Recent advances in multimode microfluidic separation of particles and cells

Microfluidic separation of particles and cells is crucial to lab‐on‐a‐chip applications in the fields of science, engineering, and industry. The continuous‐flow separation methods can be classified as active or passive depending on whether the force involved in the process is externally imposed or internally induced. The majority of current separations have been realized using only one of the active or passive methods. Such a single‐mode process is usually limited to one‐parameter separation, which often becomes less effective or even ineffective when dealing with real samples because of their inherent heterogeneity. Integrating two or more separation methods of either type has been demonstrated to offer several advantages like improved specificity, resolution, and throughput. This article reviews the recent advances of such multimode particle and cell separations in microfluidic devices, including the serial‐mode prefocused separation, serial‐mode multistage separation, and parallel‐mode force‐tuned separation.


INTRODUCTION
Over the past three decades, microfluidic devices have been increasingly used for various chemical, environmental, biomedical, and industrial applications because of their enhanced efficiency and accuracy at a reduced cost as compared to the benchtop counterparts [1][2][3][4]. Separating target particles or cells from a heterogeneous mixture in a continuous flow is often a necessary step in these labon-a-chip applications [5][6][7][8]. It can be based upon either the difference in intrinsic properties (e.g., particle/cell size, shape, and density) or extrinsic labeling (e.g., fluorescence (FACS)-and magnetic (MACS)-activated cell sorting), which has thus far been implemented by the use of either an externally imposed or an internally induced force field [9][10][11][12]. The former type of so-called active separation methods relies on acoustic [13], electric [14], magnetic [15], or optical [16] field-driven cross-stream motion of particles/cells (i.e., acoustophoresis [17], dielectrophoresis (DEP) [18], magnetophoresis [19], and optophoresis [20] in order) to manipulate them to different flow paths. These methods have the advantages of controllability and specificity but often suffer from low throughput because of the limited response rate of the external force field [21]. In contrast, the passive methods for particle and cell separation exploit the flow-induced lift and/or drag forces to direct particles/cells toward differential equilibrium positions [22,23]. This type of separation covers inertial [24][25][26]/viscoelastic [27][28][29] microfluidics, deterministic lateral displacement (DLD) [30], pinched flow fractionation (PFF) [31], hydrophoresis [32], and hydrodynamic filtration [33]. These methods are simple to use and capable of offering (relatively) high throughput but usually suffer from the drawback of limited resolution and specificity [34].
A brief summary of the equations for the forces or other important parameters underlying the active and passive methods for microfluidic particle and cell separation is presented in Table 1. The resulting particle migration velocity, , can be developed assuming the Stokes drag, = 6 , balances each of these listed forces, where is the fluid viscosity, and is the particle radius. To obtain the motion trajectory of particles and cells for understanding and predicting their separation processes, we also need to solve the Navier-Stokes equations for the fluid velocity field, [35,36]: where is the fluid density, is the time coordinate, is the hydrodynamic pressure, is the stress tensor, and is the external force acting on the fluid, such as the gravity and electric and magnetic forces.
The majority of current microfluidic separation of particles and cells has been realized using only one of these active or passive methods. Such a single-mode process is usually limited to one-parameter separation (primarily size-based), which often becomes less effective, or sometimes even ineffective, when dealing with real samples because of their inherent heterogeneity. Integrating two or more separation methods of either type has been demonstrated to offer several advantages, such as improved specificity, resolution, and throughput, other than the potential to achieve a multiparameter separation. This article is aimed to review the recent advances in such multimode separations of particles and cells in microfluidic devices. We note that there have been three review articles dedicated to this topic. Yan et al. [49] reviewed in 2017 the continuous-flow cell separation using hybrid microfluidics that integrates an active method with a passive method. They did not include other combinations such as activeactive and passive-passive configurations. Al-Ali et al. [50] later reviewed only DEP-based active and passive hybrid systems for the microfluidic manipulation of particles. They did not include those systems that are based upon a non-DEP active method (e.g., magnetophoresis and acoustophoresis) or purely passive. In another article, Cha et al. [51] reviewed multiphysics microfluidics that inte-grates multiple functional processes into one platform for cell manipulation and separation. They mainly focused on the inertial lift, elastic, dielectrophoretic, magnetic, and acoustic forces, which are combined via either cascaded connection or physical coupling for multiphysics techniques. They did not include those microfluidic systems built upon other passive (e.g., PFF and hydrophoresis) or active (e.g., optical) separation methods. In addition, there are a few other review articles that each used only one section to briefly discuss the hybrid microfluidic systems for particle and cell separation [52][53][54].
This article is aimed to provide a more comprehensive review of the recent advances in the multimode microfluidic separation of particles and cells than those in the literature [49][50][51][52][53][54]. It is organized as follows and illustrated using a vertical block list in Figure 1. Section 2 reviews the serial-mode prefocused particle and cell separation methods, which are divided into three subsections for passive focusing-passive separation, passive focusing-active separation, and active focusing-active separation, respectively. Note that the passive focusing methods here are all sheath-free. Section 3 reviews the serial-mode multistage particle and cell separation methods, which are also divided into three subsections for passive separationpassive separation, passive separation-active separation, and active separation-active separation, respectively. Section 4 reviews the parallel-mode force-tuned particle and cell separation methods, which are again divided into three subsections for active force-tuned passive separation, active force-tuned active separation, and passive forcetuned passive separation, respectively. Section 5 concludes the article with a table to summarize the reported multimode microfluidic particle and cell separation methods as well as an outlook of possible directions for future research in this field.

Passive focusing-passive separation
The first category of serial-mode prefocused particle and cell separation methods is the serially integrated passive focusing and passive separation, for which we have found only one article. Tottori and Nisisako [55] proposed an inertial focusing-enabled DLD separation method for particles and cells of different sizes. Their method utilizes a single straight rectangular microchannel to inertially focus particles and cells. The authors designed two DLD devices, one with a low-aspect ratio (AR = height/width) channel for the midplane focusing (Figure 2A), and the other with a high-AR channel for the sidewall focusing ( Figure 2B). TA B L E 1 Summary of the equations for the force(s) on a particle/cell or other important parameters involved in microfluidic particle and cell separation methods.

Technique Equation/comment Symbols References
Active Acoustic = 4 3 sin(2 ) = 5 −2 3(2 + ) − 2 3 2 acoustic radiation force acoustic contrast factor acoustic energy density wave number distance from the pressure anti-node particle density speed of sound in the particle material speed of sound in the fluid Lenshof   PFF , = 1 + 2 maximum allowed width of the focused particle stream 1 , 2 radii of particles Yamada et al. [46] Hydrophoresis Particles differing in size, shape, or other physical properties are driven by the groove-induced secondary flow into different lateral positions across a microchannel Choi [47] Hydrodynamic filtration The excess fluid within one particle radius from the main channel walls is removed gradually from the side branches, causing particles to line up along the walls Yamada and Seki [48] Abbreviations: DLD, Deterministic lateral displacement; PFF, pinched flow fractionation.

F I G U R E 1
Outline of the multimode continuous-flow microfluidic particle and cell separation methods reviewed in this article.
Both devices were successfully used to separate 13 and 7 µm polystyrene (PS) particles. The midplane-focused DLD device was found to operate at a higher flow throughput than the sidewall focused one (3 vs. 1 mL/h) but with a much longer focusing channel (25 vs. 2 mm). It was further demonstrated to separate MCF-7 and blood cells.

Passive focusing-active separation
The second category of serial-mode prefocused particle and cell separation methods is the serially integrated passive focusing and active separation. This category can be further divided into three groups based on the type of the passive focusing method, including inertial focusing, viscoelastic focusing, and other passive focusing.

Passive inertial focusing
Three articles have been published on the passive inertial focusing-based active particle and cell separation methods. Zhou et al. [56] demonstrated a ferrofluid-based inertial magnetic separation technique in a straight rectangular microchannel with a subsequent expansion. Their technique integrates in series the passive inertial focusing along the channel midplane and the active magnetic deflection toward size-dependent flow paths in the channel expansion to separate diamagnetic particles by size (and potentially other parameters such as shape [57] and magnetization [58,59]) ( Figure 3A). The authors have also developed a 3D numerical model to simulate the particle focusing and separation processes, which was found to predict the separation of 10 and 20 µm PS particles with a good agreement. In a later work, Liu et al. [60] reported a similar inertial ferro-hydrodynamic cell separation method that integrates the inertial focusing in two serpentine microchannels with the diamagnetic cell deflection in ferrofluid ( Figure 3B). The authors performed a systematic optimization of their device through the separation of PS particles by size. Their device was found capable of processing 100 000 cells/s at a flow throughput of over 60 mL/h. It was also demonstrated to separate circulating tumor cells (CTC) from blood and to enrich lymphocytes directly from white blood cells (WBCs). In another work, Mutafopulos et al. [61] developed a microfluidic fluorescence-activated cell sorter that integrates the inertial (in a spiral microchannel) and sheath focusing (in a vertical flow-focusing nozzle) with traveling surface acoustic wave ( Figure 3C). Their device was demonstrated to sort about 2000 cells/s with over 90% purity using a 25 µs acoustic wave pulse. The authors measured the viability of three different live cell lines to verify the gentleness of acoustic sorting.

Passive viscoelastic focusing
Four other articles have been published on the passive viscoelastic focusing-based active particle and cell separation methods in non-Newtonian media. Kim et al. [62] developed a two-step particle separation technique that integrates elasto-inertial focusing with magnetophoretic separation. Specifically, particles in the flow of poly(ethylene oxide) (PEO)-based ferrofluid are first focused toward the channel center by the elastic lift force and then deflected toward particle size-dependent flow paths by the diamagnetic repulsion force ( Figure 4A). The authors demonstrated the separation of 5 and 20 µm PS particles in their device and also performed numerical simulations to understand the underlying physics. A similar idea was later utilized by Zhang et al. [63] to implement a sheathless separation of 5 and 13 µm nonmagnetic particles while at a flow throughput of more than one order of magnitude higher than that in Kim et al. [62]. In another study, Del Giudice et al. [64] presented a microfluidic device, where the deflection of magnetic particles in an H-shaped microchannel can be improved by pre-aligning them in the flow of viscoelastic polyacrylamide solution ( Figure 4B). The authors demonstrated that the prefocusing of particles helps enlarge the device's operative window and the fluid viscoelasticity-induced normal stresses help contrast the particle sedimentation. Their device was used to successfully separate 10 µm magnetic particles from 6 to 20 µm nonmagnetic particles, respectively. A similar device was later used by Dibaji and Rezai [65] to investigate the effects of inertial, magnetic, and elastic forces on the focusing and separation of magnetic (9 and 15 µm) and nonmagnetic (15 µm) particles in PEO solution. The authors also studied other parametric effects.  Li et al. [66] reported an acoustic FACS system that combines viscoelastic focusing and highly focused traveling surface acoustic waves to sort single cells at ∼kHz ( Figure 4C). Their device uses a serpentine microchannel to prefocus particles in the flow of PEO solution. It integrates fluorescence and forward scattering detection to the optical setup for the counting of both target and nontarget cells, enabling quantitative single-cell analysis. The authors applied their device to sort three different cell lines, the viability of which was found to experience a much smaller drop than in a commercial FACS machine. The same group [67] later developed a tunable acoustofluidic system of similar design to that of Li et al. [66] for continuous-flow particle separation based on multiphysical properties, including the particle size, density, and compressibility. Their revised device still utilizes the elasto-inertial focusing to pre-align particles but integrates a slanted interdigitated transducer to generate a tunable traveling surface acoustic wave for improved sorting metrics ( Figure 4D). The authors have successfully used their device to sort 5.26 µm from 5 µm PS particles. They have further demonstrated an effective sorting of PS, poly(lacticco-glycolic acid), and poly(methyl methacrylate) particles of equal size.

2.2.3
Other passive focusing There are three additional reports on passive noninertial/viscoelastic focusing-based active particle and cell separation methods. Luo et al. [68] presented a sheathless cell separation device that integrates sedimentation-based focusing and dielectrophoretic separation ( Figure 5A). The prefocusing of cells was achieved through gravitational sedimentation in an inlet tubing of the microchannel, which was demonstrated to work effectively for a mixture of cells with different sizes and dielectric properties over a wide range of flow rates. The dielectrophoretic force was applied at the downstream region of the microchannel through an interdigitated electrode array. The authors assessed the efficiency of their device through the sizebased separation of leukemia THP-1 cells from yeast cells and the dielectric properties-based separation of leukemia THP-1 and OCI-AML3 cells. Kirby et al. [69] reported a centrifugo-magnetophoretic platform for particle separation, where magnetic and nonmagnetic particles are all aligned along the channel wall distant from the magnet because of centrifugal sedimentation. Magnetic particles are then deflected toward the magnet at a particle size-dependent rate, whereas nonmagnetic ones remain unaffected. Liang et al. [70] developed a microfluidic cell separation system integrating the hydrophoresis and DEP modules ( Figure 5B). The hydrophoresis module focuses cells into two streams along the edges of a serpentine microchannel with ridges and trenches via a diverging fluid flow. The DEP module sorts cells based on electrophysiological properties through a chevron-shaped electrode array. The authors tested their system with mouse neural stem cells, where the astrocyte-biased cells were found to be enriched in the focused fraction through the inner channel while depleted from the unfocused fraction through the outer channels.

Active focusing-active separation
The third category of serial-mode prefocused particle and cell separation methods is the serially integrated active focusing and active separation. This category can be divided into two groups based on the type of the active focusing method, including acoustic focusing and electric field-driven focusing.

Active acoustic focusing
Two articles have been published on the use of the active acoustic method to prefocus particles and cells for subsequent active separation. Shields et al. [71] developed a microfluidic platform for the high-throughput separation of cancer cells from blood and on-chip organization of cancer cells for streamlined analyses. Their platform consists of three modules, where the first module uses an acoustic standing wave to rapidly align cells, the second module separates magnetically labeled cells from unlabeled, and the third module captures magnetically labeled cells into a periodic array of microwells with underlying micromagnets ( Figure 6A). It is capable of isolating magnetic particles and magnetically labeled cells, and processing blood at the throughput of 3.0 mL/h. In a later study, Hu et al. [72] demonstrated a hybrid acoustic-optical method for the accurate separation of leukocyte subtypes ( Figure 6B). The acoustic part in their system can function as not only a cell sorter for certain leukocyte phenotypes but also a cell focuser for the subsequent optical sorting part. The authors successfully used the acoustic-optical system to obtain highly pure lymphocytes and monocytes, which was envisioned useful as in future leukemia auxiliary diagnosis and analysis.

Active electric field-driven focusing
Two other articles have been published on the use of the active electric method to prefocus particles and cells prior to separating them. Zhu et al. [73] reported a DEPintegrated pulsed laser-activated cell sorter ( Figure 7A). Their device uses a microchannel laid out with 3D electrodes to generate a tunnel-shaped electric field profile along the channel length for a single-stream dielectrophoretic focusing of cells in high-speed flows. Once a target fluorescent cell is detected, the system triggers a nanosecond laser pulse that is focused upon a neighboring channel to produce a rapidly expanding cavitation bubble for precise sorting. The authors used their device to achieve an over 90% purity sorting of PS particles at a throughput of 1500 particle/s. In another work, Chen et al. [74] proposed a hybrid microfluidic separation method based on the combination of electric field-driven particle focusing and deflection ( Figure 7B). Their method integrates in series the vortical focusing of particle mixture into a thin stream via induced charge electroosmosis and the electrode-based dielectrophoretic separation of particles based on dielectric properties. The authors demonstrated in their device an efficient separation of yeast cells and silica particles.

Passive separation-passive separation
The first category of serial-mode multistage particle and cell separation methods is the serially integrated passive separation and passive separation. Two articles have thus far been published on this category. Xiang et al. [75] reported a two-stage microfluidic device coupling inertial microfluidics with DLD for a size-based cell separation ( Figure 8A). The first stage uses a spiral microchannel to inertially remove the overwhelming majority of background blood cells. The second stage uses a DLD channel with triangular posts to remove the residual blood cells from tumor cells. The authors demonstrated a nearly 100% separation of differently sized particles and cells in their F I G U R E 7 Serial electric field-driven focusing and active separation: (A) schematic of a dielectrophoretic focusing integrated pulsed laser activated cell sorting system; (B) schematic of a two-step microfluidic device for particle separation through induced charge electroosmotic prefocusing and dielectrophoretic separation. Source: (A) Reproduced with permission from Zhu et al. [73] Copyright © 2017, Society of Photo-Optical Instrumentation Engineers; (B) reproduced with permission from Chen et al. [74] Copyright © 2017, American Chemical Society. device. They also used the device to achieve a good separation of rare tumor cells from diluted whole blood. In an earlier study, Shen et al. [76] developed a passive multistage microfluidic device through a unique combination of inertial microfluidics and steric hindrance to achieve highthroughput and high-efficiency cell separation ( Figure 8B). They optimized the processing parameters of their device using both numerical simulations and experimental studies of fluorescence-labeled microspheres. They used the optimized device to separate tumor cells spiked into 1% hematocrit blood with over 90% cell recovery and 2 × 10 5fold rare cell enrichment at a throughput of more than 20 million cells/min.

Passive separation-active separation
The second category of serial-mode multistage particle and cell separation methods is the serially integrated passive separation and active separation. This category can be divided into three groups based on the type of the passive separation method, including the DLD separation, inertial separation, and other passive separations.

Passive DLD separation
There have been four articles on the use of passive DLD separation as the first stage prior to an active particle and cell separation. One of these papers uses both electric and acoustic separations in the second stage. Wu et al. [77] designed a microfluidic chip that integrates four sequential modules for a multi-target separation of cells and particles ( Figure 9A): the DLD module sorts particles and cells based on size and surface charge under a combined pressure and DC electric field-driven flow; the bipolar electrode module positions one fraction of the cells to the pressure nodes or antinodes of the surface acoustic wave module for compressibility and density-based separation; the DEP module sorts the rest of the cells based on dielectric properties. The authors demonstrated in their device a high-efficiency sorting of PS particles, oil droplets, and viable/nonviable yeasts. The other three articles on the DLD separation-based active particle and cell separation methods all use a magnetic field in their second stage. Toner's group [78] reported an inertial focusing-enhanced microfluidic CTC capture platform ( Figure 9B). Their CTC-iChip has the first-stage DLD array in a polycarbonate manifold and the secondstage inertial focusing/magnetic separation components in an aluminum manifold. It is capable of sorting rare CTCs from whole blood at 10 million cells/s, which can be either dependent or independent of tumor membrane epitopes and hence applicable to virtually all cancer cells. The authors demonstrated the use of their device in an expanded set of both epithelial and nonepithelial cancers, including lung, prostate, pancreas, breast, and melanoma. In a following work, the same group [79] modified the CTC-iChip architecture by the use of two-stage magnetophoresis and depletion antibodies against leukocytes to achieve 3.8-log depletion of WBCs and 97% yield of rare cells at 8 mL whole blood per hour. In a recent study, Kim et al. [80] reported an integrated basophil isolation device for the negative immunomagnetic selection of basophils from whole blood. Their device combines the DLD enrichment of WBCs from blood, the mixing of magnetic nanoparticles and negative selection antibodies with enriched WBCs, and the magnetic depletion of non-basophils ( Figure 9C). It was demonstrated to isolate basophils with over 90% purity and recovery without causing the degradation or unintentional activation of basophils. In another work, Toners' group [81] presented an automated monolithic CTC-iChip with multiplexed DLD arrays to first deplete red blood cells (RBCs) and platelets ( Figure 9D). The DLD outputs are serially integrated with an inertial focusing system to line up all nucleated cells for two-stage magnetophoretic removal of magnetically labeled WBCs.

3.2.2
Passive inertial separation Two other articles have been published employing inertial separation as the first stage prior to an active magnetic particle and cell separation method. Huang and Xiang [82] constructed a three-stage i-Mag device integrating passive inertial microfluidics with active magnetophoresis for the rapid and precise separation of tumor cells from blood ( Figure 10A). The first-stage spiral inertial sorter rapidly removes smaller RBCs. The second-stage serpentine inertial focuser aligns the larger cells, of which the magnetically labeled WBCs are then removed in the thirdstage magnetic sorter. The authors achieved in their device a separation of breast cancer cells from diluted whole blood at an efficiency of 93.84% and purity of 51.47%, where the latter quantity could be improved with the increase of the blood dilution ratio. A similar idea was later used by Nasiri et al. [83] to develop an integrated inertial-magnetophoretic cell sorting device ( Figure 10B). RBCs and smaller WBCs are first removed in a serpentine microchannel via differential inertial focusing [84]. Magnetic nanoparticle-conjugated MCF-7 cells are then separated from the larger WBCs with a permanent magnet. The authors achieved over 92% purity and recovery rate of isolated cancer cells in their hybrid device. In addition, Zhou et al. [85] presented a hybrid microfluidic cell sorting system combining inertial focusing/separation devices with an FACS device for single-cell isolation. The authors demonstrated at least 2500-fold purity enrichment of MCF-7 breast cancer cells spiked into diluted whole blood after one run of the process. However, their system required a manual collection and reinjection of the pre-enriched cell suspension from the inertial and acoustic sorting devices, respectively.

Other passive separation
There are two other reports on the use of a passive non-DLD/inertial separation as the first stage followed by an active particle and cell separation. Mizuno et al. [86] presented a microfluidic two-parameter cell sorting via the serial connection of hydrodynamic filtration and magnetophoresis ( Figure 11A). Immunomagnetic beadconjugated cells are first sorted based on size by hydrodynamic filtration, where cells are introduced into individual separation lanes and simultaneously focused onto one sidewall. Cells are then subjected to magnetophoretic separation and recovered through multiple outlet branches. The authors successfully demonstrated in their device a continuous size and surface marker-based sorting of JM and HeLa cell mixture. In an earlier paper, Moon et al. [87] reported a high-throughput microfluidic cell sorting device that combines multi-orifice flow fractionation and dielectrophoretic separation ( Figure 11B). Cells are first sorted based on size in the multi-orifice flow fractiona-tion channel via the differential inertial focusing [88]. The larger cells are then aligned along both sides of the focusing channel via first-stage positive DEP and finally separated based on the dielectric properties via second-stage DEP.
The authors demonstrated in their device a successful separation of MCF-7 cells from blood.

Active separation-active separation
The third category of serial-mode multistage particle and cell separation methods is the serially integrated two (or more) active separations. Two articles have thus far been published on this category that both use magnetic separation in the second stage. Kim and Soh [89] reported an integrated dielectrophoretic-magnetic activated cell sorter in a single microfluidic device ( Figure 12A). The target cell types are sorted based on the bound DEP and magnetic tags. The use of distinct electric and magnetic force fields completely eliminates the cross-contamination of target cell types between the two outlets. The authors achieved ∼900-fold enrichment of multiple bacterial target cell types with over 95% purity at a throughput of 25 million cells/h after a single round of separation. In a later work, Adams et al. [90] demonstrated the integration of acoustic and magnetic separation into a monolithic device ( Figure 12B). The authors tested their device using the 5 µm diameter PS particles as acoustic targets, 4.5 µm diameter magnetic particles as magnetic targets, and 1 µm diameter PS particles as non-targets. They also achieved the size, acoustic, and magnetic content-based separation by using magnetic nontarget particles. They obtained 97% purity at the acoustic outlet and 78% purity at the magnetic outlet.

Active force-tuned passive separation
The first category of parallel-mode force-tuned particle and cell separation methods is the active force-tuned passive separation. This category can be further divided into four groups based on the type of the passive separation method, including the DLD, hydrophoretic, inertial, and PFF separations.

Passive DLD separation
Four articles have been published on the use of active electric field to tune the passive DLD separation. Two of them employ an AC electric field parallel to the hydrodynamic flow direction in the DLD array. Beech et al. [91] reported the use of DEP to tune the critical particle size and open up other parameters for particle separation in a DLD device ( Figure 13A). They demonstrated that the critical particle diameter could be tuned through the application of moderate-(∼100 V/cm) and low-frequency (100 Hz) AC electric fields over the channel length ( Figure 13B). The behavior of their device was also investigated by performing simulations of particle trajectories and comparing them with experiment observations. In a later work, Ho et al. [92] used a similar method to achieve the separation of both nano-and micro-sized particles based on zeta potential (or surface charge). The authors also demonstrated a separation of nanoscale liposomes of different lipid compositions, which has strong relevance to biomedicine. They further performed a careful characterization of experimental conditions in order for an adequate sorting of different particle types. They found that the electrokinetically enhanced particle displacement increases for particles with high zeta potential at low-frequency electric fields. Another two articles use an electric field orthogonal to the hydrodynamic flow direction in the DLD separation. Calero et al. [93] proposed to combine DLD with orthogonal electrokinetic forces for enhanced particle sorting. The authors placed planar electrodes alongside the channel walls producing an electric field orthogonal to the flow direction ( Figure 13C). They examined the binary separation of 3 and 1 µm diameter particles (both smaller than the critical diameter) in electrolytes with different conductivities. They also demonstrated the ternary separation of 3, 1, and 0.5 µm particles. The same group [94] developed in a following work scaling laws for the parametric dependence of particle displacement in the DLD channel. The predicted particle deviation angles were found to agree well with the experimental data at high frequencies. At low frequencies (below 1 kHz), however, particles were found to follow the same scaling law while at much lower voltages because of the occurrence of concentration polarization electroosmosis around the insulating pillars of the DLD array [95]. In a recent paper, Calero et al. [96] explored how the application of DC-offset AC electric fields orthogonal to the fluid flow further improves the particle separation in a DLD device. They confirmed that the Faradaic processes at the electrodes in response to an offset DC field change the medium conductivity and create electric field gradients across the channel width. This results in a nonuniform electrophoretic velocity orthogonal to the flow direction, which counters particle diffusion leading to the formation of particle size and zeta potential-dependent tight bands. The authors demonstrated the fractionation of particles on the nanoscale ( Figure 13D).

Passive hydrophoretic separation
Three articles have been published on the use of active electric or magnetic fields to tune the passive hydrophoretic separation. Yan et al. [97] presented a DEPactive hydrophoretic device for particle separation and filtration, which comprises interdigitated electrodes overlaid on a hydrophoretic channel ( Figure 14A). Their device can run at the particle filtration or separation mode simply by altering the applied AC voltage to tune the lateral particle position. The authors demonstrated that 3 and 10 µm beads have close lateral positions under 24 V AC and are redirected to the same outlet for filtration. Increasing the voltage to 36 V directs 3 and 10 µm beads into different outlets for separation. In another work from the same group, Yan et al. [98] demonstrated a DEP-active hydrophoretic sorting of particles and cells based on both size and dielectric properties. Their device consists of a particle prefocusing region, where all particles are aligned along one sidewall by steric hindrance under the electric field application, and a particle sorting region, where smaller particles are directed to the other sidewall by transverse flow, whereas larger particles remain unaffected ( Figure 14B1). The authors achieved a high efficiency and purity separation of both 3/10 µm particles ( Figure 14B2) and viable/nonviable Chinese Hamster Ovary (CHO) cells. A similar DEP-active hydrophoretic device was also demonstrated by Yan et al. [99] to isolate plasma from blood. In a recent work, the same group [100] reported a high-throughput separation of magnetic and nonmagnetic particles in a groove-based hydrophoretic channel ( Figure 14C). The authors observed that magnetic particles are focused near the channel centerline by positive magnetophoresis and groove-generated microvortices, whereas nonmagnetic particles are focused along the channel sidewalls by negative magnetophoresis and hydrophoresis [101]. They demonstrated a F I G U R E 1 5 Parallel active force-tuned inertial separation: (A) schematic illustrating the mechanism of nonmagnetic particle separation and washing in an inertial ferrofluid/water coflow; (B) 3D illustration of a dielectrophoresis (DEP)-active inertial microfluidic device for particle separation. complete separation of 6 µm magnetic particles from 13 µm nonmagnetic particles over a wide range of flow rates.

Passive inertial separation
Three other articles have been published on the use of active magnetic or electric fields to tune the passive inertial separation. For the former case, Chen et al. [102] developed a microfluidic device for a simultaneous separation and washing of nonmagnetic particles in a ferrofluid/water coflow ( Figure 15A). These two operations take place in parallel in a T-shaped rectangular microchannel, enabled by the active magnetic force induced in the ferrofluid and the passive inertial lift induced in the ferrofluid/water flows. The authors demonstrated that the larger and smaller particles' exiting positions (and hence separation) in water and ferrofluid, respectively, vary with both the total flow rate and the flow rate ratio between the two streams. Kumar and Rezai [103] proposed a hybrid microfluidic technique for multiplex fractionation of magnetic and nonmagnetic particles in a simple expansion microchannel with a side permanent magnet. Their technique is based on the interaction between the flow-induced inertial force and the countering magnetic force. The authors conducted a duplex fractionation of magnetic and non-magnetic particles as well as triplex separation of 5, 11, and 35 µm magnetic particles. They further achieved a fourplex fractionation of the three magnetic particles from nonmagnetic particles with size variations (10-19 µm). In a later work, Moghadam et al. [104] used a Lagrangian-Eulerian framework to simulate particle trajectories in such multiplex inertia-magnetic fractionation technique using the ANSYS-Fluent discrete phase modeling approach. There is another recent paper on the use of electric field to tune the inertial separation. Zhang et al. [105] reported a DEP-active inertial microfluidic device for particle separation ( Figure 15B). They used a top sheath flow to push all particles toward the bottom wall of a serpentine microchannel, such that the DEP force induced by the bottom patterned interdigitated microelectrodes can effectively modify the positions of inertially focused particles [106]. The authors evaluated the performance of their device using a binary mixture of PS particles. The advantage of their device over the traditional inertial sorter lies in the real-time tuning of the target particle dimension by simply adjusting the electric voltage. Another work worthy of notice is the tunnel DEP proposed by Kung et al. [107], where four independently tunable AC signals were applied to the quadro-electrodes along a microchannel to create a tunnel-shaped electric field profile with the minimum inside the channel for a single-stream particle focusing. The location of this electric field minimum can be adjusted in two dimensions over the channel cross section by changing the voltage combinations. Particles and cells experiencing negative DEP are focused at the location of electric field minimum in high-speed flows regardless of their types and sizes. Such tunnel DEP can serve as an upstream particle focusing module in serial-mode separation devices [73]. It may also use multifrequency electric fields to focus cells toward dielectric property-dependent locations for separation and enrichment.

Passive PFF separation
There have been three reports on the use of an active force field to tune the passive PFF separation. Lee et al. [108] reported an improved particle separation method, where the original particle position with respect to the wall in the PFF channel can be modulated by an optical scattering force orthogonal to the flow ( Figure 16A). The authors demonstrated in their device a continuous separation of PS particles with diameters of 2, 5, and 10 µm. The achieved particle separation distance was found 15 times larger than that obtained in a pure PFF device. The authors also performed theoretical calculations to interpret their experimental observations. In another work, Zhou and Wang [109] developed an acoustic bubble-enhanced PFF, which utilizes the microbubble streaming flow to increase the particle separation distance ( Figure 16B). The particle position and separation distance in their device can be adjusted by changing the driving voltage. The authors demonstrated a separation of 10 and 2 µm diameter PS particles in a 60 µm-wide pinched segment at different buffer/particle solution flow rate ratios. They also investigated several other parametric effects. In a recent paper, Khashei et al. [110] combined the passive PFF and active insulator-based DEP methods to increase the particle separation efficiency at low DC voltages ( Figure 16C). The authors used COMSOL simulations to optimize the channel geometry and applied voltage for particle separation with insignificant Joule heating effects. They achieved a continuous separation of 1.5 and 6 µm PS particles under a total applied voltage of 25 V.

Active force-tuned active separation
The second category of parallel-mode force-tuned particle and cell separation methods is the active force-tuned active separation. Two articles have thus far been published on this category that both combine the use of magnetic and acoustic fields for enhanced particle manipulation and separation. Wiklund et al. [111] combined ultrasonic standing wave and DEP in a microfluidic chip for both a high-precision handling of individual particles and a highthroughput handling of particle ensembles ( Figure 17A). The principle is based on the competition between longrange ultrasonic and short-range dielectrophoretic forces in a fluid flow. The ultrasound is produced by a transducer placed on top of the chip, whereas DEP is generated by co-planar microelectrodes placed on the bottom channel surface. The authors demonstrated various particle manipulation functions, including the separation of 15 and 10 µm beads, where the smaller beads pass by the electrodes while the larger ones are trapped and can be released individually. In a recent work, Tayebi et al. [112] demonstrated a size-based separation of submicrometer particles as well as subpopulations of extracellular vesicles in a combined surface acoustic wave and DEP device ( Figure 17B). The interdigital transducer placed beneath the fluid flow path directly couples acoustic energy and electric field into the fluid, enabling the manipulation of submicrometer particles. The authors demonstrated in their device that 300 nm particles can be separated from 100 nm ones. The authors further sorted exosomes (<200 nm) from microvesicles (>300 nm) with good purity and recovery.

Passive force-tuned passive separation
The third category of parallel-mode force-tuned particle and cell separation methods is the passive force-tuned passive separation. Total eight articles have been published on this category that each utilizes a passive lift or sedimentation to tune the PFF separation. For the former case, Lu and Xuan [113] suggested that the flow-induced passive inertial lift in a confined microchannel can be exploited to significantly increase the particle displacement in PFF because of its strong size dependence ( Figure 18A). Such inertia-enhanced PFF (iPFF) technique offers at least one order of magnitude higher particle throughput than the traditional PFF for the same sheath flow rate ( Figure 18B). Moreover, it can work effectively for the Reynolds number spanning more than one order of magnitude. In a later work, Wang et al. [114] utilized iPFF to separate microalgae cells by size. The best separation was achieved when the Reynolds number and sheath-to-sample flow rate ratio both reach approximately 10. The authors obtained 90% recovery rate and 86% purity for Tetraselmis sp. in contrast to 99% recovery rate and purity for Chlorella sp. In a more recent paper, de Timary et al. [115] studied experimentally and numerically the influence of inertia on the particle dynamics in iPFF. In another work, Lu and Xuan [116] presented an elasto-inertial PFF (eiPFF) technique for particle separation in viscoelastic PEO solutions via a combined action of elastic and inertial lift forces ( Figure 18A). This technique was found to provide a better separation resolution than iPFF and work most efficiently for the Reynolds number of order 1. Lu and Xuan [117] later utilized eiPFF to achieve a high-purity separation of PS particles by shape, which was found to fail with iPFF ( Figure 18C). Interestingly, they found that the elasto-inertial deflection of peanut particles can be either greater or smaller than that of equal-volumed spherical particles, which was later utilized by Lu et al. [118] to demonstrate a sheath-free separation of particles by shape.
Two other articles have reported the use of particle sedimentation to tune the PFF separation. Sunahiro et al. [119] developed a microfluidic device that combines the centrifugal force with PFF to separate particles by size and density ( Figure 18D). The centrifugal force arises from the device rotation, which is responsible for both the fluid transportation and particle migration perpendicular to the flow direction. In a following work from the same group, Morijiri et al. [120] expanded such sedimentation-based PFF using both the particle inertia and device rotation schemes. The former scheme utilizes the inertial force of particle movement in a curved microchannel. The authors successfully demonstrated a continuous sorting of PS and silica particles according to size and density with both schemes. In another work, Lu et al. [121] built upon an earlier developed DC electroosmotic flow-controlled PFF [122] and proposed to exploit the wall-induced noninertial lift for an enhanced separation of particles by size. This electrically originated dielectrophoretic-like lift force arises from the asymmetric electric field around a near-wall particle [123,124]. The authors demonstrated their electric PFF through both a binary and a ternary separation of PS particles by size. They also developed a semi-analytical model to simulate and understand the particle separation process.

CONCLUSION AND OUTLOOK
We have reviewed the recent advances in multimode continuous-flow microfluidic separation of particles and cells using three sections: (1) The serial-mode prefocused separation methods cover passive focusing-passive separation, passive focusing-active separation, and active focusing-active separation. This type of method has the strength of enhanced separation resolution and purity as compared to the involved passive or active separation alone because of the alignment of the particles and cells prior to be separated. It is, however, often limited to one-parameter separation due to the essentially singlemode passive or active separation. More specifically, the passive focusing-passive separation configuration has the simplest structure and operation within the type but lacks the flexibility as the sample flow rate is the only controllable parameter and is often limited to the morphology-based (e.g., size) separation only. Both the passive focusing-active separation and active focusing-active separation configurations have the potential to expand the separation parameter from morphology to, for example, electrical, magnetic, acoustic, or optical properties because of the added controllability and specificity of active separation. (2) The serial-mode multistage separation methods cover passive separation-passive separation, passive separation-active separation, and active separationactive separation. This type of method is capable of offering two (or more)-parameter separation and is more suitable for processing complex samples than either stage of the involved passive or active separation alone. It, however, usually requires the use of sheath buffer to prefocus the particles and cells in order for the first-stage separation to take place effectively. More specifically, the passive separation-passive separation configuration has the similar strengths and limitations to those for the serial-mode passive focusing-passive separation configuration. The passive separation-active separation configuration has the potential to improve the flow throughput of the active separation and as well enable the separation based on both morphology and electrical, magnetic, acoustic, or optical properties. The active separationactive separation configuration can further improve the separation specificity because of the enhanced selectivity of electrical, magnetic, acoustic, and optical properties over the morphology of particles and cells. (3) The parallel-mode force-tuned separation methods cover passive force-tuned passive separation, active force-tuned passive separation, and active force-tuned active separation. This type of method enhances the separation efficiency of the involved passive or active separation alone because the additional force can favorably change the separation parameter (e.g., reduce the particle and cell diameter that is separable) and provide an extra dimension of controllability. It is, however, still limited to one-parameter separation in most cases. More specifically, the passive force-tuned passive separation configuration has the advantages of easy operation/simple structure along with the disadvantages of limited controllability and separation parameters, like those for the serial-mode passive focusing-passive separation and passive separationpassive separation configurations. The active forcetuned passive and active separation configurations are both capable of manipulating the corresponding separation via an additional particle migration perpendicular to the flow path that depends on the morphology and/or the electrical, magnetic, acoustic, or optical property of particles and cells.
A summary of these integrated particle and cell separation methods in microfluidic devices is presented in Table 2, where the related information on the sample (type), medium, separation tag, (flow) throughput, and efficiency (or purity) is also included for each demonstration. We observe three trends/deficits from this table and suggest that they may represent the future research directions in the field: (1) No more than 10% of the reported multimode separations can offer a throughput of 10 mL/h or more, among which only two separations are able to attain 50 mL/h throughput [60,83]. Even the latter number is still insufficient for some biomedical, environmental, and industrial applications [125,126]. There are potentially two approaches to improving the throughput with one being to parallelize the active separation for matching in series the flow rate of high-throughput inertial/elastic microfluidics [127], and the other being to parallelize the inertial or viscoelastic microchannel flow [128] in the active force-tuned passive separation. (2) All but one [75] of the reported multimode separations with 10 mL/h throughput are based on magnetization. The next active separation method with the potential to attain high throughput is based on DEP. It is therefore worthwhile to pay more attention upon the production of high-quality magnetic particles for MACS or biocompatible magnetic liquids (e.g., ferrofluid) for diamagnetic particle/cell separation [129]. It is also beneficial to improve the fabrication of microelectrodes as well as the production of high-quality dielectrophoretic tagging particles. (3) Only four of the reported multimode separations are capable of working with submicron or nanoscale particles [92,93,96,112], all of which are associated with DEP. This is because DEP can be enhanced relatively easily by increasing the imposed voltage or decreasing the working distance at the micro/nanoscale. It is therefore again important to improve the fabrication and patterning of microelectrodes as well as the exploration of nonlinear electrokinetics [130][131][132]. Another approach worthy of notice is the recently developed oscillatory inertial [133] and viscoelastic [134] microfluidics that have been demonstrated for the focusing and separation of submicron and nanoscale particles, respectively. This approach may be integrated with other reported submicron/nanoparticle separation methods [135][136][137][138] into one microfluidic device for improved separation. In addition, the nearfield photonic manipulation of nanoparticles [139] may become an integrated part of a multimode separation system.

A C K N O W L E D G M E N T S
This work was supported in part by Clemson University through a SEED grant and Creative Inquiry Program (X. X.), and by University 111 Project of China under Grant number B08046 (Y. S.).

C O N F L I C T O F I N T E R E S T S TAT E M E N T
The authors have declared no conflict of interest.

D ATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.