Viscosity of Plasma as a Key Factor in Assessment of Extracellular Vesicles by Light Scattering

Extracellular vesicles (EVs) isolated from biological samples are a promising material for use in medicine and technology. However, the assessment methods that would yield repeatable concentrations, sizes and compositions of the harvested material are missing. A plausible model for the description of EV isolates has not been developed. Furthermore, the identity and genesis of EVs are still obscure and the relevant parameters have not yet been identified. The purpose of this work is to better understand the mechanisms taking place during harvesting of EVs, in particular the role of viscosity of EV suspension. The EVs were harvested from blood plasma by repeated centrifugation and washing of samples. Their size and shape were assessed by using a combination of static and dynamic light scattering. The average shape parameter of the assessed particles was found to be ρ ~ 1 (0.94–1.1 in exosome standards and 0.7–1.2 in blood plasma and EV isolates), pertaining to spherical shells (spherical vesicles). This study has estimated the value of the viscosity coefficient of the medium in blood plasma to be 1.2 mPa/s. It can be concluded that light scattering could be a plausible method for the assessment of EVs upon considering that EVs are a dynamic material with a transient identity.


S1.1.1 Dynamic light scattering.
Dynamic light scattering, DLS, enables the determination of the relaxation time,, relaxation rate, , or diffusion coefficient, D, of particles by calculating the correlation function of the scattered electric field, g 1 (t), from the measured correlation function of the scattered light intensity, G 2 (t), and applying the relation In equation S1, q is the scattering vector, given as q = (4π n 0 λ 0 )sin(θ 2 ⁄ ) ⁄ , where n 0 is the refractive index of the medium,  0 is the wave-length of light and  is the angle of observation. The values of  (or D) can be determined by the inverse Laplace transform program CONTIN developed by Provencher [1]. In case that samples are polydisperse, a multi exponential fit is used to fit the g 1 (t) function and determine the exponents. The result of such procedure is the distribution of intensity over  (or  or D) for the species in solution, which is then converted into the intensity weighted size distribution (i.e. over the hydrodynamic radii, For the herein studied samples, all size distributions were multimodal, exhibiting 2-4 peaks. The mean R h values for the peaks determined at  = 90° and their origin are reported in the main paper. An example of such multimodal distribution is shown in Figure S1, where each scattering species of the radius R h,i is represented by its relative contribution or amplitude, A i ( i ), to the total intensity of scattered light.
R h values of particles that are larger than  0 /20 or are polydisperse can significantly depend on the angle of observation. In some cases, peaks of different populations can merge into one(broad)-peak result at some angle but are well separated at another. Multiple angle analysis is therefore needed for a proper insight. The 90° angle is useful because of diminished effect of unpolarised light at this angle and because scattering of large particles dominates at small angles. This allows large populations to be analysed at angles smaller than 90°, while populations of small particles are better treated at larger angles.
When angular dependence is observed in R h values, zero angle hydrodynamic radius (R h,0 ) is needed for particle shape characterization through the shape parameter  (= R g /R h,0 ; R g is the radius of gyration; see below). Forward light scattering measurement ( = 0°) is problematic, therefore, R h,0 is normally determined by extrapolation of the experimental data obtained at higher angles to  = 0°. It is important to keep in mind that extrapolated values can be falsely high when analysing samples that contain more than one population [2], and overestimations can occur also due to the polydispersity of the particle's within one population [3] or due to the presence of aggregates and other large particle contaminations [4]. The deviation of R h to higher values in our samples was very evident in unfiltered samples (see Figure 1 in the main script). To avoid overestimations, we used linear function to determine the R h,0 of the population of large particles by considering the experimental data measured at   60° in extrapolating them to  = 0°. Figure S1. A) The intensity-weighted distribution of the hydrodynamic radii (R h ) obtained at  = 90° and T = 25°C for the thawed sample ES1. The Y-axis shows the relative contributions (A i ( i )) of particles of size R h,i to the total LS intensity. The measured correlation function of the intensity of scattered light, G 2 (t), is shown in the inset of Figure S1A; B) The lng 1 (t) plot for ES1 (the lines indicate slopes corresponding to populations of vesicles) ; C) intensity-, mass-, and number-weighted distributions for the same sample.

S1.1.2 Static light scattering.
The integral time averaged intensities, I(θ)  I(q), were recorded in the -range from 30° to 150° simultaneously with the correlation functions. Intensities measured in counts of photons per second (cps) were normalized with respect to the Rayleigh ratio, R, of toluene and converted into the absolute intensity units given in cm -1 . The intensities of the individual populations (peaks) obtained by the below outlined procedure (see equations S4 and S5) were analysed separately for the angular dependency.
Measurements of the angular dependency of scattered light enable the determination of the form factor, P(q), of colloidal particles in solution. P(q) is defined as: where I(0) and I(q) are the scattering intensities at θ (or q) = 0 and at an arbitrary θ (q), respectively. From P(q), the radius of gyration, R g , of particles, was obtained.
In order to calculate the R g of the scattering particles, i.e. extracellular vesicles in the present study, which are represented by peaks 1 and 2 in R h distributions (populations 1 and 2; the corresponding R h designations are ℎ 1 and ℎ 2 , respectively; see manuscript and Figure S1), the total LS intensity is split into contributions, I 1 (q) and I 2 (q), resulting from these two peaks. The splitting is based on the intensity weighted distributions of the relaxation times. Each scattering species of the radius R h,i is represented in this distribution by its relative contribution, i.e. by the amplitude A i ( i ), to the total intensity of scattered light, I t (q) (see Figure S1). The relative amplitudes of the two peaks are then estimated using the procedure outlined in references [5][6][7] as and converted into the LS intensities for each population of particles. The time-averaged intensities of peaks 1 and 2, I 1 (q) and I 2 (q), respectively, are calculated as: These data are then treated separately for the angular dependency. Details of this procedure can be found elsewhere [5][6][7].
In view of the small size of exosomes (population 1), the intensity of scattered light does not significantly depend on the angle. Therefore, a trustable R g determination is not possible.
It is well-known that large particles scatter more light than the small ones and they dominate in the overall scattering, as can be seen also for our samples in Figure S1a. Intensity weighted distributions of R h , such as in this figure, are therefore often transformed into the mass weighted distributions by taking into account that the scattering intensity is proportional to molar mass, M i , and concentration, c i , of a particle. M i is further proportional to the particle radius, R i . The following proportionality thus applies: ∝ or ∝ / , where the exponent a depends on the particle shape (it is for example a = 3 for a hard sphere). Based on this relationship, the relative weight content of particles can be estimated from: Figure S1 shows an example of the transformation of intensity weighted distribution ( Figure   S1A) into the mass weighted distribution ( Figure S1B) for the same sample.
Generally, functions for R g derivation are valid for particles that are small in comparison to the wavelength of light, i.e. that fulfil the criterion qR g < 1.5, which was the case with particles of population 1. To meet the same criterion, small angle scattering is relevant for R g determination and structural characterization for the population 2 (larger vesicles). Either first (for R g  25 nm; peak 1) or second (for higher R g , peak 2) order fit was used to fit the P(q) as a function of q 2 (Eq. S3) according to equation S3a:

S1.1.3.1
Form factor of a vesicle. The most relevant shape for a vesicle is that of a hollow sphere surrounded by a shell, as sketched in Scheme S1.

Scheme S1.
Form factor of a monodisperse spherical vesicle was derived as the difference between the form factors of 2 solid spheres with radii R i (inner radius of the vesicle) and R o (outer radius of the vesicle) as [8][9][10]: where j i (z) is the Bessel function of the first order.
By introducing the dimensionless parameters = and = (giving = ), the final expression for P(q) gets the form [8]: is usually presented in the form of a Kratky plot, i.e. the dependence of (qR g ) 2 × P(q) on qR g . This graph has an advantage that the very low scattering intensities at large scattering angles are amplified by the weight (qR g ) 2 , making differences between various spheroidal structures more evident. Figure S2 shows the dependence of (qR g ) 2 × P(q) on qR g for a hard sphere, a hollow sphere with an infinitely thin shell and for a vesicle with a shell of thickness corresponding to l = 0.75. Characteristic oscillations in these functions are clearly seen. Figure S2. The Kratky plot (dependence of (qR g ) 2 P(q) on qR g ) for a hard sphere (l = 0), hollow sphere (l = 1) and a vesicle with l = 0.75. For comparison, Zimm, Guinier and Debye-Bueche functions are plotted as well (see text).

S1.1.3.2 Some other form factors.
For the present study, form factors of some other particle architectures appeared useful. The data for the population 2 particles often agreed with the function (Equation S3d) suggested by Debye and Bueche [11,12], and Isihara [13], for the description of microgel-like particles with a core-shell structure having a spherical distribution of points around the centre of gravity such as in gels, networks, cross-linked or branched polymers etc., with a higher mass density in the centres: A useful scattering function (Equation S3e) for globular structures having a spherical distribution of points around the centre of gravity with the Gaussian probability is also the Guinier function: These P(q) functions are also plotted in Figure S2 in the form of Kratky plots.
For polydisperse samples, the peaks for spherical shapes in the Kratky plots broaden and make the difference between hollow and solid spheres difficult to distinguish [3]. In addition, aggregation and polydispersity also affect the results. Often, the experimental data in the Kratky plot show an upward curvature at higher q (or qR g ), which is a characteristic feature of aggregation [14], and positive deviations from the theoretical function for a vesicle towards the Debye and Bueche [14,15].

S1.2 Flow cytometry (FCM)
Particle which enabled the successful calibration of the laser settings. Then we used 460 nm beads to adjust the voltage (channels) and set the gate, which was a reference to the EVs of interest in the isolate samples. To exclude debris from the analysis, we first measured filtered buffer (PBS) and set the trigger so that the buffer events were excluded. We selected the hyper log (hlog) settings to all channels and set the trigger on SSC to 4 and shut off the secondary trigger. The final result of the measurement ( Figure S3) is the total number of directly detectable particles in the sample isolate, since we did not stain our samples nor use any antibodies. This method was shown previously to give clinically relevant results [16][17][18].  Figure S4 represents representative correlation functions for the examples of three different types of samples -exosome standard, blood plasma and EV isolate from the same blood plasma sample, and the calculated intensity distributions calculated from them. Figure S4: A) measured correlation functions (G 2 (t)), B) calculated g 1 (t) and C) intensity distributions (Ai(τ)) for examples of exosome standard, blood plasma and EV isolate from it.

S2.1 Batch DLS/SLS analysis.
In Tables S1-4 results of the DLS/SLS analysis of exosome standard (Table S1) and EVs in blood plasma of three healthy donors are collected. values (ρ corr ) were calculated based on the approximation  0 = 1.2 mPas for the plasma medium viscosity. ND = not defined NC = no correction. For diluted samples such as EV isolates it was assumed that dilution was sufficient to overcome the viscosity effect and  0 = 0.9 mPas (water at 25°) was employed in evalouation of R h . Due to the unknown value of effective viscosity of the medium for small particles, results of population 1 were never corrected. values (ρ corr ) were calculated based on the approximation  0 = 1.2 mPas for the plasma medium viscosity. ND = not defined NC = no correction. For diluted samples such as EV isolates it was assumed that dilution was sufficient to overcome the viscosity effect and  0 = 0.9 mPas (water at 25°) was employed in evalouation of R h . Due to the unknown value of effective viscosity of the medium for small particles, results of population 1 were never corrected.

S2.2 Effects of filtration
In sample preparation immediately before the DLS and SLS measurements, filtration and centrifugation are routinely used to eliminate large particle contaminations (for example dust particles) that can represent a substantial interference in LS due to their strong contribution to the measured LS intensity. However, the resistance of EVs to sheer stress, which they experience during filtering, is not very well known, and so the decision whether samples should be subjected to it or not is a delicate one.
In our blood plasma samples, centrifugation was used for the preparation of plasma from the blood (see Methods Section in the manuscript), which means that most of the large contaminants were removed already in the preparation step. Sometimes centrifugation may not be sufficiently efficient, and therefore filtering is routinely employed just prior to the LS measurements, in particular in case of water suspensions. To evaluate how the filtering procedure affects the analysis of vesicles, we analysed unfiltered and filtered samples. The DLS results (R h in dependence on q 2 ) and the SLS results (R = (R − R 0 ) 1 in dependence on q 2 ) for unfiltered and filtered exosome standard are shown in Figure S5.
Scattering was stronger from the unfiltered samples; the ratio between the scattered and the incident light intensity at 90° was up to 25 % higher in the unfiltered samples than in the filtered ones. This difference in intensity was evidently smaller in samples with added trehalose (see Section S2.1.2): in this case, intensity was only up to 6 % higher in the unfiltered samples. In the unfiltered samples, larger R h,0 was generally determined for the population 2 (see Figure S5A), if measurements at all angles were considered in the extrapolation to zero angle. In case of filtered samples, the upward curvature of R h values at small angles was lost and similarly, the dependency of the excess LS intensity on q 2 became linear after filtration, whereas it was curved for the unfiltered sample (see Figure S5B). All this suggests that R h,0 in unfiltered samples is overestimated due to the large particle contaminations. In SLS analysis, filtering usually decreased the small angle error of the measurement, but the observation that P(q) turns to higher values at larger angles (larger qR g ) was present also in the filtered samples (se Figure 2 in the manuscript). The shift of P(q) to higher values at larger angles is reported in the literature as well [14] and is, similarly to the present study, explained as a feature of particle aggregation.
To inspect the origin of changes in the unfiltered and filtered samples as observed by DLS/SLS, the samples were analysed also by AF4/UV-MALS and AFM. Some changes in size distributions of the population of larger particles in the filtered and unfiltered samples were confirmed by these two techniques as well (see Figures 1 and 3 in the manuscript).
In AF4/UV-MALS, the peak due to larger particles was narrower in the filtered sample. The value of R g for larger particles in the unfiltered (filtered) sample as calculated from AF4/UV-MALS was 93 (84) nm, which is lower than the R h,90 value for larger particles as obtained by batch DLS (R h,90 = 100-120 nm), and also lower than R g calculated from SLS (R g = 130-150 nm).
While both, batch DLS/SLS and AF4/UV-MALS showed differences between the unfiltered and filtered sample, filtration did not eliminate the discrepancy of size parameters obtained by these two techniques, suggesting a large effect of polydispersity on the batch DLS/SLS results. Polydispersity therefore may explain the change of regression curve of R h vs. q 2 on Figure S5A, which was not shifted in parallel toward higher values in unfiltered sample, but rather it exhibited a greater slope (while the data in the  range 80-100° remained similar to these for the filtered sample).
AFM was employed to reveal potential morphological changes of vesicles in exosome standards caused by filtration ( Figures 1A and 1B in the main script). Similar spheroidal shape of particles was observed in both AFM micrographs so as the grouping of the vesicles into aggregates. Particles' size in filtered and unfiltered ES was similar, but particles in the filtered sample seemed deformed. It has to be noted, however, that the area inspected by AFM was very small and may not be representative for the whole sample. Besides, reproducible preparation of samples for AFM is difficult. The above differences may be associated with these particular experimental procedures in AFM. Finally, we concluded that AFM results confirmed our anticipations obtained from scattering techniques, i.e. large polydispersity and possible aggregation of vesicles. Based on results of all three techniques, degradation of vesicles upon filtration could not be excluded.
All in all, in our study, the batch DLS results for R h,0 determination before and after filtration agreed very well if only angles larger than 50° were taken into account in the extrapolation procedure to zero angle (c.f. Figure S5A). Because we cannot ascertain if the decrease in the mean size and change in the form factor (P(q)) of larger particles in filtered samples originates from the removal of contaminations, degradation of the large vesicles into smaller ones or dissociation of exosomes from some aggregates, we suggest that unfiltered samples can be analysed if there is a risk of sample change due to filtration. However, small angles should be excluded in the analysis of the DLS data to prevent an overestimation of R h,0 .

S2.3 Effects of freezing and trehalose addition
Freezing is a convenient and widely used storage method for biological samples such as blood plasma, but cycles of freezing and thawing may induce vesicle rifting. Fresh and frozen samples were therefore analysed by DLS/SLS to evaluate the effect of sample change due to these procedures on results of the analysis. In our study, the short-time freezing did not substantially change the size and shape of particles as demonstrated by DLS/SLS analysis, but the decrease of the scattering intensity suggests that this procedure could cause some vesicle disintegration and accompanying protein release.
The ratio between scattered and incident light intensity was consistently lower in the thawed samples of the exosome standard, and oppositely, higher in the samples of blood plasma, while the size parameters and the -ratio remained similar within the limits of the experimental error (see data for the thawed samples in Tables S1-4; total LS intensity is not reported). Considering these results, the size characterization after one cycle of freezing and thawing was taken as still reliable. The decrease in light intensity ratio in samples of thawed exosome standards was attributed to some rifting of vesicles, whereas its increase in thawed blood plasma samples may be due to the aggregation of vesicles and/or proteins.
Parallel samples were prepared with trehalose to assess its potential benefit for cryoprotection and aggregation prevention. In the results reported in Tables S1-4, the differences between samples with and without trehalose were not significant enough to prove its function.
It is possible that a higher concentration of trehalose would have a different effect.

S2.2.1 Preliminary sedimentation test
Based on the literature reports [19][20][21][22] it was predicted that after 15 min centrifugation at 10 000 × g at 4 °C particles larger of 200 nm will accumulate in the pellet and after 1 h at 100 000 × g the vesicles of 30 nm would be precipitated too. Plasma sample was centrifuged in sever cycles to monitor particle sedimentation. The fractions obtained in experiment (Exp1) are described below. Pellet II: 2× 250 μL of the sample (pellet + remaining plasma) left in the tube after the second centrifugation was resuspended in 800 μL of PBS-citrate buffer.
Pellet III: 2 × 200 μL of the sample (pellet + remaining plasma) left in the tube after the last cycle was resuspended in 800 μL of PBS buffer.
Pellet IV: 2 × 150 μL of the sample (pellet + remaining plasma) left in the tube after supernatant removal was resuspended in 800 μL of PBS buffer.
Exp1. sV -EV isolate (aliquots 5-I and 6-I): 1.2 mL of PBS were added to pellets of Exp1. sIII (2 × 100 μL) after the first ultracentrifugation (1 h at 100 000 × g), resuspended and centrifuged again for 1 h at 100 000 × g. The supernatants (2 × 1.2 mL) were discarded. 1.2 mL PBS was added to remaining pellets. Samples were resuspended and centrifuged again for 1 h at 100 000 × g. 1 mL of supernatants were discarded, while 0.4 mL PBS was added to the remaining 0.2 mL of pellet in each tube and resuspended.
Exp1.s0 and the supernatants I-IV were measured for their density and viscosity. The results are given in the Table S5. The samples were analysed by DLS (results in Table S6). All the methods suggest that the final separation of the sample was hardly changed even after 4 cycles of ultracentrifugation at 100 000 × g.  Figure S6: After 8 h of ultracentrifugation at 100 000 × g, blood plasma separates into five distinguishable layers.  Figure S6) 1.008 0.94 Yellow supernatant (c.f. Figure S6) 1.021 1.32 Total supernatant (colourless s. + yellow s.)

S2.2.2 Plasma fractions obtained after 8h of ultracentrifugation at 100 000 × g
1.019 1.21  1 -Initial plasma sample (final cell debris removed by 15 min centrifugation at 1500 × g one day after blood donation) 2 -Initial plasma sample (final cell debris removed by 30 min centrifugation at 5000 × g three days after blood donation) 3 -Colourless supernatant 4 -Yellow supernatant 5 -Pellet in yellow supernatant 6 -Pellet in yellow supernatant (sample was diluted with pbs 1:1 before ultracentrifugation) 7 -Gel pellet M -Size marker (size in kDa is annotated on the picture) Typical plasma proteins: afibrinogen. b -IgG. ctransferrin. dalbumin. e -apoE. flight chains of IgG, apoA1. g -apoB. Bands were assigned according to their mass and comparison of their electrophoretic mobility described in literature.