DM: type 2 diabetes mellitus.