Parallelized continuous flow dielectrophoretic separation of DNA

Numerous microfluidic separation applications have been shown in the past years providing a fast analysis of biological samples like DNA or proteins. Microfluidic separation based on dielectrophoresis (DEP), that is the migration of a polarizable object in an inhomogeneous electric field, provides numerous advantages. However, the main drawback of DEP separation devices is that they are not sufficient for large-scale sample purification due to the lack of high sample throughput. In this work, we present for the first time a microfluidic device with two parallelized dielectrophoretic separations of (biological) samples smaller than 1 µ m. The separation is carried out by means of insulator-based DEP, that is an insulating ridge reduced the flow through height and thus created a nanoslit at which the selective DEP forces occur. The device consists of a cross injector, two parallel operation regions and separate harvesting reservoirs where the samples are collected. Each DEP operation region contains an insulating ridge. We successfully demonstrate the separation of 100 and 40 nm beads and 10 and 5 kbp DNA with a separation purity of more than 80%. This states the proof-of-concept for up-scaling of dielectrophoretic separation by parallelization. As the present technique is virtually label-free, it offers a fast purification, for example in the production of gene vaccines

respect to sample size and electrochemical parameters, can be controlled by the applied electric fields, DEP is a very versatile method in separation applications.
So far, two main concepts for DEP separation were presented in literature: First, batch processing, that is, sample plugs are inserted and separated one after another [14][15][16][17][18][19]; second, continuous processing, that is, sample mixtures are continuously flushed into separation regions and exposed to selective forces [20][21][22][23][24][25][26][27][28][29][30][31][32][33]. Continuous separation applications provide enormous potential for highthroughput separations. Up-scaling of the throughput of continuous flow separation is possible either by increasing the flow rate, which in turn leads to the need of higher electric fields to dielectrophoretically separate the sample, or by the parallelization of separation. Ref. [34] demonstrated a high-throughput DEP application with parallelization. The authors used an array of 6000 traps to trap cells dielectrophoretically and fuse them.
Though microfluidics offer new methods for the separation or analysis of biomolecules, there are some limitations that inhibit the use of microfluidic separation in wide fields of research, medicine or production. The most prominent challenge is the volume and sample throughput in microfluidic separation applications [6,35,36]. This is due to the small channel dimensions. Choe et al. recently discussed factors that have to be considered along up-scaling of microfluidic separation. For instance, sensitive biological samples might be negatively affected or damaged by high shear forces. Furthermore, the separation resolution depends on the balance of forces acting on the samples, for example, selective gradient forces versus general migration forces [6]. Thus, high-throughput separation is a research field of high relevance.
Faraghat et al. presented the high-throughput DEP separation of cells [31]. Their device consists of a stacked layer of electrodes with a channel through which the sample flows, that is, the cells pass alternating electrodes and are affected by either positive or negative DEP forces that trap or push them away. Nevertheless this device is working in a continuous mode, the throughput of the separation is limited due to a limited number of cells that can be trapped permanently at the electrodes. Therefore, no longterm separation, for example, for several hours, is possible. The high-throughput DEP device presented by Tada et al. is limited as well for the same reason [30]. Hence, a different approach for continuous separation is needed to purify or sort large sample volumes.
The separation needs to result in spatially separated continuous sample streams that can be directed into separated harvesting reservoirs. Wu et al. recently presented a dielectrophoretic separation device that is working continuously, that is, the separated species, cells and particles are deflected into separate harvesting channels [32]. The separation was conducted at microelectrodes. Zhang et al. exploited a combination of inertial flow and insulatorbased DEP separation in serpentine channels to achieve a continuous separation of micrometre beads [27]. The DEP force strongly correlates with the sample size; thus, higher electric fields and larger gradients are necessary to manipulate small samples like DNA. Therefore, the approach of Zhang et al. would not be capable to separate DNA species.
In the present work, we exploit the concept of parallelization of dielectrophoretic separation to increase the throughput. Our device was based on the general continuous flow separation concept presented earlier in Refs. [20,24], that is, the separation channel contained a ridge that creates a nanoslit at which the separation occurs. We used two parallel separation channels as proof-ofconcept to demonstrate increase of the sample throughput. To the authors' knowledge, this is the first demonstration of increase of throughput in insulator-based dielectrophoretic separation of DNA. This paves the way to applications in the purification of gene vaccines along the production.

THEORY
In this work, the general concept of continuous flow separation relied on insulator-based DEP. The DEP forces occur at a curved ridge that spanned the microchannel laterally and created a nanoslit, see Figure 1. DEP is the migration of a polarizable object in an inhomogeneous electric field. The electric field was created by applying a voltage ( ) = + sin( ) to the microfluidic channel [20]. Advantages of the insulator-based DEP are that the electric field is homogeneous over the height of the nanoslit and that no electrochemical effects appear in the region of separation because the electrodes are far away immersed in the reservoirs [11]. The resulting potential dielectrophoretic energy can be written as [12] = − 1 2 ⃗ 2 with being the polarizability of the object, and electric field ⃗ . The corresponding dielectrophoretic force is as follows [12]: The polarizability of a homogeneous, spherical, polarizable object scales with the particle size according to [37]  The sample migration is driven by electrokinetic effects, that is, electrophoresis and electroosmotic flow, and superimposed pressure-driven flow. Additionally, dielectrophoretic forces occur in the vicinity of the nanoslit. Hence, the dynamical behaviour is governed by linear and non-linear electrokinetic effects and hydrodynamic forces.
Under the given experimental conditions, that is, ≪ , the linear electrokinetic effects, that is, electroosmotic flow and electrophoresis, are controllable solely by the direct current (DC) electric field. The migration of the samples governed by electric fields can be described with ⃗ = ( + ) ⃗ + ∇| ⃗ | 2 , with , , are the electroosmotic, electrophoretic and dielectrophoretic mobilities, respectively [21]. The fluid flow additionally is controlled by the applied hydrostatic pressure. The dielectrophoretic migration is governed by the alternating current (AC) electric field. Thus, the relative strength of the driving forces, and therefore the samples' paths, can be adapted during experiments by tuning the applied DC and AC voltage amplitudes and the respective hydrostatic pressures. As the dielectrophoretic component scales with the gradient of the electric field, it does only play a significant role at spatial inhomogeneities; thus, in the vicinity of the nanoslit, for more information, we refer to Ref. [21].
Looking at the samples migration at the nanoslit, two cases can be distinguished: First, the DEP force does not overcome the driving forces and/or the thermal motion and, therefore, the sample passes the nanoslit unaffected. Second, the DEP trapping force overcomes the driving forces and the thermal motion and the sample migrates, once trapped, towards the opposite channel side where it escapes by thermal motion, see Figure 1. Therefore, samples differing in polarizability, for example, due to size, are spatially separated the downstream of the nanoslit.

Device
Fabrication of the insulating ridge required a two-step contact lithography of the master wafer, which was moulded by poly(dimethylsiloxane) (PDMS) soft lithography to fabricate the final device as described elsewhere [38]. The channels were filled with working buffer, either Milli-Q water (conductivity: 4.59 µS/cm) or 1 mM phosphate buffer (pH 7.4, 0.2 mM NaCl; conductivity: 2.72 mS/cm) (Fluka, Germany). The buffer additionally contained 1 mM ethylenediaminetetraacetic acid (Fluka, Germany) for the DNA experiments. Unspecific sample adhesion and loss is an issue in purification applications. Therefore, pluronic F 108 (Sigma-Aldrich, Germany) was added to all buffers to reduce unspecific adsorption of analytes to the surfaces and for control of electroosmotic flow [39]. Although the visual inspection of the whole device after the experiments with nanobeads yielded some adhered nanobeads, we could not observe unspecific binding of DNA in the microfluidic channels. A poly(methyl methacrylate) (PMMA) block with integrated reservoirs was placed on the top of the chip to enlarge the reservoirs and simplify handling. Platinum electrode wires were integrated in the reservoirs of the PMMA block for electric contacting. Thereby, the electrodes were immersed in the microfluidic reservoirs.

Sample preparation
The fluorescently labelled nanoparticles ( DNA was stained with either YOYO-1 (Molecular Probes, USA) or BOBO-3 (Thermo Fisher Scientific Inc., Germany), ratio dye to DNA base pairs 1:10, before starting the experiments. The sample stock solutions were diluted to several 10 pM with working buffer. Transport and injection of the samples were based on electrokinetics (electrophoresis and electroosmotic flow) and pressuredriven flow. Evaluation of the video microscopy data revealed a sample migration velocity of 18 µm/s, which corresponds to a flow rate of 20 nl/min.

Set-up
The measurements were performed on an inverted fluorescence microscope as described before [20]. An ADwin-Gold II function generator (Jäger, Germany) with a high voltage amplifier AMT-1B60-2 was used to generate the AC voltages. Three power supplies (HCL 14-12500, FUG, Germany) generated additional DC voltages. The voltage signals were controlled by a customized LabVIEW 9.0.1 program. An MFCS pressure control system (Fluigent, Germany) was connected to the reservoirs in the PMMA holder via PTFE tubes, providing pressure-driven flow additionally to the electrokinetic migration.

Experiment performance
After the device was fabricated, the parameters to drive the sample to the ridge were determined. This was done iteratively probing electrokinetic migration only, pressure driven only and electrokinetic plus pressure driven. A successful separation, that is, baseline-separated resolution of the samples, was achieved. Purification was possible only if the samples were continuously driven from the sample reservoir towards the ridge, with the sample stream covering less than a quarter of the channel width. Selective dielectrophoretic forces were observed under the following conditions: sample velocity of 18 µm/s, AC voltages of 360 V at 250 Hz for DNA and 1000 V at 50 Hz for nanobeads. The flow and distribution of the sample in the channel were observed with a fluorescence microscope. As the field of view was (significantly) smaller than the channel widths, cf. Figure 3, the channels were scanned during the experiment to monitor the sample flow and separation success at the two ridges. The process of image and data acquisition is described in detail in Ref. [38]. Briefly, the separation channels were scanned with a velocity of 10 µm/s while taking fluorescence video microscopy images (10 fps). Afterwards, the images are subdivided in several regions of interest to analyse the fluorescence intensity.

Device concept
The separation of samples was by means of insulator-based DEP. Insulating ridges, spanning the channel laterally, were placed in the microfluidic channel providing an inhomogeneous electric field [21]. This work aimed at a higher throughput implementing a concept that relied on separation not at just one nanoslit but at several in parallel channels. Thus, the number of separation channels, that is, channels containing an insulating ridge each, was increased. Therefore, basically the sample turnover increases by a factor of n for n parallel separation channels. When designing the device for parallel separations, some preliminary conditions have to be considered. First, the resistances of all branches have to be the same; otherwise, the samples would follow the path of minimal resistance [40,41,24] and a controlled migration of the sample through the device is hardly achievable. Second, the strength of the electric field in the nanoslit, and thus, the dielectrophoretic potential strongly depends on the channel geometries like length, width and height of the micro-and nanochannels [23]. The electric resistance in a microfluidic channel can be calculated exploiting the assumption that the electric resistance is proportional to the hydrodynamic resistance, and thus ∝ ⋅ℎ , with , and ℎ being the channel length, width and height, respectively [21]. The electric field in the nanoslit was calculated with the geometrical factor , which, again, depends on the channel geometries and was calculated by using an equivalent circuit model. During designing the channel geometry, the chip material also has to be considered with regards of the channel stability and maximum aspect ratio. In this work, the chips were made from a stacked layer of soft and hard PDMS, see before. Though wide channel cross sections are advantageous for high electric fields at the nanoslit, they also tend to collapse. Therefore, the width of the channels was designed to be as large as possible and as small as necessary. In this work, the maximum channel width was limited to 800 µm, at a channel height of 5 µm. The minimal electric field for successful separation was known from previous separation experiments [20,21,23,24]. Hence, the device layout was designed such that the necessary field was achieved for sufficiently low AC voltages. The electric field in the nanoslit was calculated with LTspice exploiting an equivalent circuit diagram, see electrospray ionization (ESI) Figure S1. The DEP trapping at the nanoslit depends on the electric field strength and the migration along the insulating ridge, once trapped, relies on the design of the ridge. Here, we used the same design as for the successful separation in Ref. [21], that is, an asymmetric s-shaped ridge was used. In order to ensure that the sample did not remix but flew into distinct harvesting reservoirs, we placed an additional channel wall in the device, downstream of the nanoslit, see Figure 1.
The last point that needs to be considered during designing the device is the ability to further up-scaling of the system. For instance, the number of in-and outlets should be minimized. Ideally, the device consists of one inlet, where the sample mixture is placed, and one outlet reservoir for each sample species. Thus, a device separating two species ideally has three in-and outlets in total that must be connected to the outer world. The sample is focused and controlled hydrodynamically or electrokinetically by using additional channels in most continuous flow applications though [42][43][44]. One point that can affect the number of reservoirs is the number and alignment of parallel separation channels. More than two parallel channels either have increasing numbers of in-and outlet reservoirs or 'crossing' channels. The crossing channels must not be connected fluidically. A solution to this is the use of stacked layers, consisting of the else crossing fluidic channels, which are then in separate layers. Anderson et al. used a three-dimensional microfluidic system in which microfluidic channels were guided through two different layers and thus could cross each other [45]. Here, we used for proofof-concept of sufficient parallel separation two separation channels.

Sample flow
During designing the microfluidic channels, we considered the critical points mentioned earlier. For instance, the number of reservoirs was reduced to a minimum with two inlet reservoirs, one for the sample mixture and the other to focus and control the sample flow, and two outlet reservoirs. The two species of the sample flew to separate harvesting channels downstream of the ridge. Though the lengths of the harvesting channels differed significantly, see Figure 2, the flow resistance was equal. A control of the flow behaviour of test beads revealed that the samples could be led towards the nanoslits as a narrow stream that covered less than a quarter of the separation chan- nel width. Moreover, both, sample velocity, 18 µm/s, and concentration, were the same along both paths. We tested different driving forces to lead the sample mixture towards the nanoslits. Electric migration, that is, electroosmotic flow and electrophoresis, pressure-driven flow and a combination of both were tested. The time until the sample reached the ridge for the first time after starting the experiment was quite long with about 15 min, which was due to the long fluidic channels, see Figure 2. Thus, to speed up the process, we always used a high pressuredriven flow to move the sample close to the separation channel, switched off the pressure and used the respective driving force afterwards.

Nanobead separation
For proof-of-concept of parallel separation, we performed the separation of a mixture of 100 and 40 nm beads at two nanoslits. The beads were fluorescently labelled with different dyes (515 and 605 nm fluorescence wavelength) enabling to distinguish the two species along the fluidic path. The Joule heating is a significant and well-known problem in electrokinetically driven microfluidic chips [46][47][48]. The effect is proportional to the electric current in the system, which can be controlled by the ionic strength of the used buffers. Deionized water exhibits the lowest ionic strength of aqueous solutions. The first separation experiments were performed with deionized water to minimize Joule heating effects, because those should be minimized to prevent the degradation of the chip and bubble formation in the liquid.
We observed an unexpected behaviour of the beads at the nanoslit. For instance, the nanobeads were repelled from the nanoslit, after AC voltage was switched on, and did not flow along the ridge towards the opposite channel side. Both nanoparticle species, 100 and 40 nm, exhibited this behaviour.
The observed behaviour can be explained when having a closer look at the Debye layer thickness of the buffers. The Debye layer in the deionized water is large enough to overlap in the 520 nm high nanoslit. This can be confirmed by calculating the Debye length with = 2.1⋅10 −10 √ , with = 1 2 ∑ 2 being the ionic strength, concentration and the valency of ions in solution, respectively [49]. The Debye layer in Milli-Q water was calculated to 100 nm, yielding an 8% overlap of the electrical double layers at 250 nm, respectively. Therefore, it is plausible that the nanobeads were repelled from the ridge due to ion-concentration polarization as described in Refs. [49][50][51]. Thus, controlled dielectrophoretic control was not possible in deionized water as the gradient in the electric field only occurs in the vicinity of the nanoslit. Therefore, the measurement buffer was exchanged with 1 mM phosphate buffer to further evaluate this behaviour. The behaviour of the beads was as expected in the buffer with slightly higher ionic strength, that is, the nanobeads were no longer repelled from the ridge but deflected along the ridge at sufficiently high AC voltages. Hence, all consecutive separation experiments with nanobeads were performed in 1 mM phosphate buffer.
In Figure 3, the separation of a mixture of 100 and 40 nm beads at the two nanoslits is shown. Due to the different labelling of the nanobeads, those were observed separately with respective filter sets. As expected, the larger 100 nm beads were affected by negative DEP and deflected along the ridge towards the opposite channel wall, whereas the smaller 40 nm beads passed the ridge unaffectedly. Though a fraction of the 100 nm beads did not completely migrate to the opposite channel side, the larger fraction of the 100 nm beads was deflected beyond half of the channel width, and thus, passed the separating wall in the channel. Therefore, both particle species could be harvested in separate outlet reservoirs and high separation efficiency was achieved.

DNA separation
After the general proof-of-concept with nanobeads, we tested the simultaneous separation of 10 and 5 kbp DNA at two nanoslits. Both DNA species were stained with the same fluorophore, YOYO-1, to ensure that the different DEP migration behaviour was due to the different DNA sizes only. This was necessary because DEP migration is sensitive to bound molecules as shown in literature [ 17,22]. Because of the same fluorescence labelling, the two species could not be distinguished during separation. Thus, we first performed an experiment with the separate DNA species before the separation experiment was conducted. The results revealed that the 10 kbp DNA was completely deflected at the nanoslits towards the opposite channel side if 360 V AC voltage at 250 Hz was applied. The smaller 5 kbp DNA passed the nanoslit unaffectedly at the same parameters. In Figure 4, the separation of a 10 and 5 kbp DNA mixture is shown. Only one molecule stream is visible upstream of the ridge. However, two distinct streams were formed downstream of the ridge that flew to respective harvesting outlets.
Analysis of the fluorescence intensities revealed that spatial separation resolutions of 2.18 and 3.23 were achieved in both separation channels, see Figure 5. That is a baseline-separated resolution; thus, both sample streams are completely separated per definition. However, even lower separation resolution would be acceptable as long as both sample species follow distinct paths into the respective harvesting reservoirs. Repetitions of this experiment in other devices also led to baseline-separated resolutions (data not shown). Hence, the separation was of high reliability and reproducibility. A close look at the fluorescence intensities in Figure 5 reveals partial fluorescence that seems to be outside of the channel. This was due to convolution effects, see also Ref. [38].
We also conducted repeated separation experiments over several hours to evaluate if the separation process remained stable. The observed sample stream of the deflected 10 kbp DNA slightly broadened during the long-time separation. Nonetheless, sufficient separation resolution, that is, the two DNA species flew into separate reservoirs, was observed even if the separation was continued over hours, see ESI Figure S2. The main effect that counteracts separation in electric fields is temperature rise due to the Joule heating, which was significantly reduced by the pressure-driven flow. Therefore, the buffer in the nanoslits and their vicinity was exchanged permanently.
Despite the clearly differing responses of the DNA samples in separate measurement, we also conducted a separation of DNA mixtures with different labellings. For that purpose, the large DNA species was labelled with YOYO-1, whereas the small DNA species was labelled with BOBO-3. Therefore, each species could be detected with distinct fluorescence filter sets in the same separation experiment. The observations revealed that the smaller 5 kbp DNA passed the nanoslits unaffected or was deflected at the most to half of the channel width, whereas the larger 10 kbp DNA was deflected towards the opposite channel wall along separation. Further analysis of the fluorescent video microscopy images revealed that at least 80% of the 10 kbp DNA was deflected and thus separated from the DNA mixture, see ESI Figure S3. Thus, the purity of the 10 kbp DNA stream was about 100%, and the purity of the 5 kbp DNA stream was at least 88%.

CONCLUDING REMARKS
The aim of this work was a proof-of-concept of a microfluidic device that provides parallel, continuous flow separation of samples by means of DEP. Important aspects were that the sample flow could be well controlled; sufficient parameter of the applied voltages needed to be determined that lead to a specific and efficient separation of the samples; the separation had to be stable for hours.
In the present work, we have shown an improved microfluidic device with simultaneous, continuous flow separation by means of non-invasive and virtually labelfree DEP in two parallel separation channels. The continuous flow separation approach presented in this work exhibits several advantages: fast separation time; accessibility to nm-sized biological objects, like DNA; baselineseparated resolution and an on-line adaptability of selectivity parameters via applied AC and DC voltages [24].
Appropriate designing of the fluidic channel dimensions led to well-controlled sample flow and sufficient dielectrophoretic trapping potentials. The device was successfully tested for the separation of 100 and 40 nm beads as well as 10 and 5 kbp DNA. The results revealed that baseline-separated resolution was achieved, that is the separated samples were directed into distinct reservoirs where they could be harvested. With these results we could demonstrate a successful proof-of-concept of improvement of the separation throughput of dielectrophoretic separation. The minimum purity achieved was 88%.
In future, we will further increase the number of parallel separation channels. If the total number of reservoirs should remain the same as of now, layer stacking, cf. Ref. [52], will be used. With that approach, the presented upscaling of dielectrophoretic separation throughput paves the way to applications in the purification of gene vaccines along the production.

A C K N O W L E D G M E N T S
The authors especially thank Dario Anselmetti for fruitful discussions. Proofreading is gratefully acknowledged to Daniel Wesner.
Open access funding enabled and organized by Projekt DEAL.

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.