Early Cancer Detection via Multi-microRNA Profiling of Urinary Exosomes Captured by Nanowires

Abstract

Multiple microRNAs encapsulated in extracellular vesicles (EVs) including exosomes, unique subtypes of EVs, differ in healthy and cancer groups of people, and they represent a warning sign for various cancer scenarios. Since all EVs in blood cannot be transferred from donor to recipient cells during a single blood circulation, kidney filtration could pass some untransferred EVs from blood to urine. Previously, we reported on the ability of zinc oxide nanowires to capture EVs based on surface charge and hydrogen bonding; these nanowires extracted massive numbers of microRNAs in urine, seeking cancer-related microRNAs through statistical analysis. Here, we report on the scalability of the nanowire performance capability to comprehensively capture EVs, including exosomes, in urine, extract microRNAs from the captured EVs in situ, and identify multiple microRNAs in the extracted microRNAs differing in noncancer and lung cancer subjects through machine learning-based analysis. The nanowire-based extraction allowed the presence of about 2500 species of urinary microRNAs to be confirmed, meaning that urine includes almost all human microRNA species. The machine learning-based analysis identified multiple microRNAs from the extracted microRNA species. The ensembles could classify cancer and noncancer subjects with the area under the receiver operating characteristic curve of 0.99, even though the former were staged early.

Figure 1

Nanowire-assisted extraction of exosomal miRNAs in urine. (A) Schematic illustrations for miRNA extraction from urine using a nanowire device, nanowire-extracted miRNA detection using microarrays with 2565 species probes, identification of microRNA ensembles based on fluorescence intensity analysis of each miRNA species using a logistic regression-modeled classifier, pathway analysis of identified miRNA ensembles, and lung cancer detection using identified miRNA ensembles. (B) Schematic illustrations for EV analysis in raw urine. (C) Schematic illustrations for EV analysis in flow-through urine from the device. (D) EV size distribution of urine samples after ultracentrifugation (UC). (E) EV size distribution of UC of urine flow-through from the device. (F) EV concentration of UC of urine and UC of urine flow-through from the device. (G) Membrane protein expression levels of UC of urine and UC of urine flow-through from the device. The line below the x-axis represents the captured antibody type, and the line above it indicates the corresponding fluorescent-labeled antibody. For instance, the CD81 line above the CD63 line signifies that EVs captured with the anti-CD63 antibody were detected using the fluorescently labeled anti-CD81 antibody. (D-G) Error bars show the standard deviation for a series of measurements, N = 5.

Figure 2

Comprehensive extraction of urinary miRNAs. (A) Box plot of extracted miRNA species using nanowire-based extraction from urine (pink, this work) and UC-based extraction from serum (orange, data from reference). Colored box lengths represent the interquartile range (first to third quartiles); the line in the center of each box represents the median value, and the bars show the data range (maximum to minimum). Error bars show the standard deviation for a series of measurements, N = 200 and N = 4046 for nanowire and UC methods, respectively. (B) Histogram of nanowire-extracted urinary miRNA species found in at least one sample (pink) and in all samples (cyan). (C) Pie chart of nanowire-extracted urinary miRNA species and p values showing statistically significant differences between lung cancer and noncancer subjects. (D) Scatter plot of fluorescence intensity of miRNAs with p values of less than 0.0005 extracted from lung cancer urine samples vs noncancer urine samples. Each point corresponds to a different miRNA species. The boundary between pink and cyan represents the same level of miRNA expression for the two samples. Error bars show the standard deviation for a series of measurements (N = 100). (E) Volcano plot highlighting significant miRNA species. Each point corresponds to a different miRNA species. The x- and y-axes represent the logarithm of the fluorescence ratio between lung cancer and noncancer subjects and the logarithm of p values showing statistically significant differences between lung cancer and noncancer subjects, respectively. The featured miRNAs have −log₁₀(p value) of more than 3.30 and log₂(fold change) of more than 1.00 or less than −1.00. (F) Comparison between fluorescence intensities of featured miRNAs (54 species) in lung cancer subjects extracted from samples of tissues and urine. Error bars show the standard deviation for a series of measurements (N = 8).

Figure 3

Identifying urinary miRNA ensembles. (A) Summary of analytical approaches used to classify and identify samples based on urinary miRNA ensembles. (B) Cancer risk scores and (C) AUROC curve for 100 lung cancer and 100 noncancer subjects obtained using one urinary miRNA ensemble. Since the classifier model is based on a logistic regression, the threshold for lung cancer risk is 0.5; among noncancer subjects the value is below 0.5; and for cancer subjects, it is more than or equal to 0.5. N/A represents stage as unknown. (D) Heat maps of logarithmic fluorescence intensity vs the one urinary miRNA ensemble used in the classification of Figures 3B and 3C for lung cancer (upper) and noncancer (lower) subjects. Red and blue indicate 100 and 0 overlapped data, respectively.

Figure 4

Pathway analysis and early staged cancer detection using identified miRNA ensembles. (A) Top 10 pathways with the largest number of total overlaps with miRNA targets from three groups (up- and down-regulated, nonsignificant) are shown in a bar plot, with the x-axis showing the extent of overlaps for each list and colors showing the significance in enrichment analysis. The p values were adjusted for multiple comparisons by using the Benjamini Hochberge procedure. (B) Summary of analytical approaches used to classify and identify stage I samples based on urinary miRNA ensembles. (C) Cancer risk scores and (D) AUROC curve for 24 stage I lung cancer and 25 noncancer subjects obtained using one urinary miRNA ensemble. This ensemble was identified using miRNA expression data from lung cancer subjects, excluding those of stage I, and 75 noncancer subjects.