Biomarkers from circulating neutrophil transcriptomes have potential to detect unruptured intracranial aneurysms

Abstract

Background: Intracranial aneurysms (IAs) are dangerous because of their potential to rupture and cause deadly subarachnoid hemorrhages. Previously, we found significant RNA expression differences in circulating neutrophils between patients with unruptured IAs and aneurysm-free controls. Searching for circulating biomarkers for unruptured IAs, we tested the feasibility of developing classification algorithms that use neutrophil RNA expression levels from blood samples to predict the presence of an IA.

Methods: Neutrophil RNA extracted from blood samples from 40 patients (20 with angiography-confirmed unruptured IA, 20 angiography-confirmed IA-free controls) was subjected to next-generation RNA sequencing to obtain neutrophil transcriptomes. In a randomly-selected training cohort of 30 of the 40 samples (15 with IA, 15 controls), we performed differential expression analysis. Significantly differentially expressed transcripts (false discovery rate < 0.05, fold change ≥ 1.5) were used to construct prediction models for IA using four well-known supervised machine-learning approaches (diagonal linear discriminant analysis, cosine nearest neighbors, nearest shrunken centroids, and support vector machines). These models were tested in a testing cohort of the remaining 10 neutrophil samples from the 40 patients (5 with IA, 5 controls), and model performance was assessed by receiver-operating-characteristic (ROC) curves. Real-time quantitative polymerase chain reaction (PCR) was used to corroborate expression differences of a subset of model transcripts in neutrophil samples from a new, separate validation cohort of 10 patients (5 with IA, 5 controls).

Results: The training cohort yielded 26 highly significantly differentially expressed neutrophil transcripts. Models using these transcripts identified IA patients in the testing cohort with accuracy ranging from 0.60 to 0.90. The best performing model was the diagonal linear discriminant analysis classifier (area under the ROC curve = 0.80 and accuracy = 0.90). Six of seven differentially expressed genes we tested were confirmed by quantitative PCR using isolated neutrophils from the separate validation cohort.

Conclusions: Our findings demonstrate the potential of machine-learning methods to classify IA cases and create predictive models for unruptured IAs using circulating neutrophil transcriptome data. Future studies are needed to replicate these findings in larger cohorts.

Keywords: Inflammation; Intracranial aneurysm; Machine learning; Neutrophils; Transcriptomics.

Fig. 1

Neutrophil RNA expression differences between patients with intracranial aneurysms (IA) and IA-free controls, feature selection for classification model creation, and model training. a Transcriptome profiling demonstrated 95 differently expressed transcripts (q-value < 0.05) between patients with IA and controls. Of these, 26 had a false discovery rate (FDR) < 0.05 and an absolute fold change ≥ 1.5 (in red). b Principal component analysis (PCA) using these 26 transcripts demonstrated general separation between samples from patients with IA (60%, circled in red) and those from controls (80%, circled in blue). c Estimation of model performance during leave-one-out (LOO) cross-validation in the training cohort demonstrated that most models performed with an accuracy of 0.50–0.73. Among the classification models, diagonal linear discriminant analysis (DLDA) had the highest combination of sensitivity, specificity, and accuracy (0.67, 0.80, 0.73 respectively). d Receiver operating characteristic (ROC) analysis using classifications in the training dataset showed that the models had areas under the curve of 0.54 (support vector machines [SVM]) to 0.73 (DLDA). (F-C: fold-change; ABS: absolute value; Cosine NN: cosine nearest neighbors; NSC: nearest shrunken centroids)

Fig. 2

Performance of the four classification models during model testing. a PCA using the 26 transcripts showed general separation between patients with IA (100%, circled in red) and controls (80%, circled in blue). b Validation of the classification models in an independent testing cohort of patients demonstrated that DLDA had the best performance, with sensitivity, specificity, and accuracy of 0.80, 1.0, and 0.90, respectively. c ROC analysis in the testing cohort also showed that DLDA had the best area under the curve (AUC) (0.80)

Fig. 3

Assessment of model performance by LOO cross-validation of all data, and positive predictive value (PPV)/negative predictive value (NPV). a Estimation of model performance showed that the models performed with an accuracy of 0.63–0.80. DLDA had the highest combination of sensitivity, specificity, and accuracy (0.65, 0.95, 0.80, respectively). b ROC analysis demonstrated that the models had AUC of 0.68 (NSC) to 0.84 (DLDA). c Plot showing the PPV of all models across all possible prevalence. The blue region in the figure represents the range of IA prevalence reported in the current literature. The best performing model (DLDA) had the highest PPV, and cosine NN demonstrated the poorest PPV. d The DLDA model also had the best NPV, but only slightly better than that of the cosine NN, NSC, and SVM models

Fig. 4

Validation of RNA-Sequencing data for seven transcripts by quantitative polymerase chain reaction (qPCR). Six of seven differentially expressed transcripts in samples from patients with and without IA were also differentially expressed in neutrophils in the qPCR in an independent cohort. This demonstrates consistent expression differences between patients with IA and controls in ~ 86% (6/7) of the tested transcripts

Fig. 5

Comparison of fold-change in expression in patients with “small” (< 5 mm) IAs vs. control and patients with “large” (≥ 5 mm) IAs vs. control. The plot shows the fold-change (F-C) in expression of the 26 classifier transcripts identified in the training cohort (n = 30—black line) compared to those for “small” IAs (vs. control—green) and “large” IAs (vs. control—orange). Expression changes were more pronounced in both the positive and negative direction in patients with larger IAs. Fold-changes across all 26 transcripts in the “large” group were on average 24% higher than those for the training cohort, while fold-changes for the “small” group F-C were on average 35% lower