Electronic sorting and recovery of single live cells from microlitre sized samples

Abstract

Sorting and recovering specific live cells from samples containing less than a few thousand cells have become major hurdles in rare cell exploration such as stem cell research, cell therapy and cell based diagnostics. We describe here a new technology based on a microelectronic chip integrating an array of over 100,000 independent electrodes and sensors which allow individual and parallel single cell manipulation of up to 10,000 cells while maintaining viability and proliferation capabilities. Manipulation is carried out using dynamic dielectrophoretic traps controlled by an electronic interface. We also demonstrate the capabilities of the chip by sorting and recovering individual live fluorescent cells from an unlabeled population.

---

Introduction

Rare cell populations such as adult stem cells, circulating fetal cells in maternal blood and natural therapeutic immune cells have generated a lot of hype and hope for new cell based diagnostics and therapies.^1,2 As they are present in exquisitely small quantities in samples available in limited supply, identifying, selecting and recovering them alive represents a major challenge in these highly promising fields. Strategies for isolating such cells often include a first enrichment step by magnetic or fluorescent cell sorting followed by manual cell picking and single cell analysis.³ To replace this labor intensive manual isolation of candidate cells, miniaturized devices are being developed to analyze and sort samples containing less than a few thousand cells where conventional cell sorters require tens of thousands of cells. One of the most promising means of cell friendly manipulation is dielectrophoresis (DEP).⁴ The development of static three dimensional DEP cages⁵ has demonstrated that live individual cells may be trapped into closed potential cages generated by non-uniform electrical fields. This technology has been applied to cell sorting as well as to single cell manipulation. Other cell sorters based on DEP and/or on electroosmosis have been described.^6–13 All rely on the association of electrical fields with other emerging technologies such as microfluidic laminar flows and optical tweezers^8,12,14 for efficient cell manipulation.

We present here a device in which trapping, manipulation and motion of a variety of cell types are electronically controlled by dynamic dielectrophoretic traps generated by a microelectronic CMOS silicon chip. The experimental results demonstrate the performance of the system in isolation and recovery of individual fluorescent K562 cells from a bulk of unlabeled cells while maintaining viability, DNA integrity and cell proliferation capacity.

Methods

Technology

Standard 0.35 µm 2P 3M CMOS technology silicon chips were manufactured according to our design by Austriamicrosystems AG, Austria. The 320 × 320 array is addressed in a random access mode, through the row and column decoders, for both actuation and sensing. Device architecture and circuit implementation details are described in ref. 15 and ref. 16. The final standard passivation layer protecting the electrodes was removed by Reactive Ion Etching. Double sided adhesive tape 100 µm thick was cut to our specifications using laser micromachining (Leperck, France) to yield a double chamber design with a 2.9 µl chamber connected by a 300 µm wide channel to a 0.6 µl chamber. Polycarbonate (PC) substrates were layered with low temperature process ITO as described¹⁷ and 700 µm diameter ports were mechanically drilled using an LPKF ProtoMat C100HF instrument. Printed circuit boards (PCB) were manufactured by Tecnomec srl, Italy according to our specifications. Device packaging was carried out as follows: the silicon chip was wirebonded to the PCB by Mesatronics SA (France) and wires were protected with a drop of epoxy glue. A drop of thick conductive glue (Epotecny E206) was placed on the PCB over the lid connector. After removing its first protective film, the double sided tape was manually placed over the respective inlet ports on the PC lid. The second protective film was removed and the lid was manually glued to the silicon. Efficient contact between the lid and the conductive glue was controlled visually.

Bench setup

The chip was connected to its control board and placed on the motorized stage (Ecodrive from Märzhäuser) of a BL51 Olympus microscope equipped with a Mercury lamp, a standard FITC fluorescence filter and a CCD camera (Orca ER, from Hamamatsu). Software controlling the connections and communication between the electronic board and chip was run on a standard PC computer. The electronic board was powered by a 12 V generator (E3630A from Agilent). A second computer was used to capture images, video sequences and control the motions of the motorized stage of the microscope.

Cell sample preparation

All cell lines were incubated at 37 °C under 5% CO₂. The K562 and Jurkat cell lines were cultured in RPMI Glutamax medium (Invitrogen) supplemented with 10% FCS. The adherent HeLa cell line was cultured in DMEM Glutamax medium supplemented with 10% FCS. HeLa cells were put into suspension by incubating the cell monolayer with trypsin (Invitrogen) in phosphate buffer saline (PBS, from Sigma). Red blood cells were prepared by diluting a pinprick drop of fresh blood in 280 mM mannitol supplemented with 1 mM EDTA (ethylenediaminetetraacetic acid, from Sigma). White blood cells were isolated from fresh donor blood by density centrifugation on Ficoll for 30 min at 1800g. The white blood cells were recovered above the cushion and were washed three times in RPMI culture medium. Cells were fluorescently labeled using CFSE (carboxy-fluorescein diacetate, succinimidyl ester from Molecular Probes) by incubating the cell suspension in PBS buffer containing 5 µm CFSE for 15 min.

Viability, proliferation and DNA damage assays

An aliquot of the sample was kept in low conductance buffer and at room temperature as a control throughout the experiment. After DEP manipulation, cells were recovered from the chip and viability, DNA damage and proliferation were compared to the control. Viability was evaluated using a standard trypan blue assay (Invitrogen). Effect of DEP on proliferation of cultured cell lines was assayed by putting recovered cells and control cells back into culture and comparing cell counts over 6 days. DNA oxidative damage and breaks were evaluated by a modified Comet assay.¹⁸

DEP manipulation of cells

Prior to the experiment, cells were washed three times and resuspended at the desired concentration in 280 mM mannitol supplemented with 6% FCS or 6.25mM KCl at a conductivity of 780 µS cm⁻¹. For RBC manipulations, 0.5 mM EDTA was added to the above suspension buffer to avoid aggregation. The cage pattern was activated through the software using field values of 40 kHz, 3.3 V peak to peak amplitude and a lid gain of 3.75. The cell suspension was then loaded into the larger chamber by micropipetting a few microlitres of suspension in inlet 1 (see Fig. 1) until fluid emerged from outlet 2. An air/water meniscus formed at the exit of the connecting channel and the sample containing cells did not penetrate into the recovery chamber. The recovery chamber was then filled with buffer alone with a micropipette through inlet 3 until fluid appeared at outlet 4. While filling the second chamber, buffer also entered the loading chamber through the channel and advantageously pushed back any cells that were in the vicinity of the channel entrance. Cell organization and fluorescence were observed through the microscope. Readouts from the capacitive and optical sensors gave positions of wall boundaries and of the occasional air bubble. Coordinates of cells of interest were determined by operating the motorized stage and converting metric coordinates to electrode coordinates on the 320 × 320 grid. After entering the initial coordinates of each cell of interest, cells were sent to new coordinates in the recovery chamber and pathways were automatically calculated by the software to move around static cages, chamber walls and through the connecting channel. In order to recover cells after sorting, the loading chamber was sealed by inserting plastic tips into inlet 1 and outlet 2. The recovery chamber was then emptied by aspirating its contents with a micropipette through outlet 4. This step procedure avoided aspirating cells from the loading chamber while pipetting from the recovery chamber.

Fig. 1 Illustration of the DEParray prototype with a close-up photograph in A and a cross section sketch in B. The device consists of an 8 × 8 mm² core CMOS silicon chip and a polycarbonate lid with a conductive ITO layer. These elements are bonded together by a 100 µm thick double sided adhesive tape which also defines the walls of two chambers connected by a channel. Inlets/outlets to each chamber were drilled beforehand in the polycarbonate lid (numbered in white in panel A). Inlets 1, 3 and outlets 2, 4 give access to the loading and recovery chambers. Cells are loaded into the device using standard micropipette tips shown in B. The device is wirebonded to a printed circuit board (PCB) and the ITO is connected to the PCB with a drop of conductive glue deposited during the packaging process.

Results

The microsorter device

The DEParray prototype is a hybrid silicon/plastic device consisting of a core silicon chip, a double sided structured adhesive tape and a conductive polycarbonate lid. Device packaging is illustrated in Fig. 1 and described in detail in Methods. The 0.64 cm² silicon chip hosts over 100,000 microsites arranged as a planar array of 20 µm square electrodes.¹⁵ On one half of the chip, capacitive sensors use the superficial microelectrodes and the conductive lid for capacitive sensing of the volume between the silicon microsite and the lid. On the other half of the chip, the sensor is optical and is embedded between neighboring electrodes to measure variation of light absorption due to the presence of a particle over the gap. The silicon chip is mounted with a water-resistant adhesive tape which acts as a wall and spacer for the manipulation chambers and with a conductive indium tin oxide (ITO)/polycarbonate (PC) lid bearing fluidic inlets and outlets. These ports are compatible with micropipette filling and recovery using standard disposable plastic tips, making loading and recovery of cell suspensions a straightforward and easy process. Before and after use the device was rinsed in water with mild dish washing detergent followed by ethanol.

Configuring dynamic DEP traps

A 3D DEP trap may be configured at any position on the 320 × 320 electronic array by applying a sinusoidal voltage to the center electrode and to the conductive ITO/PC lid, and a counterphase sinusoidal voltage to the 8 surrounding electrodes. Multiple traps may either be configured at will over the whole surface of the array or a regular pattern of traps may be configured by repeating a basic frame over the whole chip (illustrated in Fig. 2A). Setup parameters include frequency and amplitude of the sinusoidal voltage as well as the amplification factor for the voltage applied to the lid. The following field conditions were applied for cell trapping: 800 kHz, 3.3 V peak to peak, a lid gain of 2. Cells were suspended in a low conductance 280 mM mannitol buffer supplemented with 6% fetal calf serum (FCS) or 6.25 mM potassium chloride to adjust the conductivity of the buffer to 780 µS cm⁻¹. FCS offered the added advantage of limiting adhesiveness of certain cell lines to the device's surfaces but in all the following experiments could be replaced with potassium chloride to yield identical results.

Fig. 2 Cell organization in a DEP trap array. Panel A illustrates a standard grid pattern of cages over a 23 × 19 subset of the electrode array. One cage is generated every 4 electrodes by applying a sinusoidal waveform to the lid and to every fourth electrode (in black) and a counterphase waveform to all their surrounding electrodes (in dark grey). Panel B shows a photograph taken through a standard microscope of an area of the device a few seconds after loading K562 cells at 8 × 10⁵ cells ml⁻¹. Cell sorting: Panels B, C and D shows a sequence of photographs taken from a film captured through the microscope where 5 cells (arrowed) are moved independently and simultaneously from their initial coordinates in the grid to the recovery chamber. Panel E shows the computed pathways for each cage: optimum cage pathways are automatically calculated to move around the static cages of the pattern (represented by small dots) and obstacles (walls, outlets and inlets represented with a checkered motif) detected by the optical and sensor arrays. Numbers identify each cage and its pathway. Black lettered rectangles in panel E correspond to the microscope fields presented in panels B (start), panel C (after approximately 30 motion steps) and panel D (end). The full video is available as ESI.†

A variety of cell types such as laboratory cell lines as well as primary cells was tested in the device. Cells were loaded into the larger chamber of the chip while the electrode array was configured with a regular grid pattern depicted in Fig. 2A. We have found that activating the grid pattern before loading the sample limits cell adhesion and aggregation on the chip surface. All cell types got trapped in less than 30 seconds in the grid, illustrating that DEP forces have an effect on cells 40–50 µm away from the cage itself. Standard laboratory cell lines such as K562 (Fig. 2, panel B) and Jurkat (not shown) were recovered and tested for proliferative capacity: after being submitted to DEP fields for 30 min, cells were put back into culture and counted over 6 days versus a control culture. Proliferation was consistently comparable to the control culture for both cell lines (Fig. 3A). Our simple washing procedure with water and mild detergent followed by an ethanol rinse was sufficient to guarantee sterile culture conditions as we observed no microbial contamination in any of our multiple postDEP culture experiments. Adherent cell lines, such as HeLa cells, could also be manipulated after trypsinization, as well as primary cells, such as red (RBC) or white (WBC) blood cells freshly prepared from donor blood (not shown).. The viability of all nucleated cells was tested systematically after each DEP manipulation, for all cell lines as well as in all buffer conditions. Cell viability was consistently comparable to the control cell suspension. For example, representative counts on WBCs shows 84–92% viability after 30 min of DEP manipulation, comparable to the viability of the same cells measured before the experiment. We analyzed possible DNA damage using a modified Comet assay.¹⁸ The standard Comet assay tests for DNA fragmentation due to cell apoptosis. By adding a DNA repair enzyme (Fpg: formamidopyrimidine–DNA glycosylase), the modified Comet test detects slighter damage such as DNA oxidation. In these assays, we found no increase in damage on the cells' DNA: all comets were classified as class I (very small damage) using the method described in ref. 19 and no change of class was seen between control cells and those having undergone DEP. Results were analyzed further by determining mean tail intensities from two independent samples having undergone 30 min of DEP versus a control (Fig. 3B). The data confirms no significant increase in damage on cells having undergone DEP. All these results are coherent with previous data from studies exploring the effect of DEP on cells.⁴

Fig. 3 Study of DEP effects on cells. Panel A shows results from a representative proliferation experiment carried out on K562 cells which were manipulated under DEP fields for 30 min before recovery and culture in a microplate format. Between 80–100 cells were seeded in each well and proliferation was followed by counting cells every 24 h. On the third day of culture, cells were diluted 3 fold in fresh medium and put back into culture. Line curves correspond to cells which underwent DEP manipulation (▲) compared to those which remained in culture medium for 30 min (◊) and to cells that were suspended in low conductance buffer without being submitted to DEP (□). Panel B shows results from a Comet assay carried out on adhesive HeLa cells. After electrophoretic migration, the tail intensities were measured for 50 cells and mean and standard error of mean were calculated. Black bars correspond to the test including the Fpg enzyme for the detection of oxidative damage and grey bars correspond to the test without Fpg for the detection of DNA breaks and alkali sensitization. Error bars represent standard error of mean. Two samples of HeLa cells were submitted independently to DEP (test 1 and 2) and are presented versus a control of cells which were directly taken from the cell culture.

Depending on cell concentration and cell size, the average number of cells per cage could be modulated from 0 to more than 10. At 5 × 10⁶ cells ml⁻¹, an average of 5–6 RBCs cluster into each cage (not shown). The optimal cell concentration to obtain a homogeneous distribution of a single cell per cage was found to be 8 × 10⁵ cells ml⁻¹ (Fig 2 panel B) and this concentration was used in the following experiments to manipulate, sort and recover individual cells from a cell suspension. Taking into account a 2.9 µl chamber volume, this translates into a cell load of ∼2,320 cells for approximately 4,500 cages generated by the 4 × 4 grid pattern illustrated in Fig. 2A. This concentration corresponds to about 0.5 cell per cage, which is consistent with a statistical prevalent distribution of a single cell per cage.

Moving cells with electronic control alone

Motion is carried out by applying the same configuration of the sinusoidal waveforms to the neighboring electrodes in the direction of the desired motion. A new DEP cage is generated in this position and the cell is pulled to this new location. The principle of these moving DEP cages is described in ref. 20. Once cells have organized into the grid pattern, any number of cages may be moved independently to new locations. Start and final coordinates are entered into the control software. The software allows automatic computation of optimized pathways, moving around the pattern cages and avoiding collision with other particles. Experimental results showed that cells can be moved at an average speed of 1 mm per minute (approximately one electrode every second). Considering the overall size and configuration of the double chamber chip, this means that a particle can be moved from any location on the chip to its new location in less than 15 minutes. As all the selected cells are moved in parallel with no incidence on the speed of motion of each individual one, all motions can be completed in the same timeframe whatever the number of moving cells.

Sorting and recovering individual cells

The simple manual mounting process of the adhesive tape and lid does not require precise alignment of all components as positions of obstacles such as walls and air bubbles, as well as position of the connecting channel between chambers, are determined easily by using the embedded chip sensors. Both the optical and capacitive sensors detect the presence of tape or air bubbles over the silicon chip and generate a 320 × 320 pixel image in which each pixel corresponds to a single microsite on the chip and thus precise coordinates of each element can be determined directly and integrated into the control software. Walls, ports and even air bubbles, although uncommon, are identified and the control software automatically computes motion pathways to avoid these spaces. Thus, if particles need to be moved from one chamber to the other, the particles will automatically be routed through the connecting channel and around the static grid pattern.

Before filling the chip, a regular pattern of cages is activated every 4 electrodes as illustrated in Fig. 2A. The cell sample is then loaded into the larger chamber as described in Methods. While cells organize into the pattern, capacitive and optical readouts return a grey scale image of the wall/port layout (not shown), which defines the restricted spaces on the chip.

In the experiment described in Fig. 2, a heterogeneous cell sample was prepared by mixing K562 cells stained with CFSE (carboxy-fluorescein diacetate, succinimidyl ester) with unlabeled cells in a 1 ∶ 4 proportion. This cell suspension was loaded into the larger chamber (Fig 2B). Illumination was switched to fluorescence and positive labeled cells in the observation field were pinpointed and their coordinates were determined using a motorized stage as described in Methods. These coordinates, as well as final coordinates in the recovery chamber were entered into the control software which generates automatically the optimized pathways (Fig. 2E). In this experiment, 5 fluorescent cells were chosen. Fig. 2C shows a photograph taken during the sorting experiment after approximately 30 motion steps. Cells 2, 3 and 5 have clustered together into a single cage as their respective pathways have joined together after an identical number of motion steps (Fig. 2C). These cells move then as a triplet to the recovery chamber where their respective pathways diverge again. Thus, the cell cluster partially dissociates into separate cages: cell 2 moves to its final coordinates, whereas cell 3 stays with cell 5 and the final position of cage 3 is empty (Fig. 2D).

After all sorting experiments, once all labeled cells had been moved to the recovery chamber, its content was pipetted and placed in the bottom of a microwell plate. Fluorescently labeled cells were counted through a standard fluorescence microscope. The recovery rate of these selected cells is typically between 50% and 80% after having targeted and moved 2 to 10 cells to the recovery chamber.

Discussion

The presented data undoubtedly demonstrates how our microsystem answers the challenge of manipulating and sorting small cell samples. In contrast to standard requirements of flow cytometers, we can manipulate exquisitely small samples of only a few thousand cells with no dead volume, and observe cells continuously over time. Our technology is compatible with specific phenotypic fluorescent markers by direct microscope observation through the transparent lid. Moreover, our device offers sorting with single cell resolution and live cell recovery with no detectable effect on cell DNA, viability or proliferation capacity.

The system shows great flexibility in the type of cell it can manipulate, from cell sizes between 5 to 20 µm, to adherent or non-adherent cultured cell lines, as well as primary cells isolated from fresh clinical samples. Other dielectric particles may be manipulated, in particular polystyrene beads in the same size range. Larger particles may also be manipulated by clustering electrodes together. For example, a larger cage can be generated by applying a counterphase sinusoidal voltage to a 2 × 2 electrode pad and an in-phase voltage to the 12 surrounding electrodes. This cage can efficiently trap and move particles of 50 µm (data not shown). Today, our current device and operating system is restricted to a cell load of a little over 2,000 cells in a 4 × 4 grid. In a number of applications, this amount of cells is obtained after enrichment steps such as immunodepletion or density centrifugation and our technology offers a high performance solution to automatic isolation of single cells of interest in that small population. In fact, even 25,000 cells can be manipulated on the current chip by modifying the routing procedure and opening paths in a denser 2 × 2 array to channel cells to the recovery chamber. An even higher cell load can be obtained by shrinking electrode size or increasing chip surface. For example, an electrode feature size of 10 µm sufficient for a great number of eukaryotic cells over a surface of 1 cm² brings the number of active cages to 250,000 in a 2 × 2 grid, the equivalent of 100 current chips with no manufacturing difficulty. Automatic recognition of cells of interest from fluorescence images needs then to be implemented in a similar configuration to that seen on automatic microscope platforms with commercially available cell recognition software. In contrast to standard microscope images, the initial organization of all cells in the device into a regular predetermined matrix will greatly facilitate the subsequent automated recognition of labelled cells.

Our technology offers several advantages over other emerging technologies based on either dielectrophoresis, electroosmosis or optical tweezers.^6–10,12,13 Foremost, our device shows great flexibility and any zone of the chip can be programmed to become a virtual analysis window or a channel. Operation requires electronic control alone through a standard PC computer. Furthermore, as motion and sorting is carried out in a stationary liquid environment, there is no need of fluid flow control, or of hydrodynamic focusing, no need of fluidic particle transport or of fluidic switching in order to efficiently sort individual particles from a sample. Moreover, our multiplexed array allows parallel and independent manipulation of hundreds of cells at a time as compared to other systems where parallelization is obtained either by multiplying the number of channels, or by small arrays of immobile traps (for example of vertical cavity surface emitting lasers (VCSELs)²¹). The motion at an average speed of 1 mm min⁻¹ may appear relatively slow in regard to speeds obtained with particle fluid transport but the capacity of simultaneously moving all cells of interest at that same speed wraps up a typical sorting experiment on a 6 × 6 mm² chip in 15 minutes. There are no microstructures or moving microactuators that can get clogged and this is a major advantage when working with a complex medium such as biological fluids. Even if a cell aggregate does form and occupies an area on the chip, software programming makes it immediately possible to move around the aggregate and continue the experiment. From a detection point of view, the conductive lid is completely transparent to the standard fluorescent labels and any analyses routinely carried out in a slide and coverslip format can be done through the ITO lid.

The use of silicon microfabrication techniques not only offers low cost mass production but also allows the implementation of other functions on the chip. Today, our sensing arrays need to be upgraded to gain sensitivity and reliably detect a single cell. The whole manipulation will then be carried out automatically as numerical feedback on these sensor arrays will allow the control system to identify the cells of interest (for example through embedded optical detection instead of through an external fluorescence microscope). The control system would then program the appropriate cages to trap these cells and then proceed to move them simultaneously to the recovery chamber by avoiding other particles on the way, the latter being detected by capacitive sensing.

Other protocols may also be implemented on the chip. We are developing applications and protocols in which the user may choose to bring a specific cell into contact with other cells to study single cell–cell interactions, and/or then might choose to fuse this cell with a liposome carrying drugs or DNA (and also trapped in a DEP cage). He may then choose to move the cell to a new analysis chamber for a closer phenotypic study and finally may want to recover the cell for further analysis (such as single cell PCR or clonal amplification). Any of these protocols may be configured by software programming alone with no modification of the chip and hardware design.

Acknowledgements

This project was funded by the European Commission (MeDICS project, EU grant IST-2001-32437), and is supported by Italian Ministry of Research under FIRB project. The authors acknowledge Massimo Rubboli for cell-routing software development and Sylvie Sauvaigo for assistance in the Comet assays.

References

1. F. Z. Bischoff, M. K. Sinacori, D. D. Dang, D. Marquez-Do, C. Horne, D. E. Lewis and J. L. Simpson, Hum. Reprod. Update, 2002, 8, 493–500.
2. S. S. Farag, J. B. VanDeusen, T. A. Fehniger and M. A. Caligiuri, Int. J. Hematol., 2003, 78, 7–17.
3. R. Lovell-Badge, Nature, 2001, 414, 88–91.
4. S. Archer, T. T. Li, A. T. Evans, S. T. Britland and H. Morgan, Biochem. Biophys. Res. Commun., 1999, 257, 687–98.
5. T. Müller, G. Gradl, S. Howitz, S. Shirley, T. Schnelle and G. Fuhr, Biosens. Bioelectron., 1999, 14, 247–256.
6. J. Voldman, M. L. Gray, M. Toner and M. A. Schmidt, Anal. Chem., 2002, 74, 3984–90.
7. Y. Huang, S. Joo, M. Duhon, M. Heller, B. Wallace and X. Xu, Anal. Chem., 2002, 74, 3362–71.
8. C. Reichle, K. Sparbier, T. Muller, T. Schnelle, P. Walden and G. Fuhr, Electrophoresis, 2001, 22, 272–82.
9. X. B. Wang, J. Yang, Y. Huang, J. Vykoukal, F. F. Becker and P. R. Gascoyne, Anal. Chem., 2000, 72, 832–9.
10. A. Y. Fu, C. Spence, A. Scherer, F. H. Arnold and S. R. Quake, Nat. Biotechnol., 1999, 17, 1109–11.
11. J. Suehiro and R. Pethig, J. Phys. D: Appl. Phys., 1998, 31, 3298–3305.
12. S. Fiedler, S. G. Shirley, T. Schnelle and G. Fuhr, Anal. Chem., 1998, 70, 1909–15.
13. P. C. Li and D. J. Harrison, Anal. Chem., 1997, 69, 1564–8.
14. J. Enger, M. Goksor, K. Ramser, P. Hagberg and D. Hanstorp, Lab Chip, 2004, 4, 196–200.
15. N. Manaresi, A. Romani, G. Medoro, L. Altomare, A. Leonardi, M. Tartagni and R. Guerrieri, IEEE J. Solid-State Circuits, 2003, 38, 2297–2305.
16. A. Romani, N. Manaresi, L. Marzocchi, G. Medoro, A. Leonardi, L. Altomare, M. Tartagni and R. Guerrieri, Proceedings of the International Solid State Circuit Conference, 2004, vol. 1, pp. 224–225.
17. I. Chartier, C. Bory, A. Fuchs, D. Freida, N. Manaresi, M. Ruty, J. Bablet, K. Gilbert, N. Sarrut, F. Baleras, C. Villiers and L. Fulbert, Proceedings of the Microfluidics, BioMEMS, and Medical Microsystems II Conference (SPIE), 2004, vol. 5345, pp. 7–16.
18. S. Sauvaigo, C. Petec-Calin, S. Caillat, F. Odin and J. Cadet, Anal. Biochem., 2002, 303, 107–9.
19. A. Collins, M. Dušinská, M. Franklin, M. Somorovská, H. Petrovská, S. Duthie, L. Fillion, M. Panayiotidis, K. Rašlová and N. Vaughan, Environ. Mol. Mutagen., 1997, 30, 139–146.
20. G. Medoro, N. Manaresi, M. Tartagni and R. Guerrieri, Proceedings of the IEEE International Electronic Devices Meeting, 2000, pp. 415–418.
21. R. A. Flynn, A. L. Birkbeck, M. Gross, M. Ozkan, B. Shao, M. M. Wang and S. C. Esener, Sens. Actuators, B, 2002, 87, 239–243.

---

Footnote

† Electronic supplementary information (ESI) available: Video captured through a microscope where 5 cells are moved independently and simultaneously from their initial coordinates in a grid to a recovery chamber (see Fig. 2 for details). See DOI: 10.1039/b505884h

---