In this study, we have designed a CTCs isolation and characterization platform called CTC 100TM platform (Figure 1a). The platform is based on a microfluidics chip which could enrich CTCs within 15 minutes from 4 mL blood. A clinical protocol based on this platform is also established: (a) 4 mL venous blood sample was drawn from patients; (b) PBMC layer was collected after density gradient centrifugation; (c) CTCs was enriched from the PBMC by the microfluidic based platform; and (d) CTCs phenotype/genotype analysis was carried out by immunofluorescence and digital qPCR (Figure 1b).
Separation of CTCs by Inertial Focusing in Microfluidic Channels
Inertial focusing is a widely used working principle in microfluidics for processing fluids containing particles with different mechanical properties, including size, shape, and deformability[8]. Current CTCs separation platforms, some based on inertial focusing, still have drawbacks such as sample blockage, low recovery rate, low recovery purity, poor cell viability, etc[9]. To overcome the current limitations, in this study, we designed a novel inertial focusing-based structure which is composed of a spiral microchannel with an isosceles trapezoidal cross-section. The curvature radius of the chip is 15 mm, and the main channel width is 1 mm. There are two inlets for sample and buffer and two outlets for CTCs collection and waste output (Figure 2a-d). Initially, the sample and the buffer solution are pumped into the channel from the sample inlet and the buffer inlet respectively. When the cells of different sizes flow in the microchannel, they experience inertial lift forces (FL) from the nature of laminar velocity, which is essentially the net of two forces[9]: shear-induced lift force (FIL) and wall-induced lift force (FWL) (Figure 2e), as defined in Eq. (1):
where ρ is the density of the fluid, Um is the maximum velocity, Dh is the hydraulic diameter of the channel, CL is the lift coefficient, and ap is the particle diameter. In addition, the spiral geometry of the microchannel could produce two eddy currents (Dean flow) with opposite rotation directions, where both currents are vertical to the main flow direction (Figure 2f). This produce a third drag force (Figure 2f) from Dean flow on the cells and leads to a situation when the cells are flowing along the main flow direction, they will also migrate to a specific equilibrium position determined by the ratio of the three forces (Rf) (Figure 2e, 2f) , as defined in Eq. (2):
where R is the hydraulic diameters of the curvature. Since all these forces are the functions of the cell size, CTCs and normal blood cells migrate laterally to different equilibrium points of the cross-section. Such self-ordering offers an opportunity to separate CTCs from blood cells according to size differences. By adding bifurcation, CTCs and WBCs are subsequently separated from each other (Figure 2c).
Early studies have demonstrated the feasibility of such flow configuration for CTCs isolation, but the recovery rate was severely compromised due to the abundance of red blood cells (RBCs) in peripheral blood. In this work, we addressed this problem by obtaining the PBMC (peripheral blood mononuclear cell) from the blood sample firstly by density gradient centrifugation. The collected PBMC was then used for CTCs isolation by the microfluidic chip with a size threshold of 15 μm. As previous studies reported that density gradient centrifugation may cause the loss of tumor cells, we first evaluated the loss of tumor cells during this process. 100 Hela cells pre-stained with CellTracker dye were added into 4 mL (25 cells/mL) whole blood for density gradient centrifugation. Hela cells were then counted and the results showed that the total Hela cell number was 91 ± 2 in the PBMC layer (Figure 2g). This result demonstrated that density gradient centrifugation does not cause significant loss of tumor cells. The result also showed that there was a broad range of around 4 - 12 × 106/mL WBCs numbers in the PBMC layer (Figure 2g). We next evaluated the influence of WBCs count on CTCs isolating. Around 100 pre-stained Hela cells were added to 4 mL PBS (25 cells/mL) with different densities of WBCs (4 × 106 cells/mL, 8 × 106 cells/mL, 12 × 106 cells/mL), then the Hela cells were collected by the microfluidic chip. The results showed that under different WBCs densities, the Hela cell recovery rates were not significantly changed (Figure 2h, i).
In the spiral microfluidic chip, the smaller particles move from the outer side of the channel to the inner side, and then return to outer position again with a certain tendency of dispersion (ie they have experienced a complete Dean cycle). As the flow continues, small particles approach slowly to the outer wall, while large particles were gradually focused to the equilibrium position near the inner wall. This process could be affected by several parameters of the microchannel. We firstly performed experiments to determine the optimal parameters of the chip, including the width of channel, width ratio of bifurcation, channel height and buffer flow rate. We used 9 μm microbeads to represent the WBCs, and 15 μm microbeads to represent CTCs. 1ml of 15 μm microbeads (105/ml) and 1ml of 9 μm microbeads (106/ml) were mixed and added into the sample inlet. The fluid velocity was controlled by a micro-syringe pump.
Previous studies had reported several inertial focusing-based microfluidic chips for CTCs enrichment[3, 10, 11], most of these chips have a channel width between 100 μm and 600 μm with the sample processing time range from 8 minutes to 20 minutes. We firstly compared the recovery rate and processing time of microfluidic chips with different channel width, 500 μm and 1000 μm, while the other parameters are set to be the same which is 190 μm in channel height, 1:4 in width ratio of the bifurcation and flow rate of 3 mL/min. As shown in Figure 3a, the recovery rate was 84.8 ± 3.4% and 91.2 ± 2.4% respective, while the processing time was around 1 hour when the channel width was 500 μm, much longer than the microfluidic chip with a channel width of 1000 μm, which cost less than 4 minutes. So we firstly set the channel width to be 1000 μm due to the much shorter processing time and better recovery rate.
We next optimized the width ratio of the bifurcation to enrich CTC effectively. Three different width ratios were compared, 400:600, 200:800, and 100:900, with the recovery rates of 90.6 ± 3.6%, 91.2 ± 2.4%, and 73.2 ± 2.5%, respectively (Figure 3b). Therefore, the optimal output channel ratio was determined as 200:800. We then compared the height of the channel, 180 μm, 190 μm, and 200 μm, and the results showed that when the height is 190 μm, the 15 μm microbeads could be most effectively separated from the 9 μm microbeads (Figure 3c). The flow rate of buffer in the channel is also an important factor for the separation efficiency of the microfluidic chip, as it decides the number of dean cycles that the cells experience in the fixed-length channel. While the flow rate in the sample inlet was fixed at 600 μL/min, we compared three different flow rates of the buffer and found that 3mL/min had the best recovery rate for the 15 μm microbeads (Figure 3d).
Enrichment and separation of CTCs require efficient removal of the normal blood cell. Previous study found that, compared to the rectangle channel, the right-angled trapezoidal channel can change the shape of the velocity field and push the smaller WBCs closer to the outer wall, while the larger CTCs migrate to a position closer to the inner wall[12-14]. However, it is unclear whether isosceles trapezoidal channel is better than rectangle type in CTCs enrichment. We next compared the separation performance of the two types of microfluidic chips, with the parameters of the rectangle channels shown in Figure 3e. CellTracker dye pre-stained Hela cell were spiked into whole blood samples from healthy donors to create different tumor cell density (1 - 25 cells/mL) and isolated by the two kinds of chips. In the trapezoidal type, the average overall recovery rate of CTCs was 87.08 ± 8.65%, independent of the CTCs density in blood (Figure 3f). On the other hand, the recovery rate of the rectangle channel was lower than 80% under different Hela cell density (Figure 3g). The collected Hela cell purity was also evaluated as it is an important index for CTC downstream analysis. As shown in Figure 3h, the output purity of microfluidic chip with trapezoidal channel was 0.28 ± 0.06% when the sample density was as low as 1cell/mL (200–400 WBCs mL−1, ∼4 log depletion of WBCs). The CTCs purity increased when there were more Hela cells in the samples. The purification capability of rectangle channel was also lower than the trapezoidal type (Figure 3i), where the purity was 0.11 ± 0.02% when the sample density was 1cell/mL. As the CTCs number in cancer patients is usually between 0 - 200 cells/7.5mL[15], we compared the recovery rate and output purity when the Hela cell density range from 5 cell/mL to 25 cells/mL. The results showed that microfluidic chip with the trapezoidal channel has statistically better performance than with the rectangle channel (Figure 3j, 3k). From the aspect of CTC purification, the CTCs output purity of the microfluidic chip with the trapezoidal channel was nearly two times higher than with the rectangle channel. The flow track of Hela cells in the different microfluidic chips were also recorded by the microscope and showed that the flow tracks of Hela cell were more focused in the trapezoidal channel than in the rectangle channel (Figure 3l, 3m).
Integrity and Viability of CTCs after Microfluidic Processing
We next evaluated the performance of this microfluidic platform for tumor cell lines of various cancer types. 10 cells/mL samples of different tumor cell lines were prepared by adding around 40 pre-stained Hela cells, SK-Br-3 cells, H1650 cells or PC-3 cells into 4mL healthy blood, respectively. After density gradient centrifugation, the PBMC layer were collected for tumor cell isolation. As shown in Figure 4a-b, the recovery rates of different tumor cell lines were 86.2±9.7%, 88.9±5.3%, 84.9±7.8%, and 81.6±7.9%, respectively. The output purity of the collected tumor cells were 2.41±0.31%, 2.52±0.24%, 2.43±0.44%, 2.39±0.26% respectively. To identify whether the collected cells are indeed tumor cells, various techniques including fluorescence in situ hybridization (FISH), immunofluorescence (IF) and RT-PCR/qRT-PCR can be used[16-19]. In this study, we also established the IF protocols to identify the collected tumor cells with the criteria defined as follows: PanCK+, DAPI+, CD45- cells with nucleus to cytoplasmic ratio greater than 0.8. Figure 4c-d showed that the collected SK-Br-3 and Hela cells could be accurately identified by the method. The integrity of the isolated tumor cells was observed under the microscope (Olympus IX73), using the Hela cells as an example, there was no obvious morphology change after the separation (Figure 4e)
We next assess the tumor cell viability after microfluidic separation. Breast cancer cell line (SK-Br-3) was spiked into 4 mL healthy blood and enriched by the microfluidic protocol. The viability of collected tumor cells at different time points after the CTCs enrichment was assessed by AO/PI staining. The results showed that, at the time just after the enrichment, most collected tumor cells are AO positive and PI negative (Figure 4f), where the relative living cell ratio is 95.05 ± 3.54% in the control group and 90.55 ± 4.79% in the tested group (n=5). The relative living cell ratio decreased to around 41.2-49.0% 48 hours after the CTCs enrichment and maintained at similar level until the 7th days (n=5). For the control group, the relative living cell ratio decreased to 58.2-61.5% from 48 hours to 7 days (n=5). This indicate that microfluidic processing only caused minor increased in cell death, where a large percentage of the collected tumor cells maintained the ability to divide and proliferate.
The above results demonstrated that our microfluidics chip could separate CTCs with a higher recovery rate, higher output purity while maintaining the integrity and viability of the enriched tumor cells.
Clinical Validation of the Microfluidic CTCs Platform
To validate the clinical utility of the protocol, 4 mL blood sample was obtained from each of the 567 patients of various cancer types, stages, and treatments, plus 78 normal subjects for CTCs enrichment and characterization (Table 3, Table S1, S2). The ability of the CTCs 100 platform in differentiating normal, early and late stages of samples based on CTC enumeration was evaluated, and the feasibility of detecting specific protein markers and target gene mutations in the enriched CTCs was also demonstrated.
EMT of CTC is a complex process that occurs in the process of tumor metastasis. It was reported to be controlled by the down-regulation of epithelial markers EpCAM or PanCK and the up-regulation of mesenchymal markers such as vimentin and N-cadherin. Due to the EMT process, CTCs could be subtyped into 3 classes, including epithelial, mesenchymal and mixed types. The proportion of different CTCs subtypes had been demonstrated to be clinically valuable, especially in the prognosis aspect[16, 20]. In this study, we choose PanCK and N-cadherin as the biomarkers for CTC subtyping based on the previous studies. As shown in Figure 5a, Cells with DAPI+/CD45-/PanCK+/N-Cadherin- were defined as epithelial CTCs. Cells with DAPI+/CD45-/PanCK-/N-Cadherin+ were defined as mesenchymal CTCs. Cells with DAPI+/CD45-/PanCK+/N-Cadherin+ were defined as E\M mixed CTCs. Cells with and DAPI+/CD45+/PanCK-/N-Cadherin- were defined as WBCs.
According to the above definition, CTCs were enriched and identified in 567 patients with CTCs counts ranging from 0-520 CTCs/4mL. The CTCs positive rate (≥1/4mL) was as high as 95.1% and most patients have more than 4 CTCs in their 4mL blood samples (Figure 5b), The CTCs positive rate was much higher than the detection rate reported in the studies that using other platforms, which ranged from 17% in early stage cancer to 75% in metastatic cancer patients[21-26]. The receiver operating characteristic (ROC) curve analysis of CTCs counts for differentiating cancer patients from healthy donors showed that the area under curve (AUC) was 0.956. The threshold analysis by Youden Index method suggested 4 CTCs/4mL of blood sample as the optimal cut-off value for predicting cancer disease (Figure 5c). Using this cutoff value, the sensitivity and specificity of the test were 86.2% and 91.0%, respectively. Previous studies suggested EMT phenomenon confers CTCs with enhanced cell mobility, metastatic properties and resistant to therapies[27]. In this study, the number and percentage distribution of CTCs subtypes in various cancer types were shown in Table 4, and the distribution characteristics will be further analyzed in the future for the value of tumor diagnosis and treatment.
We also compared the CTCs in 78 healthy donors with the enrolled 128 non-small cell lung cancer patients which included 62 early-stage patients and 66 late stage patients. The CTCs count in healthy donors, early-stage NSCLC (I and II) and late-stage NSCLC (III and IV) were 0.9 ± 1.6/4mL, 8.4 ± 5.3/4mL and 26.4 ± 12.4/4mL respectively (Figure 5d). All the early-stage patients had CTCs and most of them (49 patients) had more than three CTCs. Interestingly, CTCs could also be identified in healthy donors (Figure 5b, d), this may due to the presence of aging non-tumor cells and cells undergoing apoptosis[15], but the number was significantly lower than that in cancer patients (P<0.05; Figure 5d). The CTCs number in late-stage lung cancer patients were significantly higher compared with that in early-stage patients and healthy individuals (Figure 5d). These results implicated that the CTCs count captured by CTC 100 platform could serve as a biomarker for cancer diagnosis and monitoring.
Downstream Analysis of the Enriched CTCs
Over the last decade, the immune checkpoint inhibitors (ICI) and target therapy has improved progression-free and overall survival of cancer patient[28-30]. By analyzing and identify the specific biomarkers, it can provide precise and individualized treatment to cancer patients. The expression of several proteins has been recognized as a suitable predictive biomarker for monitoring tumor response to ICI or target therapy, such as PD-L1, HER2 and VEGF[31-33]. These biomarkers are traditionally assessed via tissue-based technique of immunohistochemistry (IHC). However, tissue biopsy is invasive and difficult to obtain, and may cause high bleeding risk which often prevent additional biopsies for immunohistological evaluation of these biomarkers[18]. These disadvantages prevent the dynamic monitoring and immune/target therapeutic selection during clinical practice[34].
Surface proteins of CTCs have been established as suitable biomarkers for ICI or target therapies. A correlation between PD-L1 expression in CTCs and primary tumor was reported[35], where the expression of PD-L1 in CTCs has been utilized as a biomarker for response to ICI, such as nivolumab and pembrolizumab. The predictive value of HER2 expression in CTCs was also evaluated in breast cancer and revealed that CTC-based HER2 positive patients had higher survival under HER2-target therapy[36]. VEGF is another important protein involving in the angiogenesis of cancer, an critical event in tumor progression and metastasis[37], where the expression of VEGF in CTCs of patients could reflect the metastatic potential of cancer and help clinical decisions.
As demonstrated in the previous section, the CTCs enrichment platform developed in this study can preserve the integrity of CTC during the isolating process. This ensures the feasibility of downstream analysis of CTCs which originate from the tumor tissue. In this study, we successfully detected three clinically valuable protein biomarkers on CTCs, including PD-L1, HER2 and VEGF from patients with non-small cell lung cancer (NSCLC), breast cancer and colon cancer (Figure 6a-c).
CTC PD-L1 status was further assessed in the 62 late staged NSCLC patients to compare with their tissue-based PD-L1 assessment. Previous works suggested that anti-tumor therapies could change the PD-L1 expression in cancer cells[38, 39]. In this study, as the pathological PD-L1 tests were done a few months ago and most patients received anti-tumor therapies during this period, we divided the NSCLC patients into two groups based on whether the detection interval between tissue-based PD-L1 test and the CTCs based PD-L1 analysis was less than 6 months (n=30) or greater than 6 months (n=32). The correlation between the PD-L1 positive percentage of CTCs, which is calculated by the number of PD-L1 positive CTCs/ the number of total CTCs, with the tumor proportion score (TPS) of pathological PD-L1 results was then analyzed by the Pearson’s correlation test. In the group where the detection interval was shorter than 6 months, the results showed that the PD-L1 positive percentage of CTCs was associated with TPS score (R=0.54, P=0.002, Figure 6d), which suggested that CTCs-based PD-L1 analysis could be an alternative method for PD-L1 assessment. There was no such association when the detection interval was longer than 6 months (R=0.19, p=0.31, Figure 6e), these may reflect that the long-term anti-tumor therapy had significantly changed the PD-L1 status of cancer cells, suggesting CTCs-based PD-L1 testing may be a choice when the immunotherapy was considered while the latest pathological results were longer than 6 months ago.
Understanding the molecular basis and oncogenic drivers of cancer is crucial to targeted therapies for cancer patients[40], such as EGFR, KRAS and BRAF mutations. EGFR mutations are found in 10% to 20% of lung adenocarcinomas[41] and has been found to be predictive of NSCLC response to erlotinib[42], a target drug that inhibit the tyrosine kinase activity of EGFR . KRAS and BRAF mutations are another two widely assessed mutations in various cancer types, including pancreatic cancer, colon cancer, melanoma, etc. Assessment of KRAS status is mandatory in patients with late stage colorectal cancer (CRC) before applying targeted therapy[43]. BRAF mutation are present in up to 8% of cancers[44], specific BRAF inhibitors, such as dabrafenib and vemurafenib, were approved for used in metastatic colon cancer and melanoma[45]. However, the decision to start target-based cancer therapies rely on the analysis of specific gene mutations which is difficult to detect in metastatic or relapse cases years after the primary cancer diagnosis and surgical resection[46]. CTCs provide a valuable source for studying the whole genomic characterization of cancer as they originate from both primary and metastatic tumors, and genetic analyze of CTCs for the presence or absence of key mutations may provide important clinical information over the course of treatment to guide clinicians on when to stop or change the treatment plan[47-49]. However, most current available CTCs captures devices is not suitable for downstream genetic analyze due to the difficulty to detach CTCs from filter, morphological deformation of CTCs, and very low CTCs purification, etc[50]. The CTC enrichment platform developed in this study has the ability to enrich intact viable CTCs as unfixed cell in solution without the requirement of complex high-resolution imaging techniques or the use of expensive antibodies, which allowed for the genetic analysis of CTCs by conventional qPCR method. We also established a workflow for the genetic analyze of the collected CTCs by the integration of the CTCs enrichment platform, single cell picking system and qPCR (Figure 7a-d).
We firstly used this workflow on tumor cell lines firstly, including H1650, PANC-1 and HT-29, to assess whether the most common genetic mutations of EGFR (19del), KRAS (G12D) and BRAF (V600E) [44, 51, 52], could be detected. As the above clinical test demonstrated that most cancer patients have more than 4 tumors cell, about 4 tumor cells were spiked into 4 ml healthy blood (1 cell/mL) for cell isolation. All the enriched tumor cells were then collected by the single cell picking system, transferred into a PCR tube and tested by the qPCR together, the representative qPCR curves were shown in Figure 7e-g. The results showed that genetic mutations of EGFR, KRAS and BRAF could be successfully detected (Table 5) in the collected tumor cells even there was only one tumor cell (sample number 4 and 15). The CTCs count and the genetic mutations of CTCs were then assessed in cancer patients who were confirmed to have the mutations based on tissue testing, including 5 EGFR 19del mutation positive NSCLC samples, 5 KRAS G12D mutation positive PAAD patients and 5 BRAF V600E mutation positive CRC patients. As shown in Table 6, the genetic mutations can be reliably detected from most of the patients. Only one exception is the CTCs sample from a PAAD sample (sample number 6) which had negative result possibly due to tumor heterogeneity. These results demonstrated the great prospects of CTC genotyping, which may provide an alternative option for genetic testing and monitoring in patients with advanced malignancy.