Extracellular vesicles (EVs) are, nowadays, one of the most promising biological constructs [1]. Previously considered as cellular waste ^[2][3][2,3], new research suggests that their specific composition can be used to determine the status of cancer, auto-immune or cardiovascular diseases from mainly non-invasive liquid biopsies ^{[4][5][6][7][8]}[4,5,6,7,8]. Moreover, it is possible to turn them into targeted drug delivery systems for theranostic application with biochemical engineering methods ^{[9][10][11][12][13][14][15][16][17][18]}[9,10,11,12,13,14,15,16,17,18]. To exploit these diagnostic and therapeutic opportunities, efficient methods for separating EVs from biological solutions must be found. The limitations and drawbacks of conventional separation methods have been overcome in recent years using new microfluidic systems and highly efficient separation principles, such as affinity binding. Therefore, theis practical researchview focuses on the methodologies that are currently available to create new devices for EV separation using state-of-the-art technologies.

To separate EVs, it is important to know how they are defined and which properties they possess. The International Society of Extracellular Vesicles (ISEV), which accumulates all research results about EVs, defines them as “particles naturally released from the cell that are delimited by lipid bilayer and cannot replicate” ^[19][20][19,20]. These lipid bilayer vesicles can further be divided by their biogenesis and properties into exomeres (Ø < 50 nm), exosomes (Ø 30–150 nm), ectosomes or shedding microvesicles (Ø 100–1000 nm), apoptotic bodies (Ø 1000–5000 nm), migrasomes (Ø 500–3000 nm) and large oncosomes (Ø 1000–10,000 nm) [21]. Although more knowledge has recently been accumulated on all subtypes of EVs, current research is mainly focused on exosomes as a promising EV species ^[1][22][1,22]. The ISEV also recommends the use of the term small extracellular vesicles (sEVs) to describe exosomes.

Exosomes are formed inside the cell by inward budding of the endosomal membrane, and therefore represent a snapshot of the current status of the donor cell. After an intermediate state as intraluminal vesicles within multivesicular bodies, they are exocytosed. These vesicles are then referred to as exosomes. Outside the cell, exosomes can act as intercellular transporters, carrying their protected cargo to other cells [23]. Depending on the location of the cell, it is, thus, possible to isolate exosomes from nearly all accessible body fluids, such as systemic body fluids (e.g., blood, breast milk, follicular fluid, seminal fluid, serum, urine) and proximal body fluids (e.g., cerebrospinal fluid, saliva, sweat, tears) to analyze their composition [24].

The ability to determine the current status of a cell using exosomes via minimal to non-invasive liquid biopsies has great potential, especially for cancer diagnostics. However, the exploitation of this potential is hindered by several factors that complicate the separation of exosomes. One factor is the heterogeneity of the sites of origin, which significantly alters the membrane composition and cargo of exosomes. A second factor includes the impurities that interfere with efficient purification depending on the biofluid. Two examples are nucleic acids and lipoprotein particles (LPs) [25]. LPs are present in human serum and have a similar size and density range to exosomes [26]. Therefore, the separation of exosomes and LPs with methods that utilize these properties is difficult. Thirdly, the time-related degradation of exosomes, also referred to as biofouling, is challenging [27]. This makes a fast purification process necessary to achieve a high yield. Lastly, the general physical properties of the exosomes should be mentioned. They have a buoyant density of 1.10–1.14 g cm⁻³, an overall negative charge and, as already mentioned, a diameter in the nanometer range. Thus, there are high demands on the sample processing and measuring instruments.

2. Microfluidics—How to Build an Affinity Exosome Separation Chip

Microfluidic devices for the separation of exosomes have several advantages over conventional separation methods. Due to miniaturization, the required sample and buffer volumes are reduced. Reactions with immobilized ligands on the channel walls of a microfluidic device benefit additionally from the large surface-to-volume ratio. An increase in the reaction rate leads to a reduction in the separation or analysis time ^[28][82]. A reduction in the process time reduces the time-dependent degradation of exosomes, and thus reduces the variation in the separation results. By copying the process path several times in one device so that it can be run in parallel, this effect can even be multiplied ^[29][60]. Another advantage of the small distances in microfluidic devices is that thermal transport is faster than in larger scale operations. The improvements in the miniaturization made it possible to integrate multiple operations into one microfluidic device ^[30][83]. This enables an easier automation of the sample processing and reduces the amount of equipment, labor, and risk of cross-contaminations. These advantages can lead to high reproducibility and accuracy ^[31][84]. Overall, microfluidic devices are ideally suited for applications in a POC environment ^[32][85]. Microfluidic devices are defined as devices for the manipulation of fluids at the microscale level that utilize channels with diameters between 10 and 200 µm. While the field of microfluidic devices once started with capillary systems for gas chromatography and electrophoresis, the focus of this technology nowadays lies more on the chip format ^[33][86]. Examples of this format are microreactors, organ-on-a-chip, and LOC or µTAS devices ^[34][57]. Multiple manufacturing methods have been developed and optimized for different materials. Typical processes are based on mold manufacture, including mechanical (micro-cutting; ultrasonic machining), energy-assisted methods (electro-discharge machining, micro-electrochemical machining, laser ablation, electron beam machining, focused ion beam (FIB) machining) and traditional micro-electromechanical systems (MEMS) processes ^[35][87]. In Researchersthis review, we focus on the manufacturing methods labelling, molding, and especially 3D-printing. These approaches do not require excessive amounts of equipment or clean-room conditions and are, therefore, better suited for low-cost POC devices. It is crucial for successful manufacturing to find the optimal material for the target application. Important properties for the selection and comparison of materials are as follows: durability, ease of fabrication, transparency, biocompatibility, chemical compatibility with the implied reagents, meeting the temperature and pressure conditions needed for the reaction, and the potential of the surface functionalization ^[36][88]. Commonly used materials are glass, metal, silicone, low temperature cofired ceramics (LTCC) and polymers. Polymers have become increasingly popular in recent years, thanks to their low cost and the ease with which they can be used to build microfluidic devices ^[37][89]. Common representatives of this group are polymethylmethacrylate (PMMA), copolymers and cyclo-olefin polymers (COPs/COC) and polydimethylsiloxane (PDMS). A comparison of the different materials can be found in the recent paper of Niculescu et al. ^[36][88].

2.1. Manufacturing Methods of Microfluidic Devices

LTTC tapes, PMMA and COC can be used in a manufacturing process called laminating. This technique is characterized by multiple, differently cut layers that are piled up and bonded together ^[38][90]. The combination of these patterns in different layers creates microchannels. To cut the layers, a knife plotter or a laser cutter can be used ^[35][87]. Laser cutters are expensive but can realize high accuracies of ± 25 µm ^[39][91]. The cutting pattern for the different layers can be designed using a CAD software. After all the layers are cut, they must be aligned. A common method for the alignment is the use of alignment holes that are at the same spot in every layer. In the final step, the aligned layers must be bound together. The chosen bonding method is responsible for the pressure resistance of the microfluidic device. A bonding method that works with nearly all materials is the use of adhesive ^[40][92]. Unfortunately, it is prone to break upon high pressure (>5 bar), uneven bonding and the risk of channel clogging by residues of the adhesive. In contrast, thermal bonded layers can withstand higher pressures ^[40][92]. In this method, the layers are melted together at temperatures near the glass transition temperature. A disadvantage of this method is that unwanted deformation and bubbles can occur when the layers cool down. An alternative method for the manufacturing of microfluidic devices is molding. Molding-based methods can be divided into replica molding, injection molding and hot embossing ^[41][93]. Especially for the fast design of prototypes, replica molding is the method of choice and will, therefore, be discussed in detail in this chapter. For replica molding, a template (also called master or negative form) must be manufactured that contains the negative structures of the microfluidic chip ^[42][94]. The negative structures can be created by photolithography or high precision micro-milling (HPMM). Two common materials for the photolithography are silicones or photoresist materials (e.g., SU-8). For HPPM, harder materials, such as brass, are suitable. A more modern method for creating the template is nanoimprint lithography (NIL) ^[43][95]. This method achieves accuracies in the nanometer range. After the template is manufactured, the material for the microfluidic device is poured on it and cured. The standard material for this purpose is PDMS ^[44][96]. Due to its low surface energy, PDMS can enter even into structures down to 0.1 µm and is easily removable from the template once it is cured. Additional features of PDMS are its hydrophobicity, good biocompatibility, optical transparency, high elasticity, and low price. Therefore, it is used as the standard material for replica molding of microfluidic devices ^[45][97]. After separating the molded target structure from the template, open channels are left in the material. To seal these structures, multiple molded layers can either be stacked on top of each other to generate three-dimensional structures or the open structures can be closed by bonding the layer to a glass layer (e.g., anodic bonding). The latter method allows the direct visual tracking of the fluids. Multiple methods are available to bind stacked layers. With two PDMS layers, it is possible to add a saturated base on one layer and saturated curing agent on the other to form an irreversible connection. Additionally, the binding can be achieved via thermal or plasma treatment. The combination of photolithography and molding with PDMS results in accuracies of appx. ± 10 µm. Due to the elastomeric character of PDMS, this method is often referred to as soft lithography in the literature. Using 3D printing to manufacture microfluidic devices is another option. Three-dimensional printing enables the fast, and cost-effective creation of three-dimensional structures in just one step. In general, 3D printing processes can be divided into extrusion-based manufacturing, stereolithography (SLA) and inkjet printing. Furthermore, there are four approaches to the printing process. In the direct approach, the microfluidic device is printed completely alone by a 3D printer. In the mold-based approach, the template for the molding process is created with the help of a 3D printer. This can replace the conventional photolithographic process. By using a 3D printer, no additional expensive equipment is needed, and manufacturing time can be reduced. In the hybrid approach, only some parts, such as the channels, are created using 3D printing. These parts are then bound to classic materials (e.g., PDMS or glass), combining a quick manufacturing process with the beneficial properties of the standard materials, such as optical transparency. The continuous improvement in 3D printing methods has led to high accuracy manufacturing of units, such as the 15 µm × 15 µm valve, printed by Sanchez Noriega et al. ^[46][98]. As 3D printing of microfluidic devices is a highly dynamic field in which several working groups are active, the reader is referred to current resviearchw papers and books for a more detailed overview, e.g., ^[47][48][49][99,100,101]. Microfluidic devices can additionally be made from chromatographic paper. The so-called paper analytical devices (PADs) do not require any pumps due to their ability to transport the fluid via capillary forces. With a modified inkjet printer, wax can be applied in customizable patterns. The wax can penetrate the paper fibers and builds a hydrophobic barrier ^[28][82]. Lee et al. describe that different ligands can be coupled to selectively bind target structures ^[50][102]. Examples for PADs are pregnancy tests and more recently, coronavirus tests ^[51][52][103,104].

2.2. Structures and Modifications for Microfluidic Devices

For the design of microfluidic devices that are usable for the separation of exosomes, several structures need to be implemented. These structures can be divided into general structures, which are present in all microfluidic devices and specialized structures for the realization of a certain separation principle. The main general structure in a microfluidic device is a channel. In their diameters, channels of microfluidic devices can range from the nanometer scale up to hundreds of micrometers. The accuracy of the channel diameter and its surface structure is strongly dependent on the manufacturing method. If possible, round profiles should be preferred over rectangular profiles due to their lower hydraulic resistance ^[53][105]. The arrangement and modification of the microchannels allows various fluid manipulations. Passive hydrodynamic mixing can be realized by implementing specially designed channel routes (e.g., 3D serpentines) or by adding obstacles, such as pillars, into the flow path ^[54][106]. Valves are essential units to control fluid pathways ^[55][107]. They can be realized with two intersecting channels in different layers. Therefore, one channel is placed in a pneumatic layer and the other one in the main layer above or below. When the pressure in the pneumatic channel increases, the elastic material is pressed into the other channel and blocks it ^[28][82]. Multiple other microfluidic units for general operations, such as detectors, flow chambers and pumps, have been developed in recent years ^[56][108].

2.2.1. Design of Physical Property-Based Microfluidic Devices

It is not only affinity separation that can benefit from implementation into a microfluidic device, but other separation methods can be adapted into the microfluidic world. One example is a microfluidic device for the separation based on the physical properties of exosomes ^[57][45]. These methods are sometimes referred to as label-free separation methods compared to affinity-based methods, which are usually annotated as label-based. Strictly speaking, this terminology is not correct, as it does not consider affinity methods in which no ligands are bound to the final purified exosomes. Their main advantage is the reduced risk of the contamination of the purified exosomes with ligands. One of the physical property-based methods is microfluidic filtering ^[58][109]. Liu et al. utilized this technique by developing a microfluidic device (ExoTIC) that contained a nanoporous membrane ^[59][110]. The simple filtration enabled a separation of exosomes from plasma, urine, and lung bronchoalveolar lavage fluid. Another method based on the physical properties is called deterministic lateral displacement (DLD), describing the effect when fluids containing nanoparticles are pushed through a narrow pillar array. Hattori et al. designed a device with micropillar (diameter: 2.4 µm and gap: 2.6 µm) and nanopillar (diameter 0.47 µm and gap: 1.0 µm) arrays to successfully separate exosomes ^[60][111]. Liu et al. designed a microfluidic device with straight microchannels with a high-aspect ratio (height: 50 µm and width: 20 µm) for viscoelastic flow sorting of exosomes. This separation method is based on the “particle migration caused by size-dependent elastic lift forces in a viscoelastic medium” ^[61][112]. To achieve a highly viscoelastic fluid character, they added 0.1% polyoxyethylene (POE) to a serum and cell culture media sample. To apply external force field-based methods into microfluidics, specialized units have to be implemented in the microfluidic device. One example is the alternating current electrokinetic (ACE) microarray chips developed by Ibsen et al. ^[62][29]. It utilizes the dielectrophoretic (DEP) force generated with a platinum electrode to accumulate exosomes in DEP high-field regions. Another example is the integration of two sequential surface acoustic wave (SAW) microfluidic modules by Wu et al. ^[63][113]. These modules were able to create an acoustic radiation force, which deflected blood cells in the first module and MVs and apoptotic bodies in the second module towards a waste channel, and therefore enabled the separation of exosomes. A more detailed overview of physical-based separation methods was recently published by Hassanpour Tamrin et al. ^[64][52].

2.2.2. Design of Affinity-Based Microfluidic Devices

The high selectivity of affinity binding has led to the development of special designs in microfluidic devices to maximize the benefits of this separation principle. One approach for affinity-based separation of exosomes is the immobilization of the ligands on the inner surfaces of these devices. This review focuses on polymers as manufacturing material, because of their advantageous features for the manufacturing process. In general, the polymer surface should be altered to enhance wetting, and thereby reduce air bubble formation ^[65][114]. For affinity separation, the modification of the inert surface is essential to enable the binding of ligands. In the best case, the coupled ligands should be able to withstand the wall shear stress at higher flowrates and should not detach during storage. Therefore, covalent binding of the ligands to the surface is preferred ^[66][115]. To achieve this, it is important to know the chemical composition of the material. PDMS consists of silicon atoms to which two methyl groups are coupled. The cross-linked structure of PDMS is created via siloxane bonds. This results in the overall inert and hydrophobic properties of the material ^[67][116]. To enable the covalent coupling of ligands, hydrocarbon groups can be removed by means of oxygen plasma treatment. Afterwards, silanol or epoxy groups can bind to the oxidized PDMS surface. Following the silanization step, ligands (e.g., aptamers) can covalently bind to the surface. Finally, a blocking step must be carried out (e.g., with ethanolamine) to prevent non-specific binding. Instead of oxygen, Shakeri et al. presented a method using carbon dioxide plasma ^[68][117]. This treatment generates hydroxyl and carboxyl groups on the surface. These chemical groups can directly bind to amino acids without further treatment. A chemical method for binding antibodies to PDMS surfaces was used by Hisey et al. ^[69][118]. They first incubated the PDMS channels with 10% [3-(2-aminoethylamino)-propyl]-trimethoxysilane in ethanol. This was followed by another incubation step with 5% glutaraldehyde in distilled water. After a washing step with distilled water, the treatment made it possible to covalently bind anti-EpCAM and anti-CD9 antibodies to the channel surfaces. Nair et al. described in detail the modification of hot embossed COC to couple a ssDNA linker that contained uracil ^[70][119]. Again, the surface must be activated first. Nair et al. used UV radiation (254/185 nm) for 15 min, with a power input of 22 mW cm⁻² in combination with ozone. This leads to the formation of carboxyl groups on the surface, which are able to covalently bind to the 5′-terminus of the ssDNA linker. This linker can then be modified with specific antibodies. The coupling technique was recently used by Wijerathne et al. to manufacture a microfluidic device for the enrichment of EVs to diagnose acute ischemic strokes ^[71][120]. The uracil group contained in the linker enables simple elution after capture, as it can be digested by the addition of enzymes. An additional overview table with chip substrates and their fabrication methods for affinity separations can be found in the resviearchw paper by Bao et al. ^[56][108].

2.2.3. Enhancing the Affinity Separation Effect in Microfluidics

To enhance the affinity separation effect, surface area and flow conditions are critical parameters ^[72][73][121,122]. Therefore, multiple specialized flow routes and structures have been developed. An example of such structures are long beds filled with micropillar arrays, which decrease the diffusion distance and provide higher residence time for the affinity binding process ^[58][109]. This method was used by Wijerathne et al., who tested two types of micropillar arrays ^[71][120]. One array contained circular pillars, with a diameter of 100 µm and 15 µm spacing made from COC and another with diamond shaped pillars, with a 10 µm × 10 µm square base and 10 µm spacing made from COP. Both were functionalized with anti-CD8α antibodies coupled to oligonucleotide bi-functional linkers. The different number of parallel beds, and therefore number of pillars, made a comparison difficult. However, the diamond-shaped columns showed significantly higher flow rates, with no change in recovery. Another structure was evaluated by Hisey et al. ^[69][118]. They used a herringbone grooved surface made from PDMS that was functionalized with anti-CD9 or anti-EpCAM antibodies. The herringbone structures had a V-like shape with a length of 400 µm for one branch of the “V” and 135 µm the other, connected in a 45° angle. The positive herringbone grooves enhanced the mixing of the sample, and therefore improved the contact of the solution with the antibodies. A further way to increase the surface area and enhance the mixing was presented by Zhang et al. ^[74][123]. They manufactured a 3D porous serpentine nanostructure via patterned colloidal self-assembly. Briefly, 1 µm silica beads were placed into microfluidic chambers and then connected with (3-mercaptopropyl)-trimethoxysilane (3-MPS). This created a closely packed microbead pattern with appx. 150 nm pores. The microbeads were then activated with 4-maleimidobutyric acid N-hydroxysuccinimide ester (GMBS) and afterwards, functionalized with anti-CD81 antibodies. This method allowed the capture of exosomes from diluted plasma samples. Kang et al. utilized multiple circular chambers in their approach to increase the contact surface ^[75][34]. Their new ExoChip consisted of 30 parallel rows, with 60 connected chambers with a diameter of 500 µm ^[75][34]. Unlike their previous device (ExoChip), which had anti-CD63 antibodies coupled as ligands, the new device has immobilized annexin V. Annexin V can bind to phosphatidylserine (PS), which, as recent evidence suggests, is overexpressed on the surface of cancer-derived exosomes ^[75][34]. The new device showed significantly higher capture efficiency in comparison to the older device, highlighting the potential of the discovery of new targets on exosomes.

2.2.4. Microfluidic Design for Affinity Approaches Utilizing Beads

Beads coupled with affinity ligands have already shown good separation results in conventional separation techniques for exosomes ^[76][77][78][124,125,126]. Microfluidic separation can enhance these benefits for both non-magnetic and magnetic beads. The use of beads can also reduce the effort of the manufacturing process, by eliminating the necessity to immobilize ligands on the surfaces of the devices. One example for this is the device from Tayebi et al. ^[79][127]. They used non-magnetic microparticles with biotin-streptavidin-bound anti-CD63 antibodies for hydrodynamic trapping of exosomes. After binding the exosomes, the beads were trapped in hydrodynamic pockets during their way through the device, thus reducing the complexity of the manufacturing process. Another approach was developed by Xu et al. ^[80][128]. They combined the design of microfluidic flow routes for enhanced mixing with magnetic affinity nanoparticles. Streptavidin-coated magnetic beads were used, which were functionalized with biotinylated-mouse-Tim4-Fc. This structure can bind to PS on exosomes. To enhance the binding between exosomes and magnetic beads, the microfluidic device was manufactured with Y-shaped micropillar arrays. Comparative experiments led to optimal micropillar spacing of 50 µm and a flow rate of 0.2 µL min⁻¹. By placing or removing a permanent magnet under the micropillar area, the magnetic beads were retained or eluted. Due to the Ca²⁺-dependency of the bond between Tim4-Fc and PS, the affinity binding can easily be dissolved by adding a chelating agent. This enables non-disruptive, quick enrichment of label-free exosomes. These examples show that smart combinations of different separation principles can enhance the success of microfluidic separations, while reducing the complexity of the manufacturing process.

3. Bioassays—Validation of the Separation

After designing a microfluidic device for the separation of exosomes, it is necessary to analyze and validate the results of the separation. Determination of the size distribution and quantification of the separated exosomes can be carried out by utilizing nanoparticle tracking analysis (NTA). A disadvantage of NTA is its unspecificity towards exosomes. Thus, the quantification process can be interfered with by lipoproteins and protein aggregates. To overcome the unspecificity, multiple enzyme-linked immunosorbent assays (ELISA), fluorescent assays [26], surface-enhanced Raman scattering (SERS) approaches ^[81][129] and nanoflow cytometry setups ^[82][130] have been developed. In addition, novel methods, such as the use of electrokinetic detection with a microcapillary ^[83][131] and droplet digital ELISAs (ddExoELISA) ^[84][132], show great potential to simplify the detection process and enhance the detection limit. As already mentioned, exosomes carry proteins and RNA fragments, which can be used as prognostic and diagnostic markers for several diseases. To analyze the cargo, captured exosomes must be lysed first. Western blot (WB) analysis, sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) and nano liquid chromatography mass spectrometry (nano LC–MS) can deliver insights into the proteome and transcriptome of the captured exosomes [7]. Another specialized method to analyze the RNA cargo is (real time) quantitative polymerized chain reaction (qPCR). When developing a new device, it can be feasible to optically inspect the microfluidic device and the exosomes to evaluate the performance of the separation. This can be performed by scanning electron microscopy (SEM) or (bio-)transmission electron microscopy (TEM). The design of sensors and detectors is a large interdisciplinary field. For LOC and µTAS devices, the development of transportable detectors that can be implemented into microfluidic devices has accelerated in recent years. Since this topic is beyond the scope of this work, the reader is referred to current researchviews, e.g., ^[85][86][74,133]. However, rwesearchers want to also highlight the recent advances for extracellular vesicle detection and sensing. The newest technologies for efficient sensing include refractive index-based sensors and magneto-electrochemical sensors ^[87][88][134,135]. A great overview of the different sensing technologies, ranging from plasmon resonance, scattering, ELISA and electrochemical-based assays, is given by Im et al. ^[89][136]. Current commercially available kits for exosome isolation have been reviewed very recently by Shirejini and Inci ^[29][60]. They highlight the advantages and disadvantages of commercially available kits, such as high-throughput, cost-effectiveness, time-effectiveness and others. They studyreview membrane-based methods, precipitation-based methods, size-exclusion-based methods and immune-affinity-based methods ^[29][60]. The commercially-available products are as follows: ExoMir (Bioo Scientific) as the membrane-based method; EXO-Prep (Hasna BioMedLife Sciences), Exosome Purification Kit (Norgen Biotek), Exo-Spin Isolation Kit (Cell Guidance Systems), ExoQuick Exosome Precipitation (System Biosciences), PureExo Exosome Isolation Kit (101 Bio), miRCURY Exosome Isolation Kit (Exiqon), Total Exosome Isolation Reagent (Invitrogen), Minute High-Efficiency Exosome Precipitation Reagent (Invent Biotechnologies), RIBO Exosome Isolation Reagent (RIBO) as the precipitation-based methods; qEV (iZON Science), EVSecond (GL Sciences), ExoLutE (Rosetta Exosome Company), PURE-Evs (HansaBioMed) as the size-exclusion-based methods; Exosome-Human EpCAM isolation reagent (Thermofisher), Exosome Isolation Kit CD81/CD63 (Miltenyi Biotec), Exosome Isolation and Analysis kit (Abcam), MagCapture Exosome Isolation Kit PS (FUJIFILM Wako Pure Chemical Corporation) as the immune-affinity-based methods; ExoEasy Maxi Kit (Qiagen) and Capturem Exosome Isolation Kit (Takara Bio) as the affinity-spin-column-based methods.