The cell capture device is comprised of thousands of polydimethyl siloxane (PDMS) cell traps densely arrayed within a flow-through channel. Each cell trap consists of a weir structure that extends vertically into the channel and contains front- and backside capture cups (Fig. 1a–d). Support pillars placed on either side of the capture cups allow flow into and under the trap. We tailored the pillar heights to be slightly smaller than the cell diameter so the cells were trapped once they entered the capture cup. The support pillars also maintained proper channel height across the array once the device was bonded to a glass substrate. The cell traps were incorporated into three different devices; the largest was 8 mm × 4 mm and contained ~6000 traps (Fig. 1e). We observed that the trap spacing within the array was critical for efficient capture without clogging. With optimal column spacing (~1–1.5 cell diameters, ~20 μm) and a row spacing of 20 to 50 μm we could capture 70–90% of the cells that entered the array (Supplementary Fig. 1).

We accomplished two-cell capture and pairing using a 3-step loading protocol. We first isolated single cells in the smaller backside capture cup (Fig. 2a). Once the array was saturated, the cells were transferred directly “down” into the opposing larger capture cup (Fig. 2b). This transfer was fast (< 1s), massively parallel and highly efficient because of the laminar flow within the device (Supplementary Video 1 online). Finally, the second cell population was loaded and trapped immediately in front of the previously trapped cells (Fig. 2c). The larger frontside cup was sized to trap two cells, so additional cells traveled through the array until it was saturated. We obtained two-cell capture efficiencies up to ~80% (percentage of traps occupied by exactly 2 cells of any type), and pairing efficiencies of up to 70% (Fig. 2d, see also Supplementary Discussion online). Higher efficiencies were possible in the middle and bottom of the array where less penetration of larger cell clumps and therefore better single cell transfer occurred.

We next aimed to test the compatibility of our device with both chemical and electrical fusion protocols. We determined fusion efficiencies by imaging and quantifying both the exchange of fluorescent proteins (indicative of initiation of fusion) and plasma membrane reorganization (indicative of advanced fusion). In some experiments we determined complete fusion after prolonged culture in vitro. We paired and fused different cell types, including NIH3T3 fibroblasts, mESCs, mEFs, B cells and myeloma cells

We first explored the capability of our device to fuse cells using PEG. We flowed PEG past the cells[ please specify which cells were used in video 2], causing them to shrink from the osmotic shock (Supplementary Video 2 online). During this time the cells remained in contact and stationary within the array, demonstrating that our trap geometry can successfully immobilize the cells even though there is a substantial change in cell volume. Next, we washed the PEG out with media, causing the cells to swell back to their original size and initiate fusion. An advantage of our device is that solutions can be exchanged rapidly while the cells remain paired and in contact; therefore further doses of PEG can be applied to increase fusion efficiencies without losing cell pairing or registration. A single dose of PEG yielded 15% fluorescence exchange over CellTracker-stained 3T3 pairs and 8% membrane fusion of unstained 3T3 cells, while subsequent doses of PEG yielded up to 35% fluorescence exchange over red/green pairs and 25% membrane reorganization of unstained 3T3 cells (Supplementary Fig. 2 and Supplementary Video 3 online). Viability staining with trypan blue indicated an increase in cell death with additional doses of PEG, eventually limiting the effectiveness of subsequent doses.

We then adapted our device to be compatible with electrofusion protocols. In order to introduce electric fields we plasma-bonded the device to a glass slide containing metal electrodes (Fig. 1e). Once the cells [please specify the cells in video 4]were paired and immobilized, we flowed hypoosmolar fusion buffer past the cells, causing the cells to swell. The capture cups utilized were slightly deeper to accommodate the cells as they got larger (Supplementary Video 4 online). Again, as with the PEG protocol, the cells remained immobilized and paired as they changed size. An added benefit is that the cells are pre-aligned and in contact so no AC field is required. We analyzed membrane fusion after the electrical pulse (Supplementary Videos 4, 5 and Supplementary Fig. 3 online). We found electrofusion to be significantly more efficient than PEG (P < 0.05); a single series of pulses yielded 78% fluorescence exchange over CellTracker red/green pairs and 89% membrane reorganization of unstained 3T3 cells.

Another advantage of our device is the ability to observe the progression of fusion at the single-cell level. Before and immediately after the PEG application two distinct membranes were visible and fluorescence was still localized within each cell (Fig. 3a). After 10 minutes green fluorescence was observed within the mEF demonstrating that the cytosols of the two cells had connected and fusion was initiated. The Hoechst fluorescence was still localized in the mEF, indicating that the nucleus was intact. After 15 min the plasma membranes began to reorganize, leading to a hybrid cell containing the contents of both cells. The Hoechst fluorescence remained partitioned in the new hybrid cell, suggesting no nuclear fusion had taken place. Electrofusion followed a different timecourse (Fig. 3b–c). Interestingly, fluorescence exchange was detected within seconds after the electric pulse, and in most cases the outline of the nucleus was visible as the fluorescence first moved into the cytoplasm. This exchange of fluorescence was clear even though the cell membranes had yet to reorganize. After 10–20 min the plasma membranes began to reorganize. By immobilizing the cells we were able to observe and distinguish between the exchange of cell contents and membrane reorganization for single pairs.

Immobilizing the cells in a dense array also provides the opportunity to observe fusion for thousands of cell pairs in parallel. We used computational image analysis to monitor fluorescence exchange over the entire device in a fashion similar to a FACS plot (Supplementary Figs. 4–6 online). Immediately after the electrical pulse the red-green double-positive population increased to 53.5%. With increasing time more cells exchanged fluorescence (maximum 63.9% at t = 5 min). In addition, the amount of fluorescence exchanged also increased as shown by the red-green double-positive cell populations moving towards the center of the plot. This indicates that connections are established which allow continual exchange and eventual equilibration of cytosolic material²⁴. Slight decreases at later times were artifacts due to cells shrinking and moving out of the range of the analysis box.

Our device allows direct comparison of the fusion efficiencies of different fusion stimuli. We compared PEG and electrofusion efficiencies of properly paired cells determined by manually inspecting the images and evaluating the fluorescence exchange or membrane reorganization (Fig. 4a). Since this measurement is independent of the capture and pairing efficiency, it provides a comparison of different fusion techniques. Using PEG, we were able to initiate fusion of 39% ± 14% of cell pairs, while electrofusion resulted in a significantly higher 78 ± 12% fusion pairs (P < 0.05). These electrofusion efficiencies were comparable to those obtained when Discosoma sp. red fluorescent protein (DsRed)-expressing and enhanced green fluorescent protein (eGFP)-expressing cells were fused electrically in the device (68 ± 24%, with a single-run high of 91%), and with CellTracker-stained mESCs and mEFs (single-run value of 56%). We also placed cells in the device and cultured for three days without fusion stimulus (Supplementary Fig. 7 online); no doubly labeled cells were observed, indicating that negligible fusion occurred in the absence of fusogenic stimuli.

We compared the overall efficiencies in generating fused cells with our device to standard commercial PEG and electrofusion instruments and protocols. To compare between commercial and chip-based protocols, we primarily used a common fusion metric of fluorescence exchange that could be assessed for all protocols and has been used by others^25,26, and used membrane re-organization for the on-chip PEG experiments. Fluorescence exchange (% red-green double-positive cells over the whole cell population) was determined either by our image analysis program or by FACS while membrane re-organization was evaluated visually. A standard PEG protocol yielded 6 ± 4% fused cells using stained 3T3 fibroblasts compared with a significantly higher 25 ± 5% obtained after 4 doses of PEG in our microfluidic device (P < 0.05, Fig. 4b). When we compared electrofusion performance in a commercial system to the microfluidic device, we found significantly higher fusion efficiencies in the microfluidic device (P < 0.05). For stained 3T3 fibroblasts, we obtained 11 ± 9% fusion in the commercial Helix chamber (Eppendorf, Westbury, NY) as compared to 51 ± 16% obtained in our device, while for fluorescent-protein expressing fibroblasts, we obtained 4 ± 2% fused cells in the commercial electrofusion system and 40 ± 13% in the microfluidic device. Finally, we obtained 11 ± 4% electrofusion of stained mESCs and mEFs while 23% was achieved with our device (single run). In all cases the microfluidic device delivered a 2- to 10-fold improvement on fusion yield compared to commercial systems.

We next examined whether cells can be removed from the chip after fusion and can survive prolonged culture. We removed NIH3T3 fibroblasts after fusion in our device (Supplementary Fig. 8 online) and cultured them for 10 days. We obtained viable fused cells as demonstrated by the presence of red-green double-positive cells (Fig. 5a–b), and via FACS analysis (not shown).

Fusions of embryonic stem cells with somatic cells have been used to demonstrate the capability of ES cells to reprogram somatic cells^6,7. To show that our microfluidic device can also generate viable hybrids between mESCs and mEFs, we fused Hygromycin-resistant mESCs with Puromycin-resistant mEFs in our device and cultured them under self-renewing conditions. After 14 days under double-selection, we observed drug resistant colonies that had an ESC-like morphology and stained positive for alkaline phosphatase (Fig. 5c). Reactivation of embryonic genes, such as Nanog and Oct4, has been used to demonstrate successful reprogramming of somatic cells^7,27,28. The Puromycin-resistant mEFs carried an additional Oct4-GFP reporter in their endogenous oct4-locus, allowing us to investigate whether reprogramming as judged by the reactivation of Oct4-GFP would also occur. We were able to detect alkaline phosphatase-positive colonies that also expressed GFP, demonstrating that our device is suitable for generating viable hybrids and observing reprogramming of mEFs after fusion with mESCs (Fig. 5d).

Masters for the microfluidic device were made from SU8 (MicroChem) spun on silicon wafers using standard photolithographic techniques²⁹ (Supplementary Figs. 1 and 10 online show dimensions and masks). PDMS was poured over the master and then degassed before curing. Glass slides with electrodes were constructed from mask blanks pre-coated with chrome and photoresist (Telic), patterned by a transparency mask exposed to UV. The PDMS devices and glass slides were assembled using plasma bonding.

The devices were blocked with 7.5% bovine serum albumin and rinsed with phosphate-buffered saline before use. Cells were manually placed in the top inlet reservoir and drawn through the device at 15–50 μm/s using a syringe pump.

5×10⁵ cells of each cell type were pelleted and resuspended in ~500 μL of media, filtered through a 35 μm cell strainer (BD Falcon), and loaded into the device as described previously. PEG-1500 was put into the inlet reservoir and drawn past the cells at 0.4 μL/min for 3–5 min. The cells were washed with warm 1:1 PEG:media for 1 min, then incubated in warm media for 26 min. At t = 30 min, the cells were washed with trypan blue (10% in PBS) for 5 min, then with media for 5 min. At t = 40 min the second dose of PEG was applied, and the entire protocol was repeated for a total of 4 doses

We connected the electrodes to a power supply (Eppendorf) in parallel with a 50 kΩ resistor . After cell loading, we flushed with hypoosmolar fusion buffer at 0.4 μL/min for 10 min. The cells were pulsed at varying voltages (0.5 to 2.0 kV/cm) for 50 μs × 5 pulses. Hypoosmolar fusion buffer was flushed past the cells for an additional 10 min before being replaced with warm media. The cells were then incubated for an additional 15 min at 37 °C.

The microfluidic device was placed on an automated inverted microscope (Zeiss Axiovert 200m) fitted with a stage incubator (In Vivo Scientific) and images were acquired either every 2.5 min or 5 min. A single randomly chosen image field (~200–300 capture combs) was used for each experiment, and the size of the image field remained constant. Images were analyzed in ImageJ (http://rsb.info.nih.gov/ij/) to determine pairing efficiencies (number of traps in the field of view occupied with a single cell of one type in the bottom of the well with a second cell (or more) of the other type on top) and fusion efficiencies (based on fluorescence exchange or membrane reorganization). Fluorescence exchange efficiencies were also analyzed using an automated macro written in ImageJ.

See Supplementary Methods online for full fabrication, assembly, fusion, and imaging details.