Tunable cell separation using a thermo-responsive deterministic lateral displacement device

Abstract

Tunability in isolating target cells of varying sizes from complex heterogeneous samples is essential for biomedical research and diagnostics. However, conventional deterministic lateral displacement (DLD) systems lack flexibility due to their fixed critical diameters (D_c). Here, we present a thermo-responsive DLD micropillar array that enables tunable cell separation by dynamically modulating D_c through temperature control. Our device integrates poly(N-isopropylacrylamide) (PNIPAM) hydrogel micropillars within a PDMS-silicon microfluidic chip mounted on a Peltier element, enabling precise D_c adjustments from 0.8 to 29.0 μm within a temperature range of 20–40 °C. Transient and steady-state simulations confirmed that the silicon substrate enhances thermal performance, ensuring rapid and uniform temperature regulation. Using blood samples containing human breast adenocarcinoma cells (MCF-7), we demonstrated three separation modes: (i) major separation at 25 °C, isolating MCF-7 cells (average size: 17.6 μm) with 100% purity; (ii) selective separation at 26 °C, targeting larger MCF-7 subpopulations (average size: 18.7 μm); (iii) minimal separation at 37 °C. All processes preserved cell viability. These findings highlight the potential of our thermo-responsive DLD platform for precise, temperature-controlled cell selection, offering broad applications in biomedical research and diagnostics.

---

Introduction

Cell separation plays a pivotal role in biomedical research and clinical therapy, including cancer research, regenerative medicine, and drug discovery.¹ For instance, isolating circulating tumor cells (CTCs) from blood is clinically essential for monitoring cancer metastasis.² Similarly, the diagnosis and treatment of HIV require the effective separation of human T-lymphocytes (CD4+) from whole blood.³ Additionally, the isolation of hematopoietic stem cells (HSCs) is vital for regenerative medicine therapies, which aim to repair or regenerate damaged tissues.⁴ As the demand for understanding diverse cell types and their roles in biological systems grows, the ability to isolate specific populations or even single cells has become indispensable for both research and clinical applications.

Recent advances have enabled the development of various microfluidic techniques for cell separation,^5–7 among which deterministic lateral displacement (DLD) has emerged as a powerful method offering high-resolution, label-free, and continuous size-based separation.^8–10 DLD has been successfully applied to a wide range of targets, including CTCs,¹¹ blood cells,¹² fungal spores,¹³ droplets,^14,15 DNA,¹⁶ and exosomes.¹⁷ In DLD devices, periodic arrays of micropillars induce particle separation based on size relative to a critical diameter (D_c), which is defined by the array geometry. Particles larger than D_c are displaced laterally (bump mode), while smaller particles follow the global flow path (zigzag mode); those near D_c may exhibit mixed migration due to asymmetric flow fields.^18,19 However, conventional DLD devices are constrained by a fixed D_c, limiting their utility for complex biological samples with broad or variable size distributions. In addition, surface fouling and non-specific adsorption can alter the effective D_c during operation, degrading performance. To address these challenges, DLD systems with dynamically tunable D_c are needed for flexible optimization and sustained separation efficiency.

Recent studies have proposed various active strategies for dynamically modulating the D_c in DLD systems, broadly categorized into two approaches: applying additional forces to particles or modifying device geometry and flow patterns. Force-based methods employ dielectrophoretic (DEP) forces,^20–24 viscoelastic lift forces,^25,26 electrostatic interactions,^27,28 or acoustic waves²⁹ to manipulate particle trajectories and adjust D_c. For example, DEP forces have been used to create virtual pillar arrays or steer particles laterally by tuning voltage or flow rate,^20–24 while viscoelastic fluids and electrostatic effects allow D_c control through changes in flow rates or buffer composition.^25–28 Surface acoustic waves have also been utilized to dynamically redirect particles within the array.²⁹ Alternatively, geometry-based strategies modulate D_c by physically altering the device, such as stretching elastic poly(dimethylsiloxane) (PDMS) structures to change array spacing,³⁰ or adjusting internal flow paths through controlled flow rate ratios or channel slopes to steer streamlines.³¹

In our previous work, we demonstrated a thermo-responsive DLD array on a glass substrate using hydrogel micropillars composed of poly(N-isopropylacrylamide) (PNIPAM) within a PDMS microchannel.³² These pillars exhibit temperature-dependent volume changes due to PNIPAM's phase transition near its lower critical solution temperature (LCST), approximately 32 °C in pure water.³³ Below the LCST, PNIPAM becomes hydrophilic and swells, narrowing the inter-pillar gap (d) and lowering D_c. Above the LCST, it becomes hydrophobic and shrinks, increasing d and D_c (Fig. 1). The favorable biocompatibility and low LCST of PNIPAM make it well-suited for biomedical applications.^34,35 Unlike other active DLD methods, the PNIPAM-based approach requires no complex external field-generating equipment, offers a simpler fabrication process, and facilitates size-based separation through direct temperature-driven modulation of pillar dimensions.

Fig. 1 Tunable cell separation using a thermo-responsive micropillar array. (a) Schematic illustration of a PDMS-silicon chip mounted on a Peltier element, incorporating deterministic lateral displacement (DLD) micropillars composed of poly(N-isopropylacrylamide) (PNIPAM). (b) Magnified view of the DLD region entrance, showing the micropillar gaps and sidewalls numbered 1–16. (c) Key geometric parameters of the DLD array, with Δλ/λ = 0.125. (d) Temperature-dependent cell separation modes: major separation (left), selective separation (middle), and minimal separation (right).

While this mechanism offers the potential for flexible, size-based separation within a single device, our earlier platform had several key limitations. First, it featured only a single outlet, allowing demonstration of thermo-responsive migration mode switching but not actual separation and collection of target and non-target particles. Second, the low thermal conductivity of the glass substrate may have delayed temperature equilibration, potentially compromising the accuracy of D_c control, though this effect was not assessed. Third, the PDMS–glass interface relied on self-adhesion, restricting the system to negative pressure-driven flow, and limiting operational flexibility. Finally, the PNIPAM photoresist then available supported only ∼10 μm pillar heights, constraining throughput and precluding applications involving biological cells.

In this study, we present an improved thermo-responsive DLD device incorporating PNIPAM micropillars on a silicon substrate. This is the first such device to achieve size-based separation with distinct collection of target and non-target particles from separate outlets using PNIPAM micropillars. The use of silicon, chosen for its superior thermal conductivity, enabled more precise D_c control, as validated through simulations and experiments. Plasma bonding of the PDMS channel to the silicon substrate allowed stable, positive-pressure operation across a range of flow rates. Furthermore, by employing a new PNIPAM-based photoresist with higher polymer concentration, we fabricated micropillars up to 30 μm in height (dry state), suitable for processing diverse biological particles. Using this platform, we processed blood samples spiked with human breast adenocarcinoma (MCF-7) cells and demonstrated temperature-tunable D_c control, achieving three distinct separation modes: major isolation of MCF-7 cells, selective enrichment of larger MCF-7 subpopulations, and a minimal-separation mode. Post-separation viability tests confirmed the platform's compatibility with biological applications.

Experimental

Device design

We developed a thermo-responsive DLD device comprising PNIPAM-based micropillars confined within a PDMS-silicon microfluidic channel (Fig. 1a). The micropillars were positioned within the main channel (length: 15 mm, width: 2.3 mm, height: 30 μm), located between three inlets and two outlets. Temperature regulation was achieved using a Peltier element connected to a temperature controller. A sheath-focusing flow directed the sample solution containing cells into the micropillar array through gap numbered 4–7, where the cells began migrating near the inner right side of the main channel (Fig. 1b).

The estimation of D_c was based on the empirical formula proposed by Davis:³⁶

D_c = 1.4d(Δλ/λ)^0.48 (1)

where d is the gap size between micropillars, λ is the center-to-center distance between adjacent micropillars, and Δλ is the vertical shift between two adjacent micropillars. Fig. 1c depicts the design parameters of the DLD array. The fabricated DLD array was configured with an average pillar diameter (D_p) of 62 μm, d = 68 μm, and Δλ/λ = 0.125, yielding a D_c of 35 μm in dry condition. Each DLD section consisted of 8 columns (= λ/Δλ) and 15 rows of micropillars, with 15 identical sections arranged in series to ensure sufficient lateral displacement of particles larger than D_c.

Heat transfer simulation

Precise temperature control is essential for dynamically adjusting D_c in thermo-responsive DLD devices. Conventional substrates such as glass and PDMS exhibit low thermal conductivity, potentially delaying heat transfer and compromising temperature regulation. To assess the impact of substrate materials on fluid temperature control, 3D heat transfer simulations were performed utilizing Fluent 2025R1 (Ansys, Canonsburg, PA, USA). These simulations compared the thermal performance of silicon and glass substrates under various flow rates. Ansys Meshing was used to create the 3D model.

Transient and steady-state simulations were conducted to evaluate the thermal performance of the two substrate materials. A pressure-based solver was used in both cases, with laminar flow and heat transfer models enabled. Pressure–velocity coupling was handled using the SIMPLE algorithm. Spatial discretization schemes were consistent across simulations: second-order upwind for momentum and energy equations, second-order for pressure, and least squares cell-based for gradient calculations. The solution fields were initialized using hybrid initialization. For transient simulations, a first-order implicit time integration scheme was employed to resolve the unsteady behavior of the system. A fixed time step of 0.01 s was applied over 500 steps, yielding a total simulation time of 5 s. The operating temperature range was set to 20–40 °C, typical for biological applications. Assuming a temperature difference of 20 K during simulations, the thermophysical properties of the solid and liquid phases were treated as constant. Water (dynamic viscosity: 1.003 mPa s, specific heat: 4182.0 J kg⁻¹ K⁻¹, and thermal conductivity: 0.6 W m⁻¹ K⁻¹) was chosen as the working fluid. The substrate materials included soda glass (specific heat: 898.6 J kg⁻¹ K⁻¹, thermal conductivity: 1.0 W m⁻¹ K⁻¹) and silicon (specific heat: 697.3 J kg⁻¹ K⁻¹, thermal conductivity: 153.0 W m⁻¹ K⁻¹). These material properties were sourced from the Fluent database, while PDMS properties (specific heat: 1600.0 J kg⁻¹ K⁻¹, thermal conductivity: 0.16 W m⁻¹ K⁻¹) were obtained from literature.^37,38 Heat transfer mechanisms in the simulation included natural convection from the device surface, conduction at the solid–solid interfaces, and coupled conduction–convection at solid–liquid interfaces.

Device fabrication

PNIPAM micropillars were fabricated using lithographic techniques. A PNIPAM-based negative photoresist (Bioresist-I, thick film; Toyama Industrial Technology Research and Development Center, Toyama, Japan), featuring a higher polymer concentration than that used in our previous study,³² was spin-coated onto a silicon wafer (45 mm × 26 mm; thickness: 525 ± 25 μm) in two stages: 500 rpm for 10 s, followed by 700 rpm for 60 s. The wafer was soft-baked on a hotplate by gradually increasing the temperature from 20 °C to 100 °C at 2 °C min⁻¹, then held at 100 °C for 20 min. This coating and soft-baking process was repeated three times, resulting in a final photoresist thickness of approximately 30 μm. To pattern the micropillars, a laser-printed film mask (0.175 mm thick, 25 400 dpi resolution; Unno Giken, Tokyo, Japan) and a mask aligner (MA-10 B; Mikasa, Tokyo, Japan) were used to project the 60 μm-diameter design onto the photoresist-coated silicon wafer. Post-exposure baking was performed at 120 °C for 20 min, followed by natural cooling. Development was carried out using room-temperature pure water to remove unexposed regions, with each development cycle lasting 10 min and repeated twice. The wafer was then rinsed in 70 °C pure water for 20 min. A final hard bake was conducted on a hotplate at 80 °C for 30 min, completing the fabrication of the PNIPAM micropillars.

The PDMS layer was fabricated using standard soft lithography techniques. A 4-inch silicon wafer (thickness: 525 ± 25 μm) was coated with a negative photoresist (SU-8 3025; Kayaku Advanced Materials, MA, USA) spun at 2800 rpm for 30 s to achieve an approximate thickness of 30 μm. The wafer underwent standard photolithography to create the SU-8 mold. A mixture of PDMS prepolymer and curing agent (10 : 1 weight ratio; SILPOT 184; Dow, MI, USA) was poured over the SU-8 mold and baked at 80 °C for 2 hours to form the PDMS layer.

The fabricated PDMS channel was aligned with the silicon wafer containing the PNIPAM micropillars under an optical microscope and bonded via plasma treatment (PIB-20; Vacuum Device, Ibaraki, Japan). To enhance bond strength, the assembled device was placed on a hotplate at 80 °C overnight.

Determination of LCST

To evaluate the LCST values of PNIPAM under different solution conditions, rectangular PNIPAM patterns (5 × 5 mm, 30 μm thick) were fabricated on silicon substrates using the same procedure described above. The patterns were immersed in pure water, phosphate-buffered saline (PBS, FUJIFILM Wako Pure Chemical), or high-glucose Dulbecco's modified Eagle's medium (D-MEM; 043-30085, FUJIFILM Wako Pure Chemical, Osaka, Japan), and gradually heated from 20 °C to 40 °C using a Peltier element. Images were captured at 1 °C intervals, and the temperature at which visible turbidity first appeared was recorded as the approximate LCST.

Sample preparation

MCF-7 cells (RCB1904, RIKEN BRC Cell Bank, Ibaraki, Japan) were used in this study. The cells were cultured in D-MEM supplemented with 10% fetal bovine serum (FBS; 35-079-CV, CORNING Life Sciences, NC, USA) and 1% penicillin/streptomycin (ICN Biomedical, CA, USA). Cultures were maintained in a humidified incubator (E-22; AS ONE, Osaka, Japan) at 37 °C with 5% CO₂. Upon reaching confluency, the cells were detached using 0.25% trypsin–EDTA solution (209-16941, FUJIFILM Wako Pure Chemical), centrifuged, and resuspended in PBS. To distinguish MCF-7 cells from background blood cells, calcein-AM (C396, DOJINDO LABORATORIES, Kumamoto, Japan) was used to fluorescently stain the MCF-7 cells prior to separation.

A commercially available blood control sample (Liquichek Hematology-16T Control; BIO-RAD, CA, USA), containing human red blood cells (RBCs) and white blood cells (WBCs), was used to simulate the background blood environment. MCF-7 cells—consisting primarily of single cells with a mean diameter of 17.0 ± 2.4 μm, along with a small fraction (8.6%) of cell clusters averaging 28.7 ± 2.9 μm in diameter (Fig. S1)—were mixed with the blood sample and diluted with PBS to achieve a hematocrit (HCT) level of 5%. The final sample solutions were prepared with a cell concentration of 1.3 × 10⁶ cells per mL.

Experimental setup

A 1 mL disposable syringe (SS-01T, Terumo, Tokyo, Japan) was used to inject the sample solution, while two 10 mL disposable syringes (SS-10ESZ; Terumo) supplied PBS as the sheath fluid. Outlet collection was performed using 1 mL centrifuge tubes connected via silicone tubing. Fluid infusion into the microchannel was driven by three syringe pumps (Legato 180; KD Scientific, Holliston, MA, USA). Cell migration trajectories within the microfluidic device were observed and recorded using an upright fluorescence microscope (BX51; Evident, Tokyo, Japan) equipped with a color camera (SC2003; SWIFTCAM, Hong Kong, China). The fabricated microfluidic device was mounted on a Peltier element (CHP-44HS; Sensor Controls, Kanagawa, Japan) connected to a temperature controller (FC-3510K, Sensor Controls), enabling precise temperature regulation within the microchannel during operation.

Cell viability assay

The viability of cells after microfluidic sorting was assessed through a double-staining method with calcein-AM and propidium iodide (PI; C378, DOJINDO LABORATORIES). Calcein-AM fluoresces green to indicate live cells, while PI fluoresces red to identify dead cells. To prepare the staining solution, 5 μL of calcein-AM solution and 5 μL of PI solution were mixed with 5 mL of PBS. Cell suspensions were incubated with 5 mL of this staining solution at 37 °C for 30 min. After incubation, the cells were centrifuged, and the supernatant was carefully removed. The cells were then washed with PBS to eliminate any residual staining solution. Cells processed through the microfluidic device at 25 °C were designated as the “microfluidics group”, while stained cells that were not processed through the device and incubated at 37 °C served as the “control group”. Cell viability was evaluated under a fluorescence microscope by calculating the ratio of live cells (green fluorescence) to the total number of cells (live + dead; green + red fluorescence).

Results and discussion

Heat transfer comparison between silicon and glass substrates

To evaluate the thermal performance of silicon- and glass-based devices, 3D heat transfer simulations were conducted. Results from transient-state simulations over a 5-second interval (see Fig. S2–S5 in the SI) confirmed the superior thermal responsiveness of the silicon-based device compared to its glass counterpart.

In practical applications, particle separation is typically performed under steady-state conditions. Therefore, the steady-state thermal performance of the two substrate materials was further assessed across a range of flow rates. Total flow rates were scaled by factors of 1, 2, 4, 8, and 16, corresponding to 0.4, 0.8, 1.6, 3.2, and 6.4 mL h⁻¹, respectively. As shown in Fig. 2, the silicon substrate consistently maintained the fluid temperature at 40.0 °C even at the highest flow rate. In contrast, the glass substrate exhibited a gradual decline in heating performance, with fluid temperatures decreasing from 39.8 °C, to 38.7 °C as the flow rate increased.

Fig. 2 Steady-state simulation results of fluid temperature variation at the section 1 outlet during heating across a range of flow rates (0.4, 0.8, 1.6, 3.2, and 6.4 mL h⁻¹).

These results highlight the limitations of glass substrates in achieving precise thermal control under high-throughput conditions. In contrast, the enhanced thermal conductivity of silicon enables both precise and responsive temperature regulation, making it a more suitable choice for applications requiring accurate thermal modulation. The slower thermal response of glass may restrict its use in systems where high flow rates or rapid thermal adjustments are critical.

Characterization of unconfined PNIPAM micropillars

A periodic array of PNIPAM micropillars was fabricated on a silicon substrate as shown in Fig. S6. The micropillars had a diameter (D_p) of 62 μm in the dry state and a height of approximately 30 μm, as measured using a 3D profilometer (Dektak 150; Veeco, NY, USA). Additionally, micropillars with other diameters (40, 80, and 100 μm) and similar heights (∼30 μm) were fabricated using the same process.

To investigate the three-dimensional profiles of the PNIPAM micropillars in both dried and hydrated states, we conducted side-view observations of unconfined pillars using optical microscopy (Fig. S7). Unconfined pillars with a design diameter of 80 μm (in the dry state) were hydrated at room temperature (20 and 26 °C). The observations revealed substantial swelling in both the horizontal (parallel to the substrate) and vertical (along the pillar height) directions, including at the base region where the pillars are anchored to the silicon substrate. Notably, the footprint (or “foot”) area also expanded upon hydration, indicating that the base does not remain fixed in size.

Quantitative analysis showed that vertical swelling was more pronounced than horizontal swelling: the pillar height increased by approximately 2.6-fold at 20 °C and 1.8-fold at 26 °C, while the lateral dimensions of the head and foot regions increased by 1.6–1.7-fold at 20 °C and 1.4-fold at 26 °C. The relatively smaller horizontal expansion may be attributed to mechanical constraints or partial adhesion between the PNIPAM and the silicon substrate.

Another notable observation from the side-view images is the non-uniform diameter along the pillar height. In both the dried and hydrated states, the pillar diameter gradually increases from bottom to top. If this tapered profile is preserved in the confined device, it may result in a slight underestimation of the D_c, since D_c was calculated based on top-view measurements.

Characterization of confined PNIPAM micropillars

To investigate the temperature-dependent geometric characteristics of the micropillars confined within a PDMS channel, the devices were filled with PBS and subjected to controlled temperatures ranging from 20 °C to 40 °C using a Peltier element. Thermal equilibrium was reached within 3 minutes at each set temperature. The D_p values were measured using optical microscopy.

To achieve a wide range of dynamically tunable D_c, micropillars with high swelling ratios across the operating temperature range are essential. To identify an optimal pillar size for use in separation experiments, we evaluated the temperature-dependent swelling behavior of confined PNIPAM pillars with dry-state diameters of 40, 60, 80, and 100 μm. Fig. 3a shows representative images of the pillars at 20 °C and 40 °C, highlighting their thermo-responsive shrinkage with increasing temperature. The corresponding swelling ratios—defined as the hydrated diameter divided by the dry-state diameter—are shown in Fig. 3b. The 40 μm and 60 μm pillars exhibited higher swelling ratios compared to the 80 μm and 100 μm pillars, while the 60 μm and 80 μm pillars showed greater absolute dimensional changes. Among these, the 60 μm pillars offered the best balance between the geometric tunability and efficient device footprint utilization; thus, this diameter was selected for use in the tunable DLD array.

Fig. 3 Thermo-responsive behavior of PNIPAM micropillars with different design diameters. (a) Optical microscopy images of PNIPAM pillars with diameters of 40, 60, 80, and 100 μm. Scale bar: 50 μm. (b) Swelling ratio of different PNIPAM pillars from 20 °C to 40 °C.

Fig. 4a shows the temperature-dependent size variation of the pillars with a dry-state diameter of 60 μm. At 20 °C, the D_p was 128.4 μm, with a gap (d) of 1.6 μm, while the pitch between the pillars (λ) remained constant. As the temperature increased from 20 °C to 40 °C, D_p decreased, and d increased. A significant size change occurred between 22 °C and 29 °C due to the phase transition of PNIPAM. Beyond 35 °C, D_p stabilized at 73.8 μm despite further increases in temperature. The percentage changes in D_p across different temperature ranges during heating were 7% (20–22 °C), 14% (22–25 °C), 25% (25–28 °C), 4% (28–35 °C), and 1% (35–40 °C).

Fig. 4 Characterization of PNIPAM micropillars. (a) Top-view images of PNIPAM micropillars within the device, filled with PBS at various temperatures. (b) Variations in pillar diameter D_p, gap size d, and critical diameter D_c between 20 °C and 40 °C during heating and cooling. Red open squares and blue open rhombuses represent the heating and cooling processes, respectively.

During cooling, D_p increased while d decreased, with the most significant change observed between 29 °C and 22 °C. The percentage changes in D_p during cooling were 0% (40–35 °C), 5% (35–28 °C), 25% (28–25 °C), 14% (25–22 °C), and 6% (22–20 °C). Even after dozens of heating and cooling cycles, the PNIPAM micropillars showed no visible damage, confirming their stability and reliability for reversible thermal transitions. The most significant size change in D_p occurred between 25 °C and 28 °C, whereas previous research³² in pure water reported the largest change between 28 °C and 29 °C. This discrepancy is attributed to the presence of solutes in PBS, which reduce the LCST. Our separate turbidity analysis confirmed this, indicating approximate LCST values of 32 °C in pure water and 28 °C in PBS (Fig. S8). These results highlight the influence of solute composition on PNIPAM's thermo-responsive behavior. Thus, it is essential to characterize the LCST—or more generally, the temperature-swelling profile—of PNIPAM in the specific medium used for separation. In our case, PBS was used consistently as the working solution, supporting reproducible PNIPAM responsiveness and separation performance under the reported conditions.

The variations in D_p, d, and D_c (calculated using eqn (1)) during heating and cooling cycles are presented in Fig. 4b. As the temperature increased from 20 °C to 22 °C, D_c rose gradually from 0.8 μm to 5.2 μm. Between 22 °C and 29 °C, a significant increase in D_c was observed, from 5.2 μm to 28.0 μm. Beyond 29 °C, changes in D_c plateaued, reaching 29.0 μm at 35 °C and remaining stable up to 40 °C. Similar reversible changes in D_c were observed during cooling. Across the entire temperature range (20–40 °C), D_c varied from 0.8 μm to 29.0 μm, with the highest sensitivity to temperature changes observed between 25 °C and 28 °C.

In our separation device, the PNIPAM pillars are confined within a PDMS microchannel, and their initial (dry) height is nearly equal to the channel height (approximately 30 μm). In contrast, as described above, unconfined PNIPAM pillars exhibit significant swelling in both horizontal and vertical directions upon hydration (Fig. S7). This comparison suggests that vertical swelling of the confined pillars is partially restricted by the PDMS ceiling, resulting in upward mechanical pressure being exerted on the channel wall. To investigate the effect of this confinement, we compared the temperature-dependent changes in pillar diameters under confined and unconfined conditions (Fig. S9). The results showed that the diameters of confined pillars were slightly larger than those of unconfined ones, particularly at lower temperatures. This indicates that vertical expansion is suppressed by the PDMS ceiling, causing the swelling to be redirected laterally.

The tops of the PNIPAM pillars are not chemically bonded to the PDMS ceiling. Experimental tests confirmed that the PNIPAM material used in this study does not adhere to PDMS, even after oxygen plasma treatment. To evaluate the potential for flow over the pillars due to channel deformation, we conducted flow experiments using fluorescent nanoparticles across a range of flow rates (0.25–2.0 mL h⁻¹) and temperatures (20–40 °C) (Fig. S10). Under these conditions, we observed no particle trajectories indicative of flow over the pillar tops. This suggests that the swollen PNIPAM pillars exert sufficient upward pressure on the PDMS ceiling to effectively minimize any clearance. However, at flow rates exceeding the tested range, we anticipate that the PDMS ceiling may deform, potentially creating flow paths above the pillars. This could compromise the intended flow confinement and DLD performance under high-pressure conditions.

The responsiveness of the confined PNIPAM pillars to temperature changes was further evaluated by measuring D_p during heating (20–40 °C) and cooling (40–20 °C) cycles (Fig. 5). During heating, D_p decreased from its maximum to minimum value within 60 seconds, while during cooling, D_p increased from its minimum to maximum value within 80 seconds. Compared with our previous glass-based device,³² the PNIPAM micropillars in this study exhibited significantly increased diameter and height, resulting in an approximately 30-fold increase in pillar volume. This increase reduced the surface-to-volume ratio, which typically hinders heat transfer and water diffusion, leading to slower phase transitions in larger PNIPAM hydrogels.³⁹ However, despite these challenges, the superior thermal conductivity of the silicon substrate effectively mitigated their impact. Even with the substantial increase in pillar volume, the response time remained nearly unchanged, underscoring the critical role of the silicon substrate in enabling rapid and precise thermal transitions.

Fig. 5 Shrinkage and swelling of PNIPAM micropillars in response to temperature variations over time. (a) Heating time response (20–40 °C). (b) Cooling time response (40–20 °C). Gray filled circles indicate the pillar diameter D_p.

Effect of flow and substrate's thermal conductivity on D_c control

Flow rate can influence the behavior of confined PNIPAM micropillars through two primary mechanisms: heat transfer delay and fluid pressure-induced deformation. As flow rate increases, the fluid residence time in the channel decreases, making the actual fluid temperature more dependent on the system's thermal conductivity. Additionally, in the swollen state, the Young's modulus of PNIPAM decreases by approximately two orders of magnitude compared to the shrunk state,⁴⁰ making the pillars more susceptible to deformation under fluidic pressure.

To evaluate these effects, we measured the size deviation ratio—defined as the pillar diameter under flow divided by that under no-flow conditions at the same temperature—for PNIPAM micropillars on both silicon (Fig. 6a) and glass (Fig. 6b) substrates at flow rates of 1.0, 2.0, 4.0, and 8.0 mL h⁻¹. In all experiments, the PBS solution and ambient environment were maintained at 20 °C. After 3 minutes of thermal equilibration at each set temperature, the pillar diameter (D_p) was measured in the first section of the DLD array, where temperature and pressure effects are most pronounced.

Fig. 6 Size deviation ratio of PNIPAM pillars under various flow rates. (a) Silicon substrate device. (b) Glass substrate device.

At a flow rate of 1.0 mL h⁻¹, no significant size deviation was observed for either substrate, indicating sufficient heat transfer and negligible pressure-induced deformation. Based on 3D simulations of a single DLD section (Fig. S11), the pressure difference across the entire device at this flow rate was estimated to be 1.5–7.5 kPa. At 2.0 mL h⁻¹, the silicon-based device still showed minimal deviation. However, glass-based device exhibited a notable positive deviation in the 23–27 °C range, suggesting incomplete fluid heating due to the lower thermal conductivity of glass. Below 23 °C, deviation was minimal because the fluid and target temperatures were similar. Beyond 27 °C, the deviation decreased as the temperature approached the LCST, where PNIPAM's size becomes less sensitive to temperature changes.

At 4.0 mL h⁻¹, the silicon-based device showed a negative deviation before 30 °C, attributed to fluid pressure compressing the swollen pillars. As the temperature increased past the LCST, the PNIPAM stiffened, reducing deformation. For the glass-based device, both positive and negative deviations were observed, indicating the simultaneous influence of heat transfer delay and fluid pressure. In the glass-based device, in contrast, heat transfer delay dominated below 29 °C, leading to relatively large positive deviation. Above this point, the reduced temperature sensitivity allowed pressure effects to dominate, causing negative deviation. At 8.0 mL h⁻¹, the silicon-based device exhibited further negative deviation, while glass-based device showed reduced overall deviation, likely due to partial cancelation of heat transfer delay and pressure effects.

These results demonstrate that under steady-state conditions, glass substrate exhibit heat transfer delays, while the superior thermal conductivity of silicon makes it more suitable for thermo-responsive DLD system. Fluid pressure is also identified as a key factor affecting stability at high flow rates. Considering the compressive effects observed, the optimal operating flow rate for this system is considered to be below 2.0 mL h⁻¹.

Thermally controlled cell separation

To demonstrate the thermally tunable D_c capability of our device, MCF-7 cells were used as model particles and mixed with blood samples to prepare a sample solution. Initially, PBS was injected through inlets 1, 2, and 3 to completely fill the microchannel. Subsequently, the sample solution containing MCF-7 cells replaced PBS at inlet 2. The flow rates at the three inlets were set to 0.21 mL h⁻¹, 0.1 mL h⁻¹, and 0.09 mL h⁻¹, respectively.

At 25 °C, fluorescence trajectories and distribution of MCF-7 cells in the upstream, midstream and downstream regions revealed their migration mode through the DLD array (Fig. 7a). Meanwhile, Fig. S12a and Video S1 illustrate the migration behavior of MCF-7 cells and background blood cells in brightfield mode. At this temperature, the D_p was 102.7 μm, d was 27.3 μm, and D_c was calculated to be 14.1 μm. This temperature was chosen to minimize cell aggregation and reduce clogging risks. In the upstream region, both MCF-7 cells (including single cells and clusters) and blood cells entered the DLD array through gaps numbered 4–6. MCF-7 cells, being larger than D_c, migrated laterally in the bump mode toward the upper sidewall, while smaller blood cells retained their positions, following a zigzag migration mode. By the downstream region, MCF-7 cells were effectively displaced and exited through gaps numbered 9–16, while blood cells exited through gaps numbered 4–6 without lateral displacement. Although most MCF-7 cells were successfully directed into outlet L as intended, a small fraction entered outlet S. This is likely due to some cells having sizes near or slightly below D_c, resulting in mixed-mode migration and insufficient lateral displacement. Based on video analysis, the capture efficiency—defined as the number of MCF-7 cells entering outlet L divided by the total number of MCF-7 cells entering both outlets—was 90.4% (n = 167) (Fig. S13a).

Fig. 7 Thermally switching modes of MCF-7 cell separation from a blood sample. (a) Major separation mode at 25 °C. (b) Selective separation mode at 26 °C. (c) Minimal separation mode at 37 °C. Fluorescent migration trajectories and the distribution of MCF-7 cells are shown for the upstream, middle, and downstream DLD regions under each condition. Bright-field and fluorescent images of the separated cells are also presented. Scale bars: 50 μm.

Solutions collected from outlets L and S were centrifuged to remove excess fluid. Brightfield and fluorescence imaging confirmed separation of MCF-7 cells at outlet L with 100% purity, while background RBCs and WBCs, and a small amount of MCF-7 cells were collected at outlet S (Fig. 7a). Post-separation analysis showed a slight increase in the average size of MCF-7 single cells, with a mean diameter of 17.6 ± 2.2 μm (n = 1012), compared to the pre-separation mean of 17.0 ± 2.4 μm (n = 1022) (Fig. 8a). This increase is likely due to the removal of smaller MCF-7 cells—those with sizes below the effective D_c—which were not fully displaced and exited through outlet S.

Fig. 8 Size distributions of MCF-7 cells collected at different temperatures. (a) Outlet L at 25 °C. (b) Outlet L at 26 °C. (c) Outlet S at 26 °C. (d) Outlet S at 37 °C.

When the temperature was increased to 26 °C, the migration behavior of MCF-7 and blood cells was observed (Fig. 7b and S12b, and Video S2). At this temperature, the D_p was 94.2 μm, and D_c was calculated as 18.5 μm. Cells entered the DLD array through gaps numbered 5–7. In the midstream region, migration modes began to diverge: MCF-7 cells exceeding D_c migrated laterally in the bump mode, while smaller MCF-7 cells and blood cells followed the zigzag mode. In the downstream region, MCF-7 cells were widely distributed exited through gaps numbered 5–16, indicating mixed migration modes. Brightfield and fluorescence images confirmed that MCF-7 cells were collected at outlets L and S, while blood cells were exclusively directed to outlet S. Based on the video analysis, the capture efficiency of MCF-7 cells at outlet L was 42.8% (n = 114) (Fig. S13b). The size distribution of single MCF-7 cells collected at outlet L (18.9 ± 1.7 μm, n = 1007) was larger than that of single MCF-7 cells at outlet S (16.5 ± 1.1 μm, n = 1038), reflecting effective separation of larger MCF-7 cells into outlet L and smaller ones into outlet S (Fig. 8b and c). However, a number of single MCF-7 cells with sizes below the calculated D_c were also found in the outlet-L sample. These may have originated from partial disaggregation of clusters during post-separation handling. While most clusters were found in the outlet-L sample, a small number were also detected in the outlet-S sample, which we similarly attribute to post-separation aggregation occurring during sample handling prior to measurement.

At 37 °C, the optimal temperature for cell growth, migration trajectories of MCF-7 and blood cells were recorded (Fig. 7c and S12c, and Video S3). At this temperature, D_p was 73.9 μm, and D_c was 29.0 μm. Since the average sizes of single MCF-7 cells (17.0 μm), WBCs (10.4 μm), and RBCs (6.5 μm) were all smaller than D_c, these cells followed the zigzag migration mode without lateral displacement. Cells entered the DLD array through gaps numbered 5–6 and exited through gaps numbered 4–5 in the downstream region. Both brightfield and fluorescence imaging confirmed that the vast majority of MCF-7 cells (96.4%), along with all RBCs and WBCs, were collected at outlet S, while a small fraction (3.6%) of MCF-7 cells, observed as clusters, entered outlet L (Fig. S13c). Post-separation, the size distribution of single MCF-7 cells (17.5 ± 2.1 μm, n = 1024) was similar to the pre-separation distribution (17.0 ± 2.4 μm, n = 1022) (Fig. 8d). A small number of clusters with sizes above the calculated D_c were observed in the outlet-S sample, likely due to aggregation during post-separation handling, as previously discussed.

These results demonstrate that the device enables both effective separation and precise isolation of target cells based on size. By flexibly adjusting D_c through temperature modulation, the device alters migration modes, offering tailored separation strategies for specific applications.

In our experiments, gravitational sedimentation of cells in the syringe led to a gradual decrease in the number of cells entering the device over time. At a sample flow rate of 0.1 mL h⁻¹ and a collection duration of 20 minutes, we quantified this time-dependent decrease (Fig. S14), estimating the actual number of recovered cells to be approximately 2.3 × 10⁴, about 53% of the ideal recovery (∼4.3 × 10⁴) based on the initial concentration of 1.3 × 10⁶ cells per mL. To improve recovery efficiency and enable more stable long-term operation, a higher sample flow rate and/or continuous agitation of the upstream reservoir (e.g., syringe mixing) would be beneficial.

As discussed in the context of each separation experiment, the impact of target cell clusters (Fig. S15)—which may form either before or after microfluidic processing—depends on the separation mode. In major separation mode, where the goal is to isolate target cells from smaller background cells, clusters do not negatively impact performance. Being larger than individual target cells, clusters are also directed to outlet L via the bump mode. In contrast, in selective separation (targeting a specific subpopulation of larger cells) or minimal separation modes, clusters can introduce impurities. For example, clusters of smaller target cells may exceed the D_c and be mistakenly separated. Additionally, clusters can lead to clogging within the micropillar array, compromising flow stability and device performance. To mitigate these risks, it will be effective to minimize cluster formation by optimizing cell concentration and, where appropriate, adding surfactants to reduce cell–cell adhesion.

Cell viability assay

The impact of microfluidic separation on cell viability was assessed using a calcein-AM/PI double-staining assay. MCF-7 cells collected from the device were centrifuged to remove excess solution, and live/dead assay was performed using a fluorescence microscope (Fig. 9). Cell viability was quantified as the ratio of live cells to total cells (live + dead). The results showed no significant difference between the microfluidics group (operated at 25 °C) and the control group (37 °C). Viability was 91.3% for the microfluidics group and 92.2% for the control group, indicating that the lower operating temperature and shear forces within the microchannel during separation did not adversely affect cell viability. These findings confirm that the device's suitability for applications requiring viable target cells.

Fig. 9 Viability assessment of MCF-7 cells. Comparison of the microfluidics group (left, 25 °C) and control group (right, 37 °C). Scale bar: 100 μm.

Conclusion

We demonstrated tunable cell separation using a thermo-responsive DLD device featuring PNIPAM-based micropillars fabricated on a silicon substrate. Heat transfer simulations confirmed the superior thermal performance of the silicon substrate compared to glass, ensuring precise temperature control and rapid response times. By modulating the temperature within the range of 20–40 °C, the device enabled continuous adjustment of D_c across a broad range (0.8–29.0 μm), achieving three distinct cell separation modes: (i) major separation of MCF-7 cells from blood samples at 25 °C, (ii) selective separation of larger MCF-7 subpopulations at 26 °C, and (iii) minimal separation at 37 °C. Verification of dynamic switching performance during actual DLD separation remains an important direction for future investigation. Overall, the precision, versatility, and reliability of this platform underscore its potential for high-resolution size-based sorting, making it a promising tool for a wide range of biomedical applications.

Author contributions

Ze Jiang: methodology, validation, formal analysis, investigation, writing – original draft, visualization; Yusuke Kanno: validation, resources, writing – review & editing. Takasi Nisisako: conceptualization, methodology, validation, resources, supervision, visualization, project administration, writing – review & editing, funding acquisition.

Conflicts of interest

There are no conflicts to declare.

Data availability

Supplementary information is available. See DOI: https://doi.org/10.1039/D5LC00783F.

The data supporting this article are available from the corresponding author upon reasonable request.

Acknowledgements

This work was supported by JSPS KAKENHI Grant Number 24K21696. We appreciate Dr. Yoshiyuki Yokoyama of Toyama Industrial Technology Research and Development Center for supplying the PNIPAM-based photoresist.

References

1. A. Dalili, E. Samiei and M. Hoorfar, Analyst, 2019, 144, 87–113.
2. S. Rawal, Y.-P. Yang, R. Cote and A. Agarwal, Annu. Rev. Anal. Chem., 2017, 10, 321–343.
3. H. F. Zulfiqar, A. Javed, Sumbal, B. Afroze, Q. Ali, K. Akbar, T. Nadeem, M. A. Rana, Z. A. Nazar and I. A. Nasir, Front. Public Health, 2017, 5, 32.
4. J. Y. Lee and S.-H. Hong, Int. J. Stem Cells, 2020, 13, 1–12.
5. C. W. Shields Iv, C. D. Reyes and G. P. López, Lab Chip, 2015, 15, 1230–1249.
6. Y. Shen, Y. Yalikun and Y. Tanaka, Sens. Actuators, B, 2019, 282, 268–281.
7. N. Lu, H. M. Tay, C. Petchakup, L. He, L. Gong, K. K. Maw, S. Y. Leong, W. W. Lok, H. B. Ong and R. Guo, Lab Chip, 2023, 23, 1226–1257.
8. L. R. Huang, E. C. Cox, R. H. Austin and J. C. Sturm, Science, 2004, 304, 987–990.
9. A. Hochstetter, R. Vernekar, R. H. Austin, H. Becker, J. P. Beech, D. A. Fedosov, G. Gompper, S.-C. Kim, J. T. Smith, G. Stolovitzky, J. O. Tegenfeldt, B. H. Wunsch, K. K. Zeming, T. Krüger and D. W. Inglis, ACS Nano, 2020, 14, 10784–10795.
10. T. Salafi, Y. Zhang and Y. Zhang, Nano-Micro Lett., 2019, 11, 77.
11. H. Okano, T. Konishi, T. Suzuki, T. Suzuki, S. Ariyasu, S. Aoki, R. Abe and M. Hayase, Biomed. Microdevices, 2015, 17, 1–11.
12. D. Holmes, G. Whyte, J. Bailey, N. Vergara-Irigaray, A. Ekpenyong, J. Guck and T. Duke, Interface Focus, 2014, 4, 20140011.
13. D. W. Inglis, N. Herman and G. Vesey, Biomicrofluidics, 2010, 4, 024109.
14. G. Ji, S. Masui, Y. Kanno and T. Nisisako, Micromachines, 2024, 15, 908.
15. N. Tottori, T. Hatsuzawa and T. Nisisako, RSC Adv., 2017, 7, 35516–35524.
16. O. E. Ström, J. P. Beech and J. O. Tegenfeldt, Micromachines, 2022, 13, 1754.
17. B. H. Wunsch, J. T. Smith, S. M. Gifford, C. Wang, M. Brink, R. L. Bruce, R. H. Austin, G. Stolovitzky and Y. Astier, Nat. Nanotechnol., 2016, 11, 936–940.
18. T. Kulrattanarak, R. G. M. Van Der Sman, C. Schroën and R. M. Boom, Microfluid. Nanofluid., 2011, 10, 843–853.
19. R. Vernekar, T. Krüger, K. Loutherback, K. Morton and D. W. Inglis, Lab Chip, 2017, 17, 3318–3330.
20. S. Chang and Y. H. Cho, Lab Chip, 2008, 8, 1930–1936.
21. J. P. Beech, P. Jönsson and J. O. Tegenfeldt, Lab Chip, 2009, 9, 2698–2706.
22. D. Wang, S. Yang, N. Wang, H. Guo, S. Feng, Y. Luo and J. Zhao, Biosensors, 2024, 14, 174.
23. M. Aghaamoo, A. Aghilinejad, X. Chen and J. Xu, Electrophoresis, 2019, 40, 1486–1493.
24. J. P. Beech, K. Keim, B. D. Ho, C. Guiducci and J. O. Tegenfeldt, Adv. Mater. Technol., 2019, 4, 1900339.
25. Y. Li, H. Zhang, Y. Li, X. Li, J. Wu, S. Qian and F. Li, Sci. Rep., 2018, 8, 3618.
26. B. Senf and J. H. Kim, Micro, 2022, 2, 100–112.
27. K. K. Zeming, N. V. Thakor, Y. Zhang and C.-H. Chen, Lab Chip, 2016, 16, 75–85.
28. K. K. Zeming, T. Salafi, S. Shikha and Y. Zhang, Nat. Commun., 2018, 9, 1254.
29. D. J. Collins, T. Alan and A. Neild, Lab Chip, 2014, 14, 1595–1603.
30. J. P. Beech and J. O. Tegenfeldt, Lab Chip, 2008, 8, 657–659.
31. A. K. Barzoki and A. Shamloo, J. Chromatogr. A, 2024, 1736, 465397.
32. N. Tottori and T. Nisisako, Sci. Rep., 2023, 13, 4994.
33. K. Jain, R. Vedarajan, M. Watanabe, M. Ishikiriyama and N. Matsumi, Polym. Chem., 2015, 6, 6819–6825.
34. R. Liu, M. Fraylich and B. R. Saunders, Colloid Polym. Sci., 2009, 287, 627–643.
35. C. Barwig, A. Sonn, T. Spratte, A. Mishra, E. Blasco, C. Selhuber-Unkel and S. Pashapour, Adv. Intell. Syst., 2024, 6, 2300829.
36. J. A. Davis, Dr. Philos., Princet. Univ., 2008.
37. G. Zhang, Y. Sun, B. Qian, H. Gao and D. Zuo, Polym. Test., 2020, 90, 106670.
38. J. Wei, M. Liao, A. Ma, Y. Chen, Z. Duan, X. Hou, M. Li, N. Jiang and J. Yu, Compos. Commun., 2020, 17, 141–146.
39. Y. Pan, B. Li, Z. Liu, Z. Yang, X. Yang, K. Shi, W. Li, C. Peng, W. Wang and X. Ji, ACS Appl. Mater. Interfaces, 2018, 10, 32747–32759.
40. T. Takigawa, T. Yamawaki, K. Takahashi and T. Masuda, Polym. Gels Networks, 1998, 5, 585–589.

---