Potential of on‐chip analysis and engineering techniques for extracellular vesicle bioproduction for therapeutics

The clinical interest around extracellular vesicles (EVs) started during the year 2000s, now leading to new clinical diagnostic and therapeutic strategies. Due to their outstanding properties as biogenic drug delivery systems and as alternatives to cell therapy, the need to produce EVs with sufficiently high yield and quality for clinical use expedited the development of analytical techniques and dedicated bioproduction methods. Though preclinical studies revealed the potential of EVs to become next generation subcellular therapies that could lead to major breakthroughs in current therapeutics possibilities, they remain complex objects on both a physicochemical and biological level. Here, we review the capacity of microfluidic technologies to match EV‐based therapeutics need for clinical translation via standardized and intensified bioproduction methods. Indeed, some of the current routine tools used in bioproduction are already achieved on chips such as micromixers or particle sorting and analysis using field flow fraction or nanoparticle tracking analyzer. Also, microfluidics communities have developed a wide set of new techniques to isolate and quantify EVs, but the few that are adopted in a bioproduction workflow are well‐established since the 1990s. We first review the different EV generation and loading methods embedded on chip. We focus on EVs preparation methods, from purification to in‐line separative techniques. We finally describe the on‐chip analytical tools to analyze physicochemistry and phenotypes of EVs.


INTRODUCTION
Extracellular vesicle (EV) is a generic name used to describe a large variety of cell secreted vesicular objects (among which exosomes, microvesicles, and apoptotic bodies) with different sizes in the nanometric range, various secretion pathways, and different properties. These objects share similarities: they are all made of a lipidic membrane bilayer with membrane proteins coming from cell plasma or internal membranes, and encapsulating a fraction of its parental cell cytosol. A commonly reported marker of these objects is their expression of tetraspanins (CD63, CD9, CD81), but also parental cell specific markers that may be used to determine their origin. A detailed review of their biological origin and composition may be found in other reviews. 1 Although the EV domain has been flourishing these last two decades, there are still challenges in the field regarding mass production of EVs, their proper (subtype) isolation as well as the detailed characterization of these nanometric vesicular complexes. EVs have demonstrated a large interest as biomarkers of diseases and prognosis. 2 For now, most of the efforts of the microfluidic community have focused on developing analytical and preparative devices for diagnostics, collecting, and detecting EVs (and more specifically exosomes, i.e., d < 150 nm) out of liquid biopsies. 3 As mentioned earlier, EV-based diagnosis is historically the first application identified. Since 2015, several reviews have followed up the progression of microfluidic techniques, [4][5][6] illustrating the constant effort to provide methodological ruptures in diagnostics but also demonstrating it is a slow process. EVs are now on their way to be tested and considered as one of the potential next generation of biotherapeutics for cancer, neurodegenerative disorder, auto-immune and inflammatory diseases. 7 Preclinical studies were already published in most if not all therapeutic areas with interesting results even in intractable diseases like chronic heart insufficiency 8 or stroke. 9 EVs require to be characterized with a wide set of analytical techniques as these intermediate submicron nanosized objects combine the complexity of their size and physicochemical properties influencing their biodistribution and of their complex biological composition, containing proteins, lipids, and nucleic acids also directly impacting their therapeutic potency. This complexity can lead to unclear or mis-understandings of their therapeutic properties. For example, tumor growth has been shown to be both inhibited or promoted by EVs, sometimes even by the same cell type EV technologies are appealing for fundamental studies but require intense characterization and quality control for industrial translation and for their approval by regulatory agencies. 10,11 In 2012, Neužil et al. 12 conjectured that "these microfluidic systems (. . . ) with features and functional elements that have similar dimensions to small vesicles will allow improved handling, analysis and manipulation of exosomes". For a decade, the rapid progresses of microfluidic devices dedicated to EVs led to a flourishing literature with a wide set of reviews. Just in 2021, right after a review on separation and detection for diagnostics applications in 2020 was published, 3 a holistic review on exosomes separation 6 and another one on isolation and detection applied to personalized medicines for cancer 13 completed this effort. Microfluidics technologies indeed unlocked new research avenues for intermediate size micro/nanobiotherapeutics and nanomedicines. Current routine analytical methods such as asymmetrical flow field flow fractionation coupled with multi-angle light scattering analysis (A4F-MALS), 14 nanoparticle tracking analysis (NTA), 15 scanning ion occlusion (QNano) 16 are directly relying on milli/micro/nanofluidic devices for continuous-flow analysis and have demonstrated their interest for EV production and characterization workflow. Tangential flow filtration (TFF) offers a scalable purification method with a higher yield and less risk of aggregation compared to ultracentrifugation (UC). 17,18 Although these techniques have had a direct impact on EV research methodology, many promises of miniaturization remain unfulfilled such as optimized sample consumption, parallelization and highthroughput analysis, fast and accurate control over physical parameters that are critical in bioproduction, leaving room for improvement in current routine methods. Yet, continuous efforts in microfluidic constantly pushes the barrier of architectural and material possibilities. The high surface to volume ratio and the wide range of surface chemistry techniques available makes it particularly appealing to embed biosensors in microchannels, reducing the volume at stake (including the dead volumes) while increasing the sensor sensitivity and limit of detection (LOD). Sample can also be prepared and separated using either active methods relying on external force fields such as dielectrophoresis, magnetophoresis, acoustophoresis and optical tweezer, or passive methods which rely entirely on the channel geometry or intrinsic hydrodynamic forces, such as pinched flow fractionation, deterministic lateral displacement (DLD), flow fractionation, and inertial microfluidics. Furthermore, a whole new field of methodological research emerges with microfluidic biological models from advanced in vitro culture to organ on chip. They offer a high potential for pharmacological assays of biotherapeutics, overcoming limitations of standard in vitro tests to develop organotypic models 19 or complex interfaces like blood brain barrier. 20 Overall, these methodological breakthroughs might contribute to relate EVs phenotypes to their biological role. The integration of micrototal analysis system (namely μTAS), which shrinks a whole laboratory to a lab on a chip, in an EV bioproduction workflow is thus a challenge that could benefit both industrial and clinical translation of EV-based biotherapies.
Last but not least, microfluidics have demonstrated the capacity of chemical process intensification, as well as control of physicochemical parameters of drug delivery systems. 21,22 By their small size, microfluidic chips provide an accurate control in few seconds over parameters such as temperature, shear stress, pressure, and concentrations, which are critical in bioproduction. Mixing is a crucial step in many chemical analyses and synthesis processes, particularly in nanoparticle formation, where it determines the nucleation rate, homogeneity, and physicochemical characteristics of the products and the capacity to load it. Passive micromixers are one of the first examples to have appeared in the 1990s, based on a forced mass transport resulting from obstacles in the channel. It increases the surface of exchange between reagents and leads to fast mixing. 23 This principle led to a myriad of related technologies from passive to active mixing, mixing to loading. 24 Recent progresses in acoustofluidics have, for example, demonstrated the possibilities to form artificial vesicles in 0.8 ms leading to controlled narrower size distributions 25 and promising possibilities to load particles. At present, the historical herring-bone (HB) design developed by Stroock et al. 23 remain one of the few device used in the regulated production of lipidic nanoparticles and nanopolymers for messenger RNA (mRNA) vaccine and gene-silencing therapies. 26 The potential of microfluidics to finely tune vesiculation parameters as well as intensified drug loading make it particularly promising for producing EV-based therapeutics.
Here, we review current progress in microfluidic devices applied to produce, engineer and analyze EVs for EV-based biotherapeutics. We review the currently available method employed for controlling EVs vesiculation and loading during production. We then identify the different EV isolation methods from complex medium, the most fitted ones for purification and the methods that hyphenate separation and analysis. Finally, we summarize the different EVs detection and quantification methods that can be performed on chip.

PRODUCTION AND LOADING
EVs are continuously released in all body fluids, their secretion is even enhanced in response to physical stimulation in blood vessels. The vesiculation of EVs and their loading constitute the first steps of a bioproduction (see Figure 1). It can be triggered chemically or physically and loading can be performed indirectly on the vesiculating cells or directly on the EVs during the vesiculation or after. 27 Shear stress is a well-known trigger stimulus for EV release from endothelial as well as blood cells, such as erythrocytes and platelets. 28,29 Biomimetic microfluidic approaches for EV production have been inspired by EV release in response to cell friction against vessel walls accounting for this natural release process of EVs in the organism. Jo et al. 30 reported nanovesicle production via the extrusion of embryonic stem cells through microchannels. The obtained vesicles featured intracellular ribonucleic acids (RNAs) from producer cells and were competent to deliver this cargo to recipient cells. The mechanism underlying the shear stress-induced vesiculation relies on cell stretching and abrupt pressure change in microchannels. In a related approach (see Figure 2A), we designed a microfluidic chip composed of parallel microchannels, mimicking vascular capillaries. Cells crossing the channels at high speed deformed and shear stress lead to membrane budding and EV release while keeping the cell with a correct morphology and without nucleus destruction. Photosensitizer (PTS) loading was investigated by red fluorescence emission. Nanoparticle loading into EVs was evaluated via dark field intensity due to their light scattering properties. Noteworthy, the obtained EVs simultaneously display the PTS drug and nanoparticle cargo from their parent cells while being positive to annexin V-fluorescein isothiocyanate (FITC), which is known to stain cell membrane phosphatidylserine lipids of EVs. This work demonstrated the feasibility of inducing EV release of nanoparticle-loaded EVs by shear stress using a microfluidic device. 27 Yoon et al. 31 conceived a cell-slicing system to generate EV-like nanovesicles from cells relying on the later on spontaneous self-assembly of cell membrane fragments into vesicles. Producer cells were sliced when passing through the system constituted of silicon nitride blades of 500 nm along microchannel walls (see Figure 2B). By this method, lipid bilayer vesicles enclosed not only parental cell-derived biomolecules (proteins and RNAs) but also exogenous materials intentionally added in the medium. A proof-of-concept was carried out using fluorescent polystyrene nanobeads of 22 nm with an encapsulation efficiency up to 30%.
In a combined approach, shear stress in microfluidic devices was used concomitantly with saponin to increase permeabilization in order to enhance doxorubicin loading efficiency directly into EVs from SF7761 stem cells and U251-glioma cells. A drug loading efficiency reaching up to 30% was achieved. The authors reported an increased homotypic uptake of doxorubicin-loaded SF7761-derived EVs in comparison to heterotypic uptake by U251 cells, with a subsequent potent anti-proliferative effect on the formers. 32 Shear stress in microchips has also been combined (see Figure 2C) to acoustic stimuli in order to assemble F I G U R E 1 Work flow of extracellular vesicles (EV) bioproduction and where microfluidics provides solutions: (i) production and loading of EVs; (ii) EVs preparation by collection, isolation, and separation can be performed by three general strategies: size exclusion, filed fractionation, and affinity immobilization; (iii) EVs analysis either on-line using an external detection method or in situ embedding a biosensor biomimetic core-shell nanoparticles featuring a poly(lactic-co-glycolic acid) (PLGA) core loaded with imaging agents and a shell constituted of an exosome membrane or cancer cell membrane (both from A549 cancer cells) in comparison to a synthetic lipid-PLGA particle. These bio-camouflaged nanoparticles obtained by the combined effects of acoustic pulses and hydrodynamic mixing were produced via the immersion of the microfluidic device into an ultrasonic bath. The A549 exosome bio-camouflaged nanoparticles displayed reduced uptake by peripheral blood monocytes, resident macrophages and higher homotypic targeting to tumor cells as compared to both A549 cancer cell membrane bio-camouflaged nanoparticles and lipid-PLGA nanoparticles in mice bearing A549 and MDA-MB-231 tumors. 33 In a recent paper, a one-step microfluidic method was designed in the attempt to produce size-controlled artificial EVs containing cell-derived lipids and RNA. Kimura et al. 34 developed a passive mixer design for the production of a simple artificial EV model encapsulating small interfering RNA (siRNA). The device enabled the efficient encapsulation of the siRNA compared to standard herringbone chaotic mixers, as well as size control by varying flow conditions. The mixing was achieved in less than 3 ms for a 500 μl/min flow rate, relevant for bioproduction. As a consequence, the siRNA encapsulation efficiency was higher than 50%. siRNA-loaded artificial EVs were able to trans-fect cells, as attested by the fluorescent signal of labeled siRNA. Besides, the gene-silencing activity of these artificial EVs loaded with siGL4 was investigated in vitro using HeLa cells stably expressing firefly and Renilla luciferase. The luminescent signal knockdown evaluation indicated an efficient gene-silencing activity evidencing the potential of this approach to design functional artificial EVs.
Though microfluidic devices offer the possibility to intensify bioproduction, providing a better control on EVs size distribution, loading and yield, it remains limited for bioproduction, mostly because it deals in most cases with small volumes, that is, small flow rate.

EVs PREPARATION: PURIFICATION AND SEPARATION TECHNIQUES
In a recent review on EV-based theranostics, microfluidics was mentioned as one of the six separation strategies. 35 It actually encompasses several subsets of separation and purification strategies.

Purification techniques
Purification methods are meant to isolate one set of EVs from contaminants, among whom salts, biomolecules (proteins, lipids), large apoptotic bodies. It can be for a posteriori analysis, the purification improving the analysis performance, or for therapeutical use. A recent review has mentioned microfluidic potential for EVs isolation, for example, in blood 36 and urine 37 analysis. In the bioproduction workflow of EVs in Figure 1, reproducible and tunable isolation and purification with minimized loss of EVs is of utmost importance to fit regulation and allow clinical transfer, making TFF a preferable method compared to UC. The extension of TFF to tangential flow for analyte capture (TFAC) embeds ultrathin nanomembranes on chip, allowing EVs capture and release (i.e., purification) in up to 200 μl volumes and direct observation with NTA. 38 This method is rather adapted for a small amount of samples, requiring flow rate of 2-5 μl/min during 40 min, making it rather slow for a small volume analysis. As surface effect becomes predominant and surface to volume ratio is increased on chip, a wide set of immobilization of receptors and detection strategies are still continuously developed. 39 One possibility to perform immobilization and fluorescent detection is using embedded immunoassays, like the Exochip system. It performs a relative quantification of EVs by fluorescence and remain compatible with proteomics and transcriptomics. 40 The immobilization itself is multi-parametric, depending on chemical conditions (pH, buffer, temperature) as well as position, orientation, and density. These will drastically impact the sensitivity and LOD in the context of biosensing and maximum quantity and specificity of immobilization in any case. 41,42 A first microfluidic proof of concept for EV purification with a posteriori analysis was experimented using immunocap-ture antibodies targeting CD63 membrane markers on EVs.
The immobilized EVs amount in this approach is directly proportional to the surface area of the chip. This amount of EVs purified is sufficient to perform cancer early diagnosis on 100 μl-5 ml volumes at a flow rate of 50 μl/min, with a 94% specificity even on complex matrices such as blood or plasma. 43 This study couples affinity-based devices with herring-bone mixers, the convection force promoting in an elegant manner the interaction between the chip surface and EVs. This study led to the immobilization of up to 8 × 10 4 EVs for a 90 μl chip, which recovered sufficiently RNA for performing RNA sequencing (RNAseq) quantification on EVs. The LOD of 10 EVs∕ demonstrated by Zhang et al. 44 confirms the interest and sensitivity of the affinity immobilization approach combined with passive micromixers. It also demonstrates the possibility of refinement strategy here amplifying the immunocapture by self-assembled nanostructures (see Figure 3A), allowing an ultrasensitive detection threshold at a much slower flow rate of 0.5 μl/min for 2 μl samples from ovarian cancer patient samples. A similar type of device allowed to reveal the mRNA changes associated with acute ischemic stroke (see Figure 3B). The combination of micropillar arrays combined with immunoaffinity-based immobilization has allowed Wijerathne et al. 45 to increase the surface of interaction with samples and demonstrate the specific enrichment of 158 ± 10 nm sized EVs with 4.3 ± 2.1 × 10 9 EVs∕100 l of plasma. This EV-microfluidic affinity purification (MAP) device uses monoclonal antibodies targeting CD8 + -EVs, each device is single use as the surface chemistry cannot be regenerated. It allows to immobilize up to 2.2 × 10 11 particles in best conditions in 20 min, which is to our knowledge among the best rates achieved by microfluidic devices for EV purification. The analysis is at 5 μl/min for a ≈500 μl sample, making the purification phase rather long (100 min). The authors claim that the device is compatible with a high-scale production method (COC thermoplastic embossing), making it suitable for deployment on bioproduction site. However, it is not possible to use the current design for large volumes and clamped device are challenging to scale up, usually leading to bonding issues. Inertial microfluidics based on centrifugal force is a good option to purify microvesicles (100-1000 nm). The purification efficiency is of 90% and RNA can be extracted with this fast technique which consumes 5 μl for 1 nm. 46

Separative techniques
To separate on-chip exosomes in liquid biopsies for diagnostics applications, several techniques can be coupled with external or integrated detection. 4,47 As opposed to purification methods, separative methods are meant to decrease the complexity of the sample, revealing its subpopulations for in-line analysis. Continuous-flow fluidic devices have already emerged for separation methods and purification of EVs such as DLD where diffusion and deterministic phenomenon are in competition and lead to a separation from 20 to 110 nm sizes at a very slow flow rate of 0.2 nl/min. The field flow fractionation techniques (asymmetrical or hollow fiber) are other filter-based techniques that have also led to demonstrations for analytical quantities (<100 μg) taking advantage of a competition between a cross-flow perpendicular to the membrane and the diffusion. 14 The flow rate is between 0.1 and 5 ml/min and the volume at stake 1 μl-1 ml. Compared to size exclusion chromatography, this separative method can hardly be used for semi-preparative quantities (<1 mg). The miniaturization of these millifluidic techniques have led to higher performances of separation but led to a limited set of applications, no one applied to EVs to our knowledge. 48,49 Indeed, embedding dense membrane filters necessary for sorting items such as EVs on chip remains a technical challenge. Therefore, the microfluidic community has developed alternatives to size-based separation techniques using external fields. Interdigitated acoustic transducers have demonstrated the possibility to filter EVs function of their size in blood, the flow rate directly impacting the size cutoff (800 μl/min for a 450 nm cutoff) (see Figure 3C). 50 An elegant field-free approach has recently brought attention, based on viscoelastic forces generated by polyethylene oxide. It achieves a >90% purity of EVs in 23.3 μl/min, thus very limited in throughput.
By applying an electric field gradient, dielectrophoresis may be used to separate (up to 200 μl, even with a low 10 V/cm gradient) 51-53 based on the permittivity and size of EVs and manipulate 54 EVs based on their size although the efficiency is rather difficult to quantify. Microfluidic on-disc centrifugation systems were reported to purify and separate EVs with reported 95% recovery and 100-fold showing in a proof of concept that these methods may be integrated in μTAS systems. Separation based on viscoelastic forces allows correct separation of biological particles based on size in a large size range (from microns to 20 nm) with good scale up potential. 57,58 Acoustic flow separation allows discrimination and separation of exosome and microvesicles based on size at up to 6 μl/min but requires very fine tuning. 36,59 Our team proposed the use of nanomagnetophoresis to sort EVs loaded with magnetic nanoparticles and to quantify their magnetic load through their magnetic attraction towards a magnetized microtip on a chip (see Figure 4A). The same device could be used to assess the colocalization of fluorescent markers or drugs (such as PTS or doxorubicin) with magnetic nanoparticles in EVs, 60 that are used for longitudinal in vivo EV monitoring by magnetic resonance imaging 61,62 or magnetic EV manipulation and targeting. 60 Though these methods deal with small volumes and would be difficult to scale up for high volume for instrumental reason, they could already be well fitted for quality control of bioproductions.

EVs ANALYSIS
Though EVs can be analyzed a posteriori once prepared or trapped in a microfluidic chip following previous section descriptions, there also exists an interest to perform in situ detection using biosensors on or right outside the chip in a continuous flow. Table 1 provide a comparative of the different order of magnitude and EV parameters targeted for quantification in both approaches.

On-line detection
Conventional flow cytometry constitutes a highly sensitive technique to detect in an on-line continuous manner EVs, using for example immunocapture. 63 In our former study, our EVs were directly analyzed by imaging flow cytometry (ImageStream) which was sufficient to evaluate EV usual markers and the presence of dual cargo for microvesicles (see Figure 4B). Dudani et al. 64 developed an elegant alternative method on chip to analyze exosome sizes (120 nm) using anti-CD63-coated beads and flow focusing. The inertial shift induced by adsorbed EVs on beads allow a fast analysis at 140 μl/min of exosomes taking only several minutes without preliminar centrifugation. Friedrich et al. 65 reported an alternative flow cytometry methodology that uses epifluorescence microscopy to detect vesicles in nanochannels. Vesicle counting was possible with up to 75% particles detected with this "imaging" flow cytometer while requiring less than 20 μl, the detection being limited only by the fluorescence intensity of each vesicle, the flow rate and the tracking performances. Zhao et al. 66 reported the simultaneous measurement of CA-125, EpCAM, and CD24 using three-color epifluorescence imaging, here based on the averaged fluorescence measurement of captured EVs on aggregated EV. The LOD was around 7.5 × 10 5 EVs∕ml. Microfluidics can also be used to allow single cell or EV characterization. Trapping single tumor or immune cell using microchambers coated with collagen or anti-CD4 antibodies, Son et al. 67 then injected microbeads with anti-CD63 antibodies, a membrane protein specific to EVs, and secondary fluorescent antibodies to perform in an enzyme-linked immunosorbent assay (ELISA)-like manner characterization of EV secretion, then analyzing them on-line. Calibrating the fluorescence with anti-interferons-γ cAb-beads and knowing the concentration in CD63 per particles, They were able to measure a ≈0.2-0.9 × 10 2 particles detection of secreted EVs after 12 h. This method is particularly original as it can directly assess the vesiculation potency of cells function of their line and differentiation, a critical parameter for bioproduction. Micro-nuclear magnetic resonance (μNMR) assay was used in 2012 by Shao et al. 68 to detect EVs bound on magnetic nanoparticles by specific antibodies (see Figure 4C). It displayed a specific 10 5 -10 8 EVs dynamic range and a LOD of 10 8 EVs. The use of Raman microscopic imaging on chip is also an interesting method to characterize EV chemical composition 69 by detecting the presence of various classes of biomolecules (lipids, proteins, and oligonucleotides) although it lacks specificity. Surface enhanced Raman spectroscopy (SERS) have not yet been described in microfluidic systems, but already described in bulk and may allow more sensitive detection and characterization. 70 μNMR assay was used in 2012 by Shao et al. 68 to detect EVs bound on magnetic nanoparticles by specific antibodies (see Figure 4C). It displayed a specific 10 5 -10 8 EVs dynamic range and a LOD of 10 4 EVs. The use of Raman microscopic imaging on chip is also an interesting method to characterize EV chemical composition 69 by detecting the presence of various classes of biomolecules (lipids, proteins, and oligonucleotides). SERS have not yet been described in microfluidic systems, but already described in bulk and may allow more sensitive detection and characterization. 70 Digital microfluidics is a powerful technique that relies on the separation of each object of interest into a single droplet, in order to perform analysis inside each droplet thus increasing exponentially the throughput of analysis. As an example, digital polymerase chain reactor (PCR) can be used into microfluidic systems to detect deoxyribonucleic acid (DNA) or RNA. Wang et al. 71 has shown that digital PCR is more sensitive and consistent than classical PCR to detect urinary-derived EVs micro-RNA (miRNA). It is also used by Chen et al. 72 to detect mutated and nonmutated IDH1 mRNA from glioma-derived EVs in cerebrospinal fluid. Instead of PCR, other hybridization systems like dual probe hybridization coupled to enzymatic production of fluorescein may be used to detect mRNA down to 20 aM. 73 Technique was used to detect mutated mRNA transcripts in Ewing sarcoma, and has shown their very low abundance. Wijerathne et al. 45 used droplet microfluidics, a technique where each drop becomes an independent reaction chamber, to perform high-yield dig-ital PCR to detect mRNAs from CD8 + T-cells-derived EVs. The technique shows a good correlation between parental cells and EVs, and allows to detect a mRNA signature associated with acute ischemic stroke. Characterization of EV DNA content, and in particular the gene associated to the K-Ras protein mutational status, was proposed by Castillo et al. 74 using digital droplet PCR with good sensitivity but limited specificity. Finally, a recent team has demonstrated that using an immunomagnetic assay strategy combined with oil droplets, we can quantify the EVs concentration with a 5 log of linear range and a LOD of 10 EVs∕ making it one of the most sensitive method to our knowledge for general quantification (regardless of sizes). 75

In situ detection: Biosensors and PCR
The surfaces of microfluidic chips can easily integrate a biotransducer (electrochemical, plasmonic, quartz microbalance) coated with bioreceptors (antibodies, oligonucleotides, chelatant). The high surface to volume ratio in microchannels allows a low dead volume and high-sensitivity detection of EVs. The plasmonic device developed by Im et al. 76 correlates with calibration curves obtained with standard ELISA when combined with a signal amplification strategy coupled with Au star-shaped nanoparticles (see Figure 5A). This label-free detection method may be used to detect EVs down to 10 5 EVs/ml and coupled with detection of specific markers like HER2. 77 Obeid et al. 78 developed a NanoBioAnalytical platform coupling plasmonic detection with atomic force microscopy (AFM), allowing to quantify EVs full range of sizes from 50 nm to 1 μm with concentrating ranging within 10 7 -10 12 EVs/ml and LOD in the ng/μl range. It revealed that 95% of EVs from platelets are below 300 nm. This platform was more recently used to improve safety of EVs from plasma transfusion, avoiding the formation of neutrophil extracellular traps that can lead to lung injury. 79 At the single EV level, nanomechanical resonators with embedded fluidic channels, known as suspended nanochannel resonators (SNRs), have been designed to measure the weight of single EV with attogram (10 −18 g) precision (see Figure 5B). 80 Fibroblast-derived and hepatocyte-derived exosomes could be distinguished ( Figure 3C) with a LOD close to 5 ag. The diameter of EV can be deduced assuming a spherical shape and a constant exosome density of 1.16 g/ml throughout the populations, with a limit of diameter detection of 39 nm. In order to increase the throughput of the devices up to 40,000 particles per hour, Gagino et al. 81 designed an architecture of parallel array SNR devices with piezoresistive sensors of 100 nm thickness that allows simultaneous readout from precision. An optical lever setup detects the cantilever's motion acquired from a photodetector on a field-programmable gate array (FPGA) and then amplified and fed back to a high-current amplifier driving a piezoceramic actuator. Buoyant mass distributions of fibroblast-derived (red) and hepatocyte-derived (black) exosomes (7100 fibroblast exosomes and 9600 hepatocyte exosomes were weighed in 65and 76 min experiments, respectively). The limit of detection is depicted with a vertical dotted line close to 5 ag. (Inset) Estimation of exosome diameter by assuming a spherical shape and a constant exosome density of 1.16 g/ml throughout the populations. The limit of diameter detection is 39 nm (from Olcum et al., 80 with the permission from National Academy of Science of the United States). (C) Novel architecture of parallel array SNR devices with piezoresistive sensors of nanoscale thickness (∼100 nm) that allows simultaneous readout from multiple resonators. The throughput of 40,000 particles per hour is 10-fold higher than with a single device (from Gagino et al., 81 with the permission from American Chemical Society). (D) Surface acoustic wave (SAW) device (side view) and SAW-induced lysing of exosomes to release RNA for detection. SAWs generated at the transducer refract into the liquid bulk, inducing fluid motion, and electromechanical coupling also generates a complimentary electric wave at the surface of the substrate (adapted from Taller et al., 90 courtesy from Royal Society of Chemistry) multiple resonators (see Figure 5C). This illustrates the need for parallel array devices that can qualify/quantify a large amount of EVs with high throughput without scarifying the precision on the output parameters such as the weight or the size of vesicles.
Redox electrochemistry based on sandwiched immunoassay and including an amplification strategy has recently demonstrated an excellent LOD (5 EVs/μl) and linear dynamic range (10-10 6 EVs/μl). Another electrochemical system was described by Zhou et al. 82 using aptamer ligands to release a probe with redox activity when EVs dock to it, allowing electrochemical detection of EV concentration by electrodes. Yasui et al. 83 used a high density of ZnO nanowires in microfluidic channels to bind EVs in a non-specific manner.
Most biosensing and characterization methods are based on the preliminar in situ binding on a surface. Surfaces can be functionalized in most studies by antibodies (usually CD63, a common marker of EVs), 84 but also aptamers. 85 It implies that the EV population captured is biased by its expression of CD63, that is, a subfraction of EVs. Once captured, EVs are usually characterized by fluorescent antibodies binding biomarkers of interest (CD41, 86 CD81, CD24, EpCAM, 87 TSG101, Glypican, 88 annexin V, anti-EGFR). 89 On another side, microfluidic technologies may be used to detect the presence of nucleic acids (DNA, RNA). It implies the need of a lysis method to make these nucleic acids reachable for further analysis. Taller et al. 90 acoustofluidic device aforementioned reached a lysis rate of about 38% (see Figure 5D). Yasui et al. 83 reported the simple use of cell lysis buffer to reach a 99% lysis rate, but this method may change the properties of the media (ionic strength, pH, etc.), limiting further analysis. EVs capture and lysis was performed on chip, but PCR or isothermal amplification was performed out of the chip. In 2015, Shao et al. performed RNA extraction using a lysis buffer followed by capture of released RNAs on densely packed glass microbeads to achieve a high extraction efficiency. 91 Much fewer teams report the detection of nucleic acids on chip. Taller et al. characterized pancreatic cells media using acoustic wave lysis coupled with an ion exchange miRNA-550 specific nanomembrane, allowing quantitative sensing of miRNA through voltage measurement. 90 Another team proposed microfluidic exponential rolling circle amplification (MERCA) for specific miRNA detection in EVs. 92 Interestingly, it only required about 2 × 10 6 EVs to detect low amount miRNAs and allowed single nucleotide specificity. The RNA extraction on chip system of Shao et al. 91 followed a reverse transcription and reverse transcriptase-quantitative PCR (RT-qPCR) run using a thermocycler and a portable fluorescence detector. It displayed a 93% capture efficiency for a 10 5 EVs/μl concentration in human sera. Though no microfluidic solution has been industrially translated to this date, the recent interferometric imaging on a silicon biosensor developed by Daaboul et al. 84 demonstrate the possibility of reducing volumes and having a wide dynamic range (10 6 -10 9 EVs/ml), a low LOD (10 5 EVs/ml) while remaining specific in complex matrices.

CONCLUSION
Microfluidic techniques have proven to be of strong interest for the EV-based diagnostics, however a bridge needs to be built between the myriad of techniques developed by the lab on chip community and the current standards which are used in EVs characterization and bioproduction. As mentioned in Section 1, most of the fluidic technologies used to this end (such as NTA, passive micromixers and more recently A4F) have been existing for at least a decade, whereas a tremendous contingent of microfluidics technologies has appeared (such as miniaturized biosensors, digital microfluidics, active micromixers) promising to intensify both analytical and production methodological inside the EVs bioproduction workflow. It could improve the speed of characterization and production from days, in classical approaches, to minutes, and intensify the different steps of production. The recent survey of Royo et al. 93 on the most frequent purification methods for EVs revealed that tangential flow and microfluidics are now being used by more than 10% of respondents, confirming the growing interest of the EV community for methodological ruptures. Though there is an inherent inertia due to industrial translation, there are also several guidelines that might need to be observed by academia to accelerate this necessary methodological leap. The first is to scale up devices to tailor applications to the ranges of concentration, volume and rate of an EV bioproduction. It implies that these methods must fit in an automated, scalable workflow. Lachaux et al. 94 recently demonstrated blood oxygenation on a parallelized and stacked microfluidic chip, reaching up to 80 ml/min flow rate, paving the way toward a new approach of designing high-volume microfluidic chips. The emergence of new companies such as Astraveus also confirms there is a trend toward scaling up microfluidic technologies for bioproduction. The second is to keep technology simple, reproducible and easy to transfer from prototypes to large-scale manufacturing (i.e., affordable, cGMP materials, and fabrication methods). To this end, the consideration in terms of microfabrication process and material used is critical. The material and process questions are non-negligible regarding the application. We recently pinpointed in a methodological article the interests for prototyping projects with analytical applications to use thermoplastic elastomers such as styrenic bloc copolymers, closer from polystyrene parts used in cell culture or analytical chemistry, rather than a silicon oilbased polymer such as polydimethylsiloxane (PDMS). 95 The last is to make current existing microfluidic technologies converge. A wide range of chemical process intensification miniaturized devices have proven to be adaptable for industrial production of nanomedicines and vaccines. It offers opportunities to intensify bioproduction of EVs as well as their loading that have yet to be proven. Furthermore, there is a bridge to build between the multiple production, preparative and analytical microfluidic modules existing to allow them to be integrated in a total analytical system manner to integrate a continuous bioproduction workflow. The following decade may be a turning point for EV-based biotherapeutics and if microfluidics has a role to play, there is still room to prove itself in this domain where standardization and quality control are highly required.

A C K N O W L E D G M E N T S
The authors thank Dr. J. Branchu from Everzom for the inspiring discussions about current interest in bioproductions. Amanda K.A. Silva has received funding from the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation programme (grant agreement No. 852791). This work was supported by the Region Ile-de-France via the DIM ELICIT (EVORTEX project) and SESAME 2019-IVETh n • EX047011.