Label-free discrimination of extracellular vesicles from large lipoproteins

Abstract

Extracellular vesicles (EVs) are increasingly gaining interest as biomarkers and therapeutics. Accurate sizing and quantification of EVs remain problematic, given their nanometre size range and small scattering cross-sections. This is compounded by the fact that common EV isolation methods result in co-isolation of particles with comparable features. Especially in blood plasma, similarly-sized lipoproteins outnumber EVs to a great extent. Recently, interferometric nanoparticle tracking analysis (iNTA) was introduced as a particle analysis method that enables determining the size and refractive index of nanoparticles with high sensitivity and precision. In this work, we apply iNTA to differentiate between EVs and lipoproteins, and compare its performance to conventional nanoparticle tracking analysis (NTA). We show that iNTA can accurately quantify EVs in artificial EV-lipoprotein mixtures and in plasma-derived EV samples of varying complexity. Conventional NTA could not report on EV numbers, as it was not able to distinguish EVs from lipoproteins. iNTA has the potential to become a new standard for label-free EV characterization in suspension.

Keywords: concentration; extracellular vesicles; interferometric scattering; lipoproteins; plasma; refractive index; size.

FIGURE 1

iNTA is more sensitive than NTA for smaller‐sized EVs and LPs, and enables refractive index measurements. (a) EV, ULDL and VLDL size distributions as measured by NTA (green) and iNTA (purple). (b) Size‐RI plots for EV, ULDL and VLDL measured by iNTA. Colour bar indicates the local density of points. In the region to the right of the dashed gray line, the IQR of RI based on simulations is smaller than 0.05.

FIGURE 2

iNTA but not NTA can discriminate EVs and LPs in artificial mixtures. (a) iNTA and (b) NTA analysis of different EV‐VLDL mixtures showing mean particle concentration (filled bars indicate the measured concentration while unfilled bars indicate the expected concentration based on measurements of pure EV and VLDL samples, standard deviation is shown) and median size (box boundaries are 25–75 percentile, horizontal line is median, whiskers indicate 10–90 percentile) (n = 3). Concentration for iNTA measurements is calculated by multiplying the number of trajectories measured in 10 min by a calibration factor determined previously (see Methods). (c) Representative size‐RI plots for the different EV‐VLDL mixtures, labelled using a random forest classifier. Colour indicates probability for the particle to be an EV. Only particles with greater than 80% probability of being an EV or less than 20% probability of being an LP are included in the 1D histograms. (d) Left: Measured relative EV concentration plotted versus expected relative EV concentration. The measured concentration is calculated as C_EV/(C_EV+C_VLDL), where C denotes particle concentration determined from plots in (c) using the random forest classifier. The numbers indicate the measured relative EV concentration when no EVs were expected (∼4%) and when only EVs were expected (∼93%). Right: The extracted diameter and refractive index of EVs (red) and lipoproteins (blue). RI was calculated by considering only particles with an IQR_S < 0.05. (e) Same as (d), but using a mixture of EVs with ULDLs, VLDLs and LDLs. Expected EV concentration was determined based on iNTA measurements of the pure EV sample and of the LP mixture.

FIGURE 3

iNTA can accurately measure EVs in plasma‐derived samples with variable LP background. (a) ApoB ELISA measurements in DMC‐ and SEC‐processed samples (left and right), pre‐ and post‐prandial (green and orange). (b) Total particle concentrations measured by iNTA. (c) Representative size‐RI plots after machine learning processing (particles classified as EVs are indicated in red and LPs in blue). The numbers in the upper right corner indicate sample dilution factor by PBS prior to iNTA measurements. (d) EV concentration (top) and relative EV concentration (bottom) as determined by iNTA. Dashed line indicates the error threshold for EV classification (4%). (e)–(g) Same as (b)–(d) but for samples spiked with 5E11/mL SKMEL37 EVs. In (g) the dashed bars show the expected EV concentration based on values in (d). Mean and standard deviation are shown for all bar graphs (n = 2).

FIGURE 4

iNTA of EV samples enriched from lipemic melanoma patient plasma by SEC, DMC or DG. (a) Western blot of EV‐associated tetraspanins. Equal volumes of EV sample were loaded, with DG (6 mL) having a 12× higher plasma sample input than SEC and DMC (0.5 mL). Size‐RI plots of EVs enriched by (b) SEC; (c) DMC, arrowheads indicate the EV population; (d) DG. Numbers in the upper right corner indicate sample dilution factor by PBS prior to iNTA measurements.