Characterization of extracellular vesicles and artificial nanoparticles with four orthogonal single-particle analysis platforms

We compared four orthogonal technologies for sizing, counting, and phenotyping of extracellular vesicles (EVs) and synthetic particles. The platforms were: single-particle interferometric reflectance imaging sensing (SP-IRIS) with fluorescence, nanoparticle tracking analysis (NTA) with fluorescence, microfluidic resistive pulse sensing (MRPS), and nanoflow cytometry measurement (NFCM). Results were compared with standard EV characterization techniques such as transmission electron microscopy (TEM) and Western blot (WB). EVs from the human T lymphocyte line H9 (high CD81, low CD63) and the promonocytic line U937 (low CD81, high CD63) were separated from culture conditioned medium (CCM) by differential ultracentrifugation (dUC) or a combination of ultrafiltration (UF) and size exclusion chromatography (SEC) and characterized per MISEV2018 guidelines. Mixtures of synthetic particles (silica and polystyrene spheres) with known sizes and/or concentrations were also tested. MRPS and NFCM returned similar particle counts, while NTA detected counts approximately one order of magnitude lower for EVs, but not for synthetic particles. SP-IRIS events could not be used to estimate particle concentrations. For sizing, SP-IRIS, MRPS, and NFCM returned similar size profiles, with smaller sizes predominating (per power law distribution), but with sensitivity typically dropping off below diameters of 60 nm. NTA detected a population of particles with a mode diameter greater than 100 nm. Additionally, SP-IRIS, MRPS, and NFCM were able to identify at least three of four distinct size populations in a mixture of silica or polystyrene nanoparticles. Finally, for tetraspanin phenotyping, the SP-IRIS platform in fluorescence mode and NFCM were able to detect at least two markers on the same particle. Based on the results of this study, we can draw conclusions about existing single-particle analysis capabilities that may be useful for EV biomarker development and mechanistic studies.

Classification of extracellular vesicles (EVs) into subtypes has been proposed based on size, 85 biogenesis pathway, separation procedure, cellular or tissue origin, and function, among others 86 [1][2][3][4][5][6]. However, reproducible classification of EV subtypes will require single-particle 87 characterization techniques including phenotyping by surface molecules or molecular signatures 88 [7,8]. In this sense, current knowledge of EV subtypes could be compared with knowledge of 89 immune cells in the 1970s and early 1980s. Around that time, multiplexed flow cytometry 90 capabilities and cell sorting were developed, allowing more precise identification, 91 characterization, and molecular and functional profiling of immune cell subsets [9]. Single-92 particle technologies for much smaller biological entities will be needed to divide heterogeneous 93 EV populations into well-defined and easily recognized subgroups. 94

95
In this study, we evaluated several particle types and single-particle characterization platforms. 96 For input, we used a selection of biological and synthetic particles. EVs were separated from 97 culture medium of H9 T lymphocytic cells and U937 promonocytic cells using several methods. 98 These two cell lines were chosen because they display distinct tetraspanin levels. Specifically, 99 H9 have high CD81 and low CD63 levels, while U937 produce little CD81 but abundant CD63. 100 Mixtures of distinct sizes of synthetic silica and polystyrene beads were also tested. The 101 technology platforms (Text Box 1) were: single-particle interferometric reflectance imaging 102 sensing (SP-IRIS, NanoView) [10,11] with fluorescence, nanoparticle tracking analysis (NTA,103 ParticleMetrix) [12][13][14] with fluorescence, microfluidic resistive pulse sensing (MRPS,104 Spectradyne) [14,15], and nanoflow cytometry measurement (NFCM, NanoFCM) [16,17]. 105 106 107 Text Box 1: Evaluated Technologies Single-particle interferometric reflectance imaging sensing (SP-IRIS) captures particles (e.g. EVs) onto a chip by affinity reagents, usually antibodies, to surface antigens. Particles are imaged by interferometric reflectance for sizing and counting, and fluorescence detection may be done for up to three channels for surface antigens or internal molecules following fixation and permeabilization. Website for the platform we used: https://www.nanoviewbio.com/ Nanoparticle tracking analysis (NTA) is an optical method to track single particles and assign sizes and counts. Measuring Brownian motion allows calculation of a hydrodynamic sphereequivalent radius of each tracked particle. Additionally, fluorescence filters can be used for detection of particle-associated fluorescence moieties channels. Website for the platform we used: https://www.particle-metrix.de/en/particle-metrix Microfluidic resistive pulse sensing (MRPS) counts and sizes particles as they pass through a pore between microfluidic chambers. Occlusion of the pore results in a measurable change in electrical signal (defining an event) that is proportional to the volume of the particle. Often, this technique uses different disposable cartridge pore sizes to detect particle populations within specific size ranges. As a non-optical technology, fluorescence detection is not available. Website for the platform we used: https://nanoparticleanalyzer.com/ Nanoflow cytometry measurement (NFCM) is a flow-based technique that detects nano-sized particles by scatter and/or fluorescence. Compared with traditional flow cytometry, a smaller flow channel reduces background signal, and lower system pressure increases dwell time of particles for enhanced signal integration. Website for the platform we used: http://www.nanofcm.com/products/flow-nanoanalyzer temperature for 30 minutes and washed with phosphate-buffered saline (PBS). EVs were loaded 131 onto the column, and 0.5 mL fractions were collected by adding additional PBS to the column. 132 EV enriched fractions (SEC; fractions 7-9) were pooled and further concentrated using 3 kDa 133 MWCO Amicon Ultra-15 Centrifugal Filters to a final volume of 1 mL. 50 µL aliquots were 134 stored at -20°C for downstream assays. 135 136 Differential Ultracentrifugation (dUC): 60 mL of CCM from each cell line was centrifuged at 137 1,000 ´ g for 5 minutes at 4°C to remove cells and cellular debris and 2,000 ´ g for 10 minutes at 138 4°C to remove additional debris. The supernatant was transferred to polypropylene thin-wall 139 ultracentrifugation (UC) tubes and centrifuged at 10,000 ´ g for 30 minutes at 4°C using a 140 swinging bucket rotor (Thermo Scientific rotor model AH-629, k-factor 242, acceleration and 141 deceleration settings of 9) to pellet large EVs. Supernatant was transferred into new 142 polypropylene thin wall UC tubes and centrifuged at 100,000 ´ g for 70 minutes at 4°C using the 143 same swinging bucket rotor. The 100K pellets containing small EVs were resuspended in 1 mL 144 of PBS, vigorously vortexed, and placed on ice for 20 minutes. 50 µL aliquots were stored at -145 20°C for downstream assays. 146 147 Transmission Electron Microscopy (TEM): 10 µL freshly thawed aliquots were adsorbed to 148 glow-discharged carbon-coated 400 mesh copper grids by flotation for 2 minutes. Grids were 149 quickly blotted and rinsed by flotation on 3 drops (1 minute each) of 1´ Tris-buffered saline. 150 Grids were negatively stained in 2 consecutive drops of 1% uranyl acetate (UAT) 151 with tylose (1% UAT in deionized water (dIH2O), double filtered through a 0.22 µm filter), 152 blotted, then quickly aspirated to cover the sample with a thin layer of stain. Grids were imaged 153 on a Hitachi 7600 TEM operating at 80 kV with an AMT XR80 CCD (8 megapixel). SS and PS 154 were absorbed to grids as above, but with initial flotation for 5 minutes and imaging on a Phillips 155 CM-120 TEM operating at 80 kV with an AMT XR80 CCD (8 megapixel  and AF488 fluorophore-conjugated PS beads and a Silica Nanosphere Cocktail, respectively. 234 20 µL of each EV preparation was incubated with 20 µL PE-conjugated CD81 and 5 µL AF488-235 conjugated CD63 antibodies at 37°C for 30 minutes. After incubation, the mixture was washed 236 twice with PBS and centrifuged at 110,000 ´ g for 70 min at 4°C (TH-641 rotor, k-factor 114, 237 Thermo Fisher, using thin-wall polypropylene tubes with 13.2 ml capacity and acceleration and 238 deceleration settings of 9). The pellet was resuspended in 50 µL PBS. Events were recorded for 1 239 minute. Using the calibration curve, the flow rate and side scattering intensity were converted 240 into corresponding particle concentrations and size. 241 242 243

250
Production, separation, and quality control of input materials 251 Supernatants were collected from cultured human cell lines: H9 (T-lymphocytic) and U937 (pro-252 monocytic). EVs were separated by size exclusion chromatography and ultrafiltration or 253 differential ultracentrifugation ( Figure 1A). Marker expression and morphology were assessed 254 by WB ( Figure 1B and Supplementary Figure 1) and TEM. WB revealed characteristic CD63 255 and CD81 expression patterns, with CD81 above the limit of detection only for H9 ( Figure 1B). 256 Heterogeneous EV populations were observed by TEM in each sample preparation method 257 ( Figure 1C). Additionally, we confirmed size and purity of silica spheres and polystyrene spheres 258 using TEM ( Figure 1D In addition to particle size, we also assessed counts. For SP-IRIS, a mean of around 3000 SS 279 particles were detected per printed antibody spot ( Figure 3A), with no overall differences 280 between groups of antibody spots (i.e., three spots per chip each of three tetraspanins and an 281 isotype control; note that no differences would be expected, since particles were dried onto the 282 chips). However, per-spot events overall ranged from <2000 SS particles per spot to >4500 SS 283 particles per spot ( Figure 3A). SP-IRIS performed similarly for PS. There were no differences 284 between antibody groups, with a mean of around 1400 events/antibody spot ( Figure 3B Figure 3A and B, respectively). Some outliers were observed for CD9 (SS) or 291 CD63 (PS); however, none exceeded 1000 events. Particle concentrations were also measured by 292 NTA, MRPS, and NFCM. For SS ( Figure 3C), MRPS estimated a concentration approximately 293 one log higher than NTA (5.1´10 11 particles/mL vs. 5.4´10 10 particles/mL, respectively), with 294 NFCM in the middle (1.7´10 11 particles/mL). For PS, all three methods were in close agreement 295 ( Figure 3D). Furthermore, the measured concentration was very close to the nominal PS 296 concentration of 1´10 12 particles/mL ( Figure 3D, dotted line). 297 298 Biological particle sizing 299 EV preparations from H9 and U937 cell supernatants enriched by ultrafiltration and SEC (SEC 300 EVs) or by differential ultracentrifugation (100K EVs) were next measured using each platform. 301 For H9-derived materials, SP-IRIS returned an almost identical size distribution profile for both 302 EV enrichment methods ( Figure 4A). In contrast, NTA, MRPS, and NFCM measured more 303 particles at smaller diameters for the 100K EVs compared with the SEC EVs with roughly 304 similar particle size distributions ( Figure 4B-D). However, substantial variation between 305 replicates might limit the conclusions that can be drawn from this observation. For U937-derived 306 materials, SP-IRIS and NTA ( Figure 4E,F) detected more particles at smaller diameters from the 307 100K EVs compared with the SEC-EVs, again with roughly similar particle size distribution. 308 MRPS produced equivalent particle size distribution and particle number between the two 309 enrichment techniques ( Figure 4G ). In contrast, NFCM detected a higher particle count of 310 smaller particle diameters from the SEC EVs than the 100K EVs, with the particle size 311 distributions significantly different. Again, variability between replicates limits conclusions. 312 Overall, the results are broadly consistent with the reported power-law size distribution of EVs 313 [21,22] and the expectation that UC pellets may contain non-EV extracellular particles (EPs) 314 around the same size as EVs [1]. 315 316

Biological particle counting 317
Particle counts were next assessed. As before, we present the SP-IRIS data separately because 318 this platform does not provide an overall count, but rather a number of events detected on chips 319 printed with antibodies (shown here: to CD81 and to CD63 plus an isotype control). Consistent 320 with protein assay results, SP-IRIS shows that more H9 particles were captured by anti-CD81 321 than by anti-CD63 ( Figure 5A) and that U937 particles could be captured by CD63 capture 322 antibodies and not CD81 capture antibodies ( Figure 5B). For the remaining three platforms, 323 which measure overall concentration, several trends were apparent ( Figure 5C,D). First, for both 324 the H9 and the U937 source, and for both EV separation methods, data were consistent with the 325 results of SS counting in that NTA, NFCM, and MRPS measurements ordinally ranked from 326 lowest particles/mL to highest particles/mL. Secondly, MRPS and NFCM measured greater 327 particle concentrations for 100K EVs than for SEC EVs (corrected for processing and dilution), 328 although NTA results were similar. Finally, this is in contrast to results for the PS particles, 329 where the three techniques produced equivalent particle counts. 330

331
Single particle phenotyping by fluorescence 332 The SP-IRIS results represent a type of single-particle phenotyping since diameter is measured 333 for individual particles captured by antibodies and thus putatively positive for an antigen. 334 Captured particles can additionally be probed with fluorescently labeled antibodies. For chips 335 incubated with H9 EVs (Figure 6A,B), EVs captured by CD81 were generally positive for CD81 336 by fluorescence, and many also appeared to be CD63 positive. In contrast, CD63 capture spots 337 were largely devoid of fluorescence, as were (most) control capture spots. For chips incubated 338 with U937 EVs (Figure 6C,D), events on CD63 capture spots were also positive for CD63 by 339 fluorescence. CD81-linked fluorescence was at background levels for all spots. Note that 340 numbers of "positive" events are higher in fluorescence mode than with SP-IRIS ( Figure 5A,B), 341 likely, as discussed later, because fluorescence detection is more sensitive than reflectance 342 imaging. 343 For the two remaining platforms with fluorescence capabilities, NTA and NFCM, results are 344 shown as percent of total particles ( Figure 6E-H). Approximately 40% to 50% of detected 345 particles from H9 cells were positive for CD81 according to fluorescent NTA, while little to no 346 CD81 signal was detected for U937 materials, consistent with protein assay results. However, we 347 could not detect CD63-linked signal by fluorescent NTA for any sample. In contrast, NFCM 348 detected either CD81 or CD63 on a small percentage of particles. The percentages were similar 349 for the two tetraspanins for H9-derived particles. For U937 material, CD63-positive particles 350 were more abundant than CD81-positive particles. No major differences between the SEC and 351 100K separation methods were apparent according to these data ( Figure 6E-H). in particle concentration and size measurements. Also noteworthy is that the NFCM platform 390 distinguished subpopulations of SS particles, but that this is likely because the same beads are 391 used to calibrate the instrument. While detecting the expected concentration of high refractive 392 index PS particles, NTA was unable to resolve individual particle populations and instead 393 characterized the SS particles as a broad population distribution centered on an "average" size. 394 To be sure, it may be possible to resolve discretely sized particle populations using NTA with 395 mixtures at different ratios of sizes. We could not do so with the mixtures we used. Whether this 396 matters for biological particles is unclear. It does not seem that biological samples would contain 397 unique EV subpopulations with exquisitely defined sizes, except perhaps for samples from 398 sources infected with specific enveloped viruses. NTA does seem to be capable of detecting 399 shifts in population distributions, and this capability might be more important for biological 400 particles than resolving subpopulations. 401 On counting by SP-IRIS/fluorescence. One clear finding of our study is that, in our hands, 403 neither SP-IRIS label-free measurements nor subsequent fluorescence detection could be used 404 directly to estimate overall particle concentration. Instead, SP-IRIS is best used to understand 405 ratios within populations and for single-particle phenotyping. Even when PS beads were dried 406 onto chips, the measured concentration was approximately one-seventh of the expected 407 concentration. While uneven drying could contribute, it seems that PS particles non-specifically 408 adhered to the chip without a washing step, were undercounted slightly. For biological particles, 409 the problem is compounded since only a subset of EVs bind to any given affinity reagent "spot." 410 Binding is determined by diffusion (which is slow for EVs), presence and density of recognized 411 surface markers, and affinity characteristics of antibody-to-antigen binding. The bound 412 population of particles remaining after wash steps is only a miniscule proportion of the total in 413 the input material and cannot be used to determine overall concentration. Interestingly, 414 fluorescence results often indicated higher particle concentrations than returned by label-free 415 counting, even though particles positive for a particular antigen are expected to be only a subset 416 of the captured population (different antigens) or to approach equality (if the capture antigen is 417 targeted and antigen is abundant). Counts are higher because fluorescence detection is more 418 sensitive than label-free. That is, fluorescence detects positive particles that may be below the 419 limit of label-free detection. 420 421

Did any platforms identify differences between EV separation technologies? For both 422
biological sources of EVs, we used two methods of EV separation: dUC (100K EVs), which has 423 been the most common method for EV separation [32,33], and a combination of filtration and 424 size exclusion chromatography (SEC EVs) [34,35]. According to some evidence in the literature, 425 dUC leads to more protein contamination and aggregation and damage of EVs [36][37][38]. It should 426 be noted that alternative viewpoints can also be found [16]. However, protein particle 427 contamination might be expected to introduce more and smaller particles. This outcome is indeed 428 observed based upon TEM background and particle profile shifts towards smaller particles for 429 several of the platforms. For SP-IRIS with fluorescence, it is also interesting that tetraspanin 430 positivity is higher for samples obtained by SEC than with dUC. On the other hand, evidence of 431 aggregation by dUC is not apparent in the data presented here. We cannot rule out aggregation, 432 however, only that the techniques used here did not appear to detect it. 433 434 Single-particle phenotyping. For the three techniques with single-particle phenotyping 435 capabilities (SP-IRIS, NTA, and NFCM), each has advantages and drawbacks, as covered above, 436 all can potentially provide true single-particle phenotyping data. SP-IRIS was able to achieve the 437 most "multiplexed" detection, in that signal could be obtained above background for up to three 438 fluorescent channels. At the time of our evaluations, the NTA platform we used could not 439 perform simultaneous multi-channel measurements and thus was not a true single-particle 440 multiplexing platform. Instead, sequential filter switches were required, such that the same 441 particles could not be tracked in different channels. 442

443
In Table 2, we attempt to summarize our findings and views about the four investigated 444 techniques. Detectable size ranges for biological particles: these should be considered to be 445 rough estimates. If we accept the assumption that EVs follow a power-law size distribution (the 446 smaller, the more abundant, with lower bounds defined by membrane curvature constraints), then 447 no evaluated platform effectively detects the very smallest particles. However, SP-IRIS, MRPS,448 and NFCM appear to detect slightly smaller particles than NTA under the conditions and settings 449 we tested. For NTA, MRPS, and NFCM, linear ranges for particle concentration for all 450 instruments begin around 1´10 7 particles/mL and extend from about one order of magnitude 451 (NTA) to multiple orders of magnitude (MRPS). This spread is important, since the wider the 452 range, the fewer time-consuming concentrations or dilutions must be done to place an unknown 453 particle population into the measurable range. SP-IRIS is a special case, since particles are 454 captured by affinity, and overall concentration cannot easily be estimated. In our hands, particle 455 concentrations must be high (>>1´10 7 particles /mL) even for abundant antigens. Furthermore, 456 the optimal captured particle counts are roughly 3000 to 6000 per antibody spot (although this 457 may vary). To hit a very tight "sweet spot", many trial dilutions may be needed. Furthermore, the 458 optimal dilution may well be different for different antibodies on the chip because of different 459 percentages of EVs positive for a particular antigen, per-EV antigen abundance, and antibody 460 performance. Hence, dilutions are usually most important and time-consuming for SP-IRIS. 461 Related to dilution is the volume of input material required for a single reading. Assuming each 462 platform can measure 1´10 7 particles per mL, the required volume of a dilution at this 463 concentration ranges from 5 ul (MRPS) to around 1 mL (NTA). Of course, the actual 464 volume/number of EVs needed will also depend on the number of concentrations/dilutions 465 required to reach the measurable concentration range. The input volume difference is also 466 inconsequential for highly abundant materials, but may be important for low-abundance EV 467 samples. If done, optional calibration steps are rapid for NTA and MRPS (around 20 minutes). 468 For NFCM, we find that calibration can be as short as 20 minutes but can sometimes take longer. 469 Time for sample dilutions is most difficult to estimate, but is expected to correlate inversely with 470 the range of measurement for each platform. Read time ranges from five minute to about half an 471 hour per sample. Note that the times we indicate are for sizing and counting only. Optional 472 fluorescence measurements for the relevant platforms would in some cases add processing time 473 for antibody incubations and removal, as well as for read times (except for NFCM). For SP-IRIS, 474 we should also note that, although the total hands-on and read time is longer than for other 475 techniques, each reading includes on-chip replicates, multiple capture antibodies, and up to three 476 fluorescence readouts per capture antibody. 477

478
Costs for the platforms include initial outlay, disposable costs, and maintenance costs. For 479 acquisition, the MRPS system is most economical, while NFCM is the most expensive. For basic 480 counting and sizing, operating costs for NTA and NFCM are negligible. Adding optional 481 fluorescence increases these costs by amounts that are antibody-dependent. The MRPS system 482 uses disposable cartridges that currently cost USD 8 to USD 12 each. The SP-IRIS platform has 483 the highest disposable costs, with each sample requiring at least one chip at USD 50 to >USD 484 100 each. Since optimal dilutions are difficult to achieve and may be different for different 485 capture materials on the same chip, multiple chips may be needed for the same sample. Chips 486 also cannot be chemically stripped and re-used, at least not in our hands (Mallick and Witwer, 487 unpublished data). Furthermore, we find that chips often go unused and are thus "wasted" as a 488 result of a short shelf life of only several weeks. Since chips may take several weeks (or more for 489 custom) to procure, the three-week shelf life requires excellent planning and a lack of 490 unexpected difficulties in the laboratory; otherwise, the investment is wasted. As noted, though, 491 under optimal conditions, the platform provides multi-dimensional information that may justify 492 these costs and logistical challenges for some users. We should also mention that chips for the 493 SP-IRIS and MRPS instruments are currently available only from the instrument manufacturer 494 for that particular measurement technique. As for maintenance costs, we are unable to estimate 495 them at this time. 496 497 In conclusion: 498 • No evaluated platform is necessarily "better" or "worse" than others; rather, it is 499 important to be aware of the capabilities of each platform with respect to each particle 500 population of interest. 501 • Rather than relying on a single platform, consider using orthogonal technologies. 502 • Both acquisition and recurring costs should be considered before acquiring a platform. 503 Appropriate reference materials are needed for better evaluation of single particle phenotyping 504 capabilities, including multiplexed phenotyping.  as corresponding cell lysates from H9 and U937 using antibodies specified in Table 1 for SS using MRPS. Repeat 1 can also be found as an inset in Figure 2C.