Techniques to improve detection and analysis of extracellular vesicles using flow cytometry

Study Subjects

Study samples were obtained from subjects in the Age of Blood Evaluation (ABLE) trial 65. Intensive care unit (ICU) patients were randomly assigned to receive either fresh (< 7 days' storage) or standard (expected mean ≈ 21 days' storage) blood for transfusion. Whole blood was collected from these patients on Day 0 (before transfusion) and on Days 2, 6, and 28 post-transfusion. Some optimization steps were performed using samples from discarded Trima leukoreduction system chambers (LRSCs) from Blood Centers of the Pacific or from whole blood collected from six healthy volunteers in citrate tubes. All human subject samples were tested under an institutional review board (IRB) approved protocol and with informed consent of the subjects.

EV and Cell Processing

EVs were isolated from whole blood using a common differential centrifugation technique described in the literature 59, 66-68. Immediately after collection, tubes were centrifuged at 1,500g for 10 min to separate cells from the supernatant, then at 13,000g for 10 min to remove platelets. The supernatant was carefully removed, and this platelet-poor plasma (PPP) was used to study EV concentration and phenotype. PPP from six normal donors was combined to create a normal donor pool. Aliquots of 0.5 mL were stored at −80°C (refer to Fig. 1A for overview). Some optimization steps used EVs which had been concentrated using an additional centrifugation or filtration step. In these instances, the EVs are referred to as “concentrated” in figure legends. When centrifugation was used to concentrate EVs, 3 mL of PPP was added to 32 mL phosphate buffered saline (PBS, UCSF Cell Culture Facility, San Francisco, CA) and spun for 60 min at 100,000g. EV pellets were re-suspended in 1 mL RPMI 1640 (Invitrogen, Carlsbad, CA) and stored at −80°C. When filtration was used to concentrate EVs, 1.5 mL of PPP was added to centrifugal filters and resuspended in 400 µL PBS for immediate analysis (see below for full filtration method protocol). Peripheral blood mononuclear cells (PBMCs) were isolated from Trima LRSCs on a Ficoll-Paque PLUS density gradient (GE Healthcare Bio-Sciences, Piscataway, NJ). Aliquots of 20 × 10⁶ cells were frozen in media that contained 90% fetal bovine serum (HyClone, Logan, UT) and 10% dimethyl sulfoxide (Fisher Bio-Reagents, Pittsburgh, PA) and stored in liquid nitrogen vapor.

Antibodies

In order to examine EVs for cell of origin markers (phenotype) and immune or coagulation activation markers, we used several different fluorochrome-conjugated monoclonal antibodies, including red blood cell markers: CD108-PE and CD235a-FITC, a platelet marker: CD41a-PerCP/Cy5.5, and leukocyte markers: CD3-PerCP/Cy5.5, CD19-Alexa700, CD28-FITC, CD16-V450, CD152-APC, CD14-APC/Cy7, and CD62 L-APC. All isotype controls were matched to their respective antibodies according to their fluorochrome type, concentration, heavy chain (IgA, IgG, IgD, IgE, or IgM), subclass, and light chain class (kappa, lambda). Refer to Table 1 for a detailed summary of the antibodies that were used in the experiments described in this article.

Antibody Labeling

PPP samples were rapidly thawed and 50 or 100 µL were added to 2–5 µL of each monoclonal antibody. Prior to testing EV samples, each antibody was titrated using serial dilutions to determine the “saturating” concentration (the lowest concentration which yielded nearly maximal fluorescence). Samples were incubated at 4°C for 30 min and either filtered or re-suspended in 400 µL PBS for immediate FCM analysis.

Absolute Count Analysis

Trucount^TM tubes (BD Biosciences, San Diego, CA) with a known number of fluorescent beads were utilized for EV quantification. To each Trucount tube, 50 µL sample and 350 µL PBS were added and samples were read immediately on the flow cytometer. EV concentrations were calculated using the following equation:

(1)

Data Collection

Acquisition was performed using a 3-laser (20 mW Coherent Sapphire 488 nm blue, 25 mW Coherent Vioflame 405 nm violet, and 17 mW JDS Uniphase HeNe 633 nm red) LSR II benchtop flow cytometer equipped with FACS Diva 6.0 software (BD Biosciences). Specific cytometer parameters and filter configurations used for acquisition are outlined in Table 2. Flow cytometer setup was performed using CS&T instrument setup beads (BD Biosciences). Forward scatter (FSC) and side scatter (SSC) parameters were set to log mode and the lowest threshold allowed by the cytometer (200) was selected for each. Compensation setup was performed using AbC beads (Invitrogen) and compensation values were determined by FACS Diva software. FSC/SSC voltages were set to the highest values that excluded the majority of background noise (i.e., just below the voltage threshold at which event rate surpassed 5 events/sec while running a tube of PBS alone). Typically, this threshold occurred at FSC and SSC voltages of around 500–600 and 300–350, respectively. Rainbow fluorescent particles (Spherotech, Lake Forest, IL) were used to adjust all channel voltages between batches in order to maintain voltage consistency from run to run. Figure 1B shows the location of the EV gate in relation to 0.2, 0.24, 0.5, and 1 µm beads (0.5 mL of Megamix-Plus SSC; BioCytex, Marseille, France) beads combined with 1–2 drops of Spherotech Ultra Rainbow Fluorescent Particles). These beads cannot be used to determine EV size but are useful for showing the relative sizes of EVs detected. All samples were acquired at low sample pressure and low flow rate (∼8–12 µL/min). In pre-optimization experiments (Fig. 2 and Supporting Information Figs. 1 and 2), collection of 100,000 events was attempted for each sample, and in situations with very few events, tubes were run for at least 3 min. Post-optimization, each sample (including lysed samples) was run for exactly 1 or 2 min to allow for the subtraction of false positive events detected in the lysed sample over an equal time frame. FCS files were exported and data were evaluated using FlowJo software (TreeStar, Ashland, OR; Mac version 9.6.1 or PC version 7.6.5).

Transmission Electron Microscopy

TEM was used to visualize antibodies using negative staining. Approximately 10 µL of each sample was added to a Formvar-coated 300-mesh copper grid and allowed to adhere for 2 min at room temperature. Excess liquid was removed by blotting the edge of the grid with filter article. Next, a drop of 2% aqueous uranyl acetate solution was applied to the grid for 30 sec. The excess stain was removed as before and the specimens were examined by TEM using a JEOL JEM-1400 electron microscope.

Lysis Step

In order to discriminate between EVs and protein aggregates, we used a lysis technique similar to that described by György et al. 55 to reveal false-positive events. A non-ionic detergent, 10% Nonidet P-40 (NP-40) (New England Biolabs, Ipswich, MA) was used to lyse EVs. After an initial reading on the flow cytometer, stained EV samples were added to 20 µL of 10% NP-40 (final concentration 0.5% NP-40). Samples were then re-read and compared to the initial reading. Antibody-positive events remaining in the lysed sample were subtracted from the positive events in the initial reading to determine the proportion of true EVs.

Filtration

Antibody filtration was performed prior to staining using Ultrafree^®-MC/Durapore^®-PVDF centrifugal filter units of various pore sizes (0.1, 0.22, 0.45, and 0.65 µm, Millipore, Bedford, MA). For each panel, titrated antibodies (2–5 µL each) were combined and added to the top of a filter, the tube was centrifuged in a fixed-angle single speed microcentrifuge (Fisher Scientific, Pittsburgh, PA) for ∼ 30 sec, and the filtrate was used for staining. Post-stain filtration was performed using the same filters and same filtration process, however, the filtrate was discarded and EVs remaining on the filter surface were resuspended in 400 µL PBS and saved for flow analysis. Originally, post-stain filtration was performed at 600g for 30 sec. After testing a variety of PPP samples, it was ultimately decided that an increase to 800g for 2–5 min would be necessary to accommodate all PPP samples, some of which required a slightly higher force to move through the filter effectively.

Statistical Analysis

Nonlinear regression analysis using a semi-log line was used to assess the correlation between EV concentration and percentages of antibody-positive events. The two-way ANOVA test followed by Tukey's multiple comparisons test was used to evaluate differences in the percentages of CD14⁺ events after different sized filters were used to remove antibody aggregates in EVs and PBMCs. For EV dilution experiments, slopes were determined by nonlinear regression analysis using a log-log line, and R-squared values were determined by nonlinear regression analysis using a log–log line and slope constraint equal to −1.0 in order to assess goodness-of-fit. The t test corrected for multiple comparisons using the Holm-Sidak method was used for comparing the numbers of antibody-positive events remaining after filtration or centrifugation was used to remove antibody aggregates. The two-tailed t test for paired comparisons was used to compare the mean percentages of antibody-positive events before and after post-stain filtration. Statistical significance was defined as P < 0.05. All statistical analyses were performed using Prism 5 (GraphPad Software, La Jolla, CA).

Antibody Aggregates Fall Within the EV FSC/SSC Gate

PPP was isolated from clinical study or normal donor patient samples using a differential centrifugation process commonly found in the literature 59, 66-68 (Fig. 1A). Some of the larger EVs may have been removed with the relatively long duration centrifugation that we used. The general techniques we describe in this article, however, are applicable to all FCM analysis of EVs and not specific to EVs derived from PPP. Samples were then stained with one or more antibody-conjugated fluorochromes. When ABLE study samples were tested, which varied widely in EV concentration (EV/µL), a negative correlation between EV concentration and the proportion positive for cellular markers was found. The samples taking the longest to acquire 100,000 events on the cytometer, i.e., samples having the lowest EV concentrations, had the highest background and thus the highest number of positive events. Supporting Information Figure 1A shows the results of EV analysis of two representative donors, one with a low concentration of EVs (top row) and one with a high concentration (bottom row). Plotting results from all donors revealed a significant negative correlation between EV concentration and positive events for four markers (Supporting Information Fig. 1B). To test whether the positive events could be due to artifact, antibody alone in PBS (without EVs added) was tested, and the same pattern of positive events was observed as in the samples having low EV concentration (Supporting Information Fig. 1C). Since the antibody without EV sample showed more signal than the antibody with EV, this suggested that the signal was artifactual.

Eliminating Antibody Aggregates

To eliminate false positive events, the efficacy of filters to remove presumed antibody aggregates from the EV gate was tested. We experimented using several different sized filters (0.1, 0.22, 0.45, and 0.65 µm) and found that all filters were highly effective at removing aggregates without compromising the antibody's ability to stain PBMCs (Fig. 2). Next, to determine the most suitable method for removing antibody aggregates, we compared our filtration method against a common centrifugation method found in the literature (17,000g for 5 min) 69. Filtration was more effective than centrifugation at removing aggregates from all antibodies tested, and this was confirmed by electron microscopy (Supporting Information Fig. 2). Of note, longer centrifugation times than we used in the current experiments have been described 69, but we only tested a 5-min centrifugation in the current work in an attempt to develop a protocol suited to high-throughput sample analysis. Similar 5-min centrifugation protocols to remove antibody aggregates were recommended by other researchers as well, including 16,000g for 5 min 70 and 18,000g for 5 min 63. Antibodies filtered with smaller pore sizes were equally as effective as those filtered with larger pore sizes. We did not experience problems with clogging of the filters using any pore size during antibody filtration. One drawback to using filters for antibody aggregate removal is the cost; the retail price of each filter is ∼$2. However, because all antibodies can be combined in one tube before filtering, if 40 samples were run per batch, the cost would be 5 cents per sample. In batched analyses, the filter for removing antibody aggregates comprises a small fraction of the total cost of testing each sample.

EV Gating Strategies

The final step in successfully analyzing FCM data is setting gates to separate positive from negative events. Many researchers use isotypes to do this 60, 71, 72, though a lysis method for identifying false positive EV events has been described 55, 56. The lysis step utilizes a detergent to disrupt EVs, with immune complexes and other non-EV related events remaining after detergent lysis eliminates the EVs. In samples with paired lysed and unlysed samples, the lack of a true EV population expressing CD16 can be appreciated (Supporting Information Fig. 3A). After optimizing the use of a lysis step, lysed samples were compared to isotype stained samples for setting background fluorescence gates (Supporting Information Fig. 3B). In these examples the donor's lysed sample was superior to the isotype control for defining background fluorescence for the CD235a and CD41a antibodies because the background more closely matched that of the fully stained Ab sample. Similarly, Figure 3 shows the ability of three different negative controls (FMO, isotypes, and lysed) to accurately predict the background fluorescence of a fully stained, non-post-stain-filtered sample. FMO controls provided an appropriate indication of background fluorescence for CD41a and CD62L but not for CD235a stains. Isotype controls accurately portrayed background fluorescence in CD235a and CD41a but failed to do so for CD62L. Background fluorescence of the lysed sample, on the other hand, matched all three markers in the fully stained sample. After analyzing a large panel of samples in a similar fashion to Figure 3, lysed controls worked as well as isotype controls for setting background fluorescence gates 72% of the time (83/116) and better than isotype controls in 28% of cases examined (33/116). These results demonstrate that the selection of an appropriate control sample to use for gating can vary depending on the antibody and marker measured.

Currently, many researchers using FCM to analyze EVs use isotype controls as a means for setting positive vs. negative discrimination gates 55, 56, 60, 71, 72. However, our attempts to replicate the background of an antibody with its isotype proved to be difficult, as it was impossible to know whether the signal was true background or simply an artifact caused by the differences in spectral properties between the two stains. Indeed, a number of publications have noted similar issues associated with using isotype controls for this purpose 58, 59. Isotype gates can vary widely depending on a number of factors including: antibody supplier, fluorochrome:protein ratio, antibody concentration, propensity for aggregation, and antibody subclass 7, 48, 58. Though these variables can be accounted for/controlled to a certain degree, it is difficult if not impossible to match perfectly the background fluorescence of a fluorochrome-conjugated antibody to that of its isotype. Considering the low number of antigens per EV (and correspondingly small fluorescence signal emitted), even minute differences in background signal between an antibody and its isotype will significantly affect proper gate placement and the ability to accurately distinguish positive events. Of all of the controls we tested (FMO, unstained, isotype, and lysed), lysed samples proved to be the most consistently reliable as an indicator of background fluorescence across all markers when samples were left unwashed after staining. Other researchers have reported the use of antigen-negative EVs as negative controls for setting background fluorescence gates 58. However, because background fluorescence can often vary from individual to individual, it may be inappropriate to apply the gates created from a single sample of entirely different origin to all clinical samples in the study. Furthermore, this method requires the use of twice the amount of antibodies as well as a working bank of EVs known to be negative for all antibodies in question, which poses logistical difficulties.

The use of a lysis method is an established practice in EV analysis for identifying false positive events 55; however its utility as a replacement for isotypes in determining background fluorescence has not yet been described. In our research, we found that the lysis technique for gate placement worked as well or better than isotypes across all antibodies tested. It should be noted, however, that while using lysed samples for predicting background worked for our purposes, it may not necessarily be the best solution in all situations. One limitation of the lysis method is its inability to identify nonspecific binding of antibodies to EVs or other lipid vesicles such as chylomicrons. Furthermore, because lysis is not EV-specific, some non-EVs may become lysed and some small EVs may be resistant to lysis. For some, using a combination of the lysed sample and its isotype might work better than either alone for providing the best indication of background fluorescence for gate placement.

Washing After EV Staining

Removing unbound antibody after staining EVs often requires the use of lengthy, multi-step washing procedures such as ultracentrifugation or sucrose fractionation. In an attempt to develop a protocol suited to high-throughput sample analysis, we developed a technique using filters to wash EVs post-staining. After staining with pre-filtered antibody panels, EVs were originally added atop 0.2 μm centrifugal filters with 300 µL PBS and centrifuged at low speed (∼600g) for 10–30 sec. After testing a variety of PPP samples it was found that some required centrifugation at 800g for 2–5 min, which was the final protocol we adopted. The tops of the filters were resuspended in 400 µL PBS and analyzed using FCM. Figure 4 shows the results using antibodies for which resolution was improved after implementing post-stain filtration to remove unbound antibody. Similar results were obtained with each antibody tested (data not shown). Assay reproducibility was tested across seven different markers in three experiments performed in triplicate on PPP from a single normal donor. Coefficients of variation (CVs) for non-post-stain-filtered samples were 38.7, 15.6, 11.6, 28.9, 40.1, 12.8, and 2.4 for the markers CD14, CD16, CD19, CD152, CD235a, CD108, and CD41a, respectively. CVs for post-stain-filtered samples were 10.9, 40.7, 4.9, 11.3, 19.6, 16.4, and 7.5 for the same markers, which was not significantly different from the samples that were not post-stain filtered (P = 0.45).

Of the washing techniques currently used to remove unbound antibody post-staining, all are time-consuming, multi-step procedures not suitable for high-throughput analysis. The most common methods, ultracentrifugation and sucrose fractionation, require long processing times and expensive equipment. Density gradient techniques described in the literature require 14- to 20-h centrifugations at forces of up to 192,000g 3, 70. Here, we report the use of a novel filtration technique for reducing background fluorescence that is simple, fast, and effective. It should be noted that a significant limitation of this method is the loss of particles such as exosomes and small EVs that are small enough to pass through the 220-nm filter. This is an important consideration, as increasing evidence suggests that these very small EVs comprise the active/functional fraction of EVs as a whole, at least in some settings 68. One solution to this limitation would be to recover the filtrate, couple the smaller EVs to beads to allow washing, and analyze bead-bound small EVs; however, this would limit one's ability to measure co-expression of multiple antigens on single EVs. While the increase in signal-to-noise ratio is of obvious benefit, the loss of smaller EVs represents a significant limitation when considering any washing method 63.

Effect of EV Concentration on Assay Sensitivity

It has been described that the flow cytometer can detect multiple small vesicles simultaneously illuminated by the cytometer's laser beam, counting them as a single event 50. The phenomenon of coincidence detection would presumably be more pronounced in samples with higher concentrations of EVs, which could affect the number and type of events detected. Figure 5 shows the results of six different dilutions of PPP on EV detection using FCM. PPP from five healthy donors was stained for CD41a, then post-stain filtered and resuspended in PBS. The samples were then serially diluted, and each was read for 60 sec on a flow cytometer. While the percentage of positive events was fairly unpredictable at very low EV concentrations, the number of positive events detected within a fixed time frame decreased proportionally with dilution factor, yielding approximately the same calculated number of CD41a+ events at each dilution. CVs of CD41a+ calculated counts were 5.4, 9.8, 9.6, 16.3, and 8.4 for the five donors across all the dilutions, comparable to the CV of 7.5 for replicates of one sample tested undiluted presented in Figure 4. These data show that across the concentration range of ∼100–100,000 positive events collected, the calculated count of EVs positive for CD41a did not vary substantially, implying that co-incident particle detection did not play a significant role in detecting particles >200 nm at the concentrations tested.

Because coincidence detection is dependent on the number of surrounding EVs in a given sample, we thought it might be necessary to normalize the EV concentration of each sample prior to analyzing with FCM. However, we found no evidence to support coincident particle counting in the dilution range we tested. In our dilution experiments we found that the percentages of positive events were equal and event counts were proportional to dilution factor between 1 × 10⁵ and 1 × 10⁶ EV/µL. With each 10-fold dilution, the number of positive events detected predictably dropped by a factor of 10. At higher dilutions, however, the percentages of events staining positive for a given marker became much less reliable due to a constant number of artifactual events in the denominator, while event counts staining positive for a given marker remained consistent with dilution factor. Whereas percentage varied considerably at higher dilutions (due to the noise making up a larger proportion of the denominator at higher dilutions), the number of positive events was consistent and reliable. With the analysis of these dilution experiments, we gained a new understanding of our results and learned that the best parameter to record was the number of positive events in a set time period, not percentage of positive EV events. With this finding, we changed our recording methods to place new emphasis on number of positive events collected within a fixed time frame, rather than percentage of the total EV population collected being positive for the marker in question. Because of the influence of background noise at low EV concentrations, we concluded that counting the number of positive events within a fixed time frame yielded more reproducible results than using the percentage of the total EV population being positive for the marker in question. The CVs in these dilution experiments (between 5.4 and 16.3, mean = 9.9) were similar to the CV we observed in our reproducibility experiments (7.5), thus concentration is unlikely to play a huge role in our sample to sample variability.