Label-free enrichment of rare unconventional circulating neoplastic cells using a microfluidic dielectrophoretic sorting device

Design and validation strategy of microfluidic DEP sorting device

Design

The concept of enriching a target cell population by depleting non-relevant cells is shown schematically in Fig. 1. First, the sample is infused into the microfluidic DEP sorting device, composed of a microfluidic channel and interdigitated electrode arrays. The microfluidic device was manufactured by standard photolithographic and lift-off techniques where a silicon substrate was patterned with an interdigitated electrode layer and a polydimethylsiloxane (PDMS) layer was bonded on top to form a fluidic channel. To allow cells throughout the channel’s height to be acted upon by DEP forces, the chamber height was designed to be 25 µm or approximately twice the average size of PBMCs. Under continuous flow the sample is centrally confined by two sheath flows providing a small variation of initial speed and position. The interdigitated electrodes were designed to have a width and spacing of 35 µm, approximately 2–3 times the size of PBMCs, to minimize cells being captured on the electrodes by DEP forces (Supplementary Fig. 1). The V-shape electrode geometry was designed to deflect PBMCs in a high cell density environment by utilizing DEP forces to deflect cells along the electrodes while flow forces direct cells along the length of the device^33,34,35. Cells experiencing strong DEP forces are depleted to the side outlets, while less responsive cells are enriched in the center channel (Fig. 1a). In order to maximize the DEP response and minimize cells trapped on electrodes, a conductive buffer at 145 mS/m (approximately 0.1x PBS) was used during cell enrichment processing for both the sample and sheath inlets. The buffer was chosen to balance cell viability, DEP reactivity, and electrode corrosion^31,36. For ease of use and rapid troubleshooting the complete setup was placed under a microscope for continuous viewing where the fluidic channel is visible under a standard 5x objective lens. This setup allows for sample compatibility with genotypic or phenotypic downstream analysis (Fig. 1a) which can be helpful for understanding cellular clinical significance³⁷.

a Schematic representation of CHC enrichment from patient blood. (1) PBMCs are isolated from 2 ml of whole blood. (2) DEP microfluidic device is used to enrich for CHCs and samples is collected from outlets. (3) Downstream analysis, both phenotypic and genotypic, can be performed to obtain clinically relevant information. Created with BioRender.com. b Fluorescent images of microfluidic channel showing differential response of polystyrene beads, 10 µm (green) and 2 µm (blue). Part I represents the inlet and part II represent the outlet. Scale bar is 500 µm. c Percent of beads collected from the center outlet at an optimal condition (5 MHz), a sub-optimal condition (300 kHz), and with no applied DEP.

Validation

To demonstrate the concept of enriching by depleting a target population we used a sample mix composed of 2 µm (blue) and 10 µm (green) fluorescent polystyrene beads of identical chemistry (Fig. 1b, c). The sample mix was introduced to the inlet of the microfluidic DEP sorting device with AC voltage applied to the interdigitated electrodes, and particle movement was monitored by live imaging. We defined the optimal condition as the parameters where one type of particle was enriched by depletion of the other particle. To find this condition, the frequency was swept from 0.1 MHz to 10 MHz at 0.1 MHz increments at constant voltage. Particles were collected at the outlets and analyzed by flow cytometry.

Under the optimal frequency, both bead populations were closely aligned with the center of the channel while near the sample inlet (Fig. 1b, part I), however only the 10 µm beads were displaced further downstream by DEP forces revealing particle trajectory bias over time and space (Fig. 1b, part II). Under this condition, 95% of the 10 µm beads were depleted from the center due to the application of strong positive DEP forces, while less than 1% of the 2 µm beads were depleted to the sides (Fig. 1c). Figure 1c also shows an example at one of the sub-optimal frequencies that only 19.5% of 10 µm were depleted, suggesting that the magnitude of the DEP force was smaller than the dominant flow forces. To confirm that DEP forces were the primary driver of bead depletion, the device was run without applied DEP and resulted in 99% of both bead populations remaining in the center unaffected by drift. This workflow demonstrated that depletion of one population by DEP can result in the collection of targets with high purity under proper working parameters. To understand how this applied to a complex biological sample mix, we next tested this process on PBMCs and cell lines.

Optimization of PBMC depletion and hybrid cell line enrichment

PBMC DEP response characterization

As DEP conditions can be optimized to separate discrete physiologic objects, we set out to establish the optimal condition where hybrid cells could be enriched from PBMCs. Cells are biologic entities, thus their conductivity, permittivity, and physiological state are the main drivers of differential DEP response³⁸. Healthy PBMCs are composed of heterogeneous populations, but tend to be relatively similar in nature across individuals, therefore healthy PBMCs act as an ideal object to be optimized for depletion. PBMCs are mainly comprised of lymphocytes and monocytes. Of these, T-cells represent the majority (70–80%) of the lymphocytes, therefore we hypothesized that of all healthy PBMCs depleting T-cells would be the primary contributor of neoplastic cell enrichment. First, to understand the DEP response of PBMCs as a collective of cells and as individual subpopulations, we conducted the workflow using PBMCs isolated from whole blood (n = 5) from healthy subjects. PBMCs were diluted in DEP buffer (1 × 10⁷ cells/mL), and 1 × 10⁶ cells were loaded into the central inlet for each parameter, then subjected to device processing for at least 30 min. Cells were not responsive to DEP at low frequencies, thus we focused analyses when cells were visually responsive, between 13 and 17 MHz at a constant voltage of 8 Vpp. We selected this voltage because cells were minimally reactive below 7 Vpp but were trapped on the electrodes at voltages greater than 9 Vpp. For each frequency, cells were collected from all three outlets, then stained with antibodies against CD45 (pan-leukocyte), CD3 (T-cell receptor), CD14 (macrophage), and CD19 (pan B-cell) (Fig. 2a). Depletion of each cell type was calculated by flow cytometry analysis of the outlets where the percent of CD3⁺, CD14⁺, or CD19⁺ cells in the side channels was divided by the percent of the cell of interest from the total cells processed (CD45⁺). A similar trend was observed for all populations, where 15 MHz resulted in maximum depletion from the center channel, which was most pronounced for CD3⁺ cells (99.3%), followed by CD14⁺ cells (92.5%), and CD19⁺ cells (88.5%). Ultimately, only 1.5% of the initial number of PBMCs (CD45⁺) were collected in the central outlet (Fig. 2a). Of the subpopulations, we found CD3⁺ T-cells were the most responsive and had similar depletion levels between 14, 15, and 16 MHz at 97%, 99.3%, and 98.3%, respectively. In contrast, depletion of CD19⁺ and CD14⁺ cells significantly changed from 13 MHz to 15 MHz, suggesting that these cell populations were more sensitive to specific DEP parameters. We corroborated the previous findings by Gascoyne et al. that different cell types have different DEP response profiles^39,40,41. Not only did we observe PBMCs depletion, we found that the PBMCs composition changed post-depletion e.g. there was 7.3-fold and 4.6-fold enrichment of CD19⁺ and CD14⁺ cells in the center channel, respectively. These results suggest that across PBMC phenotypes there is a cell type-dependent DEP response that supports their differential separation and that depletion of CD3⁺ cells could drive enrichment of rare cells.

a PBMC subpopulation DEP respsone as presented by depletion of CD3⁺ (blue line), CD14⁺ (red line), CD19⁺ (green line), and CD45⁺ (purple line) cells across frequencies, 13 MHz to 18 MHz at a constant 8 Vpp by flow cytometry. b Fluorescent images near the inlet (Part I), at the center between batches of electrodes (Part II), and close to the outlet (Part III) of the microfluidic channel revealed depletion of MCF7 (green) and less reactive B16Mϕ (red) at 15 MHz and 8 Vpp. Scale bar is 500 µm. c Percent depletion of B16Mϕ, MCF7, and PBMCs from the center outlet across frequencies, 13 MHz to 19 MHz, at a constant 8 Vpp by flow cytometry. d Center enrichment heatmap with voltage (V) in the y axis and frequency (MHz) in the x axes (n = 1–5, independent experiments). e Percentage side loss heatmap with voltage (V) in the y axis and frequency (MHz) in the x axes (n = 1–5, independent experiments).

In vitro CHC DEP response characterization

Next, to understand the DEP response of malignant cells we selected an established in vitro hybrid cell line which was studied in depth previously by Gast et al. B16Mϕ–RFP, generated through the fusion of B16F10 and bone marrow-derived macrophages as a model of CHCs¹². In addition, a breast cancer cell line, MCF7, was used as a surrogate of traditional CTCs. Previously, Gast et al. characterized the phenotypic hybrid nature of in vitro derived CHCs. Conveniently these hybrids co-expressed nuclear RFP and cytoplasmic YFP facilitating their differentiation from the MCF7-GFP cells by flow cytometry¹². To optimize conditions for hybrid cell enrichment, cancer cells were diluted in DEP buffer at 1 × 10⁷/mL and only 1 × 10⁶ cells were loaded into the microfluidic DEP sorting device for each parameter. To visualize cancer cell DEP response, fluorescent images were taken at three locations (Fig. 2b), near the inlet (part I), the middle after the first batch of electrodes (part II), and close to the outlets (part III). At 15 MHz/8 Vpp, both B16Mϕ−RFP and MCF7-GFP were aligned with the central inlet flow. As the cells flowed through the device, the majority of MCF7 cells depleted to the sides while the majority of B16Mϕ remained in the center. To quantify each sorting parameter, we processed cells in the device for at least 30 min and analyzed all the outlets by flow cytometry. Across all tested frequencies MCF7 cells were more responsive than B16Mϕ cells. MCF7 cells had a minimum response at 13 MHz where 65% of the cells were depleted and a maximum response at 16 MHz where 92% were depleted to the sides (Fig. 2c). In contrast, the minimum DEP response for B16Mϕ was at 19 MHz where only 6% of the cells were depleted and the maximum DEP response was at 15 MHz where only 29.5% of cells were depleted suggesting that hybrid cells have a unique non-responsive DEP profile (Fig. 2c). In parallel, we analyzed the DEP response of healthy PBMCs (CD45⁺) under the same conditions. Again, PBMCs were highly responsive across all frequencies with a minimum response at 19 MHz with 77% of the cells depleted and a maximum response at 15 MHz with 96.5% of the cells depleted from the center channel (Fig. 2c).

Overall, we found a larger differential response between B16Mϕ-RFP and PBMCs as compared to MCF7-GFP cells and PBMCs despite their close proximity in size to PBMCs suggesting that label-free size-independent enrichment was possible within our workflow. This also suggests that B16Mϕ-RFP cells inherently have different dielectric properties likely resulting from their hybrid nature. Although numerous studies focus on CTC enrichment and isolation, these strategies rely on either the inherent large size of CTCs or specific membrane protein expression. CHCs are a promising alternative as they are more abundant than CTCs, however there is a lack of technologies focused on their enrichment, a critical factor in developing CHCs as a cancer biomarker. This data highlights the potential of using DEP to enrich for CHCs and potentially other rare cellular biomarkers that are difficult to differentiate phenotypically.

Hybrid cell line spiked into healthy peripheral blood samples

A number of technologies focused on rare cell analysis rely on the isolation of single cells, which is challenging. Recently, the development of high-throughput single-cell technologies allows for the analysis of thousands of cells and demonstrates the need for rare cell enrichment rather than pure isolation. For most enrichment strategies there is a tradeoff between total enrichment and overall cellular biomarker loss during processing due to a limit of detection (LOD) threshold. To find the optimal frequency and voltage parameters to enrich for CHCs, we evaluated the enrichment of B16Mϕ-RFP cells from PBMCs. PBMCs isolated from healthy subjects were spiked with 50,000 B16Mϕ cells (50,000 per 1 × 10⁶ PBMCs), then subjected to DEP for at least 30 min, and analyzed for cells types isolated from all outlets by flow cytometry. Enrichment of hybrids in the center channel and depletion in the side channels was calculated using flow cytometric data for each outlet (see Eq. 1 and Eq. 2). Unlike other strategies for rare cell isolation focused on the absolute number of recovered rare cells, our strategy utilized the ratio of rare cells to total PBMCs as the rating of enrichment with the goal of enriching samples above the LOD for downstream analytics. This process was repeated for six frequencies, from 13–18 MHz and three voltages for each frequency, 7–9 Vpp. A heatmap was generated to assess center channel hybrid enrichment (fold change) and side channel hybrid loss (percentage loss) taking into account both the frequency and voltage applied to the device (Fig. 2d, e). The enrichment varied from 1.4-fold and 2.8-fold at 13 MHz/7 Vpp and 19 MHz/9 Vpp, respectively, to 18.6-fold and 12.7-fold at 15 MHz/8 Vpp and 15 MHz/9 Vpp, respectively. The highest B16Mϕ enrichment identified was 18.6-fold (n = 5) at 15 MHz/8 Vpp (Fig. 2d). The voltage was a strong driver of the magnitude of DEP response with greater loss of B16Mϕ cells to the sides at higher voltages (Fig. 2e). For example, at 15 MHz, there was 6.5% loss to the sides at 7 Vpp, 18.6% at 8 Vpp, and 22.7% at 9 Vpp. Overall, we found that 15 MHz and 8 Vpp provided optimal DEP parameters for B16Mϕ hybrid enrichment and potentially for patient-derived CHCs. These trends observed follow previously describe bell-shaped DEP response curves⁴². In these models as well as in the data presented here, a narrow range of maximum DEP response can be appreciated. Overall, our unique strategy to deplete healthy PBMCs from the sample drives enrichment of the target tumor cell population.

According to current literature there are approximately 30 CHCs per 500,000 PBMCs in PDAC patients representing only 0.006% of total cells in blood¹³. This overall rarity highlights a clear need for CHC enrichment for downstream analyses such as mutational profiling using ddPCR or genomic surveying using single-cell sequencing, both of which have a limit of detection (LOD) of 0.1%^43,44,45,46. With 18.6-fold enrichment using the described workflow, the microfluidic DEP sorting device described could enrich above 0.1% LOD suggesting there is utility targeting the depletion of healthy PBMCs to facilitate a more comprehensive understanding of poorly understood circulating neoplastic cells.

Finally, since a strong electric field can induce cell death, which would disrupt downstream analyses, we assessed cell viability throughout the DEP workflow. Non-viable cells can lead to aggregation and clogging, which reduces device performance and sensitivity for isolating cellular biomarkers. The low conductive DEP buffer was designed to maintain cell viability while allowing for DEP separation. In order to study cell viability, the effect of DEP buffer alone and under applied DEP was tested. A549 cells were placed in DEP buffer and cell viability was assessed over time using propidium iodide staining via flow cytometry. A549 cells had 98% viability after 30 min in DEP buffer and 97% after 120 min (Supplementary Fig. 2). To measure cell viability post-DEP processing at 15 MHz/8 Vpp, PBMCs were stained with calcien AM from the outlets and immediately measured via flow cytometry (Supplementary Fig. 2). The side outlets had 99% viability while the center outlet had 97% viability, suggesting that enrichment via DEP is compatible with analytic techniques that require cell viability, morphology, and nucleic acid integrity to be maintained.

Label-free Enrichment of KRAS mutant cells from PDAC patient samples

To evaluate the ability of the microfluidic DEP sorting device to provide relevant clinical information when assessing peripheral blood from cancer patients, we set out to demonstrate that DEP-enriched CHCs could be used to evaluate KRAS mutational status. Whole blood from four patients with PDAC (stage III-IV) was collected. All patients had clinically confirmed KRAS^mut primary tumors (Table 1). The majority of PDAC tumors harbor KRAS mutations and its presence in circulation, from circulating cells or cell-free DNA, is clinically relevant for cancer detection⁴⁷. Moreover, our group has previously found that a subpopulation of FACS isolated CHCs have KRAS mutation as detected by ddPCR¹³. To confirm the presence of CHCs in peripheral blood specimen prior to subjection to DEP, PBMCs were isolated from whole blood and analyzed as we have previously reported^12,13. Briefly, PBMCs were adhered to glass slides then stained with antibodies against CD45 and CK, and visualized by fluorescence microscopy (Fig. 3a). The size of CHCs (n = 25) from all patients was 9.05 ± 0.89 µm, and was comparable to the size of PBMCs (n = 25) 8.57 ± 0.89 µm, confirming the non-significant difference in size between these two populations (Fig. 3b). As indicated, this negates the possibility of size-based isolation as a valid isolation method for CHCs further supporting the utility of label-free DEP enrichment¹². In addition, these samples had non-detectable levels of CTCs (CD45⁻/CK⁺) by IHC, further supporting the promising nature of CHCs as an important cancer biomarker due to their detectability and potential for harboring clinically relevant information.

a In situ immunofluorescence microscopy analysis of PBMCs before device processing, arrows indicate presence of CHCs (Hoescht⁺/CD45⁺/CK⁺). Scale bar is 20 µm. b Average diameter (µm) of CHCs and PBMCs from in situ immunofluorescence images. c Average clinical sample DNA extracted from cells collected from all outlets (n = 4, independent experiments). d Percent KRAS^mut copies for each patient (n = 4, independent experiments) from all outlets compared to average KRAS^mut copies for screen-negative controls (n = 5, independent experiments) measured by ddPCR.

To demonstrate DEP-enriched CHCs harbored mutant KRAS, a smaller aliquot of patient PBMCs were subjected to DEP. A pre-sort aliquot was retained as a control (“Pre-sort”, Fig. 3d). Specimens were loaded onto the device (1 × 10⁶ cells) and all samples were processed for at least 1 h at 15 MHz and 8 Vpp, the optimal conditions for PBMC depletion. Cells were collected from side and center outlets and each channel was subjected to DNA extraction for downstream evaluation of KRAS mutations and total DNA was quantified using Qubit^TM kits (Fig. 3c). Based on DNA concentration we found that an average of 92.6% ± 2.07 (n = 4) of the cells were depleted from the center which is comparable to previous results (Supplementary Fig. 3) but significantly different than healthy controls (p = 0.03, unpaired t test). PBMCs from cancer patients can be influenced by disease resulting in modified proportions of its subpopulation or by other conditions like immunosenescence where senescent T-cells exhibit abnormal phenotypes^48,49. These complex factors may influence PBMC DEP response and lead to the utility of such biomarkers. In this way, optimizing the system to deplete healthy PBMCs in a label-free manner may reveal other disease-related cells.

To demonstrate DEP-isolated CHCs harbor detectible KRAS mutations, we subjected isolated DNA to ddPCR. We used PBMCs of five screened negative subjects as controls. For all samples, ddPCR probes for the seven most common KRAS mutations in PDAC were evaluated. In three out of four patients analyzed, mutant KRAS alleles were identified in cells isolated from the center outlet while the side outlets only contained cells expressing wild-type KRAS (Fig. 3d). For all other outlets and the screen-negative controls, the number of KRAS mutant copies detected were below the LOD for ddPCR^43,44,45,46. PDAC3 had the highest number of KRAS mutant copies corresponding to 0.15% of the total DNA, followed by PDAC4 and PDAC1 with 0.13% and 0.1%, respectively. PDAC2 had non-detectable levels of KRAS mutations. Although our cohort is limited in number, we identified KRAS mutations in 75% of evaluated samples (p = 0.048, Fisher Exact test), which is comparable to prior studies that only detected KRAS^mut CTCs in 72% of samples⁵⁰.

The average KRAS mutant copies from the three KRAS positive samples was 0.13%. Typically KRAS positive cells are heterozygous therefore 0.26% of the cells were positive, corresponding to one KRAS mutant cell for every 385 wild-type cells post device processing^51,52,53. Previously Dietz et al. found an average of 30 CHCs (n = 5) for every 500,000 PBMCs in PDAC patients and found that only 9.1% of CHCs were KRAS mutant, therefore the enrichment presented here could be as high as 476-fold. Considering that no CTCs were found by IHC, we believe that KRAS^mut cells are likely CHCs or other important circulating tumor-derived cells. The level of enrichment presented here is sufficiently above the LOD for current single-cell technologies which could allow for a deeper understanding of circulating cancer cells and associated biology.