Using single-vesicle technologies to unravel the heterogeneity of extracellular vesicles

Extracellular vesicles (EVs) are heterogeneous lipid containers with a complex molecular cargo comprising several populations with unique roles in biological processes. These vesicles are closely associated with specific physiological features, which makes them invaluable in the detection and monitoring of various diseases. EVs play a key role in pathophysiological processes by actively triggering genetic or metabolic responses. However, the heterogeneity of their structure and composition hinders their application in medical diagnosis and therapies. This diversity makes it difficult to establish their exact physiological roles, and the functions and composition of different EV (sub)populations. Ensemble averaging approaches currently employed for EV characterization, such as western blotting or ‘omics’ technologies, tend to obscure rather than reveal these heterogeneities. Recent developments in single-vesicle analysis have made it possible to overcome these limitations and have facilitated the development of practical clinical applications. In this review, we discuss the benefits and challenges inherent to the current methods for the analysis of single vesicles and review the contribution of these approaches to the understanding of EV biology. We describe the contributions of these recent technological advances to the characterization and phenotyping of EVs, examination of the role of EVs in cell-to-cell communication pathways and the identification and validation of EVs as disease biomarkers. Finally, we discuss the potential of innovative single-vesicle imaging and analysis methodologies using microfluidic devices, which promise to deliver rapid and effective basic and practical applications for minimally invasive prognosis systems. Understanding the heterogeneity of extracellular vesicles is crucial for unraveling their functions. This review describes the benefits, challenges and contributions of the state-of-the art methods used in single-vesicle analysis.

S ince the description of 'minute bodies' found in a piece of cork by Robert Hooke in 1665 1 , both our scientific knowledge and technical abilities have increased enormously. As techniques have become more accurate and intricate, so has our understanding of biological processes and structures. Technological advances in the field of imaging have resulted in the identification of cell structures, such as the mitochondria 2 and nuclei 3 , and the discovery of different levels of cellular complexity. In 1967, Peter Wolf visualized 'platelet dust' in fresh platelet-free blood plasma using an electron microscope 4 ; thus, a mammalian vesicle-like structure was described for the first time. Gradually, these vesicles were characterized in more detail. It has been established that they are released by all kinds of cells (prokaryotic and eukaryotic) into the extracellular milleu [5][6][7][8] . The process of vesicle secretion is conserved throughout evolution, suggesting that such vesicles are likely to have specific roles in cellular and organismal development and survival 9 . Indeed, later discoveries showed that secreted vesicles participate actively in many physiological processes in mammals, for example coagulation, inflammatory response, cell maturation, adaptive immune response, bone calcification and neural cell communication, among others 10,11 .
In addition to their critical functions in normal physiology 12,13 , secreted vesicles mediate in several pathological processes 14,15 , such as the establishment of premetastatic niche during cancer progression 16,17 .
Nowadays, these secreted vesicles are extensively reported and widely known as extracellular vesicles (EVs). EVs are heterogeneous, nano-to micrometer-sized, bilayer lipid containers secreted by most cell types. They are multipurpose carriers that can contain a wide variety of cargos such as lipids, proteins, metabolites, sugars, RNA (mRNA, miRNA, siRNA) and even DNA 10 . When they are taken up by recipient cells, they trigger intracellular signaling through EV surface molecules or by the release of cargo into cell compartments via endocytic pathways. These processes can further activate downstream genetic or metabolic pathways in the recipient cell 11,18 . In mammals, EVs have been found in body fluids such as plasma, urine, saliva, breast milk and seminal fluid, among others 10 . They are classified into three different categories according to their biogenesis mechanisms and biophysical properties: • Exosomes: typically 30-150 nm in diameter, derived from intracellular endosomal compartments • Microvesicles: 100-1,000 nm in diameter, produced by outward budding and pinching-off the plasma membrane • Apoptotic bodies: 50-5,000 nm in diameter, released as blebs by cells undergoing apoptosis 10,19 In a systematic review of guidelines for this field, the International Society for Extracellular Vesicles (ISEV) 20 endorsed a categorization of EVs isolated using ultracentrifugation into large, medium and small EVs. However, it is important to note that ultracentrifugation precipitates not only vesicles but also lipoproteins, viruses, protein aggregates, ribonucleoprotein complexes and exomeres [21][22][23] , which provides an additional layer or diversity but also a bias upon analysis of EV samples. Furthermore, there is evidence that exosomes, microvesicles and apoptotic bodies contain subpopulations with unique roles in biological processes 24,25 . These subpopulations are tightly integrated with a broad range of biological processes and exhibit a wide range of functionalities, which makes them an outstanding source of potential biomarkers for early diagnosis, drug delivery systems for therapeutics, or vaccine production systems 15,[26][27][28] .
During the last few decades, the interest in EVs and their applications has grown considerably. Many articles and reviews have focused on the functional role of EV heterogeneity. Their role in specific biological processes, such as cargo trafficking or regulation of signaling pathways, and their potential as biomarkers have also been examined 11,15,25,29,30 . Nonetheless, most of these studies examine the vesicles in bulk and use ensemble-averaging assays. Although such methods have been proven useful in specific cases, it is important to realize that the extensive heterogeneity of structure, composition and function of single vesicles is masked in such assays 24,[29][30][31] . For example, the inability to detect the heterogeneity of molecular states of reaction pathways, individual proteins or nucleic acids may lead to a misinterpretation of ensemble measurements 32,33 . Recent developments in single-vesicle analysis (SVA) have opened new opportunities for the examination of heterogeneity within EV (sub)populations at the individual EV level and their characterization on the nanometer scale 9,29 . This new information is paramount for understanding the biological functions of EVs and for their potential clinical use.
Different EV populations and subpopulations can be isolated according to their physicochemical properties, yet the existing isolation technologies are intricate and still need to be further developed 24,29 . There are five main groups of techniques for sorting EV populations and subpopulations, based on ultracentrifugation, size, immunoaffinity capture, polymer precipitation and microfluidics 19 . Since each technique type sorts EVs using a different principle, each method can yield different EV subpopulations from the same sample 34,35 . Moreover, the highly concentrated EV preparations might contain contaminants, such as large protein aggregates and lipoproteins, left behind by some of the isolation techniques 36,37 . Interestingly, different approaches can also affect the physicochemical surface characteristics of EVs 38 .
In consequence, some techniques tend to enrich or discriminate against specific EV populations. As sorting EVs into populations is usually based on physical properties only, we have to assume that such classification is largely arbitrary with respect to the composition or function of EVs. Vesicle isolation and enrichment techniques can help to yield more homogeneous EV subpopulations, albeit only for particular technique-specific parameters. In conclusion, although various isolation methods could help drive EV analysis toward a single-vesicle approach 9 , many different compositions and functionalities are still expected to be found within such EV populations.
In this review, we describe the current methods used to study single-vesicles, and their contributions to the understanding of EV biology and biomarker discovery. Single-vesicle experiments can deliver direct information on the heterogeneous composition of EVs. They reveal multiple molecular states that govern EV functionality and transport and provide statistically valid information often lost in large ensemble experiments 19,29,39 .
EV heterogeneity and biomarker discovery There are many technological challenges to be met in the development of EV-based diagnostics. The relevant vesicles must be identified and isolated from complex biofluids, and a specific disease-related EV population or population mix has to be detected. Biomedical studies of EVs often focus on seeking suitable biomarkers for the diagnosis of various diseases 40,41 . For example, the functional role of EVs in various types of cancer has been extensively studied 14,16,17,27 . In this review, we concentrate on prostate cancer (PCa) diagnostics as an example application to which the SVA of EVs has made important contributions. According to the World Health Organization, PCa is among the most frequently diagnosed types of cancer, accounting for approximately a quarter of all cancer diagnoses in Europe 42 , yet the lack of sensitive diagnostic tools and insufficient knowledge of the mechanisms of cancer emergence and progression are of major concern. PCa is a heterogeneous pathological state, both in the primary tumor in the prostate tissue and at the metastatic stage. It is unfortunately not advisable to examine PCa diversity relying solely on tissue biopsies, since these are highly invasive procedures and do not guarantee an effective and reliable diagnosis [43][44][45][46] . The serum prostate-specific antigen (PSA) test-still the cornerstone of PCa screening-is particularly questionable. Up to 40% of men undergo unnecessary biopsies as a result of poor specificity of the assay.
Remarkably, liquid biopsy has a potential for marker identification and provides better evidence of PCa diversity than the conventional solid tissue biopsy. In particular, prostate-and PCa-derived EVs and concomitant markers are highly abundant in urine, blood and ejaculate samples 47,48 . Hence, these body fluids could be used for detecting and measuring the progression of the disease. For example, the EVs released by PCa cells carry unique prostate-specific membrane proteins (e.g., TMPRSS2, STEAP2, PSMA, PPAP2A) that enable the detection of pathogenic prostate EVs and their capture for ex vivo characterization 49 . Although these data show that liquid biopsies may be highly informative and minimally invasive procedures, the methods for vesicle isolation, characterization and identification for disease diagnostics remain challenging. Up to date, there are no standardized operating procedures for vesicle isolation and characterization for different types of samples and diseases, which complicates the employment of liquid biopsies as a clinical source for EV biomarkers.

SVA techniques for biological characterization of EVs
Most of the studies reviewed in this article examine the three problems that can be addressed by taking advantage of SVA techniques: (1) characterization of EV heterogeneity including subpopulations, surface (membrane protein and lipid) composition and vesicle content, (2) structural studies of EV membrane and soluble proteins and the assays to probe the metabolic activity of these proteins in a native-like environment and (3) characterization of the EV content and function depending on the cells of origin. The last task presents an interesting dichotomy: do the EVs reflect the properties of their cells of origin, or are they completely independent communication assets? On the one hand, it has been reported that EV surface and content depend on the parental cells [50][51][52] . On the other hand, some recent reports describe several EV subpopulations, with a range of different functionalities, originating from the same cell type 9,18,19,29,53,54 . Intriguingly, another recent article using SVA techniques demonstrates that the T2SS-like family of proteins is, in fact, responsible for selective cargo loading into EVs generated by the microorganism Shewanella vesiculosa 55 .

Single-vesicle techniques
As increasing numbers of researchers have highlighted the importance of accurate EV (sub)population sorting and phenotyping, so far, more than 20 new single-vesicle techniques have been developed 9,19,29 . Many of these use microfluidic devices designed to integrate various technologies to improve EV sorting and detection. Moreover, several of these methods have been used for characterizing EVs at the single-vesicle level [56][57][58][59][60][61][62][63][64][65][66][67] . Some of these techniques can directly provide information on vesicle surface, content, size and shape, while others may require an upstream physicochemical characterization of the selected EV subpopulations to conduct surface profiling, monitor the expression of biomarkers and quantify them in body fluids. These technological advances should help to design new diagnostic devices for small sample sizes, using noninvasive or minimally invasive methods.
Twelve different methods are presented in Table 1 and discussed in detail in the following sections. Some of these methods utilize labeling techniques (such as fluorescence or nanoparticle coating) to visualize the EVs, and others work as label-free systems (Fig. 1). It is important to note that, in some cases, label-free approaches might hinder the detection of EVs because they often produce weak signals, which can be enhanced using a supporting labeling technique.
Label-free methodologies Nanoparticle tracking analysis (NTA) is a technique based on the Brownian motion of microparticles in suspension, and it is used to determine the size distribution in particle populations 56,68 . In this approach, microparticles are detected by scattering the light of a laser beam, which is tracked and recorded at video frame rates. However, this approach has some disadvantages and limitations. For instance, the accurate assessment of particle size distribution requires specific track lengths, a steady temperature and a large number of replicates to provide robust results. Care should be taken when comparing different samples because variations in buffer viscosity and microparticle concentration introduce statistical errors. Moreover, the close proximity of two particles can result in overlap of the scattering signals. Accurate detection of particles with a diameter <60 nm is challenging, regardless of the NTA machine used 69 . Furthermore, vesicles cannot be discriminated from other particles, such as protein aggregates or virus particles. The vesicles can be probed specifically, and undesired particles excluded from the analysis only by employing fluorescent markers, yet only a fraction of EVs may carry known markers that can be used for labeling a specific subpopulation. General fluorescent labels (such as lipophilic carbocyanines DiO or DiI) can be used instead. However, it is important that any nonattached label be removed since this can mask the fluorescence signal emitted by labeled EVs 19 .
Raman tweezers microspectroscopy (RTM), also known as laser tweezers Raman spectroscopy (LTRS), can be employed to examine the chemical content of EVs. This approach can be used to investigate both the surface and the internal volume of single EVs, revealing specific biomolecular signatures of proteins, lipids, nucleic acids and carotenoids as major contributors 57,60,61,[70][71][72][73][74][75][76][77] . RTM is an inelastic scattering-based method. It employs a tightly focused laser beam for both optical trapping of single (or very few) vesicles in aqueous medium and excitation for subsequent Raman scattering, which provides a vibrational fingerprint from the trapped constituent biomolecules. The main inherent advantage of RTM lies in the signal linearity, which allows both qualitative and quantitative biochemical characterization of single EVs. This method is also label-free and provides data with high information content 57,71,74 . The main disadvantage is that the scattering efficiency is usually very low and thus provides a rather low level of informative Raman signal. As a result, an extended data collection time is required. Therefore, RTM, with a typical processing capacity of 0.2 particles per min, is not considered a high-throughput methodology 71 . RTM can, however, be used to obtain interesting, unique information not only for EVs 57,60,61,70-77 but also for many other bioparticles such as liposomes, lipid layers on synthetic nanoparticles and others [78][79][80][81][82][83] .
Several methods have been developed to compensate for the low Raman signal strength in RTM. For example, the vesicle concentration can be increased by drop-coating deposition of the sample, followed by drying [84][85][86][87][88] . Unfortunately, this approach results in loss of information about individual EVs, as does any other analytical study of a bulk sample. Another strategy to increase the Raman signal is to use surfaceenhanced Raman spectroscopy (SERS). In this method, EVs can be exposed to various signal-enhancing nanoparticles and/or substrates to obtain a strengthened biomolecular signal 62,[89][90][91][92][93][94][95] . The main problem of label-free SERS is that the enhancement effect depends strongly on the distance between the biomolecule and the nanoparticle/substrate, and vanishes Using an immunofunctionalized probe tip. p On glass, mica or graphite (HOPG-type). q Working concentration spreads over a rather broad range and the optimal concentration strongly depends on EVs location in the microscope field. r Signal collection layer is usually less than~100 nm thick, depending on the excitation wavelength and objective numerical aperture. s Evanescent field is created when the angle of incidence of excitation beam is larger than the angle of total reflection, so that the excitation beam does not penetrate into the sample. t miRNA, surface proteins. u Donor fluorescence disappears, acceptor fluorescence appears. v Assessment of conformational fluctuations, folding pathways, macromolecular interactions, kinetics of structural changes, etc. w When distance between donor and acceptor fluorophores becomes <8-10 nm. x Some SRM approaches damage the sample to such an extent that only one measurement is possible, and others permit several SRMimaging analyses of a sample. y EV subpopulations with a specific surface protein in common.
at distances longer than a few nanometers 89 . Therefore, this method is mainly suitable for characterization of biomolecules on the outer surface of EVs. In addition, Raman modes corresponding to molecular vibrations perpendicular to the SERS surface are preferably enhanced 89 . As a result, the overall SERS vibrational spectrum is usually somewhat distorted, lacks reproducibility and is often difficult to interpret. In electron microscopy, a beam of electrons is emitted onto a sample in a vacuum environment. The wavelength of electrons is shorter than the visible light used in optical microscopy; thus, the method gives images of much higher resolution, typically below 1 nm 19 . Cryogenic transmission electron microscopy (cryo-TEM) is among the electron microscopy methods most commonly utilized for EV characterization. In contrast to the lengthy sample preparation needed for other TEM methods (usually taking hours), no heavy metals or fixatives are added, and no dehydration steps are required. This also limits sample damage and artifact effects, but yields lower-contrast images 96 . In cryo-TEM, the samples are prepared by rapid freezing, typically with liquid   57,97 . In this process, the water vitrifies, instead of forming ordered crystals, and the native structure of EVs is preserved 98 . The first exosome visualization was achieved using cryo-EM in 2008 99 . Since then, this technique has successfully revealed EV polymorphism by imaging the membrane bilayers, EV structures and internal features of individual EVs 57,63,96,100 . Even though cryo-TEM is an extremely useful technique for high-resolution visualization of EVs, this approach is relatively low-throughput. Cryo-TEM images typically only contain a few EVs (although the throughput could be enhanced by using automated search). In addition, cryo-TEM images provide only limited information regarding EV composition. To overcome this problem, nanoparticles functionalized with immunogoldlabeled antibodies targeting markers of interest have recently been employed to characterize the biochemical composition of the EV surface 19,101 .
Yet another type of microscopy method used for SVA is atomic force microscopy (AFM), which exploits the interaction between a probing tip and a sample surface. The deflection of the probing tip caused by interaction forces is detected and recorded using a laser and a sensor 102 . AFM allows an accurate morphological and mechanical characterization of EVs; its lateral resolution is 1-3 nm, and the vertical resolution <0.1 nm 102 . Typically, visualization of a few EVs using AFM is labor-intensive and time-consuming in comparison with other microscopy methods. However, a relatively high-throughput AFM-based method has been reported that measures the size and stiffness distribution of 100 vesicles within 1 h 65 . It is important to note that the tethering surface, to which the EV is bound, strongly affects the shape of the EVs. Therefore, the vesicles must be bound to a perfectly flat surface [103][104][105] . To characterize the biochemical properties of an EV surface, either the probing tip or the surface itself can be further (immuno)functionalized 106,107 . AFM can also be coupled with infrared spectroscopy (AFM-IR), allowing simultaneous measurements with a finer spatial resolution. AFM-IR has been extensively utilized in various applications; however, few papers report its implementation in the single-EV field 66,108 . We assume this is because the weak IR signal thwarts reliable characterization of individual vesicles.
Single-particle interferometric reflectance imaging sensor (SP-IRIS) is employed in assays based on interferometric imaging. It is used to detect individual enhanced scattering signals from the bound vesicle. The signals are produced by the interference between the scattered field from a vesicle and the reference field reflected off the layered substrate 109,110 . The method can detect several surface biomarkers and simultaneously measure the size of individual EVs. It can be used to accurately count and distinguish individual vesicles, with a low level of false positives and negatives 110 . However, as the lateral resolution of the microscope (~400 nm) could accommodate several small vesicles, some detected signals could be erroneously assigned and categorized as larger vesicles instead of several smaller vesicles. This could be an issue especially in highly concentrated sample preparations 111 .

Label-based methodologies
Label-based methodologies are strongly dependent on the detection of a signal from a fluorescent protein, immuno-or lipophilic fluorophore or signal-enhancing nanoparticles. High-resolution flow cytometry (hrFC) is one of the first techniques extensively employed for individual EV analysis. hrFC can be used to quantify the size distribution and diversity of EV populations by detecting multiparametric scattered light and fluorescence emitted by the labeled vesicles. This fluorescence assay can be used to characterize the vesicle population by profiling the protein or nucleic acid content using antibody-fluorophore conjugates. However, any remaining free fluorescent dyes in the sample will cause high background fluorescence. This can be avoided by using density-based ultracentrifugation to purify labeled EVs, which leaves the nonreacted dye in the supernatant and sediments the vesicles into the pellet fraction 112,113 . Furthermore, multiple EVs (or particles) arriving simultaneously at the flow cytometer detector may be identified as single particles. This phenomenon is known as the swarm effect. The danger of such misidentification limits the concentration range within which the EVs (or particles) can be characterized effectively and makes it necessary to examine multiple diluted samples.
Fluorescence microscopy is an imaging technique particularly useful in localizing lipophilic fluorescent dyes or fluorescently labeled targets (either using fluorescent proteins or fluorescent dye-conjugated antibodies) in cells, tissues or EVs 9 . Another, rather elegant approach, now commonly used in the SVA field, is total internal reflection fluorescent (TIRF) microscopy. It can be used in an aqueous environment to image selectively the fluorescent molecules located near a highly refractive solid substance 114 . TIRF exploits the reflection of an excitation light beam at a high incident angle, typically between 60°and 80°, at which the beam of light is completely reflected by the glass-water interface. This reflection phenomenon generates a very thin electromagnetic field, called an evanescent wave, which is parallel to the substrate surface. This enables limited specimen illumination and thereby eliminates out-of-focus fluorescence and enhances the signal-to-noise ratio 115 . TIRF is predominantly used for studying intracellular single-vesicle processes such as endocytosis or exocytosis, cell-substrate contacts or internalization of plasma membrane receptors 114,115 . It can also directly localize fluorescently labeled molecules in EV preparations and allows tracking EVs in tissue preparations. However, the fluorophores can be excited only within a few hundred nanometers from the solid substrate, and the calibration of the incident angle can be difficult (depending on the setup) 114,115 . Moreover, the fluorophore instability and gradual photobleaching (although less pronounced than in other light microscopy techniques) during prolonged illumination might produce misleading results 9,116 .
Fluorescence (or Förster) resonance energy transfer (FRET) is a phenomenon where the excitation energy from a fluorophore is transferred nonradiatively to another fluorophore. This happens via resonance energy transfer at distances <10 nm. FRET imaging offers unique opportunities for the assessment of kinetic and structural dynamics and studies of the interaction and fusion events between EVs and cells [117][118][119] . Notably, this imaging-based technique is capable of producing a considerable amount of single-particle and single-vesicle fluorescence data very fast [120][121][122][123] . However, fluorescent signal fluctuations due to a low signal-to-noise ratio and poor photostability of certain dyes might lead to changes in the FRET signal that are unrelated to the biological processes. Like in other fluorescence-based techniques, the presence of multiple fluorophores within the observation volume may result in ensemble averaging of the population 119,124 .
Super-resolution microscopy (SRM) is one of the most advanced applications of fluorescence imaging. It can be used to visualize biological features smaller than the optical diffraction limit and, therefore, below the conventional optical microscopy resolution limits (which are typically limited to~250 nm axial and~500 nm lateral resolution). This attribute provides an important advantage in imaging single EVs and permits investigation of their physiological functions 67,[125][126][127][128][129][130] . Briefly, SRM techniques can be divided into two groups: (i) methods based on spatial patterning of the excitation light and (ii) methods based on single-molecule localization.
Excitation-patterning methods include structured illumination microscopy, in which the specimen is illuminated in a striped pattern 9,131 and stimulated emission depletion microscopy, which sharpens the excitation laser focus using a second laser that temporarily bleaches the fluorophores surrounding a small observation volume in the specimen 132,133 .
Single-molecule localization methods detect fluorescence emitted from spatially isolated photo-switchable or blinking fluorophores to determine their position 133 . Photoactivation localization microscopy (PALM) relies on photoactivatable fluorescent recombinant proteins 134 , whereas stochastic optical reconstruction microscopy (STORM) takes advantage of fluorophores that blink in a noncontrolled fashion. In both cases, only a small subset of fluorophores will be emitting simultaneously, allowing the precise localization of the fluorophores and reconstruction of the complete image at high spatial resolution 9 . Importantly, single-molecule localization approaches can be used in combination with TIRF, improving the signal-to-noise ratio and shortening imaging time 116 . Furthermore, all SRM techniques are based on the optics of classical diffraction-limited far-field light microscopes. This makes them compatible with existing sample preparation procedures, and also permits them access beyond the surface of a specimen. SRM techniques are also often minimally invasive 133,135 . It is important to note that most SRM approaches work only on fixed samples; thus, one should always be aware of potential artifacts introduced by the fixation method 9,136 . Another relevant aspect to consider is that, in general, the labeling of proteins using fluorescence markers or other tags might affect their localization, interaction partners and function 135 . In addition, although lipophilic or genetic labeling could allow visualization of single vesicles, lipid labeling in some cases might result in unspecific labeling or dye aggregates.
Digital droplet PCR (ddPCR) can be employed to distribute single EVs into individual droplets, which allows amplification and characterization of their genetic cargo 59,137,138 . Such cargo is usually RNA-based and mainly comprises miRNA, mRNA and noncoding RNA. In ddPCR, the EVs are tagged using anchoring molecules or antibodies and further distributed into microfluidic chambers according to their surface markers 59 . This methodology enables high-throughput quantitative analysis of EV content and can be used for the identification of biomarkers [139][140][141][142] . So far ddPCR use has been limited to validation purposes; however, it can be used as a screening technique (compromising its high-throughput capabilities) 19,143 . Interestingly, ddPCR has been already adopted for multiple mutation analysis to examine specific mutations in distinct populations of EVs. Next-generation sequencing could allow for parallel analysis of multiple mutations in many genes 144 .
Finally, SERS nanotags functionalized with biorecognition molecules (such as target-specific antibodies) can be used to bind specifically to target EVs expressing the biomarker of interest. This approach is gradually becoming an important alternative to fluorescent molecular probes 94,145-149 . The major advantage of SERS labeling lies in the superior photochemical stability of Raman reporters compared with fluorescent labels, due to the vibrational nature of the generated signal. Moreover, several high-throughput Raman/SERS screening platforms for characterizing cells and EVs have been recently reported 70,[150][151][152][153] .

Recent advances in the EV field due to SVA
The recent breakthroughs in SVA techniques help to tackle the intrinsic limitations of ensemble EV measurements and analyses. In the following sections, we review the impact of SVA method development on recent advances and discoveries in the EV field (EV characterization, internalization, the role of EVs in cell-to-cell communication and biomarker discovery). In Table 2, we provide an overview of recent scientific articles describing the characterization of EVs, including studies of their heterogeneity and phenotyping of PCa-derived EVs. Current studies of EV internalization pathways and the role of EVs in cell-to-cell communication are summarized in Table 3. SVA techniques have been used in the successful identification and validation of a wide range of biomarkers for many different diseases. As an example, Table 4 lists recent discoveries in cancer research with a strong focus on PCa.

EV characterization
Several techniques mentioned in the previous section-NTA, cryo-TEM and flow cytometry-are established as customary characterization procedures for EV studies 141 . Notably, the results obtained employing these methodologies usually require validation using a complementary technique. This is because none of them is considered the gold standard procedure, and all of these approaches come with their own challenges and limitations 154 . For example, NTA is routinely utilized to obtain the size of EVs and quantify their abundance. However, since this technique is not vesicle-specific and can detect other particles, the reliability of the results must be cross-checked with other techniques. Therefore, EV studies that draw their conclusions solely from NTA are now uncommon, especially when compared with earlier EV characterization studies (from 2012 to 2015) when NTA was commonly utilized. Currently, NTA is often used as a supplementary characterization technique instead, as it requires minimal sample preparation to introduce fluorescent markers for specific EV population studies. Likewise, labeling-based flow cytometry serves as a high-throughput sorting method for the characterization of certain EV features; however, it needs some technological improvements in separating unbound dye and dye aggregates from EVs to be utilized as a unique characterization approach 155 . Direct visualization of EVs using microscopy lets researchers assess the shape and size of vesicles in their native state 156 and in various biological processes 157 , while other techniques, such as RTM, can be employed to examine further the morphology of bioparticles without direct visualization. For example, the effect of membrane lipid composition on the shape and size of giant unilamellar vesicles was first described using Raman tweezers, demonstrating that a decrease in cholesterol concentration increases the local membrane curvature and stretches the vesicle 82 .
A large diversity in the morphology of exosomes has been reported for many types of cells 96 . For example, in the early 2010s, plasma EVs were comprehensively characterized and phenotyped using cryo-TEM in combination with gold nanoparticle-based immunolabeling 158 . This study found that  platelet-free plasma samples contain a mixture of EVs with different morphologies, including spheres, cylinders and membrane fragments that are neither tubular nor spherical. Moreover, it has been shown that EVs come in a wide range of sizes 158 . Despite this morphological heterogeneity, only a minority of EVs in plasma expose phosphatidylserine on the surface. This is at odds with the classical theory of EV formation at the cell's plasma membrane, in which the loss of phospholipid asymmetry and exposure of phosphatidylserine precedes membrane blebbing and shedding [159][160][161][162][163] . The authors have suggested that some EV (sub)populations ought to be generated and regulated via different pathways 158

NATURE PROTOCOLS
of these EVs plays a key role in Parkinson's disease progression 164 . Another study, using NTA-based technology, has identified several EV populations released by human glioblastoma cells 165 . Moreover, fluorescence-based NTA approaches are being used to examine the concentration and particle distribution of specific EV subpopulations 166,167 .
The relative chemical abundance of major biomolecules comprising EVs, namely proteins, lipids, nucleic acids and carotenoids, can be obtained from the vibrational fingerprints acquired using RTM. Tatischeff et al. have shown that Dictyostelium discoideum in two different physiological states (i.e., cell growth and starvation-induced aggregation) produce EVs with drastically different biomolecular compositions 57 . In another study, Smith et al. have categorized four EV populations according to specific protein, phospholipid and cholesterol vibrational signatures, which are shared among several cell types from different species 71 . The authors have found that human lung carcinoma A549, human hepatocarcinoma Huh-7 and mouse embryonic fibroblast 3T3 cells have similar EV populations. In contrast, Kruglik et al. have reported direct Raman evidence of pronounced biomolecular heterogeneity of single EVs in the same sample (using rat hepatocytes and human urine) 74 . In their study, the heterogeneity was determined by quantitative measurements of nucleic acid concentration within single EVs, based on the intensity of the pyrimidine ring stretching band 74 .
Further heterogeneity studies performed with cryo-TEM and AFM have described the physical characteristics of EVs. One cryo-TEM analysis has reported that the shedding process and, hence, the type of EV released is strongly affected by external stimuli, such as lipopolysaccharides or starvation conditions, in a human leukemia cell line 168 . By using the tapping mode in AFM, biomechanical properties such as elasticity, stiffness and deformability of single EVs can be assessed 169 . It has been reported that rat hepatocyte-derived EVs are more fragile and easily warped than liver-progenitor mouse EVs 170 . Another study performed using AFM-IR has allowed, for the first time, probing of molecular constituents and structures of individual vesicles 108 . In this study, the researchers were able to differentiate between the molecular compositions of EVs derived from two subtypes of placenta stem cells. Moreover, their approach has allowed discrimination between protein aggregates and EVs. The examination of DNA, lipids and proteins using AFM-IR, in just a few vesicles, has remarkable potential in early disease diagnosis 108 .
SVAs have been extensively utilized in cancer research, for example, to characterize the EVs in PCa. NTA-based research has established that cancer cells produce larger amounts of EVs than do nontumorigenic cells 47,171,172 and that low extracellular pH increases the release of EVs from cancer cells 173,174 . Interestingly, NTA studies also support the idea  179 .
In summary, the highlighted reports of single-EV characterization demonstrate that EV morphology and composition are largely independent of cell origin and that certain EV (sub) populations are involved in various diseases. These discoveries indicate that morphologically different EV populations may be distributed according to their specific function and biogenesis pathway, rather than the cell type of their origin.
EV trafficking and signaling mechanisms SVA approaches have advanced EV research by tracking particular molecules and examining the changes in cells under different conditions. Early in 2020, FRET studies in T cells showed that the concentration of free zinc in cells is a major regulator of the maturation process in insulin-storing vesicles 180 . SVA has also contributed to discoveries reporting an increase in intracellular Ca 2+ and/or protein C under certain stimulating and activating conditions in red blood cells. These conditions alter cell morphology and cause an increase in the release of microvesicles 181 . Table 3 130 . The contribution of EVs to trans-synaptic tau transmission has been confirmed using cryo-TEM in another study 183 . This single-vesicle methodology has also been utilized recently to study the near-native 3D architecture of EVs secreted after infection with poliovirus. Cryo-TEM tomography has generated images of virions and viral structures contained in EVs before cell lysis 184 . Moreover, FRET microscopy has been used to track the triad protein VOR (paramount for the transfer of EV-derived components to the nucleus) 185 . This type of research has therapeutic potential for diminishing the progression of neurodegenerative diseases (in the case of the VOR complex, by inhibiting EV-mediated intercellular communication).
Cancer mechanisms have also been analyzed using singlemolecule techniques. Mannavola et al. performed a ddPCR experiment using osteotropic melanoma cells and observed that EVs could induce the upregulation of genes such as CXCR7 186 . Thus, EVs may act as chemotaxis agents and, hence, participate in the progression of cancer; however, additional research in this field is still required to achieve better understanding of how EVs contribute to cancer 186 .
Vesicle budding and shedding and the mechanical properties of the vesicles are poorly understood. Remarkably, a recent comparative review suggests that biomechanical analysis of single EVs provides key insights into their biological structure, biomarker functions and potential therapeutic functions 169 . Sorkin et al. used AFM to study these properties in erythrocyte and EV membranes under different conditions 187 . They established that stiffness is inversely proportional to the protein-lipid ratio and linked it to several different budding mechanisms 187 . On the one hand, budding of protein-rich soft vesicles is possibly driven by protein aggregation, and on the other, budding of stiff vesicles with low membrane-protein content is likely to be driven by cytoskeleton-induced buckling 187 . A further investigation comparing EVs originating from healthy erythrocytes and from those with hereditary spherocytosis has supported these observations. It also uncovered mechanical and vesiculation differences between these two EV populations with potential use as diagnostic parameters 188 .
Vesicle endocytic pathways have been investigated primarily using microscopy-based techniques. One major EV endocytosis pathway is mediated by the formation of clathrincoated vesicles (Fig. 2b). In this process, intracellular clathrin interacts with the membrane, producing a membrane invagination that will form an endosome through which the EVs can be internalized. The disassembly of the clathrin lattice surrounding coated endosomes is a mandatory last step in their life cycle. The recruitment of auxilin and Hsc70 (fluorescently labeled) was directly visualized using an inverted fluorescence microscope equipped with TIRF hardware and described as essential for clathrin-based internalization events 189 . The clathrin-driven uncoating is a variable process in which the endocytosing clathrin-coated vesicles remain proximal to the membrane for different periods prior to the scission of plasma membrane. The dynamics of clathrin-mediated endocytosis were assayed using fluorescently tagged proteins and TIRF microscopy 190 .
Fusion states, dynamics and mechanisms of vesicle internalization during single-vesicle fusion events have been directly examined using cryo-TEM, FRET and TIRF microscopy. Characteristics and kinetics of individual fusion events can be quantified for the lipids or DNA-lipid complexes involved in the process. Different fusion pathways exist: vesicles and cell membrane merge via a direct fusion of membranes (Fig. 2a) or using protein-mediated mechanisms (Fig. 2c). These mechanisms are involved in both endocytosis and exocytosis events. Examining individual giant unilamellar vesicles by fluorescence microscopy, it has been shown that, during a direct fusion event, the hemifusion state predominates, and the fusion of two bilipid layers occurs in a single step when they are sufficiently close 191 .
During the last decade, the fusion mechanism based upon SNARE-mediated internalization pathways (Fig. 2c) has been investigated employing SVA. At the molecular level, SNARE proteins mediate vesicle fusion with the target membrane and with membrane-bound compartments. Recently, Mattie et al. visualized SNARE single-fusion events using cryo-TEM 192 . The sequence of fusion intermediates from lipid monolayers to a complete bilayer merge has been reported in homotypic vacuoles 192 . Hu et al. correlated the membrane fusion stages with a molecular mechanism using reconstituted vesicles 193 . At a single-vesicle level, they traced the lipid-mixing process using FRET microscopy and correlated it with the docking, hemifusion and full-fusion stages 193 . They also report that an optimal distance for a SNARE-mediated fusion is 5 nm. Interestingly, some regulators of fusion pathways were identified in related studies. Calcium acts as an activator of synaptotagmin-1, leading to the fusion of synaptic vesicles with the presynaptic membrane 194 . Stratton et al. identified cholesterol as an important regulator of fusion dynamics, shifting the process from hemifusion intermediates toward full-fusion membranes 195 . Large amounts of cholesterol precluster t-SNAREs, which serve as functional docking and fusion platforms. These clusters substantially affect the stability of pores by increasing the fraction of fully open pores and accelerating fusion events. Consequently, high cholesterol content triggers fast and individual SNARE-mediated fusion events 195 .
During the last decade, multimodal imaging platforms have been tested in vitro in a number of different cellular models of disease [196][197][198] . These platforms have considerable potential to be used in vivo, for instance, in mice [199][200][201][202] . However, these systems usually fail to perform single-EV tracking and have been successful only in ex vivo cultures 196 . Nevertheless, there is an animal model worth mentioning owing to its physiological characteristics and transparency. Zebrafish embryo has emerged recently as a prospective model for tracking EVs and assessing their dissemination and uptake in vivo 203,204 . In 2019, Hyenne et al. reported an approach for tracking individual circulating tumor EVs in the zebrafish embryo 58 using confocal microscopy and a combination of chemical and genetically encoded probes to image EVs in vivo. The authors described, for the first time, the hemodynamic behavior of tumor EVs and their intravascular arrest. The study shows that the endothelial cells and blood macrophages rapidly take up circulating tumor EVs. These EVs activate blood patrolling macrophages and promote metastatic outgrowth 58 . A back-to-back study performed by Verweij et al. combined the genetic labeling (using a CD63-pHluorin exosomal reporter) of specific tissues with electron microscopy to track endogenous EVs in blood and further unravel their mechanisms of biogenesis, biodistribution and target cells throughout the zebrafish embryo 203 . Intriguingly, Sung et al. recently reported a CD63-pHluorin-mScarlett fusion protein that can be used to image several stages of the exosome lifecycle in vitro 205 . This reporter likely can be used to visualize exosomes in vivo and is a prospective tool for underFstanding the physiological roles of exosomes.
EV biomarkers SVA techniques hold the capacity to discover new, specific and effective biomarkers in EVs that have been missed by ensemble methods and can be used in disease diagnostics. While the resolution and sensitivity of SVA techniques still needs improving, they provide an accurate characterization of EV subpopulations and assessment of biomarkers. Extensive research has been carried out seeking the biological biomarkers for several diseases such as PCa 47,151,172 , fibromyalgia 206 ,  189,190 . TIRF microscopy allows the examination of the clathrin uncoating process. c, SNARE-mediated membrane fusion 192 . FRET microscopy facilitates the analysis of the three main stages of endocytic and fusion pathways: docking, hemifusion and full fusion. Membrane composition enhances this fusion pathway through t-SNARE-machinery recruitment and enrichment 195 . endometrial cancer 207 , colorectal cancer (CRC) 208,209 and liverassociated diseases 210 , among others. Figure 3 summarizes recent advances in cancer research at the single-vesicle level. Currently, fluorescent assays, which are often used in cancer screening 211 , can detect miRNAs and correlate their presence with individual EVs and EV populations. Exosome-localized miRNA-21 might be used to differentiate between cancer patients and assess tumor progression and response to treatment 211 . Moreover, a specific lipid-protein signature may identify tumor-derived EVs as Raman spectra within the range of 2,800-3,100 cm −1 appear to be a distinguishing feature of genuine cancer EVs. 76 . These findings open the way for early cancer detection. Nevertheless, the biomarkers discussed here are either generic (cannot discriminate between different cancers) or come from 2D cancer model research and might not adequately diagnose clinical cancers.
The development of next-generation sequencing SVA methods has optimized the identification of cancer biomarkers. Specifically, targeted sequencing using cancer gene panels has allowed the study of EV-derived and circulating free DNA 144 , resulting in the discovery of genetic biomarkers for several diseases. For instance, nine miRNAs have been profiled in serum EVs. Using these profiles, chronic hepatitis C patients can be distinguished from healthy individuals with accuracy >95% 212 . Diagnostic opportunities presented by EV-specific genetic biomarkers have been widely reviewed [213][214][215][216] . It has been shown that cancer-derived EVs rewire and modify the premetastatic microenvironment, supporting tumor growth and metastasis during cancer progression 14,17,18,139 . This research describes a set of potential marker targets to be used as early diagnosis of PCa. The promotion and proliferation of PCa triggered by EVs produced as a result of DIAPH3 loss or growth factor stimulation have also been reported 139 . Other studies have shown that miRNA quantification in tissues can identify PCa by detecting the expression of RNU24 217 or miR-130b 218 .
Recently, several studies have reported specific and sensitive biomarkers for cancer detection 92,140,142,[219][220][221][222] . ddPCR has detected and quantified the IDH1 transcript in cerebrospinal fluid-derived EVs of patients with glioma tumors in the brain 140 . In lung cancer, 11 cancer-specific SERS signals have been obtained, allowing differentiation between EV populations derived from healthy and lung cancer cells with high sensitivity 92 . Additionally, the CD147 protein found in EVs has been identified as a biomarker for CRC diagnosis 208 . Further studies have identified BRAF and KRAS somatic mutations in plasma-derived EV populations from CRC patients 221 . Intriguingly, Melo et al. have described an explicit biomarker, glypican 1 (GPC1), found in EV populations from pancreatic cells containing different KRAS oncogenic isomers 222 . Hence, GPC1-EV identification could facilitate early pancreatic cancer diagnosis 222 . In 2020, the EV-transported HULC lncRNA (long noncoding RNA highly upregulated in liver cancer) was suggested as a chemotaxis agent for cell invasion and migration. This encapsulated HULC is a potential biomarker for human pancreatic adenocarcinoma diagnosis 142 .
The recent advances in cancer biomarker research are a proof-of-concept for noninvasive diagnostic tools based on EV fingerprinting in combination with multivariate statistical analysis 62 . The investigations in the field of noninvasive diagnostics for PCa have been stimulated by the discovery of intrinsic biomarkers detected in urinary-derived EVs. As a result of SVA research, several biomarkers have been associated with different stages of PCa. Biggs et al. have measured the levels of circulating prostate microparticles (PMPs) in plasma from PCa patients 223 and used these microparticles in a liquid biopsy platform to identify and characterize patients. This study found that subjects with an advanced and aggressive tumor (in Gleason scale, scoring 8 or higher) can be identified independently of their PSA value 223 . The lipid and surface protein signatures of prostate-derived EVs have been described using Raman and RTM 60,73,77,84 . Several characteristics have been highlighted indicating potential PCa biomarkers. For example, a shift in the structure of surface proteins from alpha-helix-rich in prostate EVs to beta-sheetrich proteins in PCa-specific EVs (isolated from blood samples) has been observed 84 61,73,77,84,92 (b), TIRF 207 (c) and ddPCR (d) 140,142,216,221 . More information for specific biomarkers discovered using each method can be found in Table 4.

Future outlook
Individually analyzed EVs provide excellent prospects for future basic and practical research with a view to halt disease progression and control cell-to-cell communication processes. To exploit the full potential of SVA techniques, biological validation and reproducibility must meet the demands of clinical applications. Each technique has its specific advantages and disadvantages, and the exact choice of the method of analysis depends on the research question, the nature of the samples and EV characteristics. These techniques still need to improve their quantitative detection power, lower their cost and increase the reliability, resolution and throughput. In addition to technologies already in use for SVA detection, we highlight several promising approaches that have yet to prove their potential in SVA.
Conventional methodologies have the potential to be applied in SVA, and flow cytometry is a good example. Its implementation using innovative approaches can provide new features and capabilities, as shown by vesicle impact electrochemical cytometry (VIEC). This electrochemistry-based flow cytometry technique uses single ruptured vesicles whose content is detected and quantified based on Faraday's law exploiting the produced oxidation current 224,225 . Extensive studies of the regulation of neurotransmitter trafficking by Ewing and colleagues, focusing on catecholamine exocytosis [226][227][228][229][230][231][232][233][234] , have demonstrated the prospective application of this approach in the EV field, highlighting electrochemical flow cytometry as a prospective asset in studies of EV functions and biology in the near future. The VIEC-based experiments have examined neurotransmitter content at a single-vesicle level in a pheochromocytoma cell line. Several studies have established that the neurotransmitter catecholamine is only partially released from the vesicles during an exocytosis event 229 . Moreover, catecholamine concentration is a key factor in regulating vesicle size since vesicular transmitter content is relatively constant and independent of vesicle size 230 . Nonetheless, many different agents may regulate exocytosis events. Using VIEC, the group later verified that the zinc and cisplatin content serve as major regulators in these processes 231,232 .
Video microscopy is another common technique that is potentially useful in EV analysis. In 2010, Zupanc et al. developed an efficient algorithm to transform video sequences into quantitative data 233 . Their work was a crucial step toward the creation of automated computer analysis and led to the development of another, more popular, methodology, NTA. In 2008, the then-emerging fluorescent ratiometric image analysis (FRIA) method was used to determine the postendocytic fate and transport kinetics of internalized cargo 234 . FRIA presented a breakthrough in this field at the time, and its application led to colocalization of EV cargo with organelle markers. However, the technique has not been very successful in further EV research, possibly owing to the emergence of other microscopy approaches such as TIRF or SRM that enable the study of EV internalization and fate with a better resolution in fluorescence images and a more straightforward analysis.
The last decade has seen the development of several techniques with primary applications in analytical fields other than EV analysis. However, some-most notably radiofrequency analysis and SP-IRIS-are applicable to single-vesicle research and could play an important role in scanning and evaluation of specific EV populations or characterizing several EV parameters in a single experiment.
Radiofrequency analysis, also known as electrically controlled tuneable broadband interferometric dielectric spectroscopy, has only been presented in several conference papers 235-238 after its first publication. In this method, specific sensors are used to perform a highly sensitive and tuneable broadband radiofrequency analysis. One study applying this method to EVs showed that the highly concentrated radiofrequency fields stimulate strong interactions between vesicles, which can be detected and quantified 239 . Specifically, the authors could detect and scan a type of EV, giant unilamellar vesicles, at multiple frequency points and determine their molecular composition. In 2017, Wu et al. reported a separation method based on acoustofluidics and created a platform employing the so-called acoustic trapping or tweezers phenomenon 240 . This technique isolates EVs from whole blood in a label-free and contact-free manner 240 . An acoustic wave falls upon a vesicle, and its scattering acts as a driving force to retain the EVs. A year later, Ku et al. demonstrated isolation and enrichment of EVs using a similar acoustic trap technology 241 . In follow-up research, the method was used to isolate RNA and sequence miRNAs from EVs. Since then, an acoustic-based microfluidic platform has been released, coupling EV trapping technology with next-generation sequencing techniques. Together, these platforms form a robust and automated strategy for biomarker discovery in small sample volumes 242 .
SP-IRIS has so far found limited use in EV research, but has the potential to fill a unique experimental niche. SP-IRIS can characterize the size and phenotype (surface biomarkers) of EVs with no need to correlate two separate measurements 64,110 . This feature provides SP-IRIS with a high throughput and substantially reduces the amount of false negatives and positives compared with techniques that assess two characteristics in individual measurements. Due to the lateral resolution of microscopy (340-435 nm), highly concentrated samples cause signal overlap and a subsequent shift in the apparent vesicle size. Strikingly, despite the microscope resolution drawback, individual Flaviviridae particles of~40 nm have been identified and characterized using SP-IRIS 111 . Besides the examples presented in Table 5, there is a commercialized platform for EV phenotyping developed by Nanoview Biosciences using SP-IRIS, highlighting its potential application in characterizing limited input EV samples. Several papers have proposed other highly promising automated on-chip platforms using SP-IRIS for EV characterization 64,111,[243][244][245] . These platforms have been combined with immunoblotting to sort and characterize EV populations from a sample and can detect size and phenotype at a single-particle level, visualizing and quantifying either viruses or single EVs, or both, in an uncharacterized sample.
One of the greatest advantages of such microfluidic platforms is that they need only very small sample volume (~20 µL) for an effective analysis 244,246 . Other techniques different from SP-IRIS have been implemented in such devices to characterize EVs from EV-regulated diseases and examine their future use as diagnostic tools. EVs derived from transfusion-related acute lung injury (TRALI) have been investigated using SP-IRIS coupled with AFM mechanics. Obeid et al. used this approach to determine that certain types of EVs trigger neutrophil extracellular traps (NETs) and that these NETs are likely to mediate in TRALI 247 .
Fluorescence microscopy coupled with on-chip nanoflow cytometry enables automated quantitative SVA of body fluid samples 246 . According to Yokota et al., the morphology and deformability of EVs from different cell lines can be investigated using nanopatterned tethering of EVs in combination with AFM 248 . A multiplexed profiling antibody-based barcoding method enables sorting and further quantification of EV populations 249 . Notably, promising minimally invasive approaches for cancer detection and analysis have also been developed. For example, a multicolor fluorescence digital PCR platform has been designed to detect the expression of cancerspecific genes with high sensitivity 250 . This platform detects miRNA, lncRNA or any other genetic biomarkers of cancer. Furthermore, several devices for PCa detection have been described. For instance, EV populations from PCa can be separated according to the zeta potential of the surface and examined using dark-field microscopy 251 . Due to the overexpression of certain markers, different numbers of antibodies are bound to the EV surfaces, modifying their surface potential. This implies that EVs are differentially separated according to this potential. The biochemical composition of each EV is profiled and quantified; hence, subpopulations can be described. Moreover, two RTM approaches that can effectively identify and quantify EVs and their surface biomarkers in several cancers have been reported 151,152 . Beekman et al. have presented a multimodal analysis platform combining Raman imaging, scanning electron microscopy, AFM and immunoblotting as the ultimate PCa diagnosis platform capable of measuring size distribution, shapes and chemical fingerprints of tumor-derived EVs 152 . The remarkable advances in single-vesicle imaging and analysis brought about by employing microfluidic devices promise to deliver rapid and effective practical applications. Moreover, such analytical systems need only very small, microliter-scale sample volumes. These innovative technologies and affiliated research pave the way toward unraveling the biological significance of EVs and using minimally invasive systems to diagnose diseases for which EVs serve as prognostic biomarkers.

Concluding remarks
EVs are key drivers of cell-to-cell communication. Understanding their biochemistry and physiological roles is paramount for unraveling biological processes such as disease progression, physiological responses or environmental regulation. EVs are also remarkably heterogeneous, which often complicates their accurate analysis by bulk and ensemble studies that often misrepresent their functionalities. Indeed, EVs comprise several populations that further branch into subpopulations according to their morphology or phenotype. These (sub)populations could have specific functions, or several (sub)populations might perform the same task as an amalgam. It is now becoming clear that not only the phenotype of an EV community but also its relative representation among the rest of the (sub)populations might be of importance. Studies using SVA methodologies represent the most encouraging attempts toward highlighting and pinpointing specific EV phenotypes within a biological system. While SVA methods are well suited for the high-resolution phenotypical characterization of EVs that is necessary for biomarker discovery, they might not be appropriate for carrying out further functional studies. So far, these methods have only been successful in identifying the uptake and fusion pathways, but future advances in system models or single-vesicle technology may encourage their wider use. Ultimately, single-vesicle techniques will provide the foundation for describing an entirely new cellto-cell communication paradigm built upon an EV-based network.