Isolation and substrate dependence on extracellular vesicle characterisation using atomic force microscopy

Extracellular vesicles are nano- to micro-sized structures that carry biomolecules between cells to coordinate cellular activity and communication. Isolation and characterisation must be standardised to better understand the role of extracellular vesicles and how they can be used for disease diagnosis. Here we use atomic force microscopy to determine the physical differences between extracellular vesicles isolated using two different methods. Extracellular vesicles were isolated using two standardised methods, vacuum filtration and syringe filtration. In addition, extracellular vesicles were immobilised to plain mica and amino-functionalised mica to observe differences in adhesion onto substrates with different hydrophobicity. The application of atomic force microscopy enabled the study of vesicle adhesion, size distribution and morphology on the two different surfaces. It was found that both the isolation method and the substrate had a considerable effect on the physical properties of the extracellular vesicles, such as root mean square roughness values and size distribution. This demonstrates the ability to use atomic force microscopy to gain a more detailed understanding of the physical features of extracellular vesicles and the influence of different isolation methods on their morphology.


Introduction
Extracellular vesicles (EVs) are bilipid spherical particles released from all cells and are involved in various pathological functions. Research on their roles in cell-cell communication and, therefore disease is still being elucidated. To characterise EVs, they must first be isolated, a source of contention since their discovery. This is because EVs are derived from complex and diverse biological fluids, such as blood plasma [1], urine [2], breast milk [3], breath condensate [4], saliva [5], or cell culture media. Within a single specific cell culture, the isolated EV population is hugely heterogeneous regarding size, surface properties, and stiffness [6]. Different isolation protocols can result in different purities and, therefore, can change the EV and how it interacts with other biological entities. With that said, most characterisation methods use the bulk population, which does not give information on the individual physical EV characteristics of the lipid bilayer membrane.
Current isolation methods for EVs from media include ultracentrifugation, size exclusion chromatography, immunoaffinity capture, and density gradient centrifugation [7,8]. As part of the isolation process, the purification step can be varied. For example, syringe filtration and vacuum filtration have both been used as valid methods of purification for EVs [9,10]. Conventionally, to characterise the purity of EVs after isolation, bulk analysis of the entire EV population or electron microscopy is used. This only provides a portion of the analysis required to determine population differences. For example, conventional size characterisation techniques for EVs include dynamic light scattering, transmission electron microscopy and nanoparticle tracking analysis (NTA), which provide sample concentration, size and morphology. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Atomic force microscopy (AFM) can be used to analyse the surface morphology or mechanical properties, which could be used in disease diagnosis applications [11]. AFM is a multifaceted technique that allows for the physical characterisation of singular EVs [12][13][14]. In addition to being able to image the particles, characterisation of size and morphology (aspect ratio) [15,16], and mechanical properties which yield elasticity, stiffness [17][18][19], and adhesion of the EVs is conferred. The versatility of AFM makes it ideal for probing heterogeneous samples. However, it suffers from complications related to tip artifacts, non-standardised protocols and analysis of data using a variety of different models [20]. This makes it a challenging technique, but recent advances have opened up opportunities for standardised analysis and operation, particularly for biological samples.
For this work, EVs were isolated using two methods, syringe filtration and vacuum filtration. The EVs derived from these two methods were compared using AFM. There was a statistically significant difference in the root mean square (RMS) roughness obtained with the height and phase images. Also, the EVs were immobilised on substrates with different functionalities, plain mica (hydrophobic) and amino-functionalised mica (hydrophilic). The EVs showed different levels of adherence that influenced the height RMS (roughness) and the size distribution of the EVs. This work demonstrates that the chosen isolation method and substrate in which EVs are immobilised can influence the physical properties of EVs.

Methods and material
Chemicals used for this work were all purchased from Merck, and AFM probes from AppNano.

Cell culture and extracellular vesicle isolation
RWPE-1 (CRL-11609, American Type Culture Collection, Cryosite, Australia) were cultivated in complete keratinocyte serum-free media (KSFM; Gibco, Thermo Fisher Scientific, Australia), with 2 mM GlutaMAX (Gibco) and were maintained at 37°C with 5% CO 2 . At 60% confluency, cells were washed with sterile phosphate buffered saline (PBS; Invitrogen, Thermo Fisher Scientific) and incubated for 48 h in supplementfree base media with GlutaMAX. The media contained no bovine pituitary extract and epidermal growth factor to remove the possibility of contaminating EVs from supplements.
EVs were isolated using two methods, syringe filtration and vacuum filtration [21,22]. All EVs were collected from the supplement-free media and centrifuged for 20min at 2000 × g. The supernatant was filtered through 0.22 μm Millex-GP polyethersulfone (PES) syringe filter (Merck Millipore, Australia). At this step, the sample was either passed through a 0.1 μm Millex-VV polyvinylidene difluoride (PVDF) syringe filter (Merck Millipore) or a Rapid-Flow filter unit 0.1 μm PES membrane (Nalgene, ThermoFisher) for the vacuum samples. The samples collected from each method were concentrated using Amicon Ultra-15 100 kDa centrifugal filters (Merck Millipore) at 4000 × g for 5-20 min at 4°C. Retentates were washed twice with 0.1 μm filtered PBS, to a final volume of approximately 200 μl and stored at −80°C.

Nanoparticle tracking
Nanoparticle tracking analysis (NTA) was performed using a Nanosight NS300 nanoparticle analyser (Malvern, ATA Scientific, Australia) and a Harvard Apparatus syringe pump module (Nanosight, Malvern). Samples were diluted 1:250 in 0.1 μm filtered PBS. Concentration was determined from 3 × 60 s captures with an sCMOS camera at an infusion rate of 50 (Nanosight measurement) and a delay of 5 s between each recording. Samples were visualised using a 405 nm violet laser. The data analysis was conducted with NTA software v3.1 after the acquisition.

Sample preparation
Mica slides were cleaved and adhered to AFM sample discs using double-sided carbon tape. The mica was rinsed with deionised water and allowed to dry. 10 μl of EV sample at a concentration range of 1-7 × 10 10 particles/ml was placed on the prepared substrate for 30 min. After the desired period of time, the samples were rinsed with 50 μl of PBS (0.1 M), followed by 50 μl of deionised water to remove excess EVs and any other contaminants. Samples were allowed to dry overnight before AFM imaging was completed.
To investigate the EV morphology upon immobilisation of a hydrophobic and hydrophilic substrate, mica was modified with (3-aminopropyl)triethoxysilane (APTES) which is positively charged. This was done using a vapour deposition method where 10 μl of APTES (as purchased) was kept in a desiccator under vacuum, along with the freshly cleaved mica surfaces for 2 h. The substrate was rinsed with water to remove unbound APTES molecules and the EVs were deposited immediately after using the same conditions as above.

Atomic force microscopy operation
Samples were imaged using the Horiba AJST-NT SmartSPM. The images were recorded at 10 μm × 10 μm, 5 μm × 5 μm and 1 μm × 1 μm with 512 points at a scan rate of 1 Hz in tapping mode. The tips used were silicon ACSTA probes (AppNano) with a spring constant of 7.8 N/m and a resonant frequency of 150 kHz and the tip radius (radius of curvature 6 nm) was calibration using a standard sample. Tip radius calibration was done using a standard sample calibration grid. All experiments were performed at room temperature and in air.

Atomic force microscopy analysis
Image analysis was performed using Gwyddion software (version 2.6, Gwyddion Team) to calculate root mean square (RMS) roughness and RMS phase of the surface of the EVs. The acquired height and phase data were preprocessed by levelling through mean plane subtraction and correction of horizontal scars caused by tip artifacts using Gwyddion's built-in tools.
For RMS roughness calculation, the pre-processed height data was processed using Gwyddion's RMS roughness function. The resulting value, representing the root mean square value of the surface roughness, was reported with the appropriate units. Similarly, the pre-processed phase data underwent processing to calculate RMS phase using Gwyddion's RMS phase function but for the apex of the sEVs, therefore the substrate had no effect on the final value. This was done using the same methods done by Sharma et al [6]. The resulting value, representing the root mean square value of the surface phase, was reported with the corresponding units.
This involved levelling of the data by mean plane subtraction and correction of any horizontal scars which are attributed to tip artifacts. As all images obtained using the AFM are a convolution between the sample and the tip, tip corrections need to be applied to calculate the radius of the vesicles. Using previous protocols published by both Vorselen et al and Ridolfi et al, the vesicle line profile (obtained through Gwyddion) was fit to a circular arc or, more simply, by using the following equation [15,20]: where R c is the radius of curvature of the vesicle, H is the height of the vesicle, FWHM is the full-width half maximum of the vesicle and R t is the radius of the tip. The original size of the vesicle (R 0 ) can then be obtained with the following formula: where H i can be estimated from the force indentation curve using the distance between the contact point and the substrate. Ridolfi et al provide a faster way of calculating this by just using the original height recorded in the image for the original vesicle radius [15]. Statistical analysis, Mann-Whitney, was conducted using statistical software (GraphPad Prism, version 9.0.2) to compare RMS roughness and RMS phase values between different samples or experimental conditions.

Isolation and characterisation of EVs derived from RWPE-1 cells
EVs were first isolated from RWPE-1 media using established methods, syringe filtration and vacuum filtration. Both syringe filtration and vacuum filtration are standardised techniques for EV isolation and have been previously reported [9,10]. The mechanisms through which the isolation methods affect the physical properties of EVs include different filtration mechanisms, the presence of contaminants, and the impact of shear forces during the isolation process. For example, the vacuum filtration process has an increased shear force applied to the EVs. These physical differences could potentially impact the biological functions or downstream applications of the EVs.

Comparing syringe filtered and vacuum filtered extracellular vesicles
Initial AFM imaging of the RWPE-1-derived EVs on plain mica showed varied sizes and physical heterogeneity, despite being from a single cell type. Figure 2 shows the morphology and size of the RWPE-1-derived EVs. Different sample preparation methods were compared to optimise imaging conditions. This included trialling different deposition times of the EVs on mica and different concentrations. In addition, sample preparation was careful to remove salts and other impurities that could cause artefacts.
In previous reports, dehydration of EVs is known to cause a 'cup-shaped' morphology [23]. For this work, AFM in air was used as it is able to still provide differences that could be used as a comparison, and is quicker than liquid AFM. The height profiles of the EVs showed spherical structures, as is seen in figure 3(A). However, the phase images showed different phase values in the centre of the EV and around it (shown in 3(B)). This is as Figure 1. EVs were isolated from spent culture media devoid of supplements that could contain exogenous EVs. Isolation was via size exclusion filtration using vacuum pressure and subsequent ultracentrifugation. Three biological replicates of EVs from RWPE1 cells. EV proteins were separated with PAGE and transferred to nitrocellulose and the following EV marker proteins were detected: CD9, integrin beta 1, Alix and TSG101. expected, as the EVs were dried on the substrate and would reflect previous work on EV morphology. This characteristic was used in part to identify EVs in the sample while imaging.
Recently, Sharma et al used the flatness of substrates to evaluate, the topographic RMS roughness to compare different isolation protocols wherein, a precipitation method showed the highest RMS value [6]. Herein, we adapt this method to determine the differences between the two isolation methods for RPWE-1-derived EVs. For the characterisation of the surface of the EVs, root mean square (RMS) roughness of the peaks and valleys measured for surfaces was used. Differences were observed in the size distribution of the EVs isolated using vacuum filtration (referred to as EV2 in figure 4) with an average size of 111nm. When compared to samples that were ultracentrifuged and then, syringe filtered using a 0.22 μm filter (referred to as EV1 in figure 4), the average size was 30 nm. Syringe filtration of EVs resulted in the isolation of smaller EVs, and potentially broken up larger EVs as sizes less than 10 nm were observed ( figure 4). This method of isolation is a common method used by researchers and therefore the results highlight the importance of standardising these processes carefully. The vacuum filtration method used to isolate EVs resulted in a larger size distribution of EVs, and there were fewer EVs observed at less than 50 nm in diameter. Size is commonly used as one form of identification of small EVs, which have a different biogenesis pathway than larger EVs. Research has focused more on small EVs as they have been shown to contain biomarkers for cancer. Figure 5(A) shows a larger range of RMS roughness mean of 0.33 nm for EVs isolated using syringe filtration (EV1). In comparison, the vacuum filtration method (EV2) had a lower RMS roughness mean of 0.22 nm and a overall smoother surface. Higher surface roughness could be attributed to multiple things, including the contents of the membrane bilayer, and any polymeric residue embedded on the surface of the EVs as part of the isolation process [24]. These have been shown to influence EV interactions with other cells, including inhibition of fusion with the cell membrane [24]. Additional evidence of this can be shown from phase roughness values ( figure 5(B)), where the mean profile RMs roughness was 0.95 and 0.32 degrees for EV1 and EV2, respectively [25]. Both the height and phase RMS data had a statistically significant difference. In contrast to topographic RMS, phase RMS reflects the phase peaks and valleys on a surface. Phase changes occur through the changes in oscillation of the probe. They can be related to the energy dissipation of the tip onto the sample and the chemical heterogeneity of the sample. Therefore, larger phase RMS roughness values mean there are differing phase  changes along the EV surface, which reflects the different materials within and on the membrane. Kikuchi et al used AFM phase imaging for quantitative imaging of individual microvesicles from bacteria [25]. Although the authors conducted AFM imaging in fluid, they were able to demonstrate a novel method to characterise compositional changes between different microvesicle populations.
Figures 5(C) and (D) show the roughness line profiles along one EV for phase differences and show a significant difference between the phase values recorded for EV1 and EV2. Figure 5 shows higher phase RMS roughness values for EVs isolated using syringe filtration, which aligns with the topographic RMS roughness results. The EV surface is therefore affected by the isolation method used, and phase RMS values can be used to assess this. The vacuum filtration uses an applied pressure and varying shear forces when compared to syringe filtration. This may result in a change in the populations physical properties, as seen in these results. This is supported by the broader size distribution of the EVs. The decreased roughness RMS and phase RMS can be attribute to a different sub-population, contamination and a change in membrane integrity. The exact factors contributing to the observed differences would need to be investigated further.

Effect of surface phobicity on extracellular vesicle adherence
EVs are known to be negatively charged on their surface. Mica is also a negatively charged and a hydrophilic substrate that is often used to deposit EVs for AFM imaging. Other alternatives include positively charged glass substrates and APTES-modified mica. Increased electrostatic interaction between the EV and the substrate would lead to increased adhesion, therefore better immobilisation for better imaging. The aspect ratio of the EVs (height/radius) was used to measure the deformation of the vesicles on a substrate. Due to their dehydration, the deformation is already high for EVs on mica and on APTES-modified mica. Here, there were no significant differences in the deformation, although it would be more valuable to do this study with the EVs in their hydrated form. There were, however, differences in the size distribution of the EVs on the two substrates, with larger EVs being observed on the APTES-mica (shown in figure 6(B). and figure 4). The most significant differences were observed in the RMS roughness values seen in figure 6(A). APTES EVs showed higher roughness values compared to mica-deposited EVs. The APTES-modified substrate itself has low RMS values and is relatively flat. Therefore, this difference can be attributed to the further flattening or deformation of the EVs on the substrate, which could lead to the contents of the EV creating a rougher surface profile.

Conclusions
EVs collected from RPWE-1 cell media were isolated using syringe filtration (EV1) and vacuum filtration (EV2). AFM was used to analyse dehydrated EVs to compare physical characteristics such as roughness and size. The use of topography RMS values to describe the roughness showed differences in the surfaces of EVs isolated using the two techniques, as well as with EVs deposited on differently charged substrates. The surface properties of EVs relate to the interactions vesicles have with other EVs and cells and also to the cell of origin. Phase RMS values also showed a statistical difference between the isolation methods, indicating that there are chemical differences on the surface of the EVs, as well as morphology. These chemical differences change based on the isolation method used but also based on the substrate used to immobilise the vesicles. As phase differences can also originate from varied stiffness values of the substrate, the phase RMS, therefore, reflects both the chemical constituents, adhesion, and stiffness of the sample. APTES functionalising of mica was done to probe the EV population and adhesion properties on the positively charged surface. It was shown that the EVs that were visualised using this method were significantly larger. The use of AFM for EV characterisation can open doors for future applications. For example, it is known that EVs can reflect the cell of origin, and cancer cells are inherently softer than their healthy counterparts. If this characteristic is passed onto the EVs via the biogenesis pathway, a method for physical property analysis for disease diagnosis could be developed [26,27].