Fast and efficient microfluidic cell filter for isolation of circulating tumor cells from unprocessed whole blood of colorectal cancer patients

Liquid biopsy offers unique opportunities for low invasive diagnosis, real-time patient monitoring and treatment selection. The phenotypic and molecular profile of circulating tumor cells (CTCs) can provide key information about the biology of tumor cells, contributing to personalized therapy. CTC isolation is still challenging, mainly due to their heterogeneity and rarity. To overcome this limitation, a microfluidic chip for label-free isolation of CTCs from peripheral blood was developed. This device, the CROSS chip, captures CTCs based on their size and deformability with an efficiency of 70%. Using 2 chips, 7.5 ml of whole blood are processed in 47 minutes with high purity, as compared to similar technologies and assessed by in situ immunofluorescence. The CROSS chip performance was compared to the CellSearch system in a set of metastatic colorectal cancer patients, resulting in higher capture of DAPI+/CK+/CD45− CTCs in all individuals tested. Importantly, CTC enumeration by CROSS chip enabled stratification of patients with different prognosis. Lastly, cells isolated in the CROSS chip were lysed and further subjected to molecular characterization by droplet digital PCR, which revealed a mutation in the APC gene for most patient samples analyzed, confirming their colorectal origin and the versatility of the technology for downstream applications.

www.nature.com/scientificreports www.nature.com/scientificreports/ mutation of the APC gene (c.4348C > T), which is highly frequent in CRC patients. Due to the limited amount of starting genetic material available, this analysis was performed by ddPCR. This APC mutation was found in 7 out of the 9 patients analyzed, which confirmed the tumor origin of the cells isolated by the CROSS chip (Fig. 4).
Clinical data correlation and overall survival. The number of CTCs enumerated by the CellSearch test was less than 3 CTCs/7.5 ml of whole blood for all samples analyzed (Fig. 3), i.e, below the established cut off for CRC using the CellSearch technology. Considering these data, patients could not be divided in different prognostic groups and all were classified as having good prognosis. However, with the CROSS chip, the CTC number obtained was higher in every patient and thus a possible correlation between CTC enumeration and disease prognosis was investigated. Patients were grouped in good or bad prognosis according to the number of isolated CTCs by the microfluidic device and using the cut-off value defined by CellSearch (<or ≥3 CTCs/7.5 ml of whole blood respectively). As illustrated in Fig. 5A, and according to a Kaplan Meier analysis with a 95% CI, a clear trend for shorter overall survival was observed for patients with ≥3 CTCs/7.5 ml of whole blood than those with <3 CTCs/7.5 mL blood, although not statistically significant (p = 0.3812). Remarkably, defining an alternative cut off of ≥7 CTCs/7.5 ml of whole blood, the CROSS chip is able to discriminate patients with good prognosis from those facing an unfavorable outcome (CTCs ≥ 7) (p = 0.0049), with a greater survival of 242 days (Fig. 5B).

Discussion
The molecular characterization of a tumor, typically performed on a tissue biopsy, can help therapeutic reasoning and significantly impact the disease outcome. However, tumor heterogeneity and fast evolving dynamics can lead to genetic alterations, making clinical decisions based on historical biopsy data suboptimal 45 . Additionally, tissue biopsies are invasive and not always available, in contrast to CTC-based liquid biopsies. Unraveling the molecular profile of CTCs can provide key information into the biology of the tumor cells and thus improve personalized therapy 2 . However, the use of CTCs as a clinical biomarker is still on hold, given their extremely low frequency and great plasticity 46,47 , making CTC isolation technically challenging. Numerous approaches have been developed to capture these rare cells 2,6,42 , but with disparate results mostly due to ambiguous CTC classification criteria and lack of standard sample preparation protocols 11 . In this work, the CROSS microfluidic filter was developed aiming at rapid and unbiased isolation of unfixed CTCs with high efficiency and purity. This system was tested in a clinical setting and compared to CellSearch using a panel of metastatic CRC patients. Lastly, cells isolated using the CROSS device were screened for the presence of a specific mutation of the APC gene, highly frequent in CRC patients, to confirm their malignant origin.
To assess the performance of the CROSS chip, spiking experiments were carried out, demonstrating a high capture efficiency (70%) and purity (7.2%), when using whole blood, in contrast to several size-based technologies that display efficiencies of 40-60% [48][49][50] . Improved recoveries (up to 95%) have been reported by other label-free methodologies, but using pre-fixed cells 51 , long processing times 52,53 , RBC-depleted 54,55 or diluted blood 56 which favor sample loss. The Vortex chip, currently commercially available (Vortex Biosciences), demonstrated high purity (57-94%) but low capture efficiency (up to 37%) of spiked MCF7 breast cancer cells when using diluted blood 57,58 . On the other hand, the ApoStream (ApoCell) was reported to capture up to 75% of spiked renal tumor cells SKOV3 with only 1% purity in processed blood 59 . Warkiani and colleagues showed recoveries of spiked cells around 85% and high depletion of WBCs (99%) for lysed blood samples processed in a curvilinear microchannel chip 60 , now commercialized as the ClearCell FX system (Clearbridge Biomedics). Still, a wide variability in CTC purity was also observed (0.1-86%) 54 . Other commercial microfiltration systems have been reported, such as ScreenCell (ScreenCell) 27 and CellSieve TM (Creatv MicroTech) 34 which offer higher throughput (>5 ml/min) 42 than the CROSS chip. Yet, cell damage, low CTC recoveries and filter clogging are major concerns of mechanical filtration devices given their high flow rate and filtration pressure 28,30 . Notably, all systems described above require sample pre-processing and struggle to discriminate CTCs from WBCs of larger size, often rendering poor www.nature.com/scientificreports www.nature.com/scientificreports/ isolation of small CTCs or large amount of WBC contamination 61 . To solve this issue, cell deformability can be further explored to discriminate CTCs from similarly sized leukocytes. In this context, capture efficiency rates of 62.5% or higher (>80%) together with minimal leukocyte contamination, have been shown with an optimization of the Parsortix TM (ANGLE) platform 48 or the Cellsee TM (Celsee Diagnostics) device 51 , using diluted or pre-fixed blood samples spiked with prostate or breast cancer cells, respectively. When using Parsortix and whole blood, capture efficiencies reached up to 70%, similar to that obtained in this work, but only for large size cells (>20 µm) 62 . Instead, with the T-24 bladder cancer cell line (average cell diameter of 18 µm) cell retention dropped to 42% 62 . Moreover, Parsortix, which has a size-restricted separation gap of 10 µm, was limited to running 4 ml of whole blood with a relatively slow sample processing speed 62 , in contrast to the 5 µm-gap of the CROSS chip that processes 7.5 ml in just 47 min. Further systems have shown good ability at isolating CTCs from whole blood, as reported using the Labyrinth chip, which in a double run isolates 91% of CTCs while retaining 663 WBCs per ml analyzed 63 .
Our geometry, combined with our surface treatment, makes for large aspect ratio slippery filters that favor CTC entrapment, while even large white blood cells can be eliminated, reaching a compromise between efficiency, speed and purity. This is due to our anisotropic 5 μm wide/20 μm high filters that, while having smaller width www.nature.com/scientificreports www.nature.com/scientificreports/  www.nature.com/scientificreports www.nature.com/scientificreports/ than other filtration-based systems, allow the cells to deform in the vertical axis and squeeze through, retaining only cells which nucleus cannot deform, favoring the trapping of larger cells with high nucleus to cytoplasm ratio. Having just one filtering row prevents air bubble formation inside the device and pre-filters layout avoids clogging that usually takes place when processing whole blood. Although cytomorphological differences in cultured cancer cells and patient derived-CTCs have been acknowledged for prostate cancer 64 , with CTCs being considerably smaller, others have shown that the SW480 cells used in this study are appropriate models to investigate size-based CTCs enrichment systems, as their median diameter (11-13 μm) is very similar to that of colorectal CTCs (11 μm, as found by CellSearch) 65 . Additionally, the filter size of the CROSS chip is smaller (5 μm) than most systems previously reported 33 , including the commercially available ISET (Rarecells; 8 μm) 26 , ScreenCell (ScreenCell; 6.5 or 7.5 μm) 27 and SmartBiopsy TM (CytoGen; 6.5 × 6.5 μm 2 square pore) 66 . Hence, it has the potential to retain even smaller/more deformable CTCs, as is the case of mesenchymal and stem-cell-like CTCs 67 . This is in agreement with a recent report that uses a similar architecture for selective CTC isolation and reports on the presence of circulating cancer stem cells 68 .
Preclinical validation of our system was next pursued through a comparative blind study between the CROSS chip and CellSearch test, the gold standard for CTCs enumeration in clinical settings. Importantly, whole blood samples were introduced and filtered directly in the CROSS chip, avoiding sample pre-processing steps performed by numerous CTC isolation techniques including CellSearch 6,51,69 . Notably, using 2 CROSS chips we were able to rapidly (in 47 minutes) process whole blood samples from a set of metastatic CRC patients, and capture a higher number of CTCs than CellSearch in all individuals analyzed, which further reinforces the clinical potential of this system. Immunofluorescence staining of trapped cells corroborated a high capture efficiency and purity of CTCs. The total number of CTCs identified in the CROSS chips was further verified and validated by a technical expert routinely involved in the analysis of CellSearch data. Of the nine patient samples analyzed using CellSearch, all patients were classified as having good prognosis, since the CTC count was below the cut off. Nonetheless, four of the nine samples (44%) had CTCs only detectable by the CROSS chip. As for the other five samples, a great discrepancy was observed in CTC enumeration, with CellSearch reporting 1-2 CTCs and the CROSS chip ranging from 2-40 CTCs (average 19.8 CTCs). These results suggest that the isolation of epithelial CTCs using the CROSS chip is more efficient (p = 0.0039) and sensitive than CellSearch. Interestingly, Vimentin+/CD45− cells were also found retained in the CROSS device, indicating entrapment of not just epithelial-like CTCs but also cells with different phenotypes. When accounted for, this will increase even further the number of isolated CTCs and provide additional information of the disease. In fact, a recent study using Parsortix described the isolation of mesenchymal-like prostate CTCs, whose number correlated with worse prognosis 70 . Other microfluidic systems, such as the Vortex HT chip 49 , Parsortix 48 and Labyrinth 63 have also reported the capture of heterogeneous CTC subpopulations expressing epithelial, mesenchymal, EMT and/or cancer stem cell markers. The capacity of the CROSS device to isolate not only single CTCs but also CTC clusters, similarly to other systems such as Parsortix 48 and ClearCell FX 60 , holds great potential as the later have been correlated with higher invasive capacity 71 .
In the scope of the comparative study herein presented, considering the established cut off for bad prognosis in CRC used by the CellSearch technology -i.e. ≥3 CTCs/7.5 ml of whole blood -, the results obtained by CellSearch were negative for all the patients analyzed. In contrast, due to the higher sensitivity, the results obtained with the CROSS chip suggest a new cut off (≥7 CTCs/7.5 ml of whole blood) that stratifies the patients in 2 very well defined populations with overall survival differences higher than 200 days. Yet, further studies on larger cohorts of patients are required to clarify the clinical relevance of this method for CRC monitoring and characterization.
Following CTC isolation, it is of outmost importance to characterize the isolated cells and confirm their tumor origin. Furthermore, this molecular and phenotypic characterization would provide clinically relevant information and be of great utility for therapeutic reasoning. Importantly, the CROSS chip allows downstream molecular analysis of the trapped cells (ddPCR, qPCR, etc), with increased sensitivity given the sample purity, as compared to other technologies 6 . To confirm the malignant origin of the cells isolated by the CROSS chip, the mutational status of APC, a tumor suppressor gene that regulates cell cycle and WNT signaling, was evaluated. This is the first time that APC status is described in CTCs isolated by a microfluidic system. For that, we selected a somatic non-sense mutation with high frequency of mutation among patients population. APC is the most frequently mutated gene in sporadic CRC, affecting up to 60% of CRC patients [72][73][74][75] . Moreover, a strong association www.nature.com/scientificreports www.nature.com/scientificreports/ between mutations in APC and other genes such as KRAS or BRAF and colon cancer initiation has in fact been established 69,[76][77][78][79] . In addition, the APC gene has been linked to tumor initiation and the frequency of mutation is maintained by the passage of benign to malignant tumors 80 . Some authors suggested that the APC mutation has a relevant role in providing a selective advantage, through the activation of the Wnt signal transduction pathway and the chromosomal instability in the tumor cell 81 . Notably, APC mutations were detected in CROSS chip-isolated CTCs from 7 out of 9 patients, by ddPCR, even using DNA amounts as low as 0,065 ng/µl. Indeed, ddPCR technique has demonstrated superior sensitivity to detect clinically relevant mutations at very low concentration in liquid biopsies from patients with different malignancies 82,83 . Thus, our findings are in agreement with the overall frequency of APC mutation in CRC 84 . Nevertheless, false-negative results cannot be ruled out due to the low amount of starting DNA. A recent work confirmed that in all CRC patients analyzed, the mutational status of APC in both CTCs and the primary tumor matched, with 60% concordance 85 . However, to the best of our knowledge this is the first study reporting the analysis of APC alterations by ddPCR in CTCs isolated from CRC patients. In a similar study, APC mutations were investigated in circulating DNA using the BEAMing technology and were detected in >60% of CRC patients 86 . On an additional note, CRC-derived CTCs isolated by the ScreenCell size-based device have also been screened for mutations in the KRAS gene using ddPCR, which were observed in 57% of the cases 87 . In fact, ddPCR is commonly used for ctDNA analysis in oncology and, for example, KRAS is analyzed in the clinical routine as it conditions treatment selection for the patients who present KRAS mutation.
In summary, although several CTC isolation systems have been described and a fraction even reached commercialization as automated platforms 42 , it should be considered that validation with clinical samples has not always been performed 88,89 . The blood of a cancer patient shows different features compared to that of a healthy donor regarding density or clotting, influencing cell isolation performance of the technologies under investigation. Furthermore, of those studies including patient samples, not all performed a comparison with the only FDA-approved technology CellSearch 66 , crucial to provide an estimated number of captured CTC for each sample, as a positive control. The CROSS chip described herein displayed higher sensitivity than the gold standard, without the need of any sample pre-processing, while allowing downstream molecular analysis, key in a clinical setting. The versatility of this low cost device is further demonstrated by its ability to process blood straight after drawing or more than 24 h post collection.
Besides allowing phenotypic and molecular characterization of captured cells, further advantages of the CROSS chip include easy cell recovery in a very small volume (the internal volume of the system is below 100 μl), by simply inverting the flow and without inducing significant cell damage, particularly important for cell culturing and further downstream applications. This particularity, can enhance the possibilities to establish CTC cultures, which has been to date challenging due mainly to low cell density 90 . Also, the CROSS chip can be used for the study of other cancer types, particularly in diseases where the scarcity of epithelial-like CTCs hinders the applicability of CellSearch. Moreover, the simple setup and protocol used, allows for the deployment of this system in any research or pathology laboratory without the need of any special equipment, and only requiring a syringe pump for CTC isolation.
The CROSS chip has shown limitations in volume capacity but, due to the inner multiplex capacity of the microfluidic systems, it is possible to redesign the system and increase the surface area of the device to analyze higher volumes of whole blood in the same or even less time, having undeniable potential for early cancer diagnosis. Also, further developments will integrate the design in an all-in-one system just comprising a single outlet in a smaller glass slide to facilitate image acquisition. Even though recent studies have shown microfluidic chips able to isolate CTCs with outstanding efficiency and purity from unprocessed blood samples 63 , our latest tests have indicated that the analysis of fresh samples renders higher isolation yields and better quality of the samples, as expected 91 .
From this work we can conclude that the CROSS chip is able to rapidly isolate unfixed CTCs from whole blood with high efficiency and purity, while enabling CTC recovery. Importantly, it shows higher sensitivity than CellSearch when isolating CTCs from metastatic CRC patients, capturing a higher number of cells, even in samples considered negative by the gold standard. The findings obtained suggest that the CROSS chip may allow a better discrimination of patients with poorer prognosis, highlighting its potential as a powerful tool for liquid biopsy studies in CRC and other types of cancer. Moreover, ddPCR confirmed the tumor origin of the isolated cells, and paves the way for further molecular downstream analysis.

Methods
Microfluidic device design and fabrication. The CROSS microdevice was designed to split the blood equally in 4 different modules (Fig. 1A). Each module is able to process a maximum of 1 ml of whole blood and contains a set of pre-filters and cell isolation filters (Fig. 1B). Across the middle section of each module, a single row of 700 anisotropic micropillars with diameter 25 μm and spaced 5 μm constitutes the cell filtering area (Fig. 1C). The gap size, geometry and aspect ratio was carefully chosen to allow blood cells to deform and gently flow through, while retaining larger or more rigid cells in the filter. The pre-filters present 120 μm gaps to prevent large clumps or debris from clogging the setup (Fig. 1D). Each microdevice holds an approximate volume of 100 μl. Cells can be retrieved from the system by simply inverting the flow. Surface coating is decisive in both cell isolation and retrieval, as it is crucial to prevent cell attachment to maximize cell purity and recovery.
The microfluidic masters were designed in 2D AutoCAD software (Autodesk, USA) and fabricated on a 200 mm silicon wafer using photolithography and deep reactive ion etching. Briefly, the silicon wafer (P/Boron, <100>, Siegert Wafer, Germany) was rinsed with deionized water, dehydrated at 150 °C and exposed to hexamethyldisilazane (HDMS, Sigma Aldrich, USA) vapour prime to improve the adhesion of the photoresist to the sample. Later, the wafer was spun coated with 2.2 μm of AZP4110 (Microchemicals GmbH, Germany), using a SÜSS MicroTec optical track (SÜSS MicroTec AG, Germany). The pattern was transferred onto the coated wafer (2019) 9:8032 | https://doi.org/10.1038/s41598-019-44401-1 www.nature.com/scientificreports www.nature.com/scientificreports/ using a Direct Write Laser system (DWL 2.0 Heidelberg, Germany) with an Hg laser energy of 95% and focus −50. Following the post bake, the exposed photoresist was developed with AZ400K (Microchemicals GmbH, Germany), and the wafer was rinsed with deionized water and dried. The pattern was then etched with sulfur hexafluoride (SF6, Sigma Aldrich, USA) by Silicon Deep Reactive Ion Etching (STPS Pegasus, United Kingdom), and exposed areas passivated with octafluorocyclobutane (C4F8, Sigma Aldrich, USA). Trench depth was measured in between steps using an optical profilometer (OPM profilometer, Oceon Optics NanoCalc XR) until the desired depth of 20 μm was reached. Residues were stripped using oxygen plasma and the master was characterized by means of Scanning Electron Microscopy (Quanta SEM, FEI, USA). Finally, the wafer was diced into the individual masters using a DAD 3350 Dicing Saw (Disco, Japan) and cleaned with Isopropyl alcohol (IPA, Sigma Aldrich, USA), rinsed with deionized water and dried at 150 °C on a hot plate.
Prior to master replication, the wafer was hydrophobized with a vapor-phase treatment in trichloro(1H,1H, 2H,2H-perfluorooctyl)silane (Sigma Aldrich, USA) for 1 h in a desiccator, and cured for another hour at 65 °C. Polydimethylsiloxane (PDMS, Ellsworth Adhesives Iberica, Spain) was mixed at 1:10 ratio, degassed, poured over the master, degassed again and cured at 70 °C for 2 h. After curing, the PDMS was unmolded and inlet and outlets were punched. Finally, clean glass slides and PDMS replicas were treated with oxygen plasma at low power for 15 s and subsequently brought in contact to produce irreversible bonding.
Upon activation with oxygen plasma, the microfluidic devices were connected to a syringe pump and filled with ethanol at 100 μl/min to enhance the wettability; then rinsed with 10 mM Phosphate Buffer Saline (PBS, Sigma Aldrich, USA), and later treated with 1% Pluronic F-127 (Sigma Aldrich, USA) overnight to avoid unspecific attachment of cells onto the channel surface. Considering its cross-shaped design, this CTC isolation device has been designated CROSS chip. (1). At the same time, the capacity of the system to deplete the WBC population was calculated by comparing the total number Calcein−/DAPI+ events against the theoretical amount of WBCs in the total volume of blood analyzed (7.5 × 10 6 WBCs/ml), according to Eq. (2). Finally, the purity of the cell population isolated in the system was determined using Eq. (3). Experiments were done in triplicate.  DNA extraction and ddPCR analysis. Extraction of genomic DNA from cells retained in the microfluidic devices was performed using AllPrep DNA/RNA Mini Kit (Qiagen, USA). Firstly, cells were lysed upon injection of a lysis buffer (Buffer RLT) at 80 µl/min followed by 5 min incubation and a second injection of the same buffer at 250 µL/min to collect all cell content. Subsequent steps were performed according to the manufacturer's recommendations. Quantification of the extracted genomic DNA was performed with the Quantifluor ONE dsDNA System using Quantus Fluorometer (Promega, USA).

CTC Isolation Efficiency
Absolute quantification of APC transcript was performed by ddPCR analysis (QX200 ™ Droplet Digital ™ PCR System, Bio-Rad, USA) at the Universitat Autònoma de Barcelona (UAB) Scientific Technical Services (Barcelona, Spain). Prior to quantification, samples were digested (HaeIII, Sigma-Aldrich, USA) and preamplified (Sso Advanced Preamp Supermix, Bio-Rad, USA). ddPCR experiments were performed using probes dHsaCP2500509 and dHsaCP2500508 for APC. The droplets were quantified using the Bio-Rad Quantisoft software. Two replicates per sample were performed. Statistical analysis. Statistical analysis was performed using GraphPad Prism software, version 6.01 (GraphPad Software, USA). The Wilcoxon signed rank test (95% confidence intervals) was used to compare CTC enumeration using CellSearch test versus the CROSS filter from the same metastatic patient, whereas Kaplan Meier method was used for survival analysis from time of sample collection. Findings of p < 0.05 were considered statistically significant.

Data Availability
The datasets generated during and/or analysed during the current study are not publicly available, being now exclusively licenced to a for-profit company, but are available from the corresponding author on reasonable request.