Introduction

Cancer liquid biopsy based on clinical effusions is established as an accurate, non-invasive approach for longitudinal disease monitoring and personalized therapeutic guidance1,2,3, particularly in enhanced malignancy diagnosis4,5. This advancement critically depends on the efficient enrichment of malignant tumor cells (MTCs) and clusters (MTCCs)6. The presence and abundance of MTCs serve as key biomarkers for cancer diagnosis, while MTCCs further indicate potential malignant metastasis7. In clinical utility, immune-affinity-based enrichment enables the detection of single MTCs, but suffers from limitations including cellular damage and the dissociation of MTCCs during processing8. Furthermore, the conjugated antibodies can mask surface protein biomarkers, impeding the development of cancer immunotherapies and antibody-drug conjugates9. Given these constraints of immune-affinity-based enrichment, a label-free microfluidic technology enabling high-fidelity enrichment of single MTCs and intact MTCCs for robust downstream analysis is clinically imperative to advance therapeutic development and precision oncology decision-making10.

To date, label-free microfluidics has provided powerful tools for enriching MTCs11,12,13. However, achieving high‑throughput, high‑purity, and concurrent enrichment of both single MTCs and intact MTCCs from large‑volume clinical samples remains a significant challenge. Most existing platforms are primarily optimized for single cells, lacking the resolution or gentle handling required for clusters14,15,16,17,18. Microfiltration techniques, for example, employed triangular19,20 or hexagonal pillars21 for rapid and straightforward enrichment of tumor cell clusters from peripheral blood in cancer patients. However, high shear forces experienced during microfiltration could damage MTCCs or dissociate them into single cells22. To eliminate this damage, deterministic lateral displacement (DLD) sorted and enriched MTCCs by size through asymmetric bifurcation of laminar flow around micro-pillar arrays23. DLDs require channel dimensions comparable to cell sizes to ensure particles bump across streamlines, and consequently, they are prone to clogging when processing large-volume samples. To minimize clogging, techniques such as acoustic24 and inertial microfluidics25,26 have been developed for continuous flow-based enrichment, but exhibit insufficient resolution to discriminate single MTCs from MTCCs within a single device.

Consequently, hybrid or cascaded microfluidics that integrate multiple separation principles have attracted growing interests for overcoming these inherent trade‑offs27,28,29. The main hybrid strategy is the coupling of immune-affinity technique and label-free microfluidics30,31,32. For instance, Mishra et al.33 developed a LPCTC-iChip in which the DLD was coupled with magnetophoresis for the tumor antigen–independent separation of circulating tumor cells (CTCs). Despite tumor antigen–independent enrichment can be realized by labeling blood cells, the disadvantages of complex operation and expensive reagents are still difficult to avoid. Another integration strategy is the cascading of passive microfluidics (e.g., multiorifice flow fractionation, spiral inertial microfluidics, and hydrodynamic-vortex separation) in series34,35,36,37. A summary of these systems is provided in Table S1. Park et al.31 utilized a microwell-on-electrode array to realize simultaneous trapping of single cells and clusters at a single cell/cluster level. However, this approach inherently prevents the subsequent physical separation or selective retrieval of individual tumor cells from the captured clusters for further analysis. Even though advances have been made, the new hybrid techniques for precise, continuous, and label-free enrichment of MTCs and MTCCs are still in urgent demands.

In this study, we report the first application of cascaded inertial microfluidics for processing malignant effusions from lung cancer patients. This approach enables high-throughput, multi‑scale enrichment of exfoliated tumor cells and intact cell clusters, thereby facilitating enhanced malignancy assessment. The cascaded inertial microfluidic system implements a two‑stage enrichment strategy: initial high‑throughput depletion of abundant background blood cells is achieved via parallelized serpentine channels, followed by multi‑scale enrichment of single MTCs and intact MTCCs using slanted spiral channels. After demonstrating the conceptual design and overall strategy, we characterize the performances of our device for the two-stage inertial focusing and sorting of simulated particle/cell samples. Finally, we employ cascaded inertial microfluidics for the label-free, high-throughput, multi-scale enrichment of single MTCs and intact MTCCs from clinical pleural effusions. This approach provides dual-cell insights to enhance malignancy diagnosis and enable actionable metastatic profiling.

Results and discussion

Microfluidic enrichment of MTCs and MTCCs from patient pleural effusions for enhanced malignancy diagnosis

The conceptual design and overall strategy for microfluidic enrichment of MTCs and MTCCs from patient pleural effusions are illustrated in Fig. 1. Pleural effusion samples with a typical volume of 50 mL per patient were collected from patients with metastatic lung cancer via standard diagnostic thoracentesis (see Methods for procedural details). Following viscous supernatant removal and PBS resuspension, a cascaded inertial microfluidic device was used to enrich MTCs and MTCCs via a two-stage, size-dependent inertial sorting strategy.

Fig. 1: Schematic illustration of high-throughput, continuous-flow, and multi-scale enrichment of viable MTCs and MTCCs from pleural effusion samples using cascaded inertial microfluidics.
Fig. 1: Schematic illustration of high-throughput, continuous-flow, and multi-scale enrichment of viable MTCs and MTCCs from pleural effusion samples using cascaded inertial microfluidics.The alternative text for this image may have been generated using AI.
Full size image

The cascaded microfluidics enable a two-stage inertial sorting via the coupling of parallelized serpentine channels and slanted spiral channels

In the first stage, parallelized serpentine channels act as a high-throughput filter. In these channels, the combined effects of inertial lift forces (FL) and Dean drag forces (FD) induce size-dependent cell focusing (Fig. S1). Small background blood cells migrate toward the sidewalls and are directed to the waste outlets, while large MTCs and MTCCs remain in the core flow and are collected for subsequent processing. In the second stage, the spiral inertial microfluidic device serves as a subsequent sorter, enabling further enrichment of individual MTCs and MTCCs, respectively. This is achieved by differentially balancing the Dean drag force and inertial lift force, which establishes distinct focusing positions for single cells and clusters according to their sizes. As a result, target components are spatially separated into different outlet streams.

Following enrichment, MTCs and MTCCs were distinguished from residual blood cells based on immunofluorescence staining (Pan-CK+/CD45−/DAPI+), which allows for the quantification of MTC and MTCC abundance in the effusion samples. An MTCC was defined as a cohesive aggregate of two or more tumor cells in direct cell–cell contact, identified as a single connected object in fluorescence images. In summary, our cascaded inertial microfluidics provide a label-free, high-throughput, and multi-scale technique for the simultaneous enrichment of single MTCs and intact MTCCs, directly supporting enhanced cytological diagnosis.

Multi-scale enrichment of simulated microparticles using cascaded inertial microfluidics

After the conceptual design, simulated particle samples were utilized to explore the effects of particle size and flow rate on the two-stage inertial sorting (Fig. 2). In this experiment, a series of microparticles (7-25 µm) served as cell proxies to simulate diverse effusion cells: normal blood cells (7-10 µm), single MTCs (15 µm), and MTCCs (25 µm). The prepared particle suspension was injected into the inlet at the flow rates of 4.8-10.4 mL/min, and subsequently sorted via four distinct outlets for recovery (Fig. 2a, b). When flowing through the first-stage parallelized serpentine channels, 7-10 µm particles migrate towards the channel walls due to the dominance of FD, where 15-25 µm particles occupy equilibrium positions near the channel centerline under the FL (Fig. 2b (i)). Figure 2c illustrates the focusing maps of 7-25 µm particles across a wide flow rate range (4.8-10.4 mL/min), with the optimum flow rate identified as 8 mL/min.

Fig. 2: Particle focusing in the cascaded inertial microfluidic device.
Fig. 2: Particle focusing in the cascaded inertial microfluidic device.The alternative text for this image may have been generated using AI.
Full size image

a, b Schematics and images illustrating the designs of the cascaded inertial microfluidic device and the performances of two-stage inertial sorting at different outlets (insets i and ii). c, d Focusing maps of 7–25 µm particles when flowing through the parallelized serpentine channels (c) and the slanted spiral channels (d). e Distributions of 15 and 25 µm particles at flow rates between 2.5 and 3.75 mL/min near the outlet of the slanted spiral channel. f Recovery rates of 7–25 µm particles at each outlet after two-stage inertial sorting

In the second-stage inertial sorting, a slanted spiral channel coupled with periodic contraction-expansion arrays was designed for multi-scale enrichment of 15-25 µm particles (simulating MTCs and MTCCs). When flowing into the spiral-contraction-expansion channels, 10-25 µm particles sequentially focused at different lateral positions under the competition between FD, FL, and local-vortex induced lift force (FV) (Fig. 2b(ii), d). Figures 2e and S2 show the focusing behaviors and equilibrium positions of 10-25 µm particles under the optimized flow conditions (i.e., a total flow rate of 8 mL/min, of which 3.5 mL/min was directed into the second-stage inlet). Particle recovery after the two-stage sorting was quantified at the optimized total flow rate of 8 mL/min (Fig. 2f). At the outlet 4, 25 µm particles (modeling MTCCs) achieved a recovery rate of 91.8% ± 6.6% (n = 3), while the outlet 3 recovered 87.4% ± 7.4% of 15 µm particles (representing single MTCs). Residual 7–10 µm particles (<10%) were efficiently removed via the waste outlet 2, confirming effective clearance of background particles. As expected, overwhelming majority of 7-10 µm particles (simulating normal blood cells) was removed via outlet 1, enabling the inertial sorting of 15-25 µm particles (simulating single MTCs and MTCCs) at the second stage.

Performance characterization of two-stage inertial enrichment

We next characterized the two-stage enrichment performances of the cascaded inertial microfluidic device by using A549 lung cancer cells (containing both single cells and clusters) and WBCs at the optimized flow rates of 8 mL/min. In these experiments, fluorescently labeled A549 cells were spiked into lysed blood samples (WBCs), and the prepared samples were subsequently injected into the inlet of the cascaded inertial microfluidic device (Fig. 3a). Figure 3b demonstrates the cell focusing behaviors near the two-stage outlets under bright-field and fluorescence imaging. After first-stage inertial sorting, A549 cells were focused near the channel centerline with majority of WBCs depletion from the waste outlet 1. Following second-stage inertial sorting, single A549 cells maintained centerline focusing, whereas A549 cell clusters were further focused adjacent to the inner wall.

Fig. 3: Characterization of the two-stage inertial enrichment using simulated cell samples.
Fig. 3: Characterization of the two-stage inertial enrichment using simulated cell samples.The alternative text for this image may have been generated using AI.
Full size image

a Overview structures of the cascaded inertial microfluidic device. b Cell focusing behaviors near the two-stage outlets at the optimized flow rate of 8 mL/min. c Cell concentrations of A549 cells and WBCs at the inlet and four outlets. d Recovery rate and purity of A549 cells and WBCs at each outlet. e Bright-field and fluorescent microscopic images illustrating cell distributions at the three outlets 2-4. f Size distributions of A549 cells (both in single and cluster forms) collected from the two target outlets 3 and 4 (p-value less than 0.01)

To quantitatively evaluate the enrichment performance, the density of A549 cells and WBCs in the initial samples, waste, and enriched suspensions were measured (Fig. 3c and Table S2). The lysed blood sample (with a WBC concentration of 4.7×105 cells/mL) spiked with A549 cells (1.5×104 cells/mL) was pumped into the device inlet. The final concentration of single A549 cells (6.3×104 cells/mL) in the enriched suspension (outlet 3) was quadruple that in the initial samples and triple that of residual WBCs (2.0×104 cells/mL). As quantitatively assessed in Fig. 3d, the cascaded inertial microfluidics achieved the high-performance enrichment of A549 cells, with single cells exhibiting a recovery rate of 75.9% ± 1.3% and a purity of 75.8% ± 1.6% at outlet 3, while cell clusters were recovered at outlet 4 with a recovery of 5.8% ± 0.9% and a purity of 79.4% ± 3.4%. The recovery rate for cell clusters is calculated based on the total number of A549 cells identified in the initial spiked sample, within which only a small proportion exists as intact clusters. Cumulatively, the cascaded microfluidic device attained a total A549 recovery of 81.7% ± 1.2% and an overall purity of 76.1% ± 1.3%. This performance compares favorably with recent label-free platforms (see Table S3 for a summary), demonstrating the excellent enrichment capability of our device for single tumor cells and clusters from complex backgrounds.

Figure 3e presents bright-field and fluorescence images of enriched cells collected from the three outlets 2-4. Blood cells were predominantly collected from outlet 2. Single A549 cells (green fluorescent-labeled) were enriched in outlet 3 but contaminated by a minor population of blood cells (red circles). High-purity large-sized single A549 cells and intact A549 cell clusters were obtained from outlet 4 using our cascaded inertial microfluidic device and can be re-cultured within 72 hours (Fig. S3). Morphometric analysis of enriched cells (Fig. 3f) revealed distinct size distributions: Single A549 cells collected from outlet 3 showed a narrow average size distribution of 14.8 ± 1.6 µm (within 11.5–17.8 µm), whereas clusters at outlet 4 exhibited a broader size variation of 15.2–33.6 µm (an average of 21.8 ± 6.7 µm). To further assess operational robustness, an additional continuous-run stress test was performed, processing 100 mL of sample at 8 mL/min. No channel clogging or outlet flow instability was observed, confirming the device’s reliability for large-volume clinical effusion processing.

MTC and MTCC enrichment from patient pleural effusions

The presence of MTCs in pleural effusions remains the gold standard for malignant effusion diagnosis, while the abundance of MTCCs serves as a further predictor of metastatic potential. However, pleural effusion cytology suffers from a non-negligible loss of sensitivity due to the obscuration of MTCs by numerous blood cells and the disruption of cellular morphology during processing. To improve the sensitivity, our cascaded inertial microfluidics was employed for the high-throughput and multi-scale enrichment of single MTCs and intact MTCCs from pleural effusion samples of lung cancer patients. Pleural effusion samples (50 mL per case) were collected from enrolled patients (n = 3) via standard diagnostic thoracentesis at Zhongda Hospital and processed within 2 h after collection. Following removal of viscous supernatant and resuspension in PBS, the prepared effusion samples were directly infused into the cascaded microfluidic device at a precisely controlled flow rate of 8 mL/min.

When flowing through the cascaded inertial microfluidic device, pleural effusion cells undergo two-stage inertial sorting via the coupling of parallelized serpentine channels and slanted spiral channels. Figure 4a illustrates the distributions of pleural effusion cells at the two-stage outlets. The majority of blood cells were removed from the first-stage outlet 1, and the residual were substantially reduced at the second-stage outlet 2. Meanwhile, centrally-focused MTCs were enriched through outlet 3, while inner-wall-adjacent MTCCs were recovered intact at outlet 4 after passing through the second-stage spiral sorter (Fig. 4a). Following immunofluorescence staining, MTCs/MTCCs were identified as DAPI+/pan-CK+/CD45−, and their counts in the enriched suspension were determined (Fig. 4b).

Fig. 4: Microfluidic enrichment of single MTCs and intact MTCCs from clinical effusions.
Fig. 4: Microfluidic enrichment of single MTCs and intact MTCCs from clinical effusions.The alternative text for this image may have been generated using AI.
Full size image

a Bright-field focusing maps depicting the cell distribution near the two-stage outlets of our cascaded inertial microfluidic device. b Fluorescent microscopic images illustrating cell distributions at the initial sample and two enriched samples collected from outlets 3 and 4. c Proportions of WBCs, single MTCs, and intact MTCCs collected from the inlet and two target outlets 3 and 4

To evaluate the enrichment performance of our cascaded microfluidic device, the proportions of MTCs and MTCCs in the initial samples and enriched samples were calculated (Fig. 4c). Cytological analysis revealed that the pleural effusion sample containing <1% MTCs/MTCCs (compared with background WBCs) was injected into the device inlet. After two-stage inertial enrichment, the sample collected from outlet 3 showed a high density of cells dominated by individual MTCs (68%), with substantially reduced WBC presence (29%). Conversely, the samples collected from outlet 4 contained relatively few cells but exhibited prominent large MTCCs (35% of total cells) and large-sized single MTCs (39%), as evidenced by distinct green fluorescent clusters in Fig. 4c. Compared with the limited throughput of conventional inertial sorters (typically ≤2 mL/min), our cascaded microfluidic device achieved an ultra-high throughput of 8 mL/min and is able to process 50 mL clinical effusion samples within 6.5 minutes. It is important to note that the clinical validation in this study is preliminary, involving effusions from only three patients. While the results robustly demonstrate the technical feasibility and enrichment capability of our platform, future studies with larger cohorts are essential to establish its diagnostic sensitivity or prognostic relevance.

Conclusion

This study demonstrates a cascaded inertial microfluidic device for high-throughput, label-free, and multi-scale enrichment of single MTCs and intact MTCCs from clinical pleural effusions. By integrating parallelized serpentine sorters (for background cell depletion) and slanted spiral microfluidics (for size-selective enrichment), the device achieves continuous processing of large-volume samples (50 mL in 6.5 min) at an ultra-high throughput (8 mL/min).

Using the cascaded inertial microfluidic device, two-stage inertial sorting achieved high recovery rates of 91.8% ± 6.6% for 25 µm particles (modeling MTCCs) and 87.4% ± 7.4% for 15 µm particles (modeling single MTCs), while reducing background particles (7-10 µm) to residual levels (<6.4% initial load). When validated with simulated cell samples, a high recovery ratio of 75.9% ± 1.3% and a high purity of 75.8% ± 1.6% were achieved for single A549 cells, while cell clusters attained a purity of 79.4% ± 3.4%.

Most importantly, clinical validation using pleural effusion samples demonstrates the excellent performances in enriching both single MTCs and MTCCs, achieving high purities of 68% for MTCs and 35% for MTCCs. For clinical translation, the device offers several practical advantages. The multilayer polymer fabrication is compatible with scalable, low-cost manufacturing. Its label-free operation, requiring no external fields or complex instrumentation, significantly reduces operational complexity. Furthermore, the enriched cells are directly compatible with standard downstream analytical protocols, such as immunofluorescence staining and microscopic examination, enabling seamless integration into existing cytopathology workflows. Future development will focus on implementing standardized disposable cartridges and automating sample handling to minimize operator dependency and enhance reproducibility.

In summary, this cascaded inertial microfluidic approach enables the efficient co-enrichment of both single MTCs and MTCCs. It presents a promising new tool for liquid-biopsy-based applications, potentially enhancing malignancy diagnosis and facilitating actionable metastatic profiling.

Materials and methods

Structural design of the cascaded inertial microfluidics

The proposed microfluidic device employs a two-stage cascaded architecture combining parallelized serpentine microchannels with a slanted spiral channel to achieve high-throughput and multi-scale enrichment of single tumor cells and clusters. The device comprises four functionally optimized modules: (1) an eight-channel serpentine sorting array (channel height: 80 μm) capable of handling large-volume samples for initial enrichment, (2) a branched flow distribution network ensuring uniform fluidic delivery to all parallel channels, (3) a hydraulic resistance matching unit precisely engineered to maintain optimal flow rates in both stages, and (4) a slanted spiral sorting module for second enrichment. The flow resistance ratio between the two outlets of first stage was set to be 9:7, which ensures the second spiral unit operates at the optimal flow rate. This resistance matching was achieved through computational fluid dynamics modeling. The optimal dimensions (length: 550 μm; height: 160 μm) for the resistance-matching channels were determined by pressure-regulated flow measurements. Furthermore, the periodic expansion structures along the spiral channel enhance Dean vortex formation for improved separation efficiency. The detailed structures of each functional modules can be found in Fig. S4a–d.

Fabrication of the integrated chip

The cascaded inertial microfluidic device was fabricated by assembling sixteen precisely aligned polymer films into four functional modules with the multilayer laser-processing technique (see Fig. S5). This fabrication technique can be briefly described as a four-step procedure: (1) each polymer film was patterned using an ultraviolet laser cutting system (TH-UV200A, Tianhong Laser). The detailed patterns of the polymer films are shown in Fig. S5a. (2) Each module had a sandwich structure consisting of three layers: top cover, bottom cover, and channel layer. To generate the enclosed channel, three laser-patterned layers were bonded through the plasma-activated bonding. (3) To form slanted spiral channels with different heights (inner wall height of 70 μm and outer wall height of 160 μm), two polymer films of corresponding thicknesses were patterned and assembled into a single layer. (4) The bonding of different modules was achieved using optically transparent PET double-sided adhesive films with a thickness of 50 μm. The complete device was assembled through sequential layer-by-layer alignment following the orders illustrated in the exploded-view schematic: serpentine sorting module (base layer), flow distribution module, hydraulic resistance matching module, and finally the spiral sorting module (topmost layer) (Fig. S5a-b). Critical alignment between layers was maintained within ±20 μm tolerance using a custom jig. This multilayer fabrication approach enables the integration of complex fluidic functions, while maintaining a total device thickness of <2 mm (Fig. S5c).

Polystyrene microparticle preparation

To evaluate the focusing and sorting performances of cells in clinical effusions, polystyrene microparticles (Thermo Fisher Scientific) with four distinct diameters (7, 10, 15, and 25 µm) were employed. The particle suspensions were diluted to a concentration of 105 particles/mL using phosphate-buffered saline (PBS, 0.01 M, Sigma-Aldrich) containing 1% (w/v) Pluronic F-127 (Sigma-Aldrich) to prevent particle aggregation.

Tumor cell culture and preparation

Human lung tumor cells (A549 cell line) were cultured in high-glucose Dulbecco’s Modified Eagle Medium (DMEM, Thermo Fisher Scientific) supplemented with 10% fetal bovine serum (FBS, Thermo Fisher Scientific) and 1% penicillin-streptomycin (Thermo Fisher Scientific). Cells were maintained at 37 °C in a humidified 5% CO2 atmosphere using an incubator (Forma 381, Thermo Fisher Scientific). Upon reaching 80-90% confluence, adherent cells were detached using 0.05% trypsin-EDTA solution (Thermo Fisher Scientific), centrifuged at 200 × g for 5 min, and resuspended in PBS buffer.

The cell concentrations used in spiking experiments were chosen within an optimized range (total cell concentration ~10⁵–10⁶ cells/mL) for inertial focusing, as established in our prior studies with similar channel geometries38,39, ensuring stable operation across the expected concentration range of processed clinical samples.

Clinical effusion sample processing

Clinical effusion samples were collected from metastatic lung cancer patients under protocols approved by the Institutional Ethics Committee (IEC) of Zhongda Hospital (approval number: 2020ZDSYLL043). To preserve cell cluster integrity, viscous supernatants were removed via gravity sedimentation without centrifugation. After 30 minutes of settling, approximately 60–70% of the supernatant was gently aspirated. The cell sediment was then resuspended by adding an equal volume of PBS, thereby maintaining the original cell concentration prior to microfluidic processing. Cellular viability >95% was confirmed by trypan blue exclusion prior to experiments.

Immunofluorescence staining and cell characterization

Following microfluidic separation, cells from all outlets were fixed onto poly-L-lysine-coated slides (P4981, Thermo Fisher Scientific) using ice-cold methanol and subjected to immunofluorescence staining. After blocking with 10% normal goat serum, tumor cells were specifically labeled with FITC-conjugated pan-cytokeratin antibody (1:100, MA5-13156), while leukocytes were identified using APC-conjugated CD45 antibody (1:50, 368512), with both incubations performed overnight at 4 °C. Cell nuclei were counterstained with DAPI-containing ProLong Gold Antifade Mountant (P36935), enabling clear differentiation of tumor cells (Pan-CK+/CD45-/DAPI+) from leukocytes (Pan-CK-/CD45+/DAPI+) under microscopy. MTCCs were defined as Pan-CK+/CD45−/DAPI+ aggregates containing ≥2 physically connected tumor cells (direct cell–cell contact) and counted as one cluster event during microscopic enumeration. This standardized protocol ensured reliable phenotypic identification, while maintaining cellular morphology for downstream analysis.

Experimental setup and image acquisition

The experimental setup comprised three integrated components: (1) a fluidic delivery unit featuring a precision syringe pump (Legato 270, KD Scientific) calibrated to maintain stable flow rates (±2% accuracy), (2) an optical imaging system combining an inverted microscope (IX71, Olympus) with a CCD camera (Retiga EXi, QImaging), and (3) an integrated microfluidic device securely mounted in a transparent acrylic fixture with laser-cut silicone gaskets to prevent leakage.

During the experiments, the system was operated at controlled flow rates of 4.8-10.4 mL/min and particle/cell trajectories were recorded near the outlet regions. Time-series images were acquired and processed to generate z-stack projections from 100 consecutive frames using ImageJ (v1.53, NIH). Cell concentrations in the initial samples and collected samples were quantified using an automated fluorescence cell counter (Countess II FL, Thermo Fisher Scientific).