Ultrahigh-throughput magnetic sorting of large blood volumes for epitope-agnostic isolation of circulating tumor cells

Workflow for Microfluidic Isolation of CTCs from Leukapheresis Products.

A typical 65-mL leukapheresis product derived from differential centrifugation of ∼5 L of whole blood consists of 3 to 6 billion WBCs, with 10 to 30 billion contaminating RBCs. We calculate that leukopaks from patients with cancer are likely to also have 100 to 20,000 CTCs, depending on the type and stage of malignancy. WBCs consist primarily of mononuclear cells, since neutrophils are depleted by centrifugal forces during apheresis, and their concentration within the leukopak ranges from 50 to 90 million cells per mL, >10-fold higher than the WBC concentration in whole blood. Depending on the apheresis settings, the concentration of contaminating platelets may also be 10-fold higher in a leukopak, compared with whole blood. Altogether, the very high number of WBCs and platelets, concentrated within a large volume of leukapheresis product, presents a major challenge, compared to standard 10-mL samples of peripheral blood that are currently used for microfluidic enrichment of CTCs.

Fig. 1A illustrates the workflow that we established for depletion of hematopoietic cells from leukapheresis products. To process these complex samples, we first labeled WBCs with a mixture of biotinylated antibodies targeting the pan-leukocyte cell surface antigens CD45, CD16, CD3, CD45RA, and CD66b. The selection of antibodies against CD45, CD66b, and CD16 was based on our previous CTC isolation studies (19, 20), while CD3 and CD45RA antibodies were added to further deplete WBCs, based on mass cytometric profiling of contaminating cells in the product (SI Appendix, Fig. S2). Using a microfluidic device termed nonequilibrium inertial separation array (29), we achieved the removal of RBCs and platelets based on their small physical size, compared with nucleated cells (SI Appendix, Fig. S3). We then tagged the antibody-bound WBCs with 1-μm streptavidin-coated superparamagnetic beads and used the high-throughput magnetic sorter chip to deplete WBCs and recover unlabeled CTCs. The whole CTC isolation process is completed within 3 h.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

High-throughput rare CTC separation from full (65 mL) leukapheresis samples. (A) Schematics illustrating the microfluidic approach for untouched CTC isolation from leukapheresis products. A leukapheresis product typically contains 3 to 6 billion nucleated cells, from ∼5 L of blood volume. We first remove RBCs and platelets from leukapheresis products using size-based inertial separation. It is followed by immunomagnetic removal of WBCs, allowing us to recover untouched CTCs without relying on antigen markers. (B) A schematic diagram illustrating the permeability-enhanced magnetic sorter which can deplete ∼3 billion WBCs per h from concentrated leukapheresis products. This chip uses two adjoining channels, one on each side of the sorting channels, which are compactly packed with soft magnetic iron particles. These high-permeability channels enhance the magnetic field gradient 35-fold, allowing us to operate at an ultrahigh throughput while deflecting every magnetically labeled cell.

Depletion of Erythrocytes and Platelets.

The inertial separation array chip is based on our previous work (29) and exploits size-dependent inertial wall lift forces, proportional to the sixth power of the cell diameter, for efficient isolation of nucleated cells into a clean buffer. This device is designed as an array of rectangular islands (200 µm long × 50 µm wide × 52 µm tall) where cells are placed in close proximity to the walls (SI Appendix, Fig. S3A). Since wall-induced lift force is highly dependent on cell size, WBCs and CTCs undergo a greater deflection away from the wall than RBCs and platelets (SI Appendix, Fig. S3A). At the end of the rectangular island, a portion of the flow in the near-wall region is siphoned (3.6% of each main channel flow) through each of numerous narrow vertical gaps between one island and the next, thus removing the smaller cells (RBCs and platelets). This siphoning is repeated over the whole array to ensure a high yield of nucleated cells in the clean buffer and the effective removal of RBCs and platelets (SI Appendix, Fig. S3 A and B).

In contrast to deterministic lateral displacement (30), where cells cross streamlines when nudged by a post, the inertial separation array described above amplifies small cell size differences by allowing inertial lift forces to push cells away from the channel walls. In addition to its intrinsically faster blood processing, clogging is eliminated due to subsecond residence times and cell-free zones near channel walls. For this purpose, we used 32 parallel devices contained in two microfluidic plastic disks, created by using injection molding of cyclic olefin copolymer (SI Appendix, Fig. S3C). The total leukapheresis sample flow rate was 73 mL/h, while buffer was injected at 348 mL/h, enabling debulking of a whole 65-mL leukapheresis product within an hour. We note that bulk methods, such as density-gradient centrifugation and RBC lysis, are also widely used for removing RBCs from leukopaks. However, given their batch processing and lower specificity, these methods suffer from 27% and 11% loss of CTCs, respectively, compared with inertial separation array devices (31). After debulking, we labeled WBCs with magnetic beads and used the magnetic sorter for depleting WBCs.

Design of an Ultrahigh-Throughput Magnetic Sorter for Depletion of WBCs.

Isolating hundreds to thousands of CTCs from a leukapheresis product requires the ability to deplete 3 to 6 billion WBCs, 50- to 100-fold more cells than devices commonly used for handling 10 mL of whole blood. We addressed this challenge by creating a permeability-enhanced magnetic sorter which utilizes the high magnetic permeability material (iron) in the vicinity of the sorting channels for enhancing magnetic gradients and throughput. This device processes 3 billion WBCs in an hour at a total (sample and buffer) throughput of 168 mL/h, offering ∼30-fold-higher cell processing capability per device than existing magnetic sorters (19, 20, 22, 32⇓⇓⇓–36).

To incorporate the high-permeability material, we created two adjoining 1,600-µm-wide channels, one on each side of the sorting channel, and packed these channels with soft magnetic iron particles of ∼40 µm in diameter (Fig. 1B). This was achieved by dispersing particles in 50% ethanol and then passing them through the empty high-permeability channels where filters at the end of the channels allow fluid to escape while retaining particles, compactly packing the channels (SI Appendix, Fig. S4). We used diamond-shaped pillar structures with 25-μm pitch as filters for trapping iron particles. These iron-filled channels can be positioned as close as 25 μm from the sorting channels without risking delamination.

A detailed labeled diagram of the magnetic sorter is shown in Fig. 2. It consists of two stages, which deflect and remove every WBC labeled with a 1-µm magnetic bead (Fig. 1B). At first, the magnetically labeled cell suspension flows into stage 1 through two sets of microfluidic inlet filters (40-μm aperture) to remove large debris or aggregates (Fig. 2 A, 1). Any slender debris smaller in diameter than the aperture of filter 1 are captured based on their length in the tortuous channel geometry of filter 2. After the filters, the cell suspension flows into the stage-1 sorting channel at 48 mL/h via two asymmetric serpentine channels, which inertially focus cells in a single file (Fig. 2 A, 2). These serpentine channels utilize a balance between shear-induced lift force and Dean flow-based drag force to focus cells near the center of the serpentine channel (37). At the core of the sorting channel, a buffer flow at 120 mL/h is provided to keep the inertially focused cells close to the channel wall, where magnetic field gradients are at maximum (Fig. 2 A, 3). As WBCs move through the deflection channel, they experience a magnetic force and are deflected toward the center of the channel into the stage-1 waste port (Fig. 2 A, 4). This provides a clog-free design, where WBCs are deflected into the core of the flow away from the walls and high-gradient regions.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

(A) A comprehensive diagram illustrating the microfluidic design of the magnetic sorter. In the first stage, cells flow through microfilters, followed by focusing into single-file streams, which are pinched close to the sidewalls of the sorting channel by buffer flow and WBCs are deflected to the waste port. The rest of the sample flows into stage 2 through a pair of inertial focusing-based microfluidic concentrators. All of the free magnetic beads and every labeled cell are magnetically deflected to the waste port in the stage 2 while the concentrated product is collected. Insets 1 through 7 show fluorescent streak images of WBCs at various positions in the magnetic sorter, from inlet to outlet. Inset 8 shows a fluorescent streak image of isolated CTCs at the product port. (B) An image of the microfabricated magnetic sorter device.

After stage 1, cells flow into stage 2 via another microfluidic filter (40-μm aperture) and two inertial-focusing-based cell concentrators. The cell concentrator works by continuously creating a cell-free region and repeated siphoning using passive flow-controlled resistance (Fig. 2 A, 5) (38). In essence, cells pass through asymmetric-inertial focusing units, which create a cell-free region due to inertial lift forces and Dean-flow-induced drag force (SI Appendix, Fig. S5). This cell-free region is siphoned away from curved focusing units by a siphoning channel, while cells pass through another focusing unit that creates a new cell-free region, which is siphoned again (SI Appendix, Fig. S5). This process is repeated over 140 units until the end of the channel achieving ∼20-fold concentration of cells (Fig. 2 A, 6). This unit serves two key purposes. First, by concentrating the cells, they are positioned close to the walls in the stage-2 sorting channel where magnetic gradients are maximal; second, after concentration, the excess cell-free fluid is removed through the stage-2 waste port (71%), thus reducing the net flow input into the stage-2 sorting channel (250 μL/min). This provides a greater residence time for cells. Six-feeder channels supply the concentrated cell suspension to the stage-2 sorting channel where any loss in cell focusing is corrected by six inertial focusing units and cells are placed in a single file close to the channel sidewalls (Fig. 2 A, 7). In the stage-2 sorting channel, every cell labeled with a bead is deflected to the waste port, while undeflected CTCs (Fig. 2 A, 8) are collected within an 11-fold reduced volume. Fig. 2B shows an image of the microfabricated polydimethylsiloxane (PDMS) magnetic sorter chip with filled high-permeability channels. The total flow rate (buffer and sample) into the chip is 168 mL/h, demonstrating a remarkable volumetric throughput.

Permeability-Enhanced Magnetophoresis.

The ultrahigh-throughput functionality of the magnetic sorter is a direct feature of the unique permeability-enhanced magnetic setup (Fig. 3A). We used a quadrupolar arrangement of rectangular (5 × 5 × 40 mm) neodymium–iron–boron (N52 grade) magnets, similar to our previously described first-generation magnetic sorter integrated into the CTC-iChip (19, 20, 22). However, the polarity of magnets is modified to the y direction to ensure that the magnetic force on the cells is directed toward the center of the sorting channel in the presence of the adjoining iron-filled channels (Fig. 3 A and B).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

(A, Top and Middle) The arrangement of permanent magnets and high-permeability channels in the magnetic sorter. We use N52 grade neodymium-iron-boron (NdFeB) rectangular magnets in a quadrupolar arrangement. (A, Bottom) The contour plot of the magnetic field intensity. The high-permeability channels and permalloy strips between magnets amplify the magnetic field in the sorting area. (B) Arrangement of magnets and corresponding contour plot of the magnetic field in the first-generation magnetic sorter used in the CTC-ichip (22). This sorter does not use high-permeability channels. (C) The magnetic field gradient in the permeability-enhanced magnetic sorter and the first-generation magnetic sorter (22). The soft magnetic iron-filled channels act as on-chip magnetic microlenses and increase the magnetic field gradient 35-fold. (D) Experimental validation of field enhancement. Using particle-tracking velocimetry, we measured significantly higher lateral deflection velocity of 2.8-μm magnetic beads with iron channels. For directly observing the sorting channel, we used a configuration with a 4-mm gap between magnets as depicted in SI Appendix, Fig. S6. (E) Vector plot of the gradient of the magnitude of magnetic field in the sorting channel. The vectors are positioned at their respective midpoints. (F) Deflection of 1-μm magnetic beads and cells in stage 1. Cells with >10 beads and most of the free beads are deflected to the waste port. (G) Lateral deflection of beads and cells in stage 2. All of the free beads and every labeled cell are deflected to the waste port.

Assuming that the superparamagnetic particles used to label cells are saturated, the lateral magnetic force on a labeled cell is directly proportional to the number of particles attached to a cell, and ∂|B→|/∂y, the gradient of the norm of the magnetic field B→ in the y direction. To achieve the deflection of a cell labeled with a single bead, increasing the magnetic field gradient is essential for improving the magnetic force and, consequently, the throughput. We therefore incorporated high-permeability channels, filled with soft magnetic iron particles, and also included a 100-μm-thick permalloy strip between the magnets. Under the action of the macro magnetic field from the rectangular magnets, these ferromagnetic microchannels are magnetized and produce a localized magnetic field that decays rapidly, resulting in a high magnetic field gradient in the sorting channel (Fig. 3 A and B). In addition to the localized field gradient from iron-filled channels, a long-range field gradient from rectangular magnets is also present in the sorting channel but it is significantly (35-fold) smaller. In essence, the high-permeability channels act as on-chip magnetic microlenses and significantly increase the magnetic field gradient. This creates a field gradient as high as 15,400 T/m in deflection channels as compared to the 440 T/m previously achieved by the CTC-iChip magnetic arrangement (20, 22), providing a 35-fold enhancement in magnetic force (Fig. 3C).

The increase in magnetic field was experimentally verified by measuring lateral deflection velocity of 2.8-μm superparamagnetic tracer particles with a high-speed camera (Fig. 3D). The lateral velocity of the magnetic particles is directly proportional to the field gradient. Given that the magnetic arrangement in the sorter does not allow visualization of the deflection channels, we created a new setup for experimental validation. A 4-mm viewing gap was created between the magnets for direct high-speed imaging of particle trajectories in 800-μm-wide channels with and without adjoining iron channels (SI Appendix, Fig. S6). Using particle-tracking velocimetry, we measured up to 54-fold higher lateral velocity with iron channels, demonstrating that the magnetic field gradients are significantly increased in the presence of the high-permeability channels (Fig. 3D).

As shown in Fig. 3 C and E, the magnetic gradient is maximal near the sidewalls of the sorting channels, and it decays progressively toward the center of the channels. Therefore, the microfluidic circuit is designed to inertially arrange and sort cells in a small near-wall region in both the stages. In stage 1, the channel width is 1,500 μm, and the cutoff for deflection is set at 240 μm from the sidewalls; in stage 2 the channel width is 1,800 μm, and the cutoff is set at 250 μm from the sidewall (SI Appendix, Fig. S7). The height of the channels was kept constant at 60 μm. As evident in the vector plot, a noticeable x component of the gradient is also present in the sidewall region (Fig. 3E), but it decays to ∼250 T/m within 100 μm from the sidewall and becomes negligible (<50 T/m) beyond 150 μm from the side wall (SI Appendix, Fig. S8A). In comparison, the y component of the gradient is more than an order of magnitude stronger in the bulk of the sorting channel. This results in a magnetic force which is predominantly in the lateral y direction in the sorting channel (Fig. 3E and SI Appendix, Fig. S8B). Thus, a cell undergoing magnetophoretic sorting mainly experiences magnetic force toward the center of the channel in the y direction, wall lift force away from the top and bottom walls, and a fluidic viscous drag force (SI Appendix, Fig. S8C). The wall lift force prevents cells from touching the top and bottom walls as they migrate toward the center of the channel.

Using a laminar velocity profile for a low-aspect-ratio rectangular channel and the magnetic force expression, we calculated the deflection of cells in both the stages. In stage 1, cells having greater than 10 beads along with most of the unbound magnetic beads are deflected to the waste port at a total flow rate of 168 mL/h (Fig. 3F). In stage 2, cells are focused 100 μm away from the walls at a flow rate of 250 μL/min. In this stage, free beads and all cells with at least one bead attached to them are deflected (Fig. 3G). To execute this high-gradient design, we brought the magnets as close as possible to the chip: The PDMS layer containing the device features was cast as a thin 1-mm layer (same thickness as the bottom glass slide), and magnets were held in a specially designed manifold (SI Appendix, Fig. S9).

We also used sequential magnetic labeling to decrease the required number of magnetic beads per cell by ∼91%, thereby lowering the cost burden of the system. This was accomplished by initially labeling WBCs with 10 beads per cell, instead of the previously used concentration of 125 beads per cell (20), thereby depleting >99.5% of WBCs within ∼40 min by using the magnetic sorter (SI Appendix, Fig. S10). The remaining 0.5% contaminating WBCs in the product were then relabeled with 125 beads per cell for 10 min and reprocessed through the magnetic sorter in less than 5 min for further removal. We performed the two-step sequential magnetic sorting for all our samples.

Isolation of CTCs.

We first tested the performance of the ^LPCTC-iChip by recovering ex vivo-cultured CTCs that had been spiked into leukapheresis-mimic samples (n = 5), which are produced by centrifuging approximately a unit of healthy donor blood (400 to 500 mL whole blood) followed by the extraction of the leukocyte-enriched layer. These samples on average contain 1.42 billion WBCs, 56.5 billion RBCs, and 16.9 billion platelets (Fig. 4A). The mean volume of the samples was 24.5 mL and the WBC concentration varied from 39.5 to 82.6 million cells per mL at an average concentration of 58.4 million cells per mL, which is ∼10-fold higher than the whole blood. The leukapheresis-mimic samples represented approximately a third of the clinical leukapheresis product in volume and the total number of nucleated cells while the concentration of WBCs was similar. In these samples, we spiked 1,000 green fluorescent protein (GFP)-expressing CTCs that had been cultured from viable CTCs enriched from a blood sample of a patient with hormone receptor-positive breast cancer (MGH-BRx-142) (39). Using this approach, we recovered 89.2 ± 5.7% of spiked CTCs and removed 99.96% WBCs (3.35 ± 0.17 log₁₀ depletion), 99.998% RBCs (4.88 ± 0.37 log₁₀ depletion), and 99.998% platelets (4.92 ± 0.15 log₁₀ depletion). We also separately quantified the performance of the inertial separation array module. It removed 99.95% RBCs (3.39 ± 0.28 log₁₀ depletion) and 99.98% platelets (3.83 ± 0.19 log₁₀ depletion). The additional depletion of RBCs and platelets was achieved in stage 2 of the magnetic sorter, since RBCs and platelets are not inertially focused due to their smaller size and removed in the waste channels.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

(A) Processed leukopak sample volumes, depletion data and cell numbers in full leukapheresis samples (shown by red symbols, n = 3), mimic samples (shown by gray symbols, n = 5), and in the isolated product. On average, we processed 64.2 ± 4.6 mL leukapheresis samples. We achieved 5.11, 3.55, and 5.08 log₁₀ depletion of RBCs, WBCs, and platelets, respectively. (B) CTC isolation yield. In leukapheresis samples, we recovered 86.1% spiked MGH-BRx-142 cells (n = 3), while mimic samples had a slightly higher yield of 89.2% cells (n = 5). (C) Immunofluorescence images of isolated spiked MGH-BRx-142 CTCs stained with EpCAM (green) and DAPI (blue). (Scale bar, 10 µm.) (D) Relative in vitro growth of isolated CTCs in product vs. control as measured by the amount of ATP present in the cells (n = 3). The inset panels show images of the cultured MGH-BRx-142 cells. (Scale bar, 100 µm.) (E) Expression of the breast lineage markers from spiked MGH-BRx-142 cells after CTC enrichment as measured by ddPCR analysis. Negative controls are healthy donor samples in which no CTCs were spiked.

Extending from the mimic samples, we processed three clinical leukapheresis samples containing 5.0 ± 1.0 billion WBCs, 92.6 ± 72.5 billion platelets, and 75.4 ± 66.5 billion RBCs, as shown by red data points in Fig. 4A. The WBCs in these samples are representative of a liter of whole blood. The average volume of these full samples was 64.2 ± 4.6 mL (Fig. 4A), into which 5,000 MGH-BRx-142 CTCs were spiked. This number of CTCs per leukopak is consistent with previous studies which processed 5% of clinical leukapheresis samples, calculating that, if technically feasible, screening of the entire leukopak would have produced >10,000 CTCs (24).

Depending on the apheresis operating conditions, the WBC concentration in these leukapheresis samples varied from 61 to 90 million cells per mL, while platelet concentrations ranged between 70 and 3,043 million platelets per mL. Even though platelet and WBC concentrations were more than 10-fold higher than whole blood, two parallel ^LPCTC-iChips effectively removed 99.97% WBCs (3.55 ± 0.26 log₁₀ purification) while recovering 4,305 CTCs out of the 5,000 spiked CTCs (86.1 ± 0.6% yield) at an average purity of 0.3% (Fig. 4B). The device also depleted >99.999% RBCs (5.11 ± 0.35 log₁₀ purification) and >99.999% platelets (5.08 ± 0.41 log₁₀ purification), demonstrating a highly efficient microfluidic removal of contaminating RBCs and platelets (Fig. 4A). The isolation process preserved the CTC morphology as demonstrated by a gallery of EpCAM- and DAPI-stained CTCs in Fig. 4C.

Postisolation, we tested a fraction of the product (40%) for in vitro culture to assess whether the microfluidic devices damage the proliferative properties of isolated CTCs. Enriched CTCs from the product proliferated comparably with control samples (Fig. 4D). We have recently described an RNA-based droplet digital PCR (ddPCR) assay for absolute quantitation of tissue-lineage-specific transcripts from CTCs in the background of normal blood cells (8, 9). The ddPCR assay confirmed the isolation of cells with intact RNA, suitable for molecular analyses (Fig. 4E). To mimic clinical situations in which CTC numbers are extremely low, such as early cancer detection applications, we tested the magnetic sorter by spiking only five GFP-labeled cells from different cell lines including the breast CTC line MGH-BRx-142, the commonly studied breast cancer line MDA-MB-231, and the prostate cancer line LNCaP, using the leukopak mimic assay (SI Appendix, Fig. S11). For precise quantitation of the number of cancer cells spiked into the leukopak mimics, cells were individually picked using single-cell micromanipulation and we conducted five independent CTC spiking experiments for each cell line. Using MGH-BRx-142, we recovered four of five spiked cells in two experiments and five of five in another three experiments. Similarly, for MDA-MB-231, we detected four of five cells in three spiking experiments and five of five cells in two other experiments. Using prostate LNCaP cells, we successfully recovered five of five spiked cells in all of the five experiments.