Circulating Cell-Free RNA in Blood as a Host Response Biomarker for the Detection of Tuberculosis

Abstract

Tuberculosis (TB) remains a leading cause of death from an infectious disease worldwide. This is partly due to a lack of tools to effectively screen and triage individuals with potential TB. Whole blood RNA signatures have been extensively studied as potential biomarkers for TB, but they have failed to meet the World Health Organization's (WHOs) target product profiles (TPPs) for a non-sputum triage or diagnostic test. In this study, we investigated the utility of plasma cell-free RNA (cfRNA) as a host response biomarker for TB. We used RNA profiling by sequencing to analyze plasma samples from 182 individuals with a cough lasting at least two weeks, who were seen at outpatient clinics in Uganda, Vietnam, and the Philippines. Of these individuals, 100 were diagnosed with microbiologically-confirmed TB. Our analysis of the plasma cfRNA transcriptome revealed 541 differentially abundant genes, the top 150 of which were used to train 15 machine learning models. The highest performing model led to a 9-gene signature that had a diagnostic accuracy of 89.1% (95% CI: 83.6-93.4%) and an area under the curve of 0.934 (95% CI: 0.8674-1) for microbiologically-confirmed TB. This 9-gene signature exceeds the optimal WHO TPPs for a TB triage test (sensitivity: 96.2% [95% CI: 80.9-100%], specificity: 89.7% [95% CI: 72.4-100%]) and was robust to differences in sample collection, geographic location, and HIV status. Overall, our results demonstrate the utility of plasma cfRNA for the detection of TB and suggest the potential for a point-of-care, gene expression-based assay to aid in early detection of TB.

Figure 1.. Plasma cell-free RNA profiling.

A) Geographic distribution of samples included in this study, which originated from two clinical studies. B) Differences in cfRNA cell-types-of-origin between the two cohorts. C) Scaled counts per million values of significantly differentially abundant genes after batch correction (DESeq2 and limma, Benjamini-Hochberg adjusted p-value < 0.01, |Log2FoldChange| > 0.5). Samples and genes are clustered based on correlation. D) Top 20 differential pathways between microbiologically confirmed TB diagnoses ranked by significance. Z-score indicated in corresponding bar. E) Flowchart of the method used to train and test machine learning classification algorithms. F) (Left) Area under the receiver operating characteristic curve (ROC-AUC) metrics for training and test sets. (Right) The test sensitivity and specificities for each model. The dotted lines indicate the optimal sensitivity (blue) and specificity (green) for a triage test.

Figure 2.. Performance of a 9-gene signature identified in plasma cfRNA.

A) Test area under the receiver operating characteristic curve (ROC-AUC) as a function of the gene added in each iteration of a greedy forward search model. B) Train and test performance of the greedy forward search algorithm in distinguishing microbiologically confirmed TB. C) Violin plot of classifier scores and confusion matrices of training and test sets using the greedy forward search algorithm. D) Abundance of the genes included in the 9-gene TB score. Outliers are indicated with arrows and values. E) Correlation of a 9-gene TB score with the Xpert Ultra Semi-quantitative Results (dotted line: classification score threshold=11.5; bars indicate 95% confidence interval). F) Correlation of a 9-gene TB score with three chest X-ray scores.

Figure 3.. Comparison of whole blood and plasma signatures of TB.

A) Overlap between top-performing whole blood signatures (blue) and the 9-gene signature (pink). Top 28 overlapping genes are shown. B) Performance comparison of whole blood signatures (blue) with the 9-gene signature (pink). Optimal triage thresholds are marked in blue, minimal triage thresholds are marked in green. C) Abundance of lung-specific markers non-programmed cell-death markers. D) Performance of GBP5 mRNA abundance in distinguishing active TB. E) Performance comparison of whole blood protein GBP5, whole blood RNA GBP5, and plasma cfRNA GBP5 in distinguishing active TB.