A Novel Automatic Platform Based on Nanostructured Microuidic Chip for Isolating and Identication of Circulating Tumor Cells

Circulating tumor cell (CTC) is commonly used as biomarker for early detection, prognostication, decision making, and evaluation of the therapeutic ecacy in cancer treatment. Unfortunately, the poor reproducibility and limited sensitivity with the CTC detection have limited its potential impact on clinical management. A reliable automated CTC detection system is therefore needed. We have designed an automated microuidic chip-based CTC detection system consisting of three main parts: V-BioChip, Cell RevealTM enrichment and staining system, and an automatic scanning and locating system. We hypothesized this novel system can reliably detect CTC from clinical specimens. SKOV3 ovarian cancer cell line was used rst to test the reliability of our system. Ten healthy volunteers, 5 patients with benign ovarian tumors, and 8 patients with epithelial ovarian cancer (EOC) were recruited after obtaining written consents to validate the CTC capturing ecacy in the peripheral blood. The capture rates for spiking test in SKOV3 cells were as 48.3% and 89.6% using EpCAM antibody alone and a combined EpCAM antibody and N-cadherin antibody, respectively. The system was sensitive to detection of low cell count and showed a linear relationship with the cell counts in our test range. The sensitivity and specicity were 62.5% and 100% when CTC was used as a biomarker for EOC. Our results demonstrated that this automatic CTC platform has a high capture rate and is feasible for detection of CTCs in EOC. were conducted by using PBMCs suspensions of cancer cells that mimic actual conditions. The donor's blood can interact with the spiked cells and markedly affect the recovery eciency. Our results show that the automated CTC platform combined with V-BioChip and Cell Reveal TM system can achieve excellent CTC capture eciency of 90% by using Combi strategy. Even in low cell count, the system also demonstrated excellent capture eciency and device linearity with R 2 of 96%.


Introduction
Despite recent advances in modern medicine, we are still facing many challenges in cancer diagnosis and treatment. Tissue biopsy is the only con rmatory diagnostic method for suspicious lesions while invasive exploratory surgery is required for cancer staging. Tissue biopsy has its own limitations due to its invasiveness and the fact that some lesions are in di cult-to-reach areas. In addition, tissue biopsy cannot assess the risk of cancer metastasis, the progression of the cancer, and the response to treatment 1 . Furthermore, it is very di cult to detect small metastatic lesions or minimal residual diseases using even the most advanced imaging technology 2 . Therefore, there is an urgent need to develop more precise methods to facilitate screening, diagnosis and treatment of cancers.
CTCs are the cells shed from the primary site of the tumor and then entered the peripheral blood circulation, which has the potential to develop into new tumor foci 3 . Although many steps involved in cancer metastasis are still unclear, CTCs obviously play a critical role in cancer metastasis 4,5 . We can repeatedly take the patient's CTCs as a "real-time liquid biopsy" to predict cancer recurrence, monitor the effect of treatment, and understand potential drug resistance mechanisms 6 . It is estimated that there are 1-10 CTCs per mL of whole blood in the peripheral blood of patients with metastatic cancer 7 . Because of the scarcity of CTCs in the blood, it is not only technically di cult to capture CTCs from the blood of cancer patients, but also requires a lot of blood for the test.
Since 1990, there have been many technological developments and applications to enrich and identify CTCs. However, even though considerable progress has been achieved in recent years, the CTC detection technology is still not widely adopted for clinical practice because of its limitations including the scarcity and fragility of CTCs, the heterogeneity of CTCs, the lack of standard detection methods, and the need for sensitivity and speci city 8 . Most laboratories detect CTC manually which has low interrater reliability and low throughput. Therefore, the key point to overcome the current limitations is to develop an automated CTC detection technology providing unbiased result and high sensitivity will bene t clinicians taking care of the cancer patients. The rapidly improving micro uidic-based technology over the last ten years can be a sensitive, e cient, and fully automated system with a portable size 8, 9 . Ovarian cancer is a common cancer in women with extremely poor prognosis. With the current diagnostic techniques, more than half of ovarian cancers are diagnosed at advanced stage 10 . Therefore, there is an urgent need for new diagnostic tools to assist in the management of ovarian cancer, including: early diagnosis, evaluation before and after surgery, prognostic judgment, detection of minimal residual disease or early recurrence, assessment of treatment response, and even drug selection. The purpose of this study is to evaluate the applicability of a newly designed, fully automated micro uidic CTC platform, in the clinical evaluation of ovarian cancer.
In the present study, we used an automatic micro uidic immunoa nity-based system to achieve reliable reproducibility with high throughput. This automated system consists of several parts, including a nanostructured micro uidic biochip, an automatic CTC enrichment and staining system, and an automatic CTC scanning and locating system. The capture e ciency of the automatic system was tested by spiking tests of SKOV3 ovarian cancer cell line. A preliminary test for clinical feasibility on epithelial ovarian cancer (EOC) from consented human volunteers was also conducted.

Results
The enrichment and staining processes were achieved within as short as 4 hours in the Cell Reveal TM system but overnight handing will not change the readouts. The automatic on-chip scanning and cell locating by CAT took about 15 minutes.
V-BioChip has excellent capture ability for target cells. Figure 1a demonstrates the gross appearance of a V-BioChip and the image of an SKOV3 cell captured by V-BioChip under a scanning electron microscope (SEM) at 5,000X magni cation (Fig 1b). It can be seen from the SEM image that there were many microvilli on the surface of the SKOV3 cell, and some microvilli rmly adhered to the nanopillars of V-BioChip. The captured SKOV3 cells had intact morphology without distortion, which means that V-BioChip can capture the targeted cells under the automatic process without obvious damage.
Because of the scarcity of CTC, capture e ciency is critical for a successful CTC test platform. With EpCAM antibody alone as the capture antibody, the recovery rate was 48.3% with a SD of 11.7%. Considering the mesenchymal characteristics of SKOV3 cells, we used a mixture of an epithelial marker antibody (anti-EpCAM antibody) as well as an EMT marker antibody (anti-N-cadherin antibody) to make a capture antibody cocktail (Combi). The Combi strategy markedly increased the capture rate to 89.6% (SD: 11.8%) (Fig 2a).
The analysis of the linear relationship between the number of spiking cells and the number of captured cells at low spiking cell numbers is also an important indicator of the capture e ciency of the CTC platform. The present study showed a high recovery rate and a linear relationship even in low spiking cell numbers (39-625). The linear regression was y=0.7888x + 1.5009 (R2= 0.9639) with an average recovery rates 73.5% (SD: 11.8%) (Fig 2b).
The results of clinical validity tests are summarized in Table 1 and Fig 3. No CTCs were detected in the peripheral blood of the ten healthy subjects. As for patients with benign ovarian tumors and EOC, the average peripheral blood CTCs were 1.40 (1.40 +1.51; ranging 0-3) and 5.88 (5.88+5.96, ranging 1-19), respectively (Fig 3a). If a cutoff of > 4 CTCs in 4 ml of blood is used to screen for EOC, the sensitivity is 62.5% and the speci city is 100%, respectively.
Our results revealed that neither healthy subjects nor patients with benign ovarian tumors could detect CD13+ CTCs in the peripheral blood. The average number of CD13+ CTCs in the peripheral blood of EOC patients was 3.375/4ml (ranging: 0-19) (Fig 3b). The results showed a sensitivity of 62.5% and a speci city of 100%, respectively, if we used the presence of CD13+ CTCs or not as the cutoff. Further analysis showed a sensitivity of 75% and a speci city of 100%, respectively, if we used the cutoff as > 4 CTCs or the presence of CD13+ CTCs in 4 ml of blood.

Discussion
In the present study, we show a fully automated CTC detection platform that can e ciently capture, identify, scan, locate, enumerate and even characterize CTCs on-chip. The development of fully automated systems has greatly increased the possibility of CTC testing being utilized in clinical practice.
This study also preliminarily tested the validity of the platform for patients with EOC, and found that its considerable potential in the detection of CTCs in EOC patients and could assist the treatment decision in EOC and even other cancers.
Due to the rarity of CTCs in the peripheral blood of cancer patients, high enrichment e ciency is critical for CTC detection platform. The design and manufacture of micro uidic chips plays an important role in the entire system. V-BioChip is an immunoa nity-based micro uidic chip which mainly based on the a nity binding between CTCs and the surface of the nanopillar on the chip. The surface of nanopillars is coated with biotinylated PLL-g-PEG + streptavidin and the target cells are pre-labeled with biotinylated monoclonal antibody. Nanotexturing improved the adhesion of microvilli and invadopodia of CTCs to the surface of nanopillars which was functionalized with streptavidin 11 .
Epithelial cell adhesion molecule (EpCAM) is a cancer-related antigen. Cancer tissues of epithelial origin often overexpress EpCAM [12][13][14] . Therefore, in the past few decades, EpCAM has been used by many enrichment technologies as the main cell membrane marker to isolate CTCs, and considerable progress has been made in the use of EpCAM-based CTC detection technologies. However, there are many reasons why EpCAM-based technologies cannot effectively detect CTCs. The expression of EpCAM in variant cancers is quite different 15 . In addition, epithelial-to-mesenchymal transition (EMT) is a key step of the metastatic cascade. This process allows epithelial cancer cells to acquire some mesenchymal characteristics while losing epithelial cell phenotype, resulting in a decrease in the cell's EpCAM expression 16,17 . For example, the expression of EpCAM in patients with ovarian cancer depends on their histological subtypes. Overall, 73% of ovarian cancers overexpress EpCAM, but serous ovarian cancers have a lower EpCAM overexpression rate (55%) 15 .
Due to the downregulation of CTC EpCAM expression during the EMT process, the use of EpCAM-based enrichment techniques may not e ciently capture the CTC populations that are currently EMT 18 . SKOV3 is an EOC cell line with mesenchymal phenotype. It is reasonable that the present study shows a low capture e ciency of 48% for SKOV3 cells when using a single epithelial antibody as the capture antibody. For cancer cells undergoing EMT, further strategies must be taken to achieve better capture e ciency.
The combined use of epidermal and mesenchymal markers not only increases the probability of capturing CTCs of different phenotypes, but also reduces the omission of EpCAM-negative CTCs. Po and his colleagues demonstrated that adding anti-N-cadherin antibodies to anti-EpCAM antibodies can increase the e ciency of capturing CTCs when using immunomagnetic beads as a strategy for isolating CTCs 19 . Another study showed that using ve different types of antibodies on the CTC platform can offer a signi cant improvement in cell-isolation e ciency, even from tiny amounts of blood (250 μL-1 mL) or in non-metastatic breast cancer 20 . The present research also reached similar conclusion as the previous studies. Using the Combi strategy by mixing an epithelial antibody and a mesenchymal antibody as the enrichment antibody combination can signi cantly overcome the adverse effect of downregulation of EpCAM expression caused by EMT on the CTC capture e ciency of the micro uidic system and achieve an optimal capture rate. Most studies were conducted by using PBMCs suspensions of cancer cells that mimic actual conditions. The donor's blood can interact with the spiked cells and markedly affect the recovery e ciency. Our results show that the automated CTC platform combined with V-BioChip and Cell Reveal TM system can achieve excellent CTC capture e ciency of 90% by using Combi strategy. Even in low cell count, the system also demonstrated excellent capture e ciency and device linearity with R 2 of 96%.
Current methods for CTCs identi cation and analysis depend on immuno uorescence technologies and uorescence microscopic imaging by characterizing the tumor cells with speci c markers. Up until now, most of the micro uidic chip-based CTCs platforms still rely on manual operations for cell screening and interpretation after immuno uorescent staining of target cells, which is time-consuming and causes inconsistent results. So far, only a limited number of automated CTCs microscope systems have been reported 21 . The automated CTCs scanning and identi cation system can complete the scanning of the target cells in a short time, which not only saves time and manpower, but also helps to establish standard processes. Our state-of-the-art automated microscope control system and CTCs scanning and identi cation system can automatically number and locate target cells in addition to automatic scanning and identi cation of target cells. Numbering and positioning functions can nd speci c target cells when manual inspection or re-evaluation is required. In addition, these functions can re-locate speci c cells on the chip in a very short period of time, which is very useful when single cell analysis is required.
In the face of the extremely poor prognosis of ovarian cancer, CTC has the potential to provide more information beyond the current testing technology to assist treatment decisions. In the past decade, there have been many studies showing the use of CTCs as a "real-time liquid biopsy" for the diagnosis and treatment of ovarian cancer. However, these studies show a low positive rate of CTCs and a low median of CTCs 8, 22-24 . The possible reason is the early onset of peritoneal dissemination in EOC, with only one third of the patient has distant metastases, leading to insu cient CTCs in the peripheral blood to be detected 25,26 . In addition, most of these studies do not report the sensitivity and speci city of the technology used 19,[27][28][29] . Therefore, the clinical application of CTCs in management of EOC still needs further con rmation.
In addition to enumeration of CTCs, single-cell phenotypic characterization can provide additional biochemical information to assist clinical diagnosis and treatment decisions 30 . Aminopeptidase N (CD13) is a surface marker of cancer stem cells (CSC). It plays a role in cancer cell invasion and is also a candidate for treatment resistance, recurrence and poor prognosis, which is highly expressed in early stage of EOC 31,32 . Van Hensbergen and his colleagues found that the expression of CD13 in EOC was associated with the histological subtype: CD13 expression in tumor cells was observed in 80% of the patients with a serous carcinoma and 100% of the patients with a mucinous carcinoma and in only 20 % of the clear cell carcinoma patients 32,33 . To the best of our knowledge, this is the rst attempt to isolate and identify CD13+ CTCs in EOC patients.
Although the case number in the present study is limited, there are several meaningful ndings. First, no matter which cutoff is used, there is a high sensitivity, which can greatly simplify the interpretation of clinicians. Second, the positive rate of CD13+ CTCs in serous adenocarcinoma is consistent with the histological study by Van Hensbergen. Third, CTCs or CD13+ CTCs can be detected even in the early stage (stage I) EOC. It means that CTCs have appeared in the blood at least in some early EOCs, indicating the possibility of micrometastasis even in the early state of EOC by surgical staging. Finally, due to the full automation of the laboratory processes, it is technically possible to simultaneously detect one or more additional cell markers in. In other words, it has the potential to provide a comprehensive cell pro ling for clinical applications, but more experimental validation is still needed.
In conclusion, as the advantage of micro uidic chips and automated systems, CTCs detection has become more and more e cient and accurate. Although many micro uidic-based methods have been developed in the past decade, most of them cannot be applied clinically due to lack of automation, low throughput, and high cost of micro uidic chips. This study shows that the automated Cell Reveal TM system has the potential for widespread clinical application to aid in the treatment decisions of EOC and even other cancers. However, in spite of the very promising preliminary results, more clinical experience is needed to reach further conclusions due to the complexity and heterogeneity of cancers.

Micro uidic chip
The V-BioChip (CytoAurora Inc., HsinChu, Taiwan) is a silicon-based chip with nano-pillar arrays on the inside of micro uidic chambers (Fig 1). The structure and production process of the predecessor of V-BioChip, Cral Chip, has been described in detail in Ma's report 11 . To improve the chip's capture e ciency, we modi ed original design of Coral Chip to adjust the distance between the nano-pillars on the micro uidic chip and the shape of the nano-pillars. In brief, metal-assisted chemical etching (MACE) technology was used to form matrix-arranged nano-pillars on the chip surface. The tip of the pillars is modi ed into a shape of volcanic cone to increase the contact surface between the target cells and the nano-pillars, in attempt to reduce cell damage. Thereafter, the polyethylene glycol-biotin (PEG-biotin) layer was modi ed on the surface of the wafer by vapor deposition method. Streptavidin, a tetrameric protein with high binding a nity to biotin 34 , was then attached to the biotin end of the PEG-biotin using liquid deposition technology. The streptavidin-biotin interaction is one of the strongest non-covalent biological interactions currently known and can markedly increase the capture e ciency to the target cells 35,36 . When the mixed cell suspension ows over the chip, the target cells will be captured by the V-BioChip, and most other cells will be washed away.

Cell spiking test
The Cell Reveal TM machine (CytoAurora Inc., HsinChu, Taiwan) was used for the enrichment and staining of the CTCs. Before the test, the V-Biochip was set up inside the machine. The mixed cell suspension of SKOV3 and PBMCs was injected into the Cell Reveal TM system, and the system automatically processed the CTC enrichment and staining procedures. The inputted cell mixture was then xed in 4% paraformaldehyde. Subsequently, 0.1% of Triton X-100 (ThermoFisher, Waltham, MA, USA) and 2% BSA (Bovine serum albumin) were added to increase the cellular permeability. The cell mixture passes through the V-BioChip at a ow rate of 0.6 mL/h, allowing the target cells to fully contact the chip to achieve an optimal capture rate. As the cell-rich centrate ows through the micro uidic chips, the streptavidin on the chips captures the target cells bound with biotinylated anti-EpCAM antibodies. The process of CTC enrichment and staining were done overnight to achieve the best staining. But this process can be completed within four hours.
In order to distinguish CTCs from white blood cells, micro uidic chips were incubated with anti-EpCAM antibody (R&D Systems, Minneapolis, MN, USA) conjugated with FITC (for the detection of CTCs), as well as anti-CD45 (Agilent, Santa Clara, CA, USA) antibody conjugated with TRITC (for the detection of white blood cells) and 4',6-diamidino-2-phenylindole (DAPI; Invitrogen, Carlsbad, CA, USA) (for the detection of nucleated cells). The CTC was de ned as an EpCAM+/CD45-/DAPI+ intact cell.
After the enrichment and staining steps, V-BioChip is sent to the CTC scanning platform for further analysis. This state-of-the-art CTC scanning platform is composed three main parts, including a modi ed upright uorescent microscope and two sets of self-developed software (CytoAcqImages system for automatic image scanning and CAT automatic cell identi cation system Under the control of CAI system, the microscope automatically focuses and scans the V-BioChip, as well as activates the exposure with a uorescent carousel (Fig 4a). A fully automatic segmented photographing method was adopted to obtain high-resolution full-area images of V-BioChips. Then, the image les were montaged together to form a whole biochip TIFF image le.
V-BioChip is pre-installed with special positioning marks. When the CAI system focuses on these marks, it also records the coordinate information to the full chip image le. When the image interpretation system reads the chip image le, it can use the coordinate information to calculate the speci c position of the speci c cell on the chip. In addition, this precise positioning function can also help to nd speci c cells on the chip again, which can assist in re-evaluating speci c cells and even further single-cell analysis.
The full chip image les were then transmitted to CAT system (Fig 4b). CAT system can identify target cells according to the immuno uorescence staining on the cells by using pre-set parameters and deep learning AI function. Count-in/ lter-out criteria were used to identify the cells while EpCAM+/CD45-/DAPI+ for CTCs and EpCAM-/CD45+/DAPI+ for WBCs (Fig 5a).
Three separate spiking tests were performed for both the single ant-EpCAM antibody and Combi strategy. The results are expressed as the mean recovery rate + standard deviation (SD).

Linearity between the numbers of captured cells and spiked cells
The relationship between the captured cell numbers and the CTC numbers was analyzed by spiking SKOV3 cells into DPBS was analyzed. We incubated 625 SKOV3 cells with 5μL cocktail capture antibodies in 200μL DPBS at 37˚C for 45min and then centrifuged with 400g for 10 min. A serial of 2-fold dilutions was performed to allow for linear regression analysis of the estimated cell numbers being 625, 312, 156, 78 and 39. The diluted cell suspensions were then injected into Cell Reveal TM System to capture SKOV3 cells by streptavidin-coated chip with a ow speed of 0.6 μL/min. The captured cells were stained by EpCAM/CD45/DAPI on chip and subsequently were scanned and counted by the automatic CAT system.
Each spiking cell number is tested twice. In addition to linear regression analysis, the recovery rate of each test was also calculated. The results of recovery rate were expressed as mean recovery rate + SD.

Feasibility study
In order to examine the clinical feasibility of the test, we recruited participants to join the study, including 10 healthy subjects without any type of ovarian tumor, 5 women with benign ovarian tumor and 8 patients with epithelial ovarian cancer (EOC). FIGO criteria were used for the staging of EOC. The study has been approved by the IRB committee in the Taiwan Adventist Hospital and Changhua Christian Hospital. All methods were carried out in accordance with relevant guidelines and regulations and each participant completed a written consent before they received the test.
In this study, anti-CD13 antibody (Abcam, Cambridge, UK) was also used for ICC staining to identify CTCs that carry CD13 (Aminopeptidase N) surface marker. CD13 is a function marker for cancer stem cell. When an intact cell shows ICC staining of CD13+/EpCAM+/CD45-/DAPI+, it is de ned as a CD13+ CTC (Fig 5b).     The automatic scanning, locating and cell analysis system (a) A Leica DM6 B upright microscope equipped with Leica HC PL APO 10x /0.45 microscope objective, spectra III light engine with wavelenght:

Declarations
380nm~750nm, Andor Zyla 4.2 sCMOS camera and Marzhauser scanning stage for 4 slides. This microscope is controlled by the CytoAcqImages system to perform automatic scanning of V-BioChip and positioning of target cells; (b) The CAT (Cell Analysis Tools) system, which can identify target cells according to the immuno uorescence staining on the cells by using pre-set parameters and deep learning AI function. Figure 5