Continuous flow separations in microfluidic devices

Nicole Pamme

Received 20th August 2007, Accepted 2nd October 2007

First published as an Advance Article on the web 2nd November 2007

DOI: 10.1039/b712784g

Biochemical sample mixtures are commonly separated in batch processes, such as filtration,

centrifugation, chromatography or electrophoresis. In recent years, however, many research

groups have demonstrated continuous flow separation methods in microfluidic devices. Such

separation methods are characterised by continuous injection, real-time monitoring, as well as

continuous collection, which makes them ideal for combination with upstream and downstream

applications. Importantly, in continuous flow separation the sample components are deflected

from the main direction of flow, either by means of a force field (electric, magnetic, acoustic,

optical etc.), or by intelligent positioning of obstacles in combination with laminar flow profiles.

Sample components susceptible to deflection can be spatially separated. A large variety of

methods has been reported, some of these are miniaturised versions of larger scale methods,

others are only possible in microfluidic regimes. Researchers now have a diverse toolbox to

choose from and it is likely that continuous flow methods will play an important role in future

point-of-care or in-the-field analysis devices.

Introduction

The development of microfluidic devices for analytical and

bioanalytical chemistry has, in the last two decades, led to

ground breaking advances in terms of the speed of analyses,

the resolution of separations and the automation of proce-

dures.1–3 Microfluidic chips have proven ideal tools to

precisely handle small volumes of samples, such as proteins

or DNA solutions, as well as cell suspensions.4,5 Additionally,

microfluidic devices can form part of portable systems for

point-of-care or in-the-field detection. In such micro total

analysis systems (microTAS) an entire analytical procedure

can be performed, including sample pre-treatment, labelling

reactions, separation, downstream reactions and detection.

However, to date this ideal case has seldom been realised.

One notable challenge is certainly the fact that on-chip separa-

tions are usually carried out as batch procedures, requiring the

precise injection of a very small amount of sample (nL or less)

into a separation channel for chromatography or electro-

phoresis (Fig. 1a).

An exciting development within the microfluidics commu-

nity has been the investigation of continuous flow separation

methods (Fig. 1b). This concept has gained momentum in

The University of Hull, Department of Chemistry, Cottingham Road,Hull, UK HU6 7RX. E-mail: n.pamme@hull.ac.uk;Fax: +44 1482 466416; Tel: +44 1482 465027

Nicole Pamme obtained aDiploma in Chemistry fromthe University of Marburg,Germany, in 1999. For herPhD she went to ImperialCollege London, where sheworked on single particleanalysis in microfluidic chipsuntil 2004. This was followedby a stay in Tsukuba, Japan asan independent research fellowin the International Centre forYoung Scientists (ICYS) atthe National Institute forMaterials Science (NIMS).In December 2005 she took

up a lectureship position at the University of Hull (UK). Hercurrent research interests include continuous flow separation inmicrofluidic devices based on magnetic forces.

Nicole Pamme

Fig. 1 (a) A typical batch separation procedure entails the injection

of a small amount of sample into a separation column and detection

after the separation has been completed, often followed by repeats

to optimise separation parameters. (b) A typical continuous flow

separation procedure: sample is injected continuously together with a

carrier liquid into a wide separation chamber, a force acts at an angle

to the direction of flow and sample components are deflected from

their flow path and thus spatially separated. Separation efficiency can

be monitored in real-time and separation parameters can be varied to

optimise conditions.

CRITICAL REVIEW www.rsc.org/loc | Lab on a Chip

1644 | Lab Chip, 2007, 7, 1644–1659 This journal is � The Royal Society of Chemistry 2007

recent years. Here, the sample is fed into a separation chamber

continuously and subjected to a force at an angle, often

perpendicular, to the direction of flow. The sample compo-

nents respond differently to this force and are thus deflected

from the direction of flow and can be collected at different

outlets. Researchers have tried a wide variety of forces to

fractionate sample components. These include the application

of electric or magnetic fields, as well as standing ultrasonic

waves and/or the intelligent design of flow obstacles. Some of

these methods are miniaturised versions of pre-existing larger

scale procedures, although with the added advantage of

favourable scaling laws. Other continuous flow separation

techniques, however, are only possible on the microfluidic

scale. In this review, the various microfluidic based continuous

flow separation methods are described and compared in terms

of performance and applicability.

Motivation for continuous flow separation

The separation of sample components is an important part of

any chemical or biochemical analysis. Common separation

processes include filtration and centrifugation and most

prominently, forms of chromatography and electrophoresis.

All these methods can be classified as batch procedures, since

in each case a particular volume of sample undergoes a

separation process and separation time is plotted versus

signal (see Fig. 1a). The efficiency is evaluated after the

separation has finished. More often than not, the separation

conditions, such as flow speed, carrier liquid composition

or instrumental parameters are not ideal and the sample

injection and separation has to be repeated several times before

optimal conditions have been found. Not only is this rather

labour and time intensive, it also poses challenges for the

combination of the separation step with upstream or down-

stream applications.

The general principle of a continuous flow separation is

shown in Fig. 1b. One of the key features is the fact that a

sample solution is introduced into the separation channel

continuously. It is not necessary to inject a small and precise

amount of sample, as in chromatography or electrophoresis. A

force acts on the sample components at an angle to the

direction of flow, often perpendicular, resulting in the sample

components taking different paths through the separation

channel and thus being spatially separated. Signal is detected

versus position. This allows for both continuous feedback for

the separation parameters and for continuous collection of

sample components.

The advantageous or characteristic features of continuous

flow separation can be summarised as follows:

(a) Continuous introduction of sample

Precise injection of small sample plugs is not required. A

relatively high throughput of sample can be achieved,

although, given the miniaturised nature of the channels,

volume throughput rarely exceeds mL min21. Another aspect

of continuous injection is that there is no capacity limit for

the separation column; in principle any volume of sample can

be processed.

(b) Continuous readout of separation efficiency

Since the separation can be monitored continuously, changes

to the separation parameters can be implemented and their

effects observed in real-time—enabling on-line feedback. Often

parameters for the fluid flow and the induced force can be

changed independently from each other.

(c) Lateral separation of sample components

The sample components are spread out orthogonally to the

direction of flow and can be continuously and independently

collected. Devices may have two outlets or multiple outlet

channels. In principle, it is possible to guide a desired sample

component for further processing downstream, whilst discard-

ing all other components. Alternatively, a sample can be

purified of undesired components and then used for down-

stream applications.

(d) Potential for integration

Since the sample is introduced continuously and the sample

components are collected continuously, separation methods

can be integrated relatively easily with upstream or down-

stream analysis steps. In other words, separation can be

performed ‘‘in-line’’ with other continuous flow processes,

such as continuous flow reactions. There is no large increase in

resistance due to a filter.

(e) Label-free

Separation is often based on intrinsic properties of the sample

components, such as their size, charge or their polarisability.

In such cases, no labelling of sample components is required.

In the literature, the term ‘‘continuous flow separation’’ is

sometimes used in different contexts. ‘‘Separation’’ often

appears in the context of ‘‘trapping’’ or ‘‘holding’’ a sample

component in a fixed position against a continuing fluid flow.

For example, magnetic particles can be held within a channel

by means of an external magnet, or a sample component can

become immobilised within a channel packing. Although these

methods are carried out under a continuous flow of carrier

fluid, the sample components themselves are not collected

continuously, and hence these methods are not reviewed here.

The term ‘‘continuous flow separation’’ is also sometimes used

for liquid–liquid phase extraction6,7 or for two-phase parti-

tioning.8,9 In these methods the separation is not based on a

specific force applied at an angle to the direction of flow.

Instead, sample components diffuse from one phase to the

other and remain in their preferred phase. Such phase

extraction methods are also not covered here.

A list of many of the on-chip continuous flow separation

methods that are covered in this review is featured in Table 1.

The large variety of forces that can be applied becomes clear

immediately. Some methods are very simple, others require

more sophisticated equipment. Examples of more simple

methods include pinched flow injection, hydrodynamic filtra-

tion or bifurcation, which make use of laminar flow behaviour

in microfluidic channels. Microfabricated obstacles, such as

pillars or dams, have been applied for controlled size

This journal is � The Royal Society of Chemistry 2007 Lab Chip, 2007, 7, 1644–1659 | 1645

separations, sometimes in connection with electric fields,

such as in the ‘‘DNA prism’’. Free-flow separation methods,

such as electrophoresis, isoelectric focussing, magnetophoresis,

dielectrophoresis and acoustophoresis, are based on homo-

geneous electric fields, pH gradients, inhomogeneous magnetic

fields, inhomogeneous electric fields or acoustic standing

waves, respectively. Other examples of applied forces include

gravity and optical pressure. The force utilised dictates the

type of sample that can be separated. The different separation

methods are discussed in detail below, arranged by type of

separation force.

Channel design and flow rates

Flow regimes in microfluidic devices are predominantly

laminar. With an appropriate microfluidic design and control

of hydrodynamic flow, it is possible to force microparticles

or cells into specific flow stream lines and to separate

these objects.

Pinched flow fractionation

Pinched flow fractionation (Fig. 2), developed by Seki and

Yamada,10 allows for separation of particles and cells based on

their size. Two laminar streams are pumped through a narrow

channel section before entering a widening channel. The

sample stream contains a particle suspension; the carrier

stream contains a buffer solution. The flow rate of the carrier

stream is higher than that of the sample stream, and hence the

microparticles are pushed against the wall in the pinched

channel segment. Small particles find themselves in a flow

stream close to the wall; larger particles have their centre of

gravity further away from the wall, it might even be outside the

original sample stream and instead within the carrier stream.

Table 1 Listing of continuous flow separation methods with details of the forces utilised, the basis of separation and the applicable samples. Thedata for flow rates and throughput were taken from the selected reference, if provided

Method Separation induced by Separation based on Sample Flow rate/throughput Ref.

Pinched flow fractionation Laminar flow regime Size Microparticles, cells 70 cm s21 1140 000 particles h21

Hydrodynamic filtration Laminar flow regime Size Microparticles 1 mm s21 14100 000s particles h21

Bifucation Laminar flow Size, shape Blood 5 mL h21 161–3 mm s21

Filtration obstacles Diffusion and obstacles Size Blood 5–12 mL min21 1910–20 mm s21

Lateral displacement Obstacle array Size Particles (0.8, 0.9, 1.0 mm) 20 mm s21 (DNA) 24DNA (61 kbp, 158 kbp) 100 pg h21 (DNA)

Brownian ratchet Diffusion and obstacles Size DNA 2 mm s21 2610 pg h21

Hydrophoreticseparation

Pressure gradient fromslanted obstacles

Size Microparticles 1 to 3 cm s21 3010 000s beads h21

DNA prism Obstacle array andelectric field

Size DNA 10 ng h21 31

Entropic trap array Dams and electric field Size DNA 10 pg h21 32

Repulsion array Dams, electric double layers Charge to size ratio Proteins 300 pg h21 32

Free-flow electrophoresis Homogeneous electric field Charge to size ratio Proteins, amino acids 6 mm s21 44

Free-flow isoelectricfocussing

pH gradient Isoelectric point Proteins and cells 9 mm s21 45

Magnetophoresis Inhomogeneousmagnetic field

Size, magnetisation Magnetic particles, cells up to 1.2 mm s21 8110 000s cells h21

Dielectrophoresis Inhomogeneous electric field Size, polarisability Microparticles, cells up to 4 mm s21 63100 000s cells h21

Acoustophoresis Acoustic pressure Size, density,compressibility

Microparticles, cells 11 cm s21 90

Optical lattice Optical force Size, refractive index Microparticles, cells 30 mm s21 9290 000 particles h21

Sedimentation Gravity Density, size Microparticles, droplets 6 mm s21 slowedto 1 mm s21

94

1646 | Lab Chip, 2007, 7, 1644–1659 This journal is � The Royal Society of Chemistry 2007

When the fluid enters the widening channel, the laminar flow

carries the two particle populations along different stream

lines. The particles are thus separated across the width of

the wide channel. If required, they can be collected further

downstream.

Initially, the separation of 15 mm and 30 mm polymer

particles was demonstrated through a pinched segment of

about 50 mm width.10 Subsequently, the method was modified

to improve performance. Outlet channels were added and

negative pressure was applied to one outlet.11 Most of the

carrier liquid was forced to leave via this outlet, whereas the

remaining flow streams could fan out more over the remaining

exits. This enabled the separation of 1 mm and 2 mm particles.

This device was also applied to the focussing of red blood cells.

In another design, flow into the outlet channels was controlled

by addition of a valve,12 which enabled the separation of 1 mm,

2 mm and 3 mm diameter particles from each other, as well as

the separation of 0.5 mm and 0.8 mm particles.

A similar concept was also published by Zhang et al.;13 their

design featured three inlet channels leading into a bent and

asymmetrically widening separation channel. By varying the

ratio of sample and buffer flow rates, microparticles could be

forced onto a specific flow stream. Any differences in initial

alignment were then enhanced by the flow profile within the

bent channel. 10 mm and 25 mm particles were separated from

each other.

Pinched flow fractionation is based on laminar flow and

channel geometry, and is suitable for the separation of

polymer particles of micrometre and sub-micrometre12

diameters as well as for biological cells.11 The separation is

based on particle size, no labelling is required and throughput

(see Table 1) can be 10 000s particles per hour. The dimensions

of the pinched segment have implications for the applicable

size range, and a significant amount of fine tuning has been

necessary for the separation of sub-micron particles

Hydrodynamic filtration

Hydrodynamic filtration is based on channel design and flow

control and can be utilised for the sorting of micrometre sized

particles.14,15 The microfluidic design features a main channel

with multiple narrow channels branching to the sides (Fig. 3).

Fluid is pumped along the main channel, so that some liquid

exits via the branch outlets. The amount of liquid leaving the

main channel depends on the flow resistance of these side

channels. When a mixture of particles is pushed through the

main channel it is important to note that smaller particles can

occupy flow streams much closer to the channel wall than

larger particles, simply because the centre of a particle cannot

come closer to the wall than its radius. In Fig. 3a, a situation is

shown where none of the particles can reach a flow stream

close enough to the wall to enter a side channel. Due to the

constant removal of liquid into the side channels, all particles

eventually become aligned against the wall of the main

channel, but they cannot enter the side channels. In Fig. 3b,

the flow resistance of the side channels is reduced and more

fluid can enter. Now, the centre of a small particle can be

positioned within a stream line that enters the side channel,

whereas the centre of a larger particle is still too far away from

Fig. 2 The principle of pinched flow fractionation for the separation

of microparticles or cells, based on laminar flow behaviour and

negligible diffusion of the particles. A buffer stream and a particle

carrying stream are forced through a narrow, pinched channel segment

before entering a widening channel. Due to the higher flow rate of the

buffer stream, the particles are pushed against the wall of the pinched

segment and are thus transported into specific laminar flow streams

based on their size. This difference is enhanced as the particles enter

the widening channel. Redrawn with permission from ref. 10,

copyright 2004, American Chemical Society.

Fig. 3 The principle of hydrodynamic filtration: (a) only a small

portion of flow enters the branching channel and neither large nor

small particles can be in a flow stream close enough to the wall of the

main channel to enter the side channel. However, after passing a series

of such branching channels, more and more fluid is removed from the

main channel and the particles are gradually aligned against the main

channel wall. (b) With more flow entering the side channels, the small

particles can now be close enough to the wall of the main channel to be

in a flow stream that enters the side channel. Large particles cannot

enter the branch channels. However, after passing several branch

points they become aligned against the main channel wall. (c) Finally,

if a relatively large portion of the flow enters the side channels, both

small and large particles can enter the side channels. (d) Design of

hydrodynamic filtration channel network for enrichment of small and

large particles. Redrawn with permission from ref. 14, copyright 2005,

Royal Society of Chemistry.

This journal is � The Royal Society of Chemistry 2007 Lab Chip, 2007, 7, 1644–1659 | 1647

the wall and hence larger particles are forced to flow straight

on. Any particles that are initially far away from the channel

wall will be brought closer to the wall each time they pass an

outlet branch, until eventually they are aligned along the wall.

Finally, if the flow into the side channels is increased further

(Fig. 3c), both large and small particles can enter the branches.

Using such a channel network, particles of different sizes can

be aligned against the main channel wall and collected into

different side channels (Fig. 3d).

In a first demonstration,14 a suspension of polymer particles

of 1 mm and 3 mm diameter was pumped through a 20 mm wide

and 7 mm deep channel featuring 50 side channels at a flow rate

of 1 mL min21. The particles were collected in different outlets

and could be concentrated 20-fold and 50-fold. Different

channel network designs were utilised for the concentration

of 1 mm and 2 mm particles and for leukocyte enrichment,

respectively. The design was further improved by splitting and

re-combining flows15 for more efficient aligning of particles

against the main channel wall. Particles of 1 mm, 2 mm and 3 mm

diameter were collected into different outlet reservoirs and

their concentrations were increased 60-fold to 80-fold.

Although this method does not allow for 100% separation

efficiency, it is a very effective way of enriching particle

concentrations according to size. Flow rates are fast (ca.

10 cm s21), and throughput can be 100 000s particles per hour.

The separation is determined purely by channel geometry,

minimal tuning is required after microfabrication and the

devices are thus simple to operate.

Hydrodynamic blood separation

Blood circulation in microcapillaries shows an intriguing

behaviour which can be utilised in microfluidic devices for

blood separation in continuous flow. Red blood cells have a

tendency to flow in the centre of the stream, blood at the edge

of the channel is thus depleted in erythrocytes and this can be

withdrawn continuously to obtain plasma with a low red blood

cell concentration (hematocrit).16 When blood encounters a

channel junction, the red blood cells have a tendency to flow

into the branch with the faster flow rate; this channel will thus

exhibit a higher red blood cell concentration, whereas the

channel with the lower flow rate will feature a lower hematocrit.

This phenomenon is sometimes referred to as ‘‘bifurcation law’’

or the ‘‘Zweifach–Fung effect’’ and has been demonstrated for

plasma separation in continuous flow.17 Another phenomenon

is referred to as ‘‘margination’’. Collisions between red blood

cells that are predominantly found in the microchannel centre

and white blood cells lead to the white blood cells rolling along

the channel wall. This process has been used for leukocyte

enrichment in continuous flow.18

Continuous flow filtration

Filtration usually means that a sample is pumped through a

membrane or a pillar array. This method is prone to clogging.

An alternative is to localise filter components, such as pillars

or posts, in-line with the direction of flow. If a particle

suspension is pumped through a channel with pillars on either

side, sample components that are sufficiently small can pass

through the gaps between the pillars into a neighbouring

receiving stream, whereas larger sample components remain in

the original sample stream. Such a diffusive filter is not prone

to clogging, since the larger size particles can continue to flow

through the device in a direction parallel to the filter.

Diffusive filters have been employed for differentiation

between red and white blood cells.19,20 Leukocytes are

spherical (6–10 mm diameter), erythrocytes are biconcave with

a diameter of about 8 mm and a thickness of about 2 mm. By

fabricating posts with appropriate gaps on either side of a

blood carrying channel, the red blood cells passed through the

gap and entered a parallel channel beyond the posts, whereas

the white blood cells remained in the central channel. Such

separations have been carried out at flow rates of 5 to

10 mL min21, equalling speeds of about 10 to 20 mm s21.

Continuous flow filtration has also been demonstrated with

an asymmetric pillar array at a channel junction featuring

narrow gaps (32 nm) for flow along the straight main channel,

but wider gaps for flow into a channel branching off at 45u.This has been applied to the separation of DNA molecules of

300 kbp (57 nm radius of gyration) and 100 kbp (22 nm

radius).21 Again, this type of filter is less prone to clogging

than a head-on filter, since larger components can flow away

into the side channel.

Another option of continuous flow filtration has been the

application of a cross-flow through shallow (3 mm) channels

running perpendicular to a 30 mm deep sample carrying

channel.22 The orthogonal flow led to continuous exchange of

medium in the main channel, which was demonstrated for a

suspension of 10 mm polystyrene beads. The device was also

tested for blood; red blood cells could pass into the side

channels and their concentration in the main channel was

reduced by a factor of 4000, whereas almost all white blood

cells were retained. A cross flow device with even shallower

side channels of 0.5 mm depth was used to obtain plasma from

blood, albeit only at 8% efficiency.23

Movement around obstacles

Obstacles can not only be used as filters to prevent or block

access. Flow around obstacles can also be used in clever ways

to separate particles or DNA molecules.

One such method has been termed ‘‘deterministic lateral

displacement’’ (Fig. 4)24,25 and has been applied to the size

separation of particles and DNA molecules by pumping

through an array of obstacles in the y-direction. The array

consisted of rows of posts with a consistent gap between the

posts in each row. The posts, however, were slightly shifted

from one row to the next, until after a few rows they aligned

again; in the example in Fig. 4, this is after three rows. Liquid

pumped through this array flowed around the posts in laminar

streams, as indicated by the different grey shades in Fig. 4. If

the liquid contained particles that were small in comparison to

the gap between the posts, then the particles followed the

direction of laminar flow. After three rows of obstacles the

particles were in the same position in the x-direction as in

the first row. If however, particles with a diameter only a little

smaller than the gap were pumped through the array of posts,

the particles had to position themselves in the middle of the

gap. The particle’s centre of gravity was thus pushed out of the

1648 | Lab Chip, 2007, 7, 1644–1659 This journal is � The Royal Society of Chemistry 2007

original flow stream into the neighbouring flow stream. After

three rows of obstacles the position of the particle had been

shifted in the x-direction, i.e. it had been laterally displaced.

With a gap between posts of 1.6 mm it was possible to

separate fluorescent particles with diameters of 0.6 mm, 0.8 mm

and 1.0 mm as well as particles 0.7 mm and 0.9 mm in diameter.24

Pressure driven flow was applied at a rate of 400 mm s21 and

the separation was detected after a running time of 40 s. By

gradually increasing the gap size from 1.4 mm to 2.2 mm

along the array, particles with diameters of 0.8 mm, 0.9 mm and

1.0 mm were separated with high resolution. The same tech-

nique was also used to separate DNA molecules of 61 kbp

and 158 kbp. The DNA containing solution was pumped via

EOF at about 20 mm s21 through an array with 3 mm gaps. The

two sizes of DNA molecules were separated from each other

within minutes.

The performance of this displacement method is certainly

impressive in terms of speed and particle size resolution. The

separation is based on inherent properties of the particles or

DNA molecules, no labelling is required. The path through the

obstacle array is pre-determined by the geometry of the array.

Each particle can only take one possible path. There is no

averaging of flow paths. Only diffusion could throw the

particle off-path. Hence, resolution of the separation is better

at faster flow rates, which makes this separation method

conceptually different from most other separation methods.

Another separation method based on an asymmetric array

of obstacles, a so-called ‘‘Brownian ratchet’’, has been

employed for the continuous flow separation of DNA

molecules based on their diffusion coefficients (Fig. 5).26–29

The DNA molecules were sufficiently small to pass through

the gaps in the array. They were pumped at relatively slow flow

rates, so that they had some time to diffuse before encounter-

ing the next row of obstacles. A large molecule (low diffusion

coefficient) would on average diffuse a much shorter distance

than a small molecule (high diffusion coefficient). Hence,

the larger DNA molecules followed the laminar flow in the

y-direction and passed almost straight through the array (solid

line). A smaller DNA molecule, however, could diffuse further

and could indeed travel far enough in the x-direction to be

forced around an obstacle (dotted line). Smaller DNA

molecules were thus likely to take a flow path shifted in the

positive x-direction and after a number of obstacle rows the

two populations of molecules would be separated. Of course,

diffusion also occurred in the negative x-direction. However,

the obstacle array was designed such, that the molecules would

have to travel a much larger distance to be displaced in this

direction (see grey cone in Fig. 5). The concept was initially

demonstrated by showing different flow paths taken by 15 kbp

and 33.5 kbp DNA.27 After improving the sample injection

method, the separation of lambda DNA (49 kbp) from T2

DNA (167 kbp) was achieved,28 as well as the separation of T2

and T7 DNA.29

Separation through a Brownian ratchet is a quite elegant

method, molecules are differentiated based on their diffusion

coefficient, which is directly related to their size. No labelling is

required. Once the obstacle array has been designed and an

optimum flow rate has been identified, the operation is very

simple. Fine tuning of the array design and flow rate for the

diffusion coefficients of the sample components however are

not trivial and a particular array design might only be suitable

for a limited range of DNA sizes. The method has two

drawbacks in comparison to lateral displacement separation

(Fig. 4). (i) Flow rates for the Brownian ratchet must be very

slow (few mm s21) to give molecules enough time to diffuse.

With an array length of 12 mm, the separation can take more

than an hour to complete,28 whereas lateral displacement

separations were completed within a few minutes. (ii) The flow

path through the Brownian ratchet is based on probability,

and it is not precisely pre-determined. Molecules within a

Fig. 5 The principle of diffusion based separation through a

Brownian ratchet. Molecules are pumped slowly in the y-direction

through an asymmetric array of obstacles. The gaps are sufficiently

large to let all molecules pass. Due to the slow flow rate, the molecules

are given some time for diffusion before encountering the next row of

obstacles. Whilst a molecule with a large diffusion coefficient might

migrate far enough to move around the obstacle (dotted line), a larger

molecule with low diffusion coefficient will be largely following the

laminar flow in y-direction (solid line). It should be noted that the

obstacles are arranged asymmetrically, the distance around an obstacle

in the negative x-direction is much larger than in the positive

x-direction (see grey cone), and hence there is a very low probability

for a molecule to take a path in the negative x-direction. Left part of

figure redrawn with permission from ref. 26, copyright 1998 by the

American Chemical Society; right part of figure redrawn with

permission from ref. 29, copyright Wiley VCH Verlag.

Fig. 4 The principle of particle separation via lateral displacement

around obstacles. Small particles follow the laminar flow streams

(shaded in grey) in the y-direction, whereas large particles are

continuously forced to change the laminar flow stream and are thus

continuously displaced in the x-direction. Redrawn from ref. 24, with

permission of AAAS.

This journal is � The Royal Society of Chemistry 2007 Lab Chip, 2007, 7, 1644–1659 | 1649

population will take a range of paths, leading to relatively wide

bands. The flow path in the deterministic lateral displacement

experiments is pre-determined and bands are narrow.

Recently, ‘‘hydrophoretic separation’’ of 9 mm and 12 mm

particles was demonstrated by implementing slanted obstacles

at the bottom and top of a microfluidic channel.30 These

obstacles generated local flow perturbations, in the form of

helical re-circulations, perpendicular to the main direction of

flow. The obstacles were arranged such that initially particles

were focussed at the bottom of the microchannel and

subsequently deviated from their direction of flow depending

on the size difference between the particle and the obstacle

gap. Large particles were found to flow in the centre of the

channel, whereas smaller particles were directed onto an

oscillating path.

Similar to the other obstacle based separation methods

mentioned above, hydrophoretic separation does not require

any labelling of sample components and the devices are

relatively easy to fabricate and to operate. Remarkably, the

sorting performance is independent of the applied flow rate

in the range of 0.7 to 3.3 cm s21. Only the geometry of the

obstacles determines the separation efficiency. This, however,

poses limitations for the range of particle sizes. For example,

with a gap of 25 mm, sorting was limited to particles between

8 mm and 12 mm in diameter.

Separations involving electric fields

Electric fields have long been popular for microfluidic

separations. Due to the large surface to volume ratios, Joule

heating can be dissipated rapidly, allowing for the application

of high electric fields and thus efficient separation of amino

acids, DNA molecules, proteins and peptides. It is not

surprising then, that electric fields have been used extensively

for continuous flow separation. For the purpose of this

review, these methods have been divided into four groups: (i)

separation involving electric fields in combination with

obstacles, (ii) free-flow electrophoresis based on charge to size

ratio, (iii) free-flow isoelectric focussing based on isoelectric

point (pI) and (iv) dielectrophoretic separation based on

polarisability.

Electric fields and obstacles

Arrays of obstacles can be combined with electric fields for the

separation of charged molecules, such as DNA or proteins. A

‘‘DNA prism’’ was developed by Huang et al.,31 consisting of a

regular array of posts over which an electric field could be

applied parallel or diagonally to the posts. By continuously

switching between two fields, E1 and E2, the separation of

DNA molecules based on their molecular mass could be

achieved as follows (Fig. 6): at the time t0 a relatively large

electric field E1 was applied diagonally to the pillars and both

DNA molecules experienced a strong electrophoretic force in

direction of the field. Then at t1 the field direction was

switched and a smaller electric field E2 was applied parallel to

the array in the x-direction. The DNA molecules experienced a

relatively weak electrophoretic force in direction of E2. The

short DNA molecule could pass through the obstacle array

easily. The larger DNA molecule however, became entangled

in the pillars. Before this long DNA molecule could

disentangle itself, the electric field was switched back to E1

and both molecules were pulled along diagonally again. The

long DNA molecule thus moved back onto its original path,

the short DNA molecule however, had now been displaced

in the x-direction, or, in other words: long DNA molecules

moved along the direction of E1 and short DNA molecules

moved along the average direction of the two electric field

vectors. Using an array of 2 mm posts with 2 mm gaps, Huang

et al. demonstrated the separation of 61 kbp, 114 kbp, 158 kbp

and 209 kpb DNA within 15 s. Applied electric fields were of

the order of 20 to 250 V cm21 for pulse durations of a few tens

of milliseconds. This separation method is rather fast and

robust and the authors report a throughput of 104 molecules

per second or 10 ng of DNA per hour.

Obstacles do not have to be in the form of pillars. Shallow

gaps can also be used very effectively as demonstrated by

Fu et al.,32 who fabricated arrays of deep parallel channels

separated by shallow nanometre gaps. Orthogonal electric

fields were applied simultaneously over this array, pulling

molecules in the x- as well as in the y-direction. Movement

in the x-direction however, was hindered by the nanogaps.

Depending on the size of the gap, the thickness of the electric

double layer (Debye layer), and the size of the sample

components, three separation regimes could be differentiated:

Ogston sieving, entropic trapping and electrostatic sieving

(Fig. 7).

An array consisting of deep trenches in the y-direction (1 mm

wide, 300 nm deep), and shallow passages in the x-direction

(1 mm wide but only 55 nm deep) was used to demonstrate

the three regimes.32 The thickness of the Debye layer was

controlled via the buffer concentration: at 130 mM the layer

Fig. 6 Separation in a DNA prism, consisting of a regular array of

posts. A large electric field is applied diagonally (E1), a small electric

field (E2) is applied horizontally. By switching between the two fields,

DNA molecules of different length are separated. Depicted are a long

(light grey) and short (dark grey) strand of DNA. When E1 is applied

at the time t0, both molecules experience a strong electrophoretic force

and migrate along the field direction. When the smaller electric field E2

is applied at t1, the molecules experience a weaker electrophoretic force

in the direction of E2. Whilst the small DNA molecule can pass

through the post array easily, the long DNA molecule becomes

entangled. When switching back to E1 at t2, the large DNA molecule is

pulled back into its original path, whereas the short DNA molecule has

been displaced in the x-direction. Figure redrawn with permission from

Macmillan Publishers Ltd.: Nature Biotechnology,31 copyright 2002.

1650 | Lab Chip, 2007, 7, 1644–1659 This journal is � The Royal Society of Chemistry 2007

was 0.8 nm thick, whereas at 1.3 mM the thickness increased

to 8.4 nm.

Ogston sieving is based on steric hindrance. This regime

occurs when the double layer is thin in comparison to the gap

and the sample components are small enough to fit easily

through the gap. DNA strands of 50 bp to 766 bp, corres-

ponding to a strand length of 16 nm to 150 nm, were

suspended in a 130 mM buffer. When introduced into the

array, the larger DNA strand would only fit through the gaps

if correctly aligned. The smaller DNA stands, however, could

pass the gaps easily and would on average travel further

into the x-direction. By applying 35 V cm21 in the x-direction

and 75 V cm21 in the y-direction, the five different strand

lengths were separated.

Entropic trapping had previously been shown as a batch

procedure in a straight channel33 and has now been adapted to

continuous flow. For this regime, the DNA molecules were

considerably larger than the gap, ranging from 2.3 kbp to

23 kbp, corresponding to radii of gyration between 140 nm

and 520 nm. The electric field in the x-direction pulled the

DNA molecules into the trap, they became stuck for some

time, until, eventually, a long enough part of the strand, a

hernia, had ventured into the gap and the entire molecule

could get drawn through the narrow slit. Interestingly, the

larger DNA molecules passed through the gaps more easily.

This was explained by that fact that a large molecule is in

contact with a large surface area of the gap region and hence

there is a higher probability for hernia formation and

thus being pulled through the gap. Electric fields of Ex =

185 V cm21 and Ey = 100 V cm21 were used to separate the

five DNA sizes in less than 1 min. DNA throughput was about

130 pg h21.

Electrostatic repulsion is dominant if the Debye layer is of

considerable thickness with respect to the gap size. This was

demonstrated for two proteins of similar size but different

charge: lectin (49 kDa, pI 8–9) and streptavidin (53 kDa,

pI 5–6). Using a 1.3 mM buffer (pH 9.4), the more negatively

charged streptavidin molecules were repelled from the gap to a

larger extent than the less charged lectin molecules. Hence,

lectin was deflected further into the x-direction and the two

proteins could be separated within tens of seconds.

The nanogap array for continuous flow separation has only

been published recently and is still in its infancy. A striking

advantage is the versatility of the method, the same chip could

be used for a variety of samples and separation regimes. The

separation speed could be tuned by two independent para-

meters, Ex and Ey. Separations could be performed within

minutes, with throughputs of 100s pg h21.

Free-flow electrophoresis and free-flow isoelectric focussing

Free-flow electrophoresis (FFE) is performed in a shallow

chamber, as shown in Fig. 8a. Buffer and sample solutions are

continuously pumped into this chamber through numerous

inlet channels and collected via outlet channels. Perpendicular

to the direction of flow, a homogeneous electric field is

applied. Charged molecules are thus subjected to two flow

vectors: the hydrodynamic flow in the y-direction and

the electrophoretically induced flow in the x-direction. The

observed flow path is the sum of these two vectors. The

direction of deflection depends on whether the molecule is an

anion or a cation; the extent of deflection depends on the

charge to size ratio. Molecules with zero net charge are not

deflected and flow straight through the chamber. Free-flow

electrophoresis was initially developed on the larger scale.34,35

Miniaturisation of the separation chamber, typically to a width

and length of several mm and a depth of a few mm, brings

similar advantages to those for capillary electrophoresis: Joule

heating can dissipate quickly due to the larger surface to

volume ratio, hence, high electric fields can be applied, and

thus separation efficiency and speed are improved.

Fig. 7 Anisotropic sieving through shallow nanometre sized gaps.

Depending on the size of the gap and the thickness of the Debye layer,

three separation regimes can be distinguished. (a) In Ogston sieving,

the electric double layer is thin compared to the gap and sample

molecules are also smaller than the gap. Passage through the gap is

based on steric hindrance, the smaller molecules (light grey) pass more

easily than the larger molecules (dark grey). (b) Entropic trapping: if

the sample contains DNA molecules that are larger than the gap, the

DNA becomes trapped at the gap, but eventually a sufficient length of

DNA will have been pulled into the gap by the electric field, that the

entire molecule squeezes itself through the gap. Counter-intuitively,

larger DNA molecules move through the gap more easily. (c)

Electrostatic sieving: when the Debye layer is thick in comparison to

the gap size, electrostatic forces become dominant. A molecule of

higher negative charge is repelled more by the gap than a molecule of

low negative charge. Figure redrawn with permission from Macmillan

Publishers Ltd.: Nature Nanotechnology,32 copyright 2007.

This journal is � The Royal Society of Chemistry 2007 Lab Chip, 2007, 7, 1644–1659 | 1651

Free-flow isoelectric focussing (FF-IEF) is based on a

similar setup to free-flow electrophoresis, and hence many

researchers have demonstrated both applications with the

same device layout. The principle of FF-IEF is shown in

Fig. 8b. A pH gradient is generated in the x-direction, fluid is

pumped in the y-direction. A sample mixture, containing

amphoteric molecules, such as proteins, is introduced over the

entire width of the separation chamber. A sample protein

entering at a location where the pH is different from its

isoelectric point (pI) exhibits a net charge, and thus experiences

an electrophoretic force. The protein migrates along the

x-direction until it reaches a point in the pH gradient where

the pH equals its pI. The protein no longer has a net charge

and hence ceases to migrate in the x-direction. The protein is

focussed at its isoelectric point and now only follows the

direction of flow. FF-IEF allows for the focussing and

concentration of proteins based on their pI. Hence, proteins

with different pIs can be separated from each other.

On-chip free-flow electrophoresis and isoelectric focussing

were first demonstrated in the late 1990s and there have been a

relatively large number of publications since then. Differences

in performance result from the chip material and most notably

from the connection between the electrode reservoirs and the

separation chamber.

The first free-flow electrophoretic devices were fabricated in

silicon.36,37 However, the maximum electric field that could be

applied to this material was 100 V cm21. Later, glass devices38–41

and hybrid devices made from glass and silicon42,43 or glass

and PDMS44,45 were introduced, in which higher electric fields

could be applied, typically between 200 V cm21 and 300 V cm21.

In one case43 the field was as high as 580 V cm21.

The most challenging aspect of free-flow electrophoresis

concerns the generation of a high electric field across the

separation chamber. This must be undertaken whilst minimis-

ing fluid exchange between the chamber and electrode

reservoirs and also minimising the detrimental effect of the

products of electrolysis (local pH changes, bubble formation).

Researchers who opted for direct contact of the electrodes with

the separation chamber fluid were forced to limit their electric

fields to a few tens of V cm21.46–48 Most researchers chose to

separate the electrode reservoirs from the separation chamber

by some form of membrane. The disadvantage here is the

voltage drop across the membrane; only a fraction of the

applied field is present over the chamber itself. A simple way of

introducing a membrane is to connect the electrode reservoir

with the separation chamber via a number of microfluidic

channels. With small cross section channels, as used by Zhang

et al.44 and Xu et al.,45 only 4.5% of the applied electric field

was transferred across the chamber. The large cross section

channels, as reported by Fonslow et al.,42 would transfer as

much as 50% of the applied field, however, at the cost of a

facilitated fluid exchange. By removing connection channels

altogether and introducing trenches with high speed laminar

flow along the electrodes in the y-direction, as much as 91%

of the applied field could be transferred to the separation

section.43 Membranes made from UV-polymerised acryl

amide,41 nitrocellulose,38 or agarose gel49 have also been

applied successfully. Perhaps the most elegant way to separate

the electrode reservoir from the separation chamber is via a

dielectric barrier, such as a thin wall of glass, as demonstrated

by Janasek et al.:50 50% of the applied field could be

transferred across the chamber and liquid interchange could

be totally avoided.

Isoelectric focussing requires the generation of a pH

gradient across the separation chamber. This is conventionally

achieved by a mixture of ampholytes that establish the

gradient as soon as an electric field is applied. The challenge

again is that the electrodes are usually positioned outside the

separation chamber. The pH gradient should ideally not

extend beyond the separation chamber. Unfortunately, this is

sometimes unavoidable, for example when channels are used

to connect the electrode reservoirs and chamber. Xu et al.45

used a wide pH gradient (pH 3 to 10), of which only a fraction

would have been over the separation chamber, in order to

focus two similarly charged proteins, angiotensin I (pI 6.69)

and angiotensin II (pI 7.75). Kohlheyer et al.41 made more

effective use of a wider pH gradient and were able to focus four

proteins with quite different pIs (4.5, 5.5, 7.6 and 8.7). A

totally different approach was taken by Cabrera and Yager.47

Having the electrodes in direct contact with the separation

medium led to electrolysis and local pH changes in the vicinity

of the electrodes. These pH changes resulted in a ‘‘natural’’ pH

gradient that could be used for isoelectric focussing. Song et al.

took yet another approach;51 they surrounded their sample

containing stream (pH 6) by two buffer streams at pH 8.

Differences in ionic strength and pH between the streams led

to a diffusion based pH gradient, which again was used for

isoelectric focussing.

FFE and FF-IEF have been applied to a variety of samples.

Free-flow electrophoresis is suitable for the separation of

molecules with different charge to size ratios. For example,

fluorescent dyes, such as fluorescein or rhodamine 110,

Fig. 8 Free-flow separation methods usually feature a shallow

separation chamber with flow pumped in the y-direction and a field

applied in the x-direction. (a) In free-flow electrophoresis, a homo-

geneous electric field is applied over the chamber orthogonal to the

direction of flow. A sample mixture is fed into the chamber through a

narrow channel, surrounded by carrier liquid. Charged sample

components are deflected from the direction of laminar flow depending

on their charge to size ratio, whereas neutral components follow the

direction of flow. (b) In free-flow isoelectric focussing, a pH gradient is

generated over the separation chamber. Amphoteric samples, such as

proteins, are injected over the entire width of the chamber. They

migrate until they exhibit zero net charge, i.e. until they have reached

the pH that equals their isoelectric point.

1652 | Lab Chip, 2007, 7, 1644–1659 This journal is � The Royal Society of Chemistry 2007

fluorescently labelled amino acids or proteins have been

separated. The throughput depends on the size of the

separation chamber and the strength of the applied electric

field. Reported separation times vary greatly, ranging from

sub-second,44,45 to several seconds,39,41,42 to minutes.36–38

Free-flow isoelectric focussing is applicable to components

with different pIs. Samples reported in the literature include

the separation of fluorescent IEF marker molecules,41

proteins,45,46,51 but also mitochondria48 and bacteria.47 Due

to their large size, the separation of the cells took minutes.

Reported focussing times for smaller sample components,

however, vary significantly, again ranging from sub-second,45

to several seconds,41 to minutes.46

Another variation of free-flow separation based on electric

fields is free-flow isotachophoresis, as demonstrated by

Janasek et al.50 It should also be noted that the term ‘‘free-

flow’’ has been adopted for many other separation methods,

for example those based on magnetic gradients,52 acoustic

pressure53 or, recently, thermal gradients.54

Continuous flow dielectrophoretic separations

Dielectrophoresis (DEP) occurs in inhomogeneous electric

fields and can be utilised for the sorting of particles or cells.

When a particle is subjected to an electric field, charges are

induced and the particle becomes polarised along the direction

of the electric field. In a homogeneous electric field, the

electrostatic forces on opposite ends of the dipole are equal

and there is no net movement. In an inhomogeneous electric

field (Fig. 9), the forces on either side are different and the

net force results in movement of the particle. The movement

can be towards or away from the high intensity field. If

the electrical polarisability of the particle is larger than the

polarisability of the surrounding buffer medium, then the

particle moves into the electric field, this is called positive

dielectrophoresis (pDEP). If, on the other hand, the polarisa-

bility of the particle is lower than that of the surrounding

medium, the particle is forced away from the electric field

towards a low intensity region. This is referred to as negative

DEP (nDEP). The dielectrophoretic force thus depends on the

difference in polarisability between particle and surrounding

medium. The force is also proportional to the volume of the

particle, as well as to the gradient of the electric field. DEP

separation can thus be based on differences in polarisability or

size or a combination of the two.

One method used to generate an inhomogeneous electric

field in a microfluidic channel has been to apply a voltage

along a channel that contained obstacles, such as ridges,55

narrowing wedges56,57 or an array of posts.58 In the vicinity of

these obstacles, the applied electric field became non-uniform

and thus dielectrophoretic forces could be generated on

particles or cells that flowed along the edges of these obstacles.

Pumping of particle suspensions was achieved via electro-

osmotic flow and particle concentrations ranged from 105 to

107 particles mL21.

With ridges restricting access to part of the microchannel,55

bacterial cells could be continuously enriched from a mixture

of 200 nm polymer particles. When an electric field of at least

3 V cm21 was applied over the channel network, the bacteria

cells were prevented from crossing the 5 mm ridge gaps due to

nDEP, while the 200 nm diameter polymer particles could

pass. In another example, a 300 mm wide channel was

narrowed to 60 mm by a rectangular obstacle.56 Polymeric

microparticles experienced nDEP when flowing through this

narrowed channel section, the DEP forces depending on

their size and the applied voltage. The authors demonstrated

controlled sorting of 5 mm, 10 mm and 15 mm diameter particles

into two downstream exits. Using the same principle, an oil

droplet can also be used as a narrowing obstacle.57 Since the

size of the droplet can be altered, the DEP forces can be tuned.

This has been utilised for the separation of a mixture of 1 mm

and 5.7 mm polymer particles, as well as a mixture 5.7 mm and

15.7 mm particles. An array of posts was used for dielectro-

phoretic focussing of 200 nm latex particles.58 The posts were

diamond shaped, with a 36 mm edge, leaving 30 mm wide and

7 mm deep channels. When an electric field of 8 V cm21 was

applied, the latex particles experienced nDEP in the vicinity of

the posts and were found to concentrate in streamlines. The

shape of the posts and orientation of the array were found to

be crucial to achieve this DEP based focussing. Very recently, a

device for three-dimensional continuous dielectrophoresis was

theoretically evaluated and demonstrated for the separation of

2 mm and 3 mm particles.59

Inhomogeneous electric fields for DEP can also be generated

by microelectrodes embedded in the microfluidic channel.

Such microelectrodes could be a single stripe or many parallel

stripes. When switched on, they generate strong gradients,

albeit usually only over a limited volume. Often electrodes are

integrated above and below a microchannel to maximise the

DEP force. When particles are pumped over the electrodes, the

non-uniform electric field induces a dielectrophoretic force

orthogonal to the direction of flow. This can be used for the

sorting of particles or cell populations. Such continuous

flow sorters are usually operated under alternating current

in order to minimise electrolysis at the microelectrodes inside

the channel.

Sorting of 4 mm and 6 mm polymer particles was achieved in

a 500 mm wide and 28 mm deep channel, with embedded Pt

Fig. 9 The principle of dielectrophoresis (DEP): when subjected to an

electric field, a particle or cell becomes polarised. If the electric field is

inhomogeneous, then the electrostatic forces on the two ends of the

dipole are not equal and a movement is induced. (a) Positive DEP

occurs when the particle exhibits a larger polarisability than the

surrounding medium. Positive DEP is directed towards the stronger

electric field. (b) Negative DEP occurs if the particle is less polarisable

than the surrounding buffer medium. Negative DEP forces the particle

away from areas of high field intensity towards areas of low intensity.

Figure redrawn from ref. 63 with permission from Elsevier.

This journal is � The Royal Society of Chemistry 2007 Lab Chip, 2007, 7, 1644–1659 | 1653

stripe electrodes that were 50 mm wide and 50 mm apart and

arranged at a 45u angle to the direction of flow.60 When 10 V

peak to peak were applied to the electrodes at a frequency of

1 MHz, the particles were subjected to a nDEP force transverse

to the direction of flow. The 6 mm particles were deflected

further from the direction of flow than the 4 mm particles and

thus the two populations could be separated at a flow rate of

about 100 mm s21. A series of gold electrodes in the shape of

trapezoids were integrated at the bottom of a microchannel

and used to sort polystyrene particles of 6 mm and 15 mm

diameter via nDEP into two different outlet channels.61 A

voltage of 8 V at 50 kHz was applied to each electrode and

particles were sorted at a flow rate of 2 mm s21. In another

device, nDEP was used to block particles from entering a

channel.62 A gate electrode was integrated above and below a

100 mm central outlet channel. When 20 V was applied at MHz

frequencies to this gate electrode, 1 mm diameter polymer

particles were prevented from entering this channel and forced

to take adjacent channels. This valve was found to be 97%

efficient at a flow rate of 500 mm s21.

Viable and non-viable cells are often found to exhibit

different polarisabilities and can be sorted using DEP.63,64 At

low frequencies, the viable cells exhibit nDEP and the non-

viable cells pDEP. At high frequencies however, these forces

are reversed. Continuous flow sorting of live and dead yeast

cells was demonstrated in a 400 mm wide and 50 mm

deep channel with embedded gold electrodes.63 When 8 V

was applied at a frequency of 5 MHz, the viable yeast cells

exhibited pDEP, whereas the non-viable yeast cells exhibited

nDEP. Thus the two cell populations could be separated at a

flow rate as fast as 4 mm s21, with a cell throughput of

1000 cells s21. In another example, viable and non-viable

yeast cells were separated using 10 mm wide stripes of Pt/Au

electrodes, with a spacing of 10 mm that was integrated in

the bottom of a 600 mm wide and 100 mm deep separation

channel.64 Again, cells were deflected from the direction of

flow based on positive or negative DEP forces. The viable cells

were found to exhibit pDEP at 14 V and an applied frequency

of 100 kHz, whereas at the higher frequency of 2 MHz, the

viable cells were found to experience nDEP forces.

Cell separation via dielectrophoresis can be enhanced using

polymer particles as labels.65,66 The DEP force on an E. coli

cell attached to a polymer particle was found to be about

100 times larger than the force on an unlabelled cell.65 With a

20 mm deep microchannel featuring stripes of gold electrodes at

the top and bottom and an ac voltage of 20 V at 0.5 MHz, it

was possible to enrich 95% of the labelled E. coli cells. The cell

suspension was pumped at a flow rate of 3 mm s21 and thus a

throughput of 10 000 cells s21 was obtained. This device was

later improved to a two stage sorting system.66 Cells labelled

with 5.6 mm polymer beads had to pass two electrode arrays

to be collected. This device was operated for a period of

three hours with a throughput of 5 6 108 cells, equalling

about 50 000 cells s21.

Continuous flow dielectrophoretic sorting has so far only

been applied to a small range of samples, mainly polymer

particles and yeast cells. The method is mild enough to prevent

damage to biological cell samples and throughput can be tens

of thousands of cells s21.

Another interesting application involving inhomogeneous

electric fields was demonstrated by Leineweber and Eijkel.67

They integrated a regular array of electrodes into a micro-

channel. When applying a voltage, each electrode influenced

the local electrolyte concentration and molecular enrichment

against a concentration gradient was obtained. This was

used for continuous flow demixing: components of a

homogeneously distributed electrolyte could be focussed

into streams.

Separations involving magnetic fields

Continuous flow separation of magnetically susceptible

materials can be achieved by applying an inhomogeneous

magnetic field perpendicular to the direction of flow (Fig. 10).

A magnetically susceptible object, such as a magnetic

microparticle or a magnetically labelled cell, is pulled into

the magnetic field. The magnetic force depends on the gradient

of the applied magnetic field and also on the volume and

magnetisation of the particle. Particles of different sizes and/or

different magnetisation can be separated. The observed

particle trajectory is the sum of two flow vectors: the

hydrodynamic velocity arising from pumping the liquid

through the microfluidic system and the magnetically induced

velocity. Balancing of these two flow vectors is required

to generate sufficient deflection. General applications of

magnetism and magnetic particles in microfluidics have been

reviewed recently.68,69

Continuous flow magnetic separation methods were initially

developed on larger scales,70–73 but have been frequently

applied to microfluidics as well. Separators either feature two

outlets (Fig. 10a) or multiple outlets (Fig. 10b). The design

with two outlets is sometimes also termed split flow thin

fractionation (see below), whereas the design with multiple

outlets has been termed free-flow magnetophoresis, in analogy

to free-flow electrophoresis (see above).

Fig. 10 Continuous flow magnetic separations: an inhomogeneous

magnetic field is applied perpendicular to the direction of flow.

Magnetic particles or magnetically labelled cells are attracted into the

field and thus deflected from the direction of laminar flow. (a) Sorting

of magnetic particles is often carried out via two outlets. (b) In free-

flow magnetophoresis, multiple outlets are used, allowing for the

separation of different magnetic particles from each other, as well as

from non-magnetic material.

1654 | Lab Chip, 2007, 7, 1644–1659 This journal is � The Royal Society of Chemistry 2007

Separation of magnetic particles from a sample suspension

using a microchannel network with two outlets have been

reported, for example by Blankenstein et al.74 and by Luke

Lee’s group.75 In these earlier examples, researchers have

utilised small permanent magnets for particle deflection.

More recently, Kim and Park used a similar setup for an

immunoassay based on magnetic nanoparticles as labels.76

Microfabricated magnets in close proximity to the channel

can generate high magnetic field gradients, but have a limited

reach and might lead to extensive heating. Recently, a very

simple method for fabricating a micro-electromagnet has

been introduced by Siegel et al.,77 who filled a channel

adjacent to the particle carrying channel with liquid solder.

After solidification, the solder acted as wire and thus as an

electromagnet for continuous sorting of magnetic particles.

Most efficient for continuous flow separations are hybrid

magnets, a combination of a larger external magnet and a

small magnetisable feature in close proximity to the micro-

channel. For example, with a microfabricated NiFe comb and

an external magnet, a three times higher gradient could be

achieved in comparison to the external magnet alone; this in

turn allowed for the application of a three times higher flow

rate for the separation of magnetic microparticles or magne-

tically labelled E. coli cells.78 In another example, nickel wires

in close proximity to the microchannel were magnetised by an

external permanent magnet for the separation of red blood

cells in continuous flow.79

Free-flow magnetophoresis devices, as shown in Fig. 10b,

feature multiple outlet channels. Thus it is not only possible to

separate magnetic from non-magnetic particles but also to

separate different magnetic particles from each other. Using a

small permanent magnet, this has been demonstrated for

magnetic microparticles,52,80 as well as for cells labelled

with magnetic nanoparticles.81 In another example, micro-

fabricated magnetic Ni-stripes were embedded at the bottom

of the separation chamber at an angle to the direction of flow.

These stripes served to deflect magnetically labelled cells from

the direction of laminar flow.82

Magnetic manipulation has some distinct advantages over

electric manipulation. The magnetic field is applied externally;

there is no need for physical contact between the magnet

and any liquid. Unlike electrically induced forces, magnetic

forces are not significantly influenced by ionic strength, pH or

surface charges. Magnetic manipulation is mild and not

usually damaging to biological materials. However, magnetic

manipulation is rarely based on intrinsic properties. Most

materials do not exhibit intrinsic magnetism and hence

must be labelled with magnetic micro- or nanoparticles.

Exceptions include de-oxygenated red blood cells79,83 and

magnetotactic bacteria,84 which can be manipulated without

prior labelling. Magnetic forces on individual nanoparticles

are too small, given that the force is proportional to the cube

of the particle radius. Individual nanoparticles can generally

not be manipulated, agglomerates of thousands of nano-

particles are usually required. Even when using magnetic

microparticles, the applied flow rates tend to be somewhat

slow, hundreds of mm s21 or at most a few mm s21. This

imposes limitations on throughput in comparison to other

continuous flow methods.

It is less commonly known that non-magnetic materials,

or more specifically diamagnetic materials, are repelled by

magnetic fields. This effect is several orders of magnitude

weaker than magnetic attraction exhibited by ferromagnetic

materials. However, it has recently been demonstrated that

6 mm polystyrene particles can be deflected in continuous

flow using the high magnetic gradients generated by super-

conducting magnets.85

Separations involving sound pressure

Continuous flow separation of particles and biological cells

can also be performed with acoustic forces generated from

ultrasonic waves. For this, a standing sound wave must be

generated over the cross section of a microchannel, i.e.

orthogonal to the direction of flow (Fig. 11). Usually the

wave is tuned, such that the node is in the centre of the channel

and two anti-nodes at the edges. Such waves can be generated,

for example, with piezoelectric transducers. Particles or cells

subjected to the sound wave experience a force, either towards

the node or towards the anti-node.

A detailed description of the theory is beyond the scope of

this review and the interested reader is referred to reviews on

this topic by Coakley,86 and recently by Laurell et al.87 Briefly,

the acoustic force on a particle depends on the properties of

the acoustic field, as well as on the properties of the particle

and its surrounding medium. The acoustic field is charac-

terised by its frequency and amplitude: the higher the

frequency and the larger the applied voltage, the higher the

acoustic force. The required frequency is predetermined by

the width of the microchannel. For example, a channel of

Fig. 11 Continuous flow separation based on sound pressure. (a) A

standing ultrasonic wave is generated over the width of a microchannel

by means of an external transducer. Particles experience a force, either

toward the pressure node or the anti-node. The magnitude of the force

is influenced, amongst other parameters, by the particle size. The

direction of the force is determined by the ratio of the particle’s density

and compressibility in relation to the density and compressibility of the

surrounding buffer medium. (b) Once aligned in the sound field, the

different particles can be collected into separate outlets.

This journal is � The Royal Society of Chemistry 2007 Lab Chip, 2007, 7, 1644–1659 | 1655

1 mm width requires a wave frequency of about 100 kHz,

whereas a channel of 10 mm width requires 10 MHz. In short,

the smaller the channel, the higher the required frequency and

in turn the larger the acoustic force generated. The properties

of the particle and the medium in which it is suspended also

have a pronounced influence. The acoustic force is propor-

tional to the volume of the particle. Furthermore, it depends

on the density and compressibility of the particle in relation to

the density and compressibility of the surrounding medium.

This ratio determines the direction of the acoustic force

towards a node or anti-node. In general, a solid particle

suspended in an aqueous medium will be pushed towards a

pressure node, whereas a gas bubble will be pushed towards

an anti-node. For a given acoustic field, particles can be

separated if they differ in size, density or compressibility or a

combination of these factors. As for all continuous flow

separation methods, the flow rate must be adjusted such that a

particle spends sufficient time in the pressure field.

Microfluidic devices with ultrasonic transducers have been

used to focus polymer particles or cells into a stream and thus

to separate these from the bulk of the suspension. Hawkes

and Coakley88 fabricated a stainless steel device with a channel

10 mm wide and 250 mm deep, over which a standing

ultrasonic wave was generated with a frequency of about

3 MHz. Yeast cells of 5 mm diameter were introduced at a

flow rate of 6 mL min21 (40 mm s21). The cells were focussed

into the centre of the channel and collected in a central outlet

channel, whereas the clarified medium was collected in the two

outer outlet channels. A throughput of 2 6 108 cells min21

was achieved.

Laurell’s group have reported on a number of ultrasonic

separation devices. In a 750 mm wide and 250 mm deep silicon

channel, a first harmonic ultrasonic wave with two nodes was

generated at a frequency of 2 MHz.89 The device featured three

outlet channels. A suspension of 5 mm diameter polyamide

particles was focussed into the two nodes and collected via the

two outer outlet channels with 90% efficiency in 2/3 of the

original volume at a flow rate of 0.1 mL min21 (9 mm s21). In

a subsequent design, a 350 mm wide and 125 mm deep channel

was fabricated and a fundamental ultrasonic wave with one

node was generated at 2 MHz.90 In this device, almost 100% of

the 5 mm polyamide particles could be collected in the central

outlet channel, i.e. into 1/3 of the original volume at a flow rate

of 0.3 mL min21 (110 mm s21). More recently, a fluidic design

with a similar separation channel cross section, but with a

more sophisticated design of inlet and collection channels,53

was utilised to separate a mixture of polystyrene particles

of 2 mm, 5 mm, 8 mm and 10 mm diameter at efficiencies

between 62% and 94%. Due to the multiple outlets, the authors

termed their method ‘‘free-flow acoustophoresis’’. The same

device was also used to separate polystyrene particles of 3 mm,

7 mm and 10 mm with efficiencies between 76% and 96%.

The separation of polyamide microparticles and of red

blood cells from blood plasma was also demonstrated by

Kapishnikov et al.91

As mentioned above, particles of the same size can be

separated if they differ in density from each other and the

surrounding medium. This was demonstrated with 3 mm dia-

meter polystyrene and PMMA particles that were suspended

in an aqueous solution of 0.22 g mL21 cesium chloride.53 The

same CsCl solution was also applied to the ultrasound based

separation of red blood cells, leukocytes and platelets.

An interesting application of continuous flow ultrasonic

separation is the purification of blood from lipid droplets

during open heart surgery. The sonic force on red blood cells

suspended in plasma pushes the cells towards pressure nodes,

whereas lipid droplets are pushed towards pressure wave anti-

nodes. Petersson et al. demonstrated how on-chip continuous

ultrasonic separation could be used for this application.90

The elegance of acoustic separations stems from fact that no

physical contact is required between the ultrasonic transducer

and the liquid. A standing wave is generated inside the channel

by an external device. Ionic strength and surface charges are

not important, and according to the literature, the ultrasonic

treatment has not been shown to damage cells or biological

material. Notably, throughput can be quite high with flow

rates of tens of mm s21. The method, however, is not

applicable to small nanoparticles or small molecules, given

that the acoustic force is proportional to the cube of the

particle radius.

Continuous flow separation via optical forces

Optical forces have been applied extensively in optical tweezers

where a focussed laser beam is utilised to trap and move

microparticles or cells, and also for particle by particle sorting

in flow cytometry. Continuous flow separation, as indicated in

Fig. 1b, can be achieved by means of an optical interference

pattern, a so-called optical lattice. A three-dimensional optical

field can be generated, such that the optical force is directed

perpendicular to the direction of flow. The optical force on a

particle or cell depends on optical polarisability. Sorting can be

achieved based on differences in particle size or refractive

index. MacDonald et al. demonstrated the sorting of two types

of particles based on refractive index.92 In an H-shaped

channel device, they introduced a buffer solution, as well as a

suspension with a mixture 2 mm silica particles (n = 1.73) and

2 mm polymer particles (n = 1.58). When passing through the

optical field, the polymer particles were deflected into the

buffer carrying stream, whereas the silica particles flowed

straight on. Sorting was achieved with almost 100% efficiency,

at a flow rate of 30 mm s21, equalling a throughput of

about 25 particles s21. The same device was also utilised for

deflecting and separating protein capsules of 2 mm from a

mixture of 4 mm diameter capsules at a flow rate of 20 mm s21.

More recently, the same group demonstrated the separation

of a mixture of four differently sized silica particles (2.3 mm,

3.0 mm, 5.3 mm and 6.8 mm).93

Optical sorting is another example of a non-invasive

method. Separation is based on intrinsic properties, no

labelling is required. The optical force can be tuned

independently to the flow rate. The method can be applied

to inorganic, polymeric and biological particles.

Separation based on gravity

Gravity is an obvious force to utilise for continuous flow

separations. However, gravity forces on polymer particles or

1656 | Lab Chip, 2007, 7, 1644–1659 This journal is � The Royal Society of Chemistry 2007

cells suspended in buffers tend to be several orders of

magnitude smaller than, for example, dielectrophoretic or

magnetic forces. The gravity force is proportional to the

particle volume and also to the difference in density between

a particle and the surrounding buffer medium. Recently,

Huh et al. demonstrated the separation of 20 mm polystyrene

particles from 1 mm particles and also from 3 mm particles

in a microfluidic device (Fig. 12) by combining gravity

forces with hydrodynamic flow.94 A sheath flow was applied

to initially focus the particles parallel to the gravitational

field. The particle stream was then directed perpendicular to

gravity into a separation channel. By gradually widening this

channel, the flow rate slowed down and sedimentation started

to take effect. Differences in particle position were then further

enhanced by the hydrodynamic flow profile and channel

geometry, similar to pinched flow fractionation10 or asym-

metric cavity separation.13 Separation was achieved in less

than 1 min, with almost 100% efficiency. The method was also

utilised to separate perfluorocarbon droplets in water.

Split flow thin fractionation

Split flow thin (SPLITT) fractionation was originally

developed by Giddings95 and has been extensively used in

millilitre volume flow cells. The method is based on forcing

sample components out of a sample stream into a neighbour-

ing buffer stream by a force applied perpendicular to the

direction of flow. This force can be based on gravity,96

electrophoresis,97 or magnetism.98 In contrast to a conven-

tional H-shaped flow cell design, as for example shown in

Fig. 10a, SPLITT devices feature two flow splitters at the inlet

and outlet of the flow cell (Fig. 13). The carrier stream is often

introduced at a faster flow rate than the sample stream, thus

focussing the sample stream into a relatively narrow band, also

referred to as the inlet splitting plane. With no force applied,

all sample components leave via the outlet opposite the inlet.

When a force is applied, any susceptible material must be

pulled beyond the so-called outer splitting plane to be collected

in the second outlet channel.

Split flow thin fractionation has also been demonstrated

in microfluidic devices.99 For example, an electrophoresis

SPLITT cell was fabricated by Narayanan et al.,100 featuring

gold electrodes at the top and bottom of a 1 mm wide

and 40 mm deep channel. By applying an electric field of

300 V cm21, a mixture of 108 nm and 220 nm polymer

particles could be separated at a flow rate of 14 mm s21.

Conclusion

There are now an immense variety of continuous flow

separation methods available to the bioanalytical chemist for

the separation of particles, cells or DNA molecules, proteins

and smaller molecules.

Comparison of performance is not always straightforward:

some methods offer high throughput, whereas others offer

high resolution; several of the microfluidic devices are simple

in terms of operation, other techniques might require a

specialist; a number of separation principles require labelling

of sample components, whilst some processes are based on

intrinsic sample properties. As always, the optimum method

will depend on the sample and the analytical task at hand.

Most publications on continuous flow separation methods

so far have been concerned with proof of principle experiments

or characterisation of performance. As yet, there are very few

cases where continuous separation methods have been

integrated within a microTAS. One exception is the pillar

based leukocyte enrichment filter by Chen et al.,20 that was

combined with downstream cell lysis and DNA purification.

However, researchers now have an extensive toolbox of

methods available, and in the near future we are likely to see

more of these continuous flow techniques integrated within

miniaturised total analysis systems, such as continuous flow

blood separation for point-of-care systems, continuous flow

DNA separation for at-the-scene forensic testing and con-

tinuous flow molecule separation for in-the-field environ-

mental detection. It is quite possible that these continuous flow

separation methods will contribute significantly to a wider

breakthrough of microfluidic technology in the market place.

Fig. 12 Separation of two particle populations based on gravity.

Initially, particles are hydrodynamically focussed parallel to the

direction of gravity and then guided around a 90u bend into a

gradually widening separation channel. As particles enter this

separation channel, flow rates slow down and sedimentation starts to

take effect. Any differences in particle position caused by sedimenta-

tion are further enhanced by the widening flow streams, thus allowing

for the separation of the two particle populations. Figure redrawn

from ref. 94 with permission from the ACS, copyright 2007.

Fig. 13 The principle of split flow thin (SPLITT) fractionation. A

particle suspension and a carrier stream are pumped into a separation

chamber, often at different velocities. Flow splitters at the entrance

and exit of the chamber smooth the flow and define an inner and outer

splitting plane. A force field, such as gravity, an electric field or a

magnetic field, is applied perpendicular to the direction of flow,

deflecting some of the particles into the carrier stream.

This journal is � The Royal Society of Chemistry 2007 Lab Chip, 2007, 7, 1644–1659 | 1657

References

1 D. R. Reyes, D. Iossifidis, P. A. Auroux and A. Manz, Anal.Chem., 2002, 74, 2623.

2 P. A. Auroux, D. Iossifidis, D. R. Reyes and A. Manz, Anal.Chem., 2002, 74, 2637.

3 G. M. Whitesides, Nature, 2006, 442, 368.4 J. Berthier and P. Silberzan, Microfluidics for Biotechnology,

Artech House, Boston, 2006.5 P. C. H. Li, Microfluidic Lab-on-a-Chip for Chemical and

Biological Analysis and Discovery, Taylor & Francis Group,Boca Raton, 2006.

6 K. Sato, A. Hibara, M. Tokeshi, H. Hisamoto and T. Kitamori,Anal. Sci., 2003, 19, 15.

7 K. Sato, A. Hibara, M. Tokeshi, H. Hisamoto and T. Kitamori,Adv. Drug Deliv. Rev., 2003, 55, 379.

8 M. Yamada, V. Kasim, M. Nakashima, J. Edahiro and M. Seki,Biotechnol. Bioeng., 2004, 88, 489.

9 K. H. Nam, W. J. Chang, H. Hong, S. M. Lim, D. I. Kim andY. M. Koo, Biomed. Microdev., 2005, 7, 189.

10 M. Yamada, M. Nakashima and M. Seki, Anal, Chem., 2004, 76,5465.

11 J. Takagi, M. Yamada, M. Yasuda and M. Seki, Lab Chip, 2005,5, 778.

12 Y. Sai, M. Yamada, M. Yasuda and M. Seki, J. Chromatogr., A,2006, 1127, 214.

13 X. L. Zhang, J. M. Cooper, P. B. Monaghan and S. J. Haswell,Lab Chip, 2006, 6, 561.

14 M. Yamada and M. Seki, Lab Chip, 2005, 5, 1233.15 M. Yamada and M. Seki, Anal. Chem., 2006, 78, 1357.16 R. D. Jaggi, R. Sandoz and C. S. Effenhauser, Microfluid.

Nanofluid., 2007, 3, 47.17 S. Yang, A. Undar and J. D. Zahn, Lab Chip, 2006, 6, 871.18 S. S. Shevkoplyas, T. Yoshida, L. L. Munn and M. W. Bitensky,

Anal. Chem., 2005, 77, 933.19 P. Sethu, A. Sin and M. Toner, Lab Chip, 2006, 6, 83.20 X. Chen, D. F. Cui, C. C. Liu, H. Li and J. Chen, Anal. Chim.

Acta, 2007, 584, 237.21 W. Hattori, H. Someya, M. Baba and H. Kawaura,

J. Chromatogr., A, 2004, 1051, 141.22 V. VanDelinder and A. Groisman, Anal. Chem., 2007, 79, 2023.23 V. VanDelinder and A. Groisman, Anal. Chem., 2006, 78, 3765.24 L. R. Huang, E. C. Cox, R. H. Austin and J. C. Sturm, Science,

2004, 304, 987.25 D. W. Inglis, J. A. Davis, R. H. Austin and J. C. Sturm, Lab Chip,

2006, 6, 655.26 T. A. J. Duke and R. H. Austin, Phys. Rev. Lett., 1998, 80, 1552.27 C.-F. Chou, O. Bakajin, S. W. P. Turner, T. A. J. Duke,

S. S. Chan, E. C. Cox, H. G. Craighead and R. H. Austin, Proc.Natl. Acad. Sci. U. S. A., 1999, 96, 13762–13765.

28 L. R. Huang, P. Silberzan, J. O. Tegenfeldt, E. C. Cox, J. C. Sturm,R. H. Austin and H. Craighead, Phys. Rev. Lett., 2002, 89,178301.

29 M. Cabodi, Y. F. Chen, S. W. P. Turner, H. G. Craighead andR. H. Austin, Electrophoresis, 2002, 23, 3496.

30 S. Choi and J.-K. Park, Lab Chip, 2007, 7, 890–897.31 L. R. Huang, J. O. Tegenfeldt, J. J. Kraeft, J. C. Sturm,

R. H. Austin and E. C. Cox, Nat. Biotechnol., 2002, 20, 1048.32 J. P. Fu, R. B. Schoch, A. L. Stevens, S. R. Tannenbaum and

J. Y. Han, Nat. Nanotechnol., 2007, 2, 121.33 J. Han and H. G. Craighead, Science, 2000, 288, 1026.34 L. Krivankova and P. Bocek, Electrophoresis, 1998, 19, 1064.35 K. Hannig, J. Chromatogr., 1978, 159, 183.36 D. E. Raymond, A. Manz and H. M. Widmer, Anal. Chem., 1994,

66, 2858.37 D. E. Raymond, A. Manz and H. M. Widmer, Anal. Chem., 1996,

68, 2515.38 M. Mazereeuw, C. M. de Best, U. R. Tjaden, H. Irth and

J. van der Greef, Anal. Chem., 2000, 72, 3881.39 H. Kobayashi, K. Shimamura, T. Akaida, K. Sakano, N. Tajima,

J. Funazaki, H. Suzuki and E. Shinohara, J. Chromatogr., A,2003, 990, 169.

40 D. P. de Jesus, L. Blanes and C. L. do Lago, Electrophoresis,2006, 27, 4935.

41 D. Kohlheyer, G. A. J. Besselink, S. Schlautmann andR. B. M. Schasfoort, Lab Chip, 2006, 6, 374.

42 B. R. Fonslow and M. T. Bowser, Anal. Chem., 2005, 77, 5706.43 B. R. Fonslow, V. H. Barocas and M. T. Bowser, Anal. Chem.,

2006, 78, 5369.44 C. X. Zhang and A. Manz, Anal. Chem., 2003, 75, 5759.45 Y. Xu, C. X. Zhang, D. Janasek and A. Manz, Lab Chip, 2003, 3,

224.46 K. Macounova, C. R. Cabrera and P. Yager, Anal. Chem., 2001,

73, 1627.47 C. R. Cabrera and P. Yager, Electrophoresis, 2001, 22, 355.48 H. Lu, S. Gaudet, M. A. Schmidt and K. F. Jensen, Anal. Chem.,

2004, 76, 5705.49 G. Munchow, S. Hardt, J. P. Kutter and K. S. Drese, Lab Chip,

2007, 7, 98.50 D. Janasek, M. Schilling, A. Manz and J. Franzke, Lab Chip,

2006, 6, 710.51 Y. A. Song, S. Hsu, A. L. Stevens and J. Y. Han, Anal. Chem.,

2006, 78, 3528.52 N. Pamme and A. Manz, Anal. Chem., 2004, 76, 7250.53 F. Petersson, L. Aberg, A.-M. Sward-Nilsson and T. Laurell,

Anal. Chem., 2007, 79, 5117.54 M. Becker, A. Mansouri, T. Matsui, J. Franzke, A. Manz and

D. Janasek, in Proceedings of the MSB conference, Vancouver,2007.

55 L. M. Barrett, A. J. Skulan, A. K. Singh, E. B. Cummings andG. J. Fiechtner, Anal. Chem., 2005, 77, 6798.

56 K. H. Kang, Y. J. Kang, X. C. Xuan and D. Q. Li,Electrophoresis, 2006, 27, 694.

57 I. Barbulovic-Nad, X. C. Xuan, J. S. H. Lee and D. Q. Li, LabChip, 2006, 6, 274.

58 E. B. Cummings and A. K. Singh, Anal. Chem., 2003, 75, 4724.59 B. G. Hawkins, A. E. Smith, Y. A. Syed and B. J. Kirby, Anal.

Chem., 2007, 79, 7291.60 J. G. Kralj, M. T. W. Lis, M. A. Schmidt and K. F. Jensen, Anal.

Chem., 2006, 78, 5019.61 S. Choi and J.-K. Park, Lab Chip, 2005, 5, 1161–1167.62 C. D. James, M. Okandan, S. S. Mani, P. C. Galambos and

R. Shul, J. Micromech. Microeng., 2006, 16, 1909.63 I. Doh and Y. H. Cho, Sens. Actuators, A, 2005, 121, 59.64 Y. L. Li, C. Dalton, H. J. Crabtree, G. Nilsson and K. Kaler,

Lab Chip, 2007, 7, 239.65 X. Y. Hu, P. H. Bessette, J. R. Qian, C. D. Meinhart,

P. S. Daugherty and H. T. Soh, Proc. Natl. Acad. Sci. U. S. A.,2005, 102, 15757.

66 P. H. Bessette, X. Y. Hu, H. T. Soh and P. S. Daugherty, Anal.Chem., 2007, 79, 2174.

67 F. C. Leinweber, J. C. T. Eijkel, J. G. Bower and A. van den Berg,Anal. Chem., 2006, 78, 1425.

68 M. A. M. Gijs, Microfluid. Nanofluid., 2004, 1, 22.69 N. Pamme, Lab Chip, 2006, 6, 24.70 R. Hartig, M. Hausmann, J. Schmitt, D. B. J. Herrmann,

M. Riedmiller and C. Cremer, Electrophoresis, 1992, 13, 674.71 L. P. Sun, M. Zborowski, L. R. Moore and J. J. Chalmers,

Cytometry, 1998, 33, 469.72 M. Zborowski, L. P. Sun, L. R. Moore, P. S. Williams and

J. J. Chalmers, J. Magn. Magn. Mater., 1999, 194, 224.73 K. E. McCloskey, L. R. Moore, M. Hoyos, A. Rodriguez,

J. J. Chalmers and M. Zborowski, Biotechnol. Prog., 2003, 19,899.

74 G. Blankenstein and U. D. Larsen, Biosens. Bioelectron., 1998,13, 427.

75 N. Chronis, W. Lam and L. Lee, in Micro Total Analysis Systems2001, ed. J. M. Ramsey, A. v. d. Berg, Kluwer AcademicPublishers, Monerey, USA, 2001, pp 497.

76 K. S. Kim and J.-K. Park, Lab Chip, 2005, 5, 657–664.77 A. C. Siegel, S. S. Shevkoplyas, D. B. Weibel, D. A. Bruzewicz,

A. W. Martinez and G. M. Whitesides, Angew. Chem., Int. Ed.,2006, 45.

78 N. Xia, T. P. Hunt, B. T. Mayers, E. Alsberg, G. M. Whitesides,R. M. Westervelt and D. E. Ingber, Biomed. Microdev., 2006, 8,299.

79 K.-H. Han and A. B. Frazier, Lab Chip, 2006, 6, 265–273.80 N. Pamme, J. C. T. Eijkel and A. Manz, J. Magn. Magn. Mater.,

2006, 307, 237.

1658 | Lab Chip, 2007, 7, 1644–1659 This journal is � The Royal Society of Chemistry 2007

81 N. Pamme and C. Wilhelm, Lab Chip, 2006, 6, 974.82 D. W. Inglis, R. Riehn, R. H. Austin and J. C. Sturm, Appl. Phys.

Lett., 2004, 85, 5093.83 M. Zborowski, G. R. Ostera, L. R. Moore, S. Milliron, J. J.

Chalmers and A. N. Schechtery, Biophys. J., 2003, 84, 2638–2645.84 D. Schuler and R. B. Frankel, Appl. Microbiol. Biotechnol., 1999,

52, 464.85 N. Hirota, A. Iles and N. Pamme, in Proceedings of microTAS

2007 Conference, ed. J.-L. Viovy, 2007, p. 212.86 W. T. Coakley, Trends Biotechnol., 1997, 15, 506.87 T. Laurell, F. Petersson and A. Nilsson, Chem. Soc. Rev., 2007,

36, 492.88 J. J. Hawkes and W. T. Coakley, Sens. Actuators, B, 2001, 75,

213.89 A. Nilsson, F. Petersson, H. Jonsson and T. Laurell, Lab Chip,

2004, 4, 131.90 F. Petersson, A. Nilsson, C. Holm, H. Jonsson and T. Laurell,

Lab Chip, 2005, 5, 20.

91 S. Kapishnikov, V. Kantsler and V. Steinberg, J. Stat. Mech.,2006, P01012.

92 M. P. MacDonald, G. C. Spalding and K. Dholakia, Nature,2003, 426, 421.

93 G. Milne, D. Rhodes, M. MacDonald and K. Dholakia, Opt.Lett., 2007, 32, 1144.

94 D. Huh, J. H. Bahng, Y. B. Ling, H. H. Wei, O. D. Kripfgans,J. B. Fowlkes, J. B. Grotberg and S. Takayama, Anal. Chem.,2007, 79, 1369.

95 J. C. Giddings, Sep. Sci. Technol., 1985, 20, 749.96 M. H. Moon, H. J. Kim, S. Y. Kwon, S. J. Lee, Y. S. Chang and

H. Lim, Anal. Chem., 2004, 76, 3236.97 C. B. Fuh and J. C. Giddings, Sep. Sci. Technol., 1997, 32, 2945.98 C. B. Fuh, Anal. Chem., 2000, 72, 266A.99 Y. H. Zhang, R. W. Barber and D. R. Emerson, Curr. Anal.

Chem., 2005, 1, 345.100 N. Narayanan, A. Saldanha and B. K. Gale, Lab Chip, 2006,

6, 105.

This journal is � The Royal Society of Chemistry 2007 Lab Chip, 2007, 7, 1644–1659 | 1659