Immuno-analysis of microparticles: probing at the limits of detection

Microparticle (MP) research is clouded by debate regarding the accuracy and validity of flow cytometry (FCM) as an analytical methodology, as it is influenced by many variables including the pre-analytical conditions, instruments physical capabilities and detection parameters. This study utilises a simplistic in vitro system for generating MP, and through comparative analysis with immuno-electron microscopy (Immuno-EM) assesses the strengths and limitations of probe selection and high-sensitivity FCM. Of the markers examined, MP were most specifically labelled with phosphatidylserine ligands, annexin V and lactadherin, although only ~60% MP are PS positive. Whilst these two ligands detect comparable absolute MP numbers, they interact with the same population in distinct manners; annexin V binding is enhanced on TNF induced MP. CD105 and CD54 expression were, as expected, consistent and enhanced following TNF activation respectively. Their labelling however accounted for as few as 30–40% of MP. The greatest discrepancies between FCM and I-EM were observed in the population solely labelled for the surface antigen. These findings demonstrate that despite significant improvements in resolution, high-sensitivity FCM remains limited in detecting small-size MP expressing low antigen levels. This study highlights factors to consider when selecting endothelial MP probes, as well as interpreting and representing data.


Results
FCM characterisation of endothelial MP. MP in resting and TNF-stimulated conditions were double labelled for PS (annexin V or lactadherin) and a surface Ag (CD105 or CD54). MP numbers positive for either or both markers were recorded and are represented in Table 1 along with the calculated percentages of positive MP for the various markers. TNF stimulation significantly increased the numbers of MP detected by the PS markers whether they appeared single or double stained (Fig. 1a). In all combinations, both lactadherin and annexin V detected similar numbers of MP as shown in Fig. 1a and Table 1. Moreover, a highly significant correlation between these numbers was observed (Fig. 1b). A small minority of the CD54 + MP were labelled with CD54 only, both in resting and stimulated conditions (~4% for all combinations), while in resting conditions an average of 95% and 96% of CD54 + MP were double labelled with annexin V and lactadherin, respectively. However, MP positive for PS only represented 62% and 59% of the total MP detected with annexin V and lactadherin, respectively. These percentages, in stimulated conditions, were 51.5% and 43.7% for annexin V and lactadherin, respectively. CD105 gave a similar labelling profile with a large majority (88-92% depending on the Annexin V is shown on the left hand side of each graph vs lactadherin on the right, CD54 is on top and CD105 below (mean ± SE; *p < 0.05, **p < 0.01, ***p < 0.0001). (b) The numbers of annexin V and lactadherin positive MP per 10 3 cells were compared between the 6 replicate wells analysed over 3 individual experiments (n = 18). Spearman's test shows a significant correlation between the two PS labels (r = 0.96; p < 0.0001). (c) Mean fluorescence intensity (MFI) data for each individual label were pooled (n = 36) and differences between resting (NS) and TNF MP were assessed (mean ± SE; ***p < 0.0001). (d) Median fluorescence intensity for arbitrary size gates (a-e) (smallest to largest) for each individual label (mean ± SE; *p < 0.05, **p < 0.01, ***p < 0.001, ***p < 0.0001).
Scientific RepoRts | 5:16314 | DOI: 10.1038/srep16314 combinations) of CD105 + MP being double labelled with either PS marker. The proportion of PS marker alone was more important within the CD105 staining (72%).
It is important to note that the percentage of the labelled population represented within the total number of events detected by the cytometer could vary significantly depending on the combinations of markers used (Table 1). CD54 and CD105 positive MP could represent as little as 5 and 3% of the total events when observed in combination with annexin V and lactadherin.
The level of fluorescence of the MP population in the resting vs stimulated conditions was then analysed regardless of whether they were double or single stained (Fig. 1c). The annexin V+ population showed a significant increase in mean fluorescence intensity (MFI) following stimulation (p < 0.0001), while the other PS marker, lactadherin showed no change. Among the surface Ag, CD54 MFI was significantly increased (p < 0.0001), while CD105 remained unchanged. When arbitrary size gates were defined and the corresponding median fluorescence recorded we observed that, in stimulated conditions, for all surface markers the level of fluorescence was significantly higher as relative size increased (Fig. 1d). The greatest differences in the level of fluorescence between relative size gates were observed with annexin V labelling with a 2.17-fold increase between gates A (smallest) and E (largest). Lactadherin had a smaller fold change between gates A and E of 1.33-fold, with greater inter-size variation seen. Although significant, levels of fluorescence of CD54 and CD105 only increase by 1.25-and 1.09-fold, respectively.
Endothelial MP size distribution. Purified MP from resting and TNF-activated hCMEC/D3 were subsequently examined by immuno-EM. Heterogeneity in intra-sample size and density is demonstrated in Fig. 2a, a representative low magnification micrograph of negatively stained MP from TNF-activated hCMEC/D3. Across the 16 experimental groups in this EM analysis, 6.87% ± 5.07 (SD) of events were eliminated for falling under the 0.1 μ m MP size limit and 0.14% ± 0.41 exceeded the 2 μ m range. Of the 1690 events which fell within the set dimensional boundaries, 2.20% ± 1.76 were within the 1-2 μ m range. MP diameters ranged from 0.11-1.93 μ m and 0.10-1.79 μ m for the maximum and minimum Feret values, respectively. A significant correlation between these measurements was observed (Fig. 2b, Spearman's coefficient r = 0.96, p < 0.0001), demonstrating that we are examining a distinct population of vesicles with proportional maximum and minimum sizes. Furthermore, the equation of the best fit line, where y = 1.2167x + 0.0167, supports that these MP are spherical and not augmented. The respective mean, median and mode of the maximum (0.34; 0.26; 0.19) and minimum (0.26; 0.20; 0.15) diameters (in μ m) indicate that MP from the hCMEC/D3 are most concentrated in the lower end of the size range. No significant differences in MP size distribution were observed between the resting and TNF activated conditions ( Fig. 2c; Mann-Whitney test, p = 0.38).
MP characterisation by quantitative immuno-EM. The four markers of interest, annexin V, lactadherin, CD54 and CD105, were examined separately on the MP surface by immunogold labelling. Positively labelled events are shown in Fig. 3a with closed arrow-heads, whilst MP negative for gold particles are marked with open arrow-heads. CD235a (Glycophorin A), an erythrocyte specific Ag (used as an isotype-matched control), and the secondary Ab alone showed no non-specific gold-labelling amongst samples. Quantification of gold particles showed a significant correlation between elevated numbers of gold bound per MP and increased MP diameter. This was true for each of the markers examined, where the Spearman correlation test gave r values of 0.68, 0.51, 0.48 and 0.40 for annexin V, lactadherin, CD54 The Mann-Whitney test confirmed this significant size difference between gold-positive and negative MP, with p < 0.0001 for each of the markers examined (Fig. 3c).
The numbers of gold particles bound to resting (NS) vs TNF derived MP (Table 2) was tested with the Kolmogorov-Smirnov test for frequency distribution. No significant differences between the two distributions were observed for lactadherin (p = 0.734) or CD105 (p = 0.998). Distinct variations were however observed in the annexin V and CD54 labelled samples, where higher numbers of gold were associated with MP from TNF activated hCMEC/D3 (p = 0.010 and p < 0.0001, respectively). Comparative FCM and immuno-EM of dual labelled endothelial MP. PS and surface Ag expression were examined in combination with a dual immunogold labelling method utilising 10 and 15 nm gold particles. Figure 4a shows representative images of MP from TNF activated hCMEC/D3 dual tagged for annexin V or lactadherin (10 nm gold, closed arrow-heads), and CD105 or CD54 (15 nm gold, open arrow-heads). The numbers of PS +, double +, and Ag + MP were enumerated and expressed as a percentage of the total number of positive MP counted (Fig. 4b,c; right EM column). A comparison of these with the FCM data (Fig. 4b,c) show that in each sample PS labels the majority (> 75%) of marker positive events. The trend for CD54+ MP to increase between resting and TNF conditions is consistent between FCM and immuno-EM, as is the percentage of CD105 which is unchanged by TNF activation.    The striking and consistent difference between the two analytic methods is the higher proportion of MP which are solely labelled for the Ag in the immuno-EM experiments. These MP are examined further in Table 3 which summarises their size distributions (range of maximum Feret diameter in μ m) and the number of gold bound per MP. The median diameter for the CD54 labelled MP is 0.273 μ m and median number of gold bound was 2, whilst a median of 1 gold was bound in the CD105 only population which had a median diameter of 0.25 μ m.

Discussion
The limitations of MP detection are persistently being tested by improvements to the physical components and analytical parameters of new generation flow cytometers. Improved sensitivity of the Gallios FCM in detecting MP has been shown previously 24 . Employing the wide 2 FS detector with triggering events set on the SS channel improved the dynamic range of detection and reduced the resolution limit to ~300 nm. Whilst the authors of this work sought to demonstrate applicability of this high-sensitivity system in clinical settings, our work aimed to provide novel assessment of a selection of currently employed probes and in turn assess the strengths and weaknesses of high-sensitivity FCM.
The technically challenging and laborious nature of quantitative immuno-EM makes it a less favourable method for routine MP analysis. The study employed a whole mount negative staining method adopted from immuno-EM works used to validate Ag expression on MP 27,28 , and is comparable to studies specifically examining exosomes 29,30 . The primary benefit of this technique is that MP can be visualised in their entirety within 2-dimensions, without the need for resin embedding and sample sectioning, which is not only time consuming but can affect both image interpretation and gold quantification 31,32 . Drawbacks include the use of chemical fixation, which compromises membrane integrity, along with the low labelling efficiency of immuno-gold tagging. Although efficiency is estimated to be as low as 10-15% it is shown to be constant and reproducible if sample processing is standardised 33,34 . This means that whilst true Ag levels cannot be quantified, the relative expression between different experimental conditions and gold tags can be evaluated.
Our analysis demonstrated that of the MP probes examined, PS markers labelled almost all of the antibody fluorescent positive MP events, making PS the most efficient MP label in this study. Although other immuno-EM studies of annexin V did not quantify the numbers of gold particles bound our results are qualitatively comparable with that of Heijnen et al. who utilised a similar methodology. Arraud et al. alternatively demonstrate that MP were either heavily labelled or not labelled at all. We believe this difference in labelling density to our study is due to the 1) their annexin V being pre-conjugated to gold nanoparticles and 2) their use of cryo-EM sample preparation and imaging methods which better preserve the membrane integrity.
While high-sensitivity FCM indicates that PS positive MP account for only ~30-55% of the total MP population depending on the vesiculation stimulus, immuno-EM data determine this level to be higher at 62-65%. reported that PS positive MP accounted for 74% of the total endothelial MP levels measured, although they also suggested there may have been contamination with platelet MP due to CD31+ /CD42-labelling 17 . A stark difference is that in our relatively 'pure' system this measure of positivity vs negativity cannot be influenced by other MP types, but rather only by the quality of the buffer (i.e. filtered vs. unfiltered PBS) and antibody (a consequence of not washing the unbound antibodies). Chandler et al. similarly observed that not all negative MP events appeared to have cellular origins 16 . As such it is necessary to comment one should take care when reporting percentage positivity/negativity FCM results. Reporting the absolute numbers of positive MP per volume of plasma would overcome this.
By high-sensitivity FCM, annexin V and lactadherin measured comparable levels of absolute MP numbers. Hou et al. report that lactadherin identifies higher MP levels than annexin V, however this observation is made after several days of storage and not on fresh samples 35 . The prominent difference that we observed between the two ligands was their labelling intensity in resting versus TNF conditions. By both FCM and immuno-EM, MFI and gold values indicated that annexin V labelling was higher on TNF MP, whilst lactadherin labelling was unchanged. The differences could not be accounted for by changes in size distribution between the two stimuli. Both ligands showed increased labelling on larger MP, as expected, however measuring the median fluorescence intensity relative to arbitrary MP size indicated that this difference is more pronounced with annexin V compared to lactadherin. In the literature it has been shown that annexin V clustering results in its preferential binding to flat membranes and that a threshold of PS must be met for adequate binding, whilst lactadherin has a preference for sharp curvatures and binds PS proportionately to the levels available 11,[36][37][38] . Therefore, from our results we propose that the increased labelling intensity of annexin V following TNF activation may be explained by, 1) the increased availability of larger and therefore flatter MP, and 2) discrete changes in PS exposure at the membrane surface which exceed the required threshold.
In this system, CD105 labelled MP represent a modest 30% of the total marker positive MP population, whilst CD54 positive MP encompass between 40-60% depending on the activation stimuli. Thus, even in resting condition, on this particular immortalised cell line, CD54 is more highly represented on the MP than the constitutively expressed CD105. Consistency in CD105 expression between stimuli and upregulation of CD54 following TNF activation were well demonstrated by MFI measurements, observations similarly described by Jimenez et al. 14 . The low FCM detection of Ag-labelled MP was not unexpected, as various papers have described similar findings 39 . For CD105, the relatively low numbers of gold particles observed on MP by immuno-EM data corroborate the low expression observed by FCM. We believe this weak staining to be a true representation of CD105 expression as a number of antibody clones were tested and the same fixation and staining conditions were suitable for the other markers examined.
Comparative assessment of FCM and immuno-EM data indicates that the greatest variation between the two quantitation methods exists in these Ag single labelled populations. Where they represent between 1.5-5% of the total marker positive population by FCM, immuno-EM indicates they in fact contribute between 3-19% depending on the marker and stimulus of choice. It cannot be excluded that both the difference in antibody clone (for CD105) and conjugation (for CD54) affects the binding affinity of these labels. Closer examination of these MP alone found that they were not only on the small end of the MP size range but also expressed very low levels of Ag. As size distribution experiments however clearly demonstrate no variation between stimulus and MP size in this in vitro system, a finding similarly has also been observed for monocytic MP 40 , it can be reasoned that the size limitation of FCM would equally impact both stimulus conditions. Thus if a constitutive marker such as CD105 was to be used to analyse MP, an accurate relative expression between conditions could be obtained. The limitation is that precise enumeration of the true MP numbers in a sample will not be accurate; however it is clear that none of the markers available will achieve this.
It may be that overcoming these FCM limitations requires alternative high throughput analysis techniques. Nanoparticle tracking analysis (NTA) is a laser based method capable of determining particle size and suspension concentration through the investigation of vesicle Brownian motions. The use of NTA as a quantitative method for MP analysis is currently gaining momentum within the field, particularly as it is able to accurately assess events within a 50-1000 nm size range, without the limitations of a detector system or refractive index mismatch 41,42 . The technique is however still in its infancy regarding the use of fluorescent markers, and considerable development remains for Ab labelling to be achieved with the ease observed in FCM assays.
To summarise, constitutive and PS markers should ideally be used in combination to quantitate endothelial MP as data obtained solely with Ag markers do not appear reflective of the true number of MP events. An optimal set up would address the issue of FCM sensitivity with multi-colour antibodies or monochrome composite markers 20 . Measures of MFI are more appropriate for the assessment of inducible markers and conclusions should not easily be drawn on the degree of marker 'negativity' without direct visual confirmation. Annexin V and lactadherin can be used interchangeably if MP raw numbers are the primary interest of the study. However if the introduction of calcium is incompatible with the system at hand (e.g. EDTA is used as an anti-coagulant) or it is necessary to evaluate the relative levels of PS between varying MP, then the use of lactadherin would be preferential.
Ultimately, probe selection for MP FCM analysis remains at the discretion of the researcher and is dependent on the overall aims, requirements and outcomes of the study. There is great necessity for researchers within the field to form standardised quantitative approaches for MP analysis. Factors that must be addressed include how we define MP (and whether this definition solely encompasses PS positive events), how we collect and store MP, which analytic parameters are used and how we report our findings. Only then can a more holistic interpretation and understanding of MP be achieved.
Samples were made to a final volume of 200 μ l with PBS and analysed on the Gallios ™ flow cytometer.
Six replicate wells were analysed across 3 separate experiments.
Employing the wide 2 forward scatter detector, samples were measured over 120 seconds at a medium flow rate. Instrument detection was triggered on side scatter (SS), which was set at a threshold of 6, the lowest possible value where neither the background nor the numbers of positive events were altered. The acquisition rate did not exceed 1000 events per second. A flow rate of 18.5 μ l/min was determined with Flowcount beads ® (Beckman Coulter) as previously described 45 .
On a logarithmic scale, forward scatter (FS) was plotted against SS. Despite varying in their light scattering properties from plasma membrane vesicles, fluorescent nano-particles (Spherotech ® beads, 0.21, 0.45, 0.88 and 1.34 μ m in diameter) were used to set the limits of the MP size gate and ensure system stability (Fig. 5a). The electronic noise of the instrument was determined with PBS ( Fig. 5b) and unlabelled cell culture supernatant, in which large debris were removed by a 1800 g 10 min spin, was used to define the analysis window of MP (Fig. 5c). In FCM, the relative size of a cell or particle is determined by the intensity of light scattering detected by the FS detector. In this manuscript we have employed both the terms "size" and "relative size" keeping this in mind. To determine the relationship between the relative size of MP and their level of fluorescence, we define arbitrary gates on the FS axis (A to E, Fig. 5d) and measured the median fluorescence intensity of each subpopulation for each marker in stimulated conditions (Fig. 1d).
Swarm detection is a major issue when small events are analysed. In addition to using a medium flow rate as described above, we have taken the following precautions to ensure that we are not counting coincident events: (1) we perform serial dilutions of our labelled supernatants or biological samples to find the optimal dilution from which we have a linear correlation between the dilution factor and the numbers of positive events detected (Fig. 5e) and (2) we use the Time of Flight (TOF) parameter of the Gallios flow cytometer, which measures the transit time of a cell or particle traversing the laser beam. This parameter allows the detection of doublets and other coincident events. Particles of a uniform size will have an equivalent TOF (Fig. 5f, single MP gate) whilst samples that are in the laser beam at the same time have less total light and take longer to traverse the laser beam (Fig. 5f, coincident events gate). Within the labelled MP, this coincident event population never exceeds 2% of the total labelled events. Figures 5g and 5h show various combinations of labelling with both PS markers and surface Ag as well as resting and stimulated conditions. FL1 (FITC) was plotted against FL2 (PE) within each sample. Samples were compensated post-acquisition against single labelled, unstained and IgG-matched controls. Buffer only controls consistently lead to 30-50 events per second. The events were disregarded in analysis as they were not positive for fluorescence and did not modulate the numbers of positive events. FL1 + /FL2-MP expressed PS alone (PS+ ), FL1 + /FL2+ MP were positive for both the CD marker and PS (double), whilst FL1-/FL2 + MP solely displayed the Ag on their surface (Ag+ ). MP numbers were expressed per 10 3 cells. To allow reproducibility between instruments, we have analysed in our acquisition protocol Spherotech RCP-30-5A beads that contain 8 levels of fluorescence. Corresponding molecule equivalent surface fluorescence (MESF) values have been calculated for both FITC and PE channels and the corresponding data are reported in Table 4.
Microparticle Purification. MP were purified from hCMEC/D3 culture supernatants as previously described 44 . Briefly, large cellular debris were removed at 1800 g for 10 min before supernatants were spun for 45 min at 18,000 g. MP pellets were resuspended by gently pipetting in 1 ml of PBS, sample purity was confirmed by FCM based on size and samples were furthermore spun for 45 min at 18,000 g.

Transmission Electron Microscopy (TEM).
Purified MP were resuspended in PBS (approx. 3000 MP/μ l as estimated by annexin V labelling by FCM) and adsorbed onto formvar-carbon coated nickel 300 mesh grids (5 μ l/grid; Proscitech, Thuringowa, Qld, Australia) for 20 min. Labelling with annexin V-Biotin (#556417, BD Pharmingen, San Diego, CA, USA) in the presence of binding buffer or lactadherin-FITC in the presence of PBS was performed for 30 min at RT prior to fixation. Grids were fixed in 2% PFA for 15 min, washed with PBS, quenched in 0.05 M glycine for 15 min, blocked in 2% BSA in PBS for 30 min and incubated in 10 μ g/ml of the following primary Ab in blocking buffer for 1 h at RT: CD105 (#sc-20072, Santa Cruz Biotechnology, Santa Cruz, CA, USA); CD54 (#IM0544, Beckman Coulter,) and CD235a (#IM2210, Beckman Coulter). Samples were washed twice in blocking buffer prior to incubation in Goat-anti-Mouse-IgG-gold 15 nm (#815-022, Aurion, Wageningen, The Netherlands) and GAM-streptavidin-gold 10 nm (#810-099, Aurion) for 1 h at RT. Grids were washed with PBS and fixed with 2% Glutaraldehyde. For lactadherin-FITC labelled samples, grids were again quenched prior to secondary conjugation with a mouse-anti-FITC-IgG-gold 15 nm (#810.233, Aurion). Samples were washed with water prior to negative staining with 2% Uranyl Acetate for 3 min at RT.
All TEM images were collected with the JEOL 1400 (JEOL, Tokyo, Japan) at 120 kV at the AMMRF facility at the University of Sydney. Single labelled samples were imaged at 60,000 X Mag whilst double labelled samples were imaged at 100,000 X Mag, as determined by optimisation experiments. Image Analysis. Image J software was utilised for all immuno-EM MP quantification and analysis.
Briefly, MP regions of interest (ROI) were manually selected and the diameter and number of gold particles bound per MP were measured. Feret's diameter was selected as the preferred measurement over Statistical Analyses. Statistical analyses were performed with Graph Pad Prism 5 (Graphpad Software Inc, CA, USA) and IBM SPSS v20.0 (IBM, NY, USA) software. In each case, data normality was determined prior to application of the appropriate statistical test. MP numbers as determined by FCM were assessed with a one-way ANOVA and Bonferroni's correction for multiple comparisons when data assumed a normal distribution or with the non-parametric Kruskal-Wallis test followed by Dunn's test for multiple comparisons. All correlations were performed with the nonparametric Spearman's test. All comparisons between two conditions were achieved with either the Students t test or Mann Whitney nonparametric test. Gold distributions were compared with the Kolmogorov-Smirnov test. All tests were two-tailed and the level of significance was set at p value < 0.05. All graphical representations were produced in Graph Pad Prisms 5. In the text standard error of the mean is denoted with SE and standard deviation with SD.