Microfluidic sorting and multimodal typing of cancer cells in self-assembled magnetic arrays

Superparamagnetic beads are extensively used in conventional immunomagnetic sorting. Upon application of an external magnetic field, each bead develops a strong magnetic moment parallel to the field. In previous magnetic sorting strategies (e.g., 5, 18, 19), the cells to be captured were mixed with a suspension of beads, and a high magnetic field gradient was used to collect the beads either statically or in a flow. Here, we transpose this paradigm, and use the dipole-dipole interactions between beads to create a posts array. Under moderate external fields (typically 10–30 mTesla), these interactions overcome Brownian motions and induce the formation of chains oriented in the field direction (20). When the suspension is enclosed in a slit-like microchannel (Hele-Shaw cell) with its main walls perpendicular to a uniform field, an array of columns is formed with a hexagonal order on short distances and glass-like order on long distances (21). The average center-to-center distance between two columns, and the distribution of pore sizes and column diameters, result from a complex interplay between the thermodynamics of bead-bead interactions, the kinetics of bead chains formation, and the interactions of the chains with the microchannel’s walls. The stability of this array decreases when the microchannel thickness and/or the column spacing increase. In practice, arrays with a spacing of 2–5 μm are easy to achieve with a good reproducibility, and were previously used in our laboratory for the separation of DNA (22, 23). In contrast, arrays with column spacing larger than 10 μm are irregular and fragile. Thus, spontaneously self-assembled magnetic arrays are not suitable for separating eukaryote cells with a typical diameter ranging from 5 μm–20 μm.

To overcome the above limitation, we propose here to use a permanent magnetic pattern with the desired organization, deposited at the bottom of the microchannel, to direct beads self-assembly (Fig. 1A). Several methods for microfabricating magnetic patterns were previously proposed (24, 25), but they require advanced facilities. We propose here a simpler alternative, based on the microcontact printing (26) of a water-based ferrofluid (“magnetic ink”) onto glass (SI Text and Fig. S1). These magnetic ink dots have a magnetic volume susceptibility, 2.33, about 10⁵ times higher than that of glass or buffer. When an external uniform magnetic field is applied, they concentrate field lines and create local gradients that act as potential wells for the magnetic beads contained in the supernatant liquid.

An integrated system was prepared in PDMS (Polydimethylsiloxane) following a process inspired from ref. 27. This system comprises a lower layer with microchannels for the transport of fluids and cells, and an upper layer containing microchannels crossing the fluid transport channels and acting as pinch valves (Fig. 1B). These valves open or close the inlet channels by application of an external pressure. To avoid flow pulses detrimental to the stability of the magnetic columns, the valves and the flow of samples were controlled by a customized high resolution pressure controller (28) (Fluigent, Paris, see SI Text).

The cell-antibodies interactions in our system occur in hydrodynamic conditions very similar to those at play in micropillar arrays, so we expect to recover here the high capture efficiency emphasized e.g., ref. 15, 29. Ephesia has four further advantages: (i) Functionalized beads can be prepared in large quantities in a batch process, yielding a lower production cost and easier quality control. Using modified beads also opens access to more elaborated antibody immobilization protocols, and to hundreds of commercially available microparticles, capitalizing the know-how of about 20 yr of research and development and clinical validation; (ii) The self-assembly process can yield convenient microstructures with an aspect ratio out of reach of the most sophisticated nanofabrication tools (30), offering new possibilities for imaging. Notably, we could achieve post arrays with 50 μm height and 4.5 μm diameter (aspect ratio 12), allowing imaging with 100X, 1.4 NA objectives, and analysis with commercial softwares, whereas the silicon devices in (15) only used 10X objectives, and even with such low aperture objectives, they required specific software developments to deal with its imaging complexity (16); (iii) Our devices can be prepared by a combination of low resolution molding or embossing and microcontact stamping, for which both low cost laboratory and high throughput industrial machines exist. Thus, the chips can be mass produced at low cost (in contrast e.g., with the devices in ref. 15), which require for each chip a large area monocrystalline silicon and deep wet etching. (iv) Finally our system can accommodate smaller volumes: our cell’s volume is around 0.5 μL and can be operated with less than 10 μL total sample volume, whereas the cell in ref. 15 has an internal volume of 60 μL. These volumes are consistent with our different objectives, CTC for ref. 15, and FNA for us.

As compared to our earlier work on self-assembly (21–23), the use of magnetic ink templates has two advantages: (i) First the magnetic potential wells impose onto the magnetic columns a perfectly predefined order, instead of the glass-like order obtained at long range with spontaneous self-assembly (21); (ii) Second, the magnetic wells stabilize the array, and allow much higher flow rates.

Two types of experiments were designed to validate the potential of this system for cell sorting. First, yield, specificity, and in situ postcapture culture were evaluated using mixtures of cells from culture cell lines: T (Jurkat cell line, ATCC TIB-152) and B (Raji cell line, ATCC CCL-86) lymphoid cells.

Second, to assess the potential of Ephesia on clinical samples, we focused on the classification of B-cell malignancies, a major type of hematological cancers. B-cell malignant tumors generally develop in lymph nodes. Cancerous B-cells may circulate in blood, populate serosal membranes, and/or give rise to effusions such as pleural effusion. The classification of B-cell cancers is complex, and diagnosis is secured by a combination of clinical observation, morphology, immunophenotyping and genetic characterization (World Health Organization (WHO) classification) (31). Briefly, diagnostic steps starts with a morphological characterization aimed at determining the presence of cancer cells. Upon positive response, a further immunophenotypic characterization is performed (most often by FC), to identify specific antigen expression at the cell surface. This procedure may require several sequential immunophenotyping experiments, and in the most unfavorable situations several hundred μL of sample, exceeding the volume of a FNA. It thus constitutes a good challenge for evaluating an Ephesia system for multimodal, high-content morphotyping and phenotyping of cancer cells from minute samples.

In order to extract from experimental data microscopic parameters useful for further optimization, a theoretical model of cell capture in our device was developed. The array is subdivided into a series of rows. A master equation predicting the time-dependent occupancy of each row of obstacles is expressed as a function of cells flow rate, a geometrical factor (collision probability) and a microscopic, biological factor (capture efficiency) is derived (see SI Text for the full model). In the limit of low occupancy (short times or samples with only a few positive cells), the model simplifies greatly, and the number of cells in row k after the entrance of the Nth cell, m_k,N, per row can be approximated by a simple exponential m_k,N ≅ c exp(-αuk) where c is a constant depending on the total number of cells, u is the collision probability per row (a hydrodynamic parameter) and α the capture efficiency. A comparison between the theory and experimental data was used to determine the microscopic capture efficiency, in the general case and in the “rare cells” limit.

The array of magnetic dots was prepared on a 0.5 × 10 mm² area. Optical imaging shows that a regular, uniform, and essentially defectless array is achieved over the whole surface (Fig. 1C). Electronic microscopy shows that dots adopt a reproducible cone-like shape (Fig. S2). Their diameter, 5 ± 1 μm, is significantly smaller than the initial size of the pins in the stamp. This size reduction and cone-like shape are consequences of hydrodynamics flows in the ink and partial dewetting during the lift-off process. They allow for a better alignment of the magnetic column on the spot’s center. Atomic Force Microscopy (AFM) measurements indicate that the top of the spot is 475 ± 25 nm high. After a baking step overnight at 150 °C the arrays could be washed and used repetitively for days.

The magnetic beads, in suspension in PBS containing 0.1% BSA at a concentration 3 mg/mL are introduced into the separation channel under microfluidic control. Flow is then arrested, and the magnetic field is applied (Fig. S3 and Movie S1). The beads self-organize into columns over the ferrofluid dots, as expected (Fig. 1C). Surprisingly, when a moderate velocity (less than 20 μm/s) was applied to the array and the magnetic field was turned off, the columns kept their cohesion and remained attached to the magnetic dots. We attribute this attachment to field-mediated bead-bead adhesion promoted by the interpenetration of polymer or protein layers at the surface of the beads, under the pressure induced by dipole-dipole interactions (30). In the present microchannel configuration, under moderate flow these cohesive columns align in the flow at the bottom of the channel. We could then characterize the columns in side-view (Fig. 1C, insert). We studied the dependence of the structure of the magnetic columns, upon the initial concentration of the beads suspension. For the beads and arrays used here (4.5 μm Dynal beads on a hexagonal array with a 40 μm center-to center spacing) the optimal beads concentration in the suspension is 3 mg/mL. For this concentration, columns have a height equal to the channel’s thickness, and are made of single aligned beads with only a few defects. At higher concentration, thicker columns are obtained (Fig. S4). When magnetic columns are assembled under optimal conditions (see SI Text), they start to detach for average flow velocities between 800 μm/s and 1 mm/s. This resistance is a considerable improvement with regards to nontemplated magnetic arrays, which are destabilized for fluid velocities around 20 μm/s. For flow velocities above 1 mm/s, no particle remains in the channel, but the template of magnetic dots is undamaged.

Magnetic beads with a diameter of 4.5 μm coated with anti-CD19 mAb (Dynabeads, Dynal) were used and self-assembled as described above. Cell suspensions of mixtures of T (Jurkat cell line) and B (Raji cell line) lymphoid cells (see SI Text) were flown in the array at a concentration of 2.10⁶ cells/mL.

Cell capture was first studied as a function of flow rate. After a prescribed capture time, the array was washed with free buffer to eliminate nonattached cells. The fraction of captured cells was quantified by visual counting. No significant change in the capture efficiency was observed for flow velocities from 50 μm/s to 700 μm/s (see SI Text and Fig. S5). This result is consistent with the optimal range of flow velocities reported earlier (15, 29). We could not investigate capture efficiency at higher flow rates, because then the magnetic array is dragged away. The experiments described below were performed at flow velocities 100 ± 10 μm/s. At this flow rate, visual observation is easy, and we did not observe any column detachment, even when the columns are saturated in cells (in this situation, columns start to detach from the posts at flows around 500 μm/s instead of 1,000 μm/s in the absence of cells, but cells remain attached to the beads). In our device, with a channel width 500 μm and a channel height 50 μm, a 100 μm/s flow corresponds to a flow rate of a few μL/ min and a throughput in the range of ten to hundred cells/s.

Two micrographs taken at the entrance of the magnetic array after the passage of pure populations of Jurkat and Raji cells, respectively, are shown in Fig. 1 D, E. Introduced Raji cells were observed as “rosettes” on the columns whereas no cell was seen in the array after passage of a population of 1,000 Jurkat cells, qualitatively demonstrating the specificity of the system (visualization in Movies S2 and S3).

Cell sorting from a mixture was also evaluated regarding purity and yield (see SI Text). Purity is defined here as: (number of positive cell captured)/ (total number of cells captured). The yield represents the total number of cells of a given type captured divided by the total incoming number of cells of that type. Table S1 summarizes the results for Raji and Jurkat cell populations, starting from initial ratios of Raji to total: 33%, 10%, and 1%, respectively. As expected it decreased by increasing the ratio of Jurkat cells (which play here the role of “contaminant” negative cells) in the sample, from typically 97 ± 1% for 33% positive cells, to a typical 85 ± 5% at one positive cell per 100. For both positive and negative sorting, yields were extremely high. Raji cells were captured with a yield of 97 ± 2% whereas Jurkat cells exited the array with a yield > 98%. Fig. 1E also shows that each trapped cell is visible individually, contrary to routine bulk experiments where cells are often entrapped in aggregates (32).

For comparison with earlier work, we also performed experiments using anti-EpCAM Dynal beads and MCF7 breast cancer cells spiked in a large excess of endothelial cells (SI Text). The capture yield was 80 ± 20%, comparable with those previously reported for the CellSearch system® (80%) (4) and the Nagrath system (> 65% in (15), and 42%–64% on clinical samples in (16)).

The model was fitted to experimental data using a unique value for r (capture efficiency) and L (number of capture sites per row) for all data. Theoretical and experimental results agree well (Fig. 2), allowing an evaluation of the coefficients L and r. The first cells entering the channel are preferentially captured on the first rows of the array. While increasing the number of entering cells, the reduction in the number of free capture sites becomes significant, and more and more cells cross first rows and are captured downstream. The saturation of the first rows of columns was just reached in our experiments, and simulations indicate that a continuing flow of the sample through the array should yield a propagating sigmoidal-like front, the amplitude of which is given by the total number of capture sites. Comparing the best-fit value L = 25 with the actual number of columns in a row, 10, one can deduce that each column is capable of capturing 2–3 cells in average, in consistency with visual observation in saturated conditions (Fig. 1E). The best-fit parameter r, 0.2 ± 0.05 for all datasets, is equal in this model to the product of the probability of collision in a row by the probability of attachment upon collision. The accurate calculation of the collision probability per row is delicate (see SI Text), but knowing the average size of lymphocytes, 10 μm, the beads size, 4.5 μm, and the beads spacing in a row (40 μm center-to-center) it should lie between 0.28 and 0.36. This yields estimated values of the probability of attachment per collision (between 0.55 and 0.8). This excellent capture efficiency is consistent with visual observations, in which the first collision most of the times leads to attachment. This efficiency, and the fact that the sample is completely depleted from positive cells long before the microchannel’s end, suggest that the system will be robust to limited changes in the array geometry, or to defects in the columns (see SI Text for further discussion).

The cells captured on the column could be kept in physiological conditions over long time periods. To that aim, complete and buffered (CO₂-independent) growth medium (Invitrogen) were continuously infused in the channel after capture. Temperature was maintained between 36 °C and 37 °C. Raji cells attached to the columns were viable as demonstrated by their ability to move and divide (Fig. 3). We could observe mitosis after 12 h of culture, consistent with the population doubling time of this cell line. The mother and daughter cells remain attached to the capture column, but they eventually displace individual beads and modify the structure of the magnetic columns. The cell on the left (Fig. 3) has extracted one bead from the column, and drives it around between different frames. This behavior is consistent with the forces involved in cell adhesion and cytoskeleton motions, in the nN range (33), since the magnetic cohesive force of our columns is rather in the upper pN range. Ephesia is thus a straightforward microfluidic method to cultivate nonadherent cell immediately after their sorting, opening the route to fundamental cell biology research directly on cancer cells from patients.

The method was validated regarding different types of lymphomas and leukemia: chronic lymphocytic leukemia (CLL) (n = number of patients = 4), mantle cell lymphoma (n = 1) and follicular lymphoma (n = 2), and two healthy volunteers (Table S2 and Fig. 4). Samples were analyzed in parallel in a blind protocol, by a combination of cytology and FC on the one hand, and by a single step capture and image analysis using the Ephesia system on the other hand, without prior knowledge of the diagnosis. Three different types of clinical samples were involved: blood (n = 7), pleural effusion (n = 1), and FNA (n = 1) from lymph nodes. Capture of B-cells in Ephesia was performed using anti-CD19 4.5 μm Dynabeads. 10 μL of sample were used for each analysis. The whole array of cells-bearing beads can be imaged in situ by epifluorescence, or immobilized by agarose, allowing for the transport and ex-situ imaging of the whole chip by confocal fluorescence. A library of images was prepared and analyzed for each sample (see SI Text). Typical “vignette” images from different clinical samples, acquired ex-situ on a A1R Nikon confocal microscope, are provided in Fig. 4. Ephesia provides the same major nucleus morphological characteristics as classical cytological staining. Normal lymphocytes and B-cell malignant cells from CLL (Fig. 4 A, B) are similar in cytological examination. This result is also observed in Ephesia, with bright field images showing a similar cell diameter, and Hoechst staining showing a nucleus with a chromatin mainly located at its periphery and absent from the center in both normal and CLL cells. Typical morphological features of nuclei observed in mantle and follicular lymphoma (nucleoli and fragmented-like nucleus respectively) are also visible by confocal microscopy (Fig. 4 C, D). Movie S4 emphasizes these nuclear characteristics by presenting the stack images of trapped DoHH2 cells (a cell line of a follicular lymphoma (see SI Text)) that exhibit large and multiple nucleolus and nucleus fragmentation. Antigen expression analysis shows a strong correlation with FC data. Healthy B-lymphocytes are CD10 negative and present a minor population of CD5 positive cells. This CD5 expression is linked to the B1 subset of B-lymphocytes (34). Malignant B-cells homogeneously express CD5 in CLL and mantle lymphoma but are CD10 negative. We observe the localization of the antigen at the membrane in these two latter conditions. As in FC data, follicular lymphoma malignant cells express CD10 but not CD5. In this case, fluorescence appears localized at the membrane and in the cytoplasm at the center of the cell. This peculiar topological distribution for CD10 might be due either to a plasma membrane invagination or to labeling of cytoplasmic CD10 after its internalization from the membrane. Because of the labeling procedures (cells are labeled first and then fixed) these two options are plausible, and will require further investigation, since this kind of information is not provided by FC.