Abstract
The increasing incidence of bladder cancer and its high rate of recurrence over a 5-year period necessitate the need for diagnosis and surveillance amelioration. Cystoscopy and urinary cytology are the current tools, and molecular techniques such as BTA stat, NMP22, survivin mRNA, and urovysion FISH have attracted attention; however, they suffer from insufficient sensitivity or specificity. We developed a novel microfluidic approach for harvesting intact urinary-exfoliated tumor cells (UETC), either individually or in clusters, in a clean and segregated environment, which is crucial to minimize cross-contamination and misreads. To reliably and accurately identify UETC, our quantitative immunoassay involved concurrent use of two oncoproteins CK20 and CD44v6 antigen. CK20 is an intermediate filament protein overexpressed in urothelial tumors, and CD44v6 is a membrane adhesion molecule closely associated with cell invasion, tumor progression, and metastatic spread. Single-cell whole-genome sequencing on 12 captured UETCs and copy number alteration analysis showed that 11/12 (91.7%) of the immunofluorescence-identified UETCs possessed genomic instability. A total of 79 patients with bladder cancer and 43 age-matched normal controls (NC) were enrolled in the study. We detected considerably higher UETC counts in patients with bladder cancer versus the NC group [53.3 (10.7–1001.9) vs. 0.0 (0–3.0) UETCs/10 mL; P < 0.0001]. For bladder cancer detection, a stratified 10-fold cross-validation of training data reveals an overall predictive accuracy of 0.84 [95% confidence interval (CI), 0.76–0.93] with an 89.8% (95% CI, 71.5%–86.4%) for sensitivity and 71.5% (95% CI, 59.7%–83.3%) for specificity. Overall, the microfluidic immunoassay demonstrates increased sensitivity and specificity compared with other techniques for the detection of bladder cancer.
Significance: A unique and promising diagnostic assay for bladder cancer is proposed with potential clinical utility as a complement for cytology. Cancer Res; 78(14); 4073–85. ©2018 AACR.
Introduction
Urothelial bladder carcinoma forms the vast majority of bladder cancer and is the second leading cause of mortality with an estimated 150,000 deaths per year (1). At initial diagnosis, about 70% of the cancer is non-muscle invasive bladder cancer but over a 5-year period with the risk of recurrence varying from 30% to 80%, about 15% progresses to muscle invasion (2). Therefore, patients with bladder cancer require long-term screening and surveillance, relying on cytology and cystoscopy with tissue biopsies as the primary detection tools. However, the low sensitivity and high interobserver variability of cytology, particularly for low-grade bladder cancer together with the high degree of invasiveness and high cost of cystoscopy are major issues for surveillance and diagnosis (3). A non-invasive and robust technique for a rapid and accurate detection of bladder cancer is highly desirable.
Numerous cell-free protein and nucleic-acid based techniques sourced from voided urine have been developed over the past 2 decades for non-invasive and high-fidelity diagnosis of bladder cancer. BTA stat (Polymedco; ref. 4), for example, detects urinary human complement factor H-related protein with a sensitivity and specificity of between 57%–83% and 60%–92%, respectively (5, 6). Another protein-based test, NMP22 BladderChek (Alere; ref. 7) detects urinary NMP22 protein, which reflects the mitotic activity of cells and is rated with an overall sensitivity of 65% and specificity of 81% (8). Survivin mRNA has been measured in voided urine as a diagnostic marker (9); however, the lack of assay standardization and inconsistent cutoff values limits its wide-spread use (3). UroVysion FISH assay detects aneuploidy in chromosomes 3, 7, and 17, as well as, the loss of 9p21 locus in urine specimens but its detection accuracy varies widely (10). Although promising, these markers are not commonly used because of the insufficient accuracy, high cost or ambiguous clinical readout (11) and to reduce their relatively high false-positive rates, they are often used in conjunction with cystoscopy (12).
Urinary-exfoliated tumor cells (UETC), which are shed directly by the growing tumor in the bladder, carry first-hand information along each step of the disease progression (13), making them a promising biomarker for bladder cancer detection. Historically, UETCs have been used as a diagnostic marker in urinary cytology for decades, but the low sensitivity and qualitative nature of the method are 2 severe setbacks. To capture UETCs, several groups have recently introduced a filtration-based method that involves the use of a polycarbonate membrane stamped with uniformly distributed perforations of 8-μm size (14, 15). However, this approach traps urinary cells of all types and all sizes that range from several microns to dozens of microns (16). This indiscriminate and accumulative trapping of cells generates 3 significant drawbacks: a built-up of cellular and noncellular debris that quickly contaminates the entire capture region making the identification of UETCs challenging; the membrane quickly becomes clogged, reducing the flowrate making the process slow and time-consuming; and the trapped cells face a higher risk of damage when buried in the debris rendering them useless for the downstream cell-based analysis.
In this work, we report on a microfluidic approach using a specially designed chip to efficiently harvest and identify UETCs of various sizes with intact cellular morphology (Fig. 1A and B). In particular, the capture of undamaged UETCs offers opportunities for an on-chip proteomic and off-chip single-cell genomic analyses for an objective diagnosis of bladder cancer. Figure 1C–E depicts the setup of a microfluidic immunoassay system (details in Supplementary Fig. S1). Furthermore, we reported the first-ever combined use of the two markers CK20 and CD44v6 employed simultaneously for the UETC-identification. We also performed a single-cell whole-genome sequencing of the marker-identified UETCs to confirm if they are indeed tumor cells.
Microfluidic immunoassay system, chip design, and performance characterization for rapid harvesting of UETCs. A, Schematic of UETC capture from the voided urine of a patient with bladder cancer. B, Single cells are marked by green arrows and clustered cells by red arrows. C–E, Setup of the microfluidic system for the urinary cell capture. F, Size distribution of bladder cancer cells. G, Characteristic dimensions and patterns of the capture chamber. H and I, Traditional cytology (H) versus microfluidic cytology (I) capture of urinary cells. Observe the clean and uncluttered field of view in the latter compared with background impurities marked by red arrows in the former. J–M, Finite element simulation of the flow inside the capture chamber: velocity profile (J); cell pathway as indicated by the flow streamlines (K); wall shear stresses from 0 to 2.98 Pa (L); and wall shear rates across the flow field from 0 to 333.9 1/s (M). N, Capture efficiency (CE) of 85.0%, 83.5 %, 76.8%, 73.8%, 67.1% under varying flow rates. O, The intra-assay variability as measured by CV values of 5 samples is <20%, indicating a high assaying consistency. P, The cell loss analysis for three batches of low cell input reveals that the microfluidic approach achieves a higher recovery factor compared with the traditional pap smear technique (77.3% vs. 67.9%; 81.6% vs. 72.0%; 82.8% vs. 74.3%); scale bar, 20 μm.
Microfluidic immunoassay system, chip design, and performance characterization for rapid harvesting of UETCs. A, Schematic of UETC capture from the voided urine of a patient with bladder cancer. B, Single cells are marked by green arrows and clustered cells by red arrows. C–E, Setup of the microfluidic system for the urinary cell capture. F, Size distribution of bladder cancer cells. G, Characteristic dimensions and patterns of the capture chamber. H and I, Traditional cytology (H) versus microfluidic cytology (I) capture of urinary cells. Observe the clean and uncluttered field of view in the latter compared with background impurities marked by red arrows in the former. J–M, Finite element simulation of the flow inside the capture chamber: velocity profile (J); cell pathway as indicated by the flow streamlines (K); wall shear stresses from 0 to 2.98 Pa (L); and wall shear rates across the flow field from 0 to 333.9 1/s (M). N, Capture efficiency (CE) of 85.0%, 83.5 %, 76.8%, 73.8%, 67.1% under varying flow rates. O, The intra-assay variability as measured by CV values of 5 samples is <20%, indicating a high assaying consistency. P, The cell loss analysis for three batches of low cell input reveals that the microfluidic approach achieves a higher recovery factor compared with the traditional pap smear technique (77.3% vs. 67.9%; 81.6% vs. 72.0%; 82.8% vs. 74.3%); scale bar, 20 μm.
Materials and Methods
Microfluidic chip fabrication
The 10,000 or so capture units in the capture chamber are designed using AutoCAD (Autodesk Inc.) and the chip fabricated via a soft lithography process (17). A layer of 20-μm thick SU-8 photoresist is uniformly spun on a silicon wafer and micro-patterns lithographically created via a chrome photomask. PDMS (Sylgard 184, Dow Corning) and its cross-linker (of ratio 10:1) are casted in the mold, degassed and cured under 60°C for 6 hours. The fabrication process is completed by peeling the PDMS from the mold and bonded to a glass slide via oxygen plasma.
Simulation of urine flow inside the microfluidic chip
A finite element analysis of the fluid flow inside the microfluidic chip is performed using the COMSOL software (COMSOL Inc.). For the flow speeds encountered, a laminar flow model should be amply adequate to mimic the flow pattern in the micro-channels. Human urine at 20°C (density = 1.015 g/cm3 and kinematic viscosity = 1.0700 cSt; ref. 18) was used as the matrix fluid with a flow rate of 500 μL/s and the outlet set as an open boundary. From our computational simulations, we obtained the shear stresses and shear rates in the capture chamber of the chip.
Patient characteristics and ethics
Two age-matched cohorts comprising cystoscopic-positive and pathologically confirmed patients with bladder cancer (N = 79) and normal control (NC) group (N = 43) sourced from the First Affiliated Hospital of Zhejiang University School of Medicine were enrolled in the study from July 2015 to December 2017. The NC group consists of non-neoplastic urinary disease individuals (N = 32) and healthy individuals (N = 11). All subjects were anonymously coded and were asked to provide a written informed consent in compliance with the local ethic regulations before enrolling in the study. The work was carried out in accordance to the Declaration of Helsinki with the research protocol approved by the Ethics Committee of the First Affiliated Hospital at Zhejiang University School of Medicine. Histologic evaluation of tumor specimens was performed by two independent cytopathologists in accordance to the 1973 WHO classification. Detailed clinical history (Supplementary Table S1A) and cystoscopy findings (Supplementary Table S1B) was collected at baseline.
Urine samples collection and processing
For all enrolled patients, 30 to 50 mL second urine of the morning was obtained before a TURBT. The urine samples were then sent to the laboratory within 4 hours of collection for the microfluidic analysis. The sample was centrifuged at 300 g for 5 minutes and pellet washed in PBS followed by a further 5-minutes centrifugation. The supernatant was then removed and the pellet resuspended in approximately 500 μL of PBS. A syringe pump was used to drive the fluid into chip at a speed of 500 μL/h.
Cell culture and cell size measurement
Three kinds of human bladder cancer cell lines (T24, 5637, and UM-UC-3) and a nonmalignant bladder epithelial cell line (SV-HUC-1) were purchased from Type Culture Collection of the Chinese Academy of Sciences, Shanghai, China. All cell lines were verified by a STR DNA profiling analysis and Mycoplasma contamination was ruled out using the Mycoplasma Stain Assay Kit (Beyotime). They were cultured at 37°C in a humid atmosphere of 5% CO2 and used for experiments within 10 passages after thawing. Typically, the T24 cells were cultured in McCoy's 5a Medium (Invitrogen) supplemented with 10% FBS (Gibco), the 5637 cells in the RPMI-1640 Medium (Gibco) supplemented with 10% FBS (Gibco) and the UM-UC-3 cells in the MEM Medium (Gibco) supplemented with 10% FBS (Gibco). The SV-HUC-1 cells were cultured in a F-12K Medium (SIGMA) supplemented with 10% FBS (Gibco). After growing to confluence, the cells were disassociated using the trypsin (Gibco), resuspended in a culturing medium and diluted to an appropriate concentration. A random portion of the cell resuspension was transferred into a 96-well plate. Microscopic images were taken (Leica, DM IL LED) with a phase contrast. The size of 250 cells were measured using a scale to generate their distribution histogram in order to correctly design the capture chamber of the chip.
On-chip immunofluorescence staining
After capture, the cells were fixed by passing a 4% PFA through the chip for 30 minutes, treated with a 0.1% Triton X-100 in PBS for 10 minutes to induce cellular permeability, followed by a PBS washing for 15 minutes and finally incubated with 5% BSA for 30 minutes to minimize the nonspecific bindings. Anti-CK20 and anti-CD44v6 primary antibodies (Abcam) were diluted in accordance to the manufacturer's instruction and infused into chip for 60 minutes at room temperature. Secondary antibodies conjugated with Alexa Fluor 488 and Alexa Fluor 594 (Invitrogen) were subsequently loaded into the chip at room temperature for 60 minutes, followed by DAPI for 20 minutes. The immunostaining process ended by washing with PBS before imaging. The flow rate was set at 500 μL/h during the entire procedure.
Identification and enumeration of UETCs and cell-line tumor cells via fluorescence microscopy
The microfluidic chip with captured cells was scanned under an inverted fluorescence microscope (Leica Microsystems, DM IL LED) and a digital camera (Leica DFC450). A putative UETC should meet the following criteria: clear fluorescence signal (DAPI+/CK20+/CD44v6+); appropriate cell size from 10 to 50 μm; identifiable cellular morphology with nucleus in a well-delimited cytoplasm. At the end, UETCs were enumerated and the count normalized to the number of UETCs per 10 mL of urine. To avoid introducing biasness into the results, investigators were not told beforehand the clinical diagnosis of patients whose UETCs were harvested from.
Pap staining and cytological evaluation process
Routine Pap Staining was performed in accordance to the standard protocol (19). For the on-slide pap smear, cells suspension was smeared and dried on the slide, fixed in 95% ethanol for 10 minutes and then, gently rinsed in running tap water for 1 minutes. The slides were then immersed in Hematoxylin for 5 minutes, washed with tap water and blued in lithium carbonate for 30 seconds. After washed in tap water and 95% ethanol, the slides were placed in OG stain for 2 minutes, dipped in 95% ethanol twice. This was followed by EA staining for 2 minutes and then, sequentially dehydrated in 95% and 100% ethanol. After two washes in xylene, the specimens were cover-slipped with a mounting media. For the on-chip pap staining, each solution was serially loaded into the microfluidic chip via a precise syringe pump control. After staining, the chip was directly examined under a microscope. All urinary cytology slides were read and independently evaluated by two cytopathologists (blinded to the clinical diagnosis) at the Pathology Department of the First Affiliated Hospital with the Zhejiang University School of Medicine.
Assessment of capture efficiency and intra-assay variability
To assess the chip performance, we define the following metric, CE = [(captured cells)/(total input cells) × 100%] and was assessed using cancer cell lines. The T24 cell suspension was serially diluted in PBS in a 96-well plate to about 200 cells/well. The exact number of cells in each well was manually enumerated under a microscope (Leica, DM IL LED) with a phase contrast. All cells in one well were then transferred and spiked into 500 μL PBS (to mimic the sampling method in actual processing of the urine). The cell resuspension was introduced into chip at several designed flow rates of 250, 500, 1000, 1500, and 2000 μl/h. After processing, we enumerated the number of captured cells in the chip to assess the CE. For each flow rate, five repeats were performed.
To investigate the repeatability and consistency of the microfluidic immunoassay, an intra-assay variability test is conducted. Stained T24 cell suspensions (DAPI+/CK20+/CD44v6+) with gradient concentrations of 10, 50, 100, 200, 500 cells in a 500 μL PBS are sampled. For each concentration, 3 replicates are tested, processed with a microfluidic chip and the corresponding CE computed. The intra-assay variability is defined by CV = [(standard deviation/ mean) × 100%] and is assessed for each replicate test. Generally, a CV < 20% indicates a high consistency between assay runs, hence, implying a reliable and reproducible assay result.
Comparison of cell loss between microfluidic immunoassay and routine pap smear
The RF provides a quantitative measure of the cell loss and is given by, RF = [(recovered cells in chip or on slide)/(total input cells) × 100%]. For the microfluidic immunoassay, the T24 cell suspension is serially diluted in a 96-well plate to the desired concentration and the exact number of cells is manually counted. The cell suspension is introduced into chip at 500 μL/h. After captured, an on-chip pap staining is performed and the number of cells enumerated on-chip to evaluate the RF.
For routine pap smear, the T24 cell suspension is serially diluted to the desired concentration in a 96-well plate and a droplet of cell suspension pipetted on an adhesion glass slide (Citoglas). The number of cells is manually counted and the slide dried at room temperature for 2 hours, followed by a routine pap staining process as described. After the whole-staining process, the cells remained on the slide are counted and the RF is determined. For each group of cell input, three repeat experiments are performed.
BTA stat and NMP22 BladderChek tests
The urine sample is also analyzed with 2 FDA-approved markers: BTA stat (Polymedco) and NMP22 BladderChek (Alere). Briefly, 3 and 4 drops of fresh urine are added. respectively, to the sample well of BTA stat and NMP22 BladderChek, and allowed for a room temperature reaction of 5 minutes for the former and 30 minutes for the latter. For both tests, a readout is assessed as positive when two visible lines are formed at both target and control zones, and negative when only the control line is formed. The absence of a visible control line implies an invalid test, necessitating a repeat test.
DNA harvesting from tumor tissue and single UETCs for whole genome sequencing
Fresh primary bladder tumor and the adjacent cancer-free bladder mucosa were harvested before administration of radiotherapy or chemotherapy. Upon collection, all specimens were snap frozen in liquid nitrogen before storage at −80°C. The gDNA of both tumor and the match normal tissue were extracted using the QIAamp DNA Micro Kit (QIAGEN) with concentration measured by Qubit 3.0 fluorometer with Qubit DSDNA BR Assay Kit (Life Technologies, Invitrogen). Typically, a total of approximately 600 ng gDNA was used for libraries construction using the NEBNext Ultra DNA Library Prep Kit for Illumina (New England Biolabs). All procedures were performed under the manufacturers' instructions.
DNA harvesting and genome sequencing of single tumor cells is challenging due to the extremely minute amount of DNA (∼5 pg/cell) involved. The first step is to recover the captured cells from the microfluidic chip. This is done by introducing a reversed PBS flow into the chip and the released cells collected for the fluorescence microscopy identification as stated in the previous section. The next step is the individual handling of the single cells using micropipettes and washing them several times in droplets of UV-exposed PBS to minimize any DNA contamination. They are moved one-by-one into separate PCR tubes containing 3 μL PBS. We employed the Multiple Displacement Amplification (MDA) method for the whole-genome amplification (WGA) of each single UETC using the QIAseq FX Single Cell DNA Library Kit (Qiagen).
To enhance success of the whole-genome sequencing, a 2-step quality control process is performed on the WGA products: assessing their concentration using Qubit 3.0 fluorometer with Qubit DSDNA BR Assay Kit and assessing their genomic integrity using a quantitative PCR (qPCR) with a check on 8 randomly selected loci (20). Typically, WGA products are deemed qualified when their DNA concentration is higher than 100 ng/μL and six out of eight loci have been amplified with a reasonable Ct number.
The whole-genome library for sequencing is constructed next. Each qualified WGA product underwent a PCR-free whole-genome library preparation based on the manufacturer's instructions for the QIAseq FX Single Cell DNA Library Kit (Qiagen). The prepared library is then sequenced on the platform illumina HiSEqXTen 2 × 150-bp PE reads at approximately 1.0 × depth, which is sufficient for our copy number analysis.
Bioinformatic analysis
The final step is to analyze the sequenced data of the whole genome. First, the raw sequence reads are preprocessed using After QC (21) to filter-out low-quality reads, trim adapters, and correct sequencing errors. Then, the clean sequence reads are aligned to the human reference genome (hg38) with Burrows-Wheeler aligner MEM (22), ordered and indexed using SAMtools (23). The copy number alteration (CNA) status of a single cell is analyzed using Ginkgo (24). Briefly, the mapped reads are binned by chromosome position, normalized for GC biases and other amplification artifacts, and then segmented to identify chromosome regions with consistent copy number status. The integer copy-number status is assigned to each segment, which allows Ginkgo to calculate hierarchical trees and heat maps from the copy number profiles of the collection of cells. The CNA status of the tissue is analyzed using Control-FREEC (25). The workflow of Control-FREEC is similar to Ginkgo, but developed specifically for calling copy numbers of tumor tissues. It uses mappability profiles to normalize read counts and correct possible contaminations from normal cells when constructing a copy-number profile of a tumor genome.
Statistical analysis
SPSS version 22.0 (SPSS, Inc.) and MATLAB version 2017a (The MathWorks, INC.) are used to analyze the data. The normality of the data initially is determined by the Kolmogorov–Smirnov test. Data with normal distribution is presented as a mean (SD); data without normal distribution is presented as a median (IQR, inter-quartile range). As the UETC counts are non-normally distributed, the Mann–Whitney test is performed for the nonparametric test between two groups. The Fisher's exact test is performed for a comparative analysis of the classified variable among the groups. The receiver operating characteristic (ROC) curves are generated from a cohort that consists of 79 patients with bladder cancer and 43 NCs. The area under the curve (AUC) is computed and the optimal cutoff (diagnostic threshold) is evaluated when the sum of sensitivity and specificity attained the maximum. To minimize the possibility of the training dataset yielding optimistic results, we performed a stratified k-fold cross validation on the computed optimal cutoff point to assess the performance of the predictive model in the new dataset (26). Briefly, the total training data is first randomly split into 10 sets (k = 10) keeping the ratio of NC to bladder cancer to be almost similar across all 10 sets, then 9 sets are used to train the model. The left-out dataset is then used to test the trained model. The whole process is reiterated until each of the dataset has been used as a testing set. The splitting of the total training dataset was done using the built-in MATLAB function crossvalind. By including a grouping vector that specifies the classification of patients, each NC/BC group was divided randomly into 10 sets and an index (ranging from 1 to 10 for k = 10) was assigned randomly to that particular group. Subjects with the same indices were then grouped together to form the 10 sets. The average AUC, sensitivity and specificity at the optimal cutoff point are finally computed. A two-tailed P < 0.05 is considered as statistically significant for all comparisons.
Results
Chip design of the capture chamber and a finite element simulation of the flow conditions
The size distribution of T24 cells is given in Fig. 1F with details listed in Supplementary Table S2A. From these data, key statistical parameters (min: 8.8 μm, max: 21.7 μm, mean: 14.8 μm, SD: 2.3 μm) were extracted and used to design the capture chamber of the chip. Furthermore, to handle the frequently encountered cell clusters (27), which could be more than double the size of individual tumor cells, the capture chamber consists of micro-units with 3 distinct sizes: 15, 30, 60 μm and they are collectively grouped in repeat patterns spread throughout the chamber (Fig. 1G). This novel design allows for a wide range of urinary cells, both single and clustered UETCs, to be captured and retrieved. In addition, the spatially gradated distribution of the micro-units enables a clean uncluttered segregation of trapped tumor cells of various sizes, making the process highly efficient as there is no debris to impede the flow: traditional cytology (Fig. 1H) versus microfluidic cytology (Fig. 1I).
A finite element simulation of the flow inside the capture chamber is depicted in Fig. 1J–M. The shear stresses generated by the flow provide an indication if a cell stays undamaged inside the capture chamber (28) for the varying flowrates studied (29). Under the 500 μL/h flowrate, the velocity increases in-between adjacent micro-units and decreases inside the area of a micro-unit (Fig. 1J). The streamlines (Fig. 1K) depict probable cell pathways during a flow. When streamlines intersect with a micro-unit, we assume that the cells are captured at that micro-unit. Furthermore, the wall shear stresses (Fig. 1L) of captured cells range from 0 to 2.98 Pa, which are safely consistent with reported stresses of 1.0 to 7.0 Pa in large arteries of humans (30). Similarly, the wall shear rates of 0 to 333.9 1/s (Fig. 1M), which are also within the reported safe range of human physiological states (from 10 1/s in veins to 2000 1/s in the smallest arteries; ref. 31). Hence, the finite element simulation gives us confidence that urinary cells have a good chance to stay intact during the microfluidic capture and this is confirmed by actual cell-line and voided urine experimentations.
Spiked cell-line experiments were carried out to evaluate the CE of the chip. We tested 5 flow rate groups that ranged from 250 to 2,000 μL/h and as expected, the CE decreases with increasing flowrates (Fig. 1N; Supplementary Table S2B). At high flowrates, many cells are damaged to bits and pieces, resulting in a significant drop in the capture efficiency. Therefore, to balance the CE with the need for the captured cells to remain undamaged, we selected the 500 μL/h flowrate as the optimal speed for our subsequent experiments.
Microfluidic chip achieves high consistency runs and minimizes cell loss
As shown in Fig. 1O, the intra-assay variability data for all 5 samples are less than 20% and this result provides an assurance of high consistency runs for good reliability and reproducibility of the microfluidic immunoassay outcomes (Supplementary Table S2C). Also, a cell loss analysis reveals that the recovery factor of the microfluidic immunoassay method is consistently higher than that of the routine pap smear approach for all 3 groups (Fig. 1P). This is especially important for minimizing cell loss and to allow for a more precise handling (32) during a sampling process of urine samples with sparse cell materials.
Combined use of CK20 and CD44v6 yields an accurate identification of UETCs
We use two oncoproteins CK20 and CD44v6 in combination to reliably and accurately identify UETCs. CK20 is an intermediate filament protein expressed in epithelial cells. It is significantly higher in urothelial tumors in comparison with normal transitional epithelium (33, 34) and has been considered as a biomarker for the 7 detection of bladder cancer (35–37); CD44v6 is a membrane adhesion molecule that is closely involved with cell invasion, tumor progression, and metastatic spread (38). The expression of CD44v6 is usually associated with stem-like cancer cells (39–41) and its role as a biomarker in bladder cancer has been widely explored (42–45). The flowchart for an immunofluorescence (IF) identification of putative UETCs using the combined CK20 and CD44v6 is given in Fig. 2A. A series of cell-line tests involving 3 human bladder cancer cell lines (T24, 5637 and UM-UC-3) and a nonmalignant bladder epithelial cell line (SV-HUC-1) was used to validate the performance of the marker-panel. The results showed an enhanced CK20 and CD44v6 expressions in bladder cancer cells compared with normal urothelial cells (Fig. 2B and C).
UETC identification flowchart, cell-line verification of proposed biomarkers, and ROC curves. A, Flowchart for immunofluorescent identification of putative UETCs. B, Cell-line test of the combined CK20 and CD44v6 to distinguish UETCs from normal urothelial cells. C, UETCs captured by the microfluidic chip observe the coarse cellular surface and eccentric nucleus in the morphology. D–F, ROC curves of the training set (red) and 10-fold cross validation (blue) for detecting different tumor grades: NCs versus all patients with bladder cancer (D); NCs versus patients with G1 bladder cancer (E); and NCs versus patients with G2 and G3 bladder cancer (F). The green dot in D–F represents the optimal cutoff point as listed in Table 2. The cross-validated ROC curves are smooth as they reflect the averaging effect of the 10-fold iterations; scale bar, 20 μm. BC, bladder cancer.
UETC identification flowchart, cell-line verification of proposed biomarkers, and ROC curves. A, Flowchart for immunofluorescent identification of putative UETCs. B, Cell-line test of the combined CK20 and CD44v6 to distinguish UETCs from normal urothelial cells. C, UETCs captured by the microfluidic chip observe the coarse cellular surface and eccentric nucleus in the morphology. D–F, ROC curves of the training set (red) and 10-fold cross validation (blue) for detecting different tumor grades: NCs versus all patients with bladder cancer (D); NCs versus patients with G1 bladder cancer (E); and NCs versus patients with G2 and G3 bladder cancer (F). The green dot in D–F represents the optimal cutoff point as listed in Table 2. The cross-validated ROC curves are smooth as they reflect the averaging effect of the 10-fold iterations; scale bar, 20 μm. BC, bladder cancer.
Baseline characteristics for subject groups and UETC count
The captured UETC (DAPI+/CK20+/CD44v6+) count is assessed from a combined cohort of the NC group (N = 43) and patients with bladder cancer (N = 79) and their baseline characteristics are listed in Table 1. As shown, the UETC count per 10 mL is significantly higher in the bladder cancer patient group compared with the NC group: [53.3 (10.7–1001.9) vs. 0.0 (0.0–3.0) UETCs/10 mL; P < 0.0001].
Baseline characteristics and UETC count of NCs and patients with bladder cancer
Patient variable . | NC . | BC Patients . | P . |
---|---|---|---|
N | 43 | 79 | |
Age | 61.4 (13.4) | 65.8 (12.4) | 0.071 |
Males, n (%) | 28 (65.1) | 61 (77.2) | 0.111 |
Body mass index | 23.9 (3.4) | 23.1 (2.9) | 0.187 |
Smoking History, n (%) | 0.884 | ||
Current | 11 (25.6) | 23 (29.1) | |
EX | 5 (11.6) | 10 (12.7) | |
Never | 27 (62.8) | 46 (58.2) | |
Drinking history, n (%) | 0.499 | ||
Current | 8 (18.6) | 18 (22.8) | |
EX | 6 (14.0) | 6 (7.6) | |
Never | 29 (67.4) | 55 (69.6) | |
Hematuria, n (%) | 4 (9.3) | 39 (49.4) | <0.001 |
Bladder irritation, n (%) | 9 (20.9) | 22 (27.8) | 0.270 |
Urine | |||
Leucocyte (/μL) | 19.3 (6.5–135.7) | 38.9 (6.9–146.6) | 0.408 |
Bacterium (/μL) | 56.1 (22.2–377.6) | 36.9 (14.58–298.3) | 0.271 |
Blood | |||
Serum creatinine (μmol/L) | 78.0 (65.0–90.0) | 85.0 (73.0–98.0) | 0.013 |
Serum urea (mmol/L) | 5.1 (3.9–6.7) | 5.9 (4.9–7.9) | 0.006 |
Serum uric acid (μmol/L) | 280 (227–346) | 335 (249–395) | 0.053 |
Recurrence, n (%) | |||
Yes | — | 27 (34.2) | — |
No | — | 49 (62.0) | — |
Unknown | — | 3 (3.8) | — |
Multifocality, n (%) | |||
Yes | — | 23 (29.1) | — |
No | — | 50 (63.3) | — |
Unknown | — | 6 (7.6) | — |
Tumor size, n (%) | |||
<2 cm | — | 24 (30.4) | — |
≥2 cm | — | 33 (41.8) | — |
Unknown | — | 22 (27.8) | — |
Grading, n (%) | |||
G1 | — | 29 (36.7) | — |
G2 | — | 12 (15.2) | — |
G3 | — | 34 (43.0) | — |
Unknown | — | 4 (5.0) | — |
Invasiveness, n (%) | |||
Yes | — | 25 (31.6) | — |
No | — | 50 (63.3) | — |
Unknown | — | 4 (5.1) | — |
UETC/10 mL, median (IQR). | 0.0 (0.0–3.0) | 53.3 (10.7–1001.9) | <0.0001 |
Patient variable . | NC . | BC Patients . | P . |
---|---|---|---|
N | 43 | 79 | |
Age | 61.4 (13.4) | 65.8 (12.4) | 0.071 |
Males, n (%) | 28 (65.1) | 61 (77.2) | 0.111 |
Body mass index | 23.9 (3.4) | 23.1 (2.9) | 0.187 |
Smoking History, n (%) | 0.884 | ||
Current | 11 (25.6) | 23 (29.1) | |
EX | 5 (11.6) | 10 (12.7) | |
Never | 27 (62.8) | 46 (58.2) | |
Drinking history, n (%) | 0.499 | ||
Current | 8 (18.6) | 18 (22.8) | |
EX | 6 (14.0) | 6 (7.6) | |
Never | 29 (67.4) | 55 (69.6) | |
Hematuria, n (%) | 4 (9.3) | 39 (49.4) | <0.001 |
Bladder irritation, n (%) | 9 (20.9) | 22 (27.8) | 0.270 |
Urine | |||
Leucocyte (/μL) | 19.3 (6.5–135.7) | 38.9 (6.9–146.6) | 0.408 |
Bacterium (/μL) | 56.1 (22.2–377.6) | 36.9 (14.58–298.3) | 0.271 |
Blood | |||
Serum creatinine (μmol/L) | 78.0 (65.0–90.0) | 85.0 (73.0–98.0) | 0.013 |
Serum urea (mmol/L) | 5.1 (3.9–6.7) | 5.9 (4.9–7.9) | 0.006 |
Serum uric acid (μmol/L) | 280 (227–346) | 335 (249–395) | 0.053 |
Recurrence, n (%) | |||
Yes | — | 27 (34.2) | — |
No | — | 49 (62.0) | — |
Unknown | — | 3 (3.8) | — |
Multifocality, n (%) | |||
Yes | — | 23 (29.1) | — |
No | — | 50 (63.3) | — |
Unknown | — | 6 (7.6) | — |
Tumor size, n (%) | |||
<2 cm | — | 24 (30.4) | — |
≥2 cm | — | 33 (41.8) | — |
Unknown | — | 22 (27.8) | — |
Grading, n (%) | |||
G1 | — | 29 (36.7) | — |
G2 | — | 12 (15.2) | — |
G3 | — | 34 (43.0) | — |
Unknown | — | 4 (5.0) | — |
Invasiveness, n (%) | |||
Yes | — | 25 (31.6) | — |
No | — | 50 (63.3) | — |
Unknown | — | 4 (5.1) | — |
UETC/10 mL, median (IQR). | 0.0 (0.0–3.0) | 53.3 (10.7–1001.9) | <0.0001 |
Abbreviation: BC, bladder cancer.
Microfluidic immunoassay of UETCs yields good detection accuracy
To evaluate if UETC count can be used as a complementary biomarker for bladder cancer detection, we performed an ROC analysis between the NC group and the different grades of bladder cancer patients: G1, G2, and G3 listed in Table 2,Table 3,Table 4. The AUC [95% confidence interval (CI)] are calculated by comparing NC group with the collective bladder cancer cohort (Fig. 2D), NC group with G1 bladder cancer patients (Fig. 2E) and NC group with combined G2 and G3 bladder cancer patients (Fig. 2F). The predictive accuracy of the training dataset, as measured by the AUC for the entire study cohort is 0.85 (95% CI, 0.78–0.92; P < 0.0001) with sensitivity and specificity at 89.9% (95% CI, 80.5%–95.2%) and 72.1% (95% CI, 56.1%–84.2%), respectively (Table 2). Furthermore, from the training dataset the optimal cutoff (Table 2) and marked by the green dot (Fig. 2D) is 1.95 UETC/10 mL urine. Through a 10-fold cross validation technique, the predictive performance of the model in new test dataset is found to be similar with an average AUC, sensitivity and specificity of 0.84 (95% CI, 0.76–0.93), 89.8% (95% CI, 71.5%–86.4%) and 71.5% (95% CI, 59.7%–83.3%), respectively. Also, as expected, the higher the grade of cancer, the more sensitive the combined CK20/CD44v6 marker is in detecting bladder cancer (94.0% G2 and G3 sensitivity vs. 90.0% G1 sensitivity in Table 2).
Training, validating and diagnostic performance of the microfluidic immunoassay method
Training dataset and validating with a 10-fold cross-validation of the total study cohort of N = 122 (79 BC and 43 NC). . | |||
---|---|---|---|
. | NC vs. BC (N = 122) . | NC vs. G1 BC (N = 72) . | NC vs. G2 and G3 BC (N = 89) . |
Training Set | |||
Optimal Cutoff Point (No. of UETCs/10 mL) | 1.95 | 0.95 | 3.50 |
AUC (95% CI) | 0.85 (0.78–0.92) | 0.78 (0.66–0.89) | 0.91 (0.85–0.97) |
P-value | <0.0001 | <0.0001 | <0.0001 |
Sensitivity (95% CI) | 89.9% (80.5%–95.2%) | 89.7% (71.5%–97.2%) | 93.5% (81.1%–98.3%) |
Specificity (95% CI) | 72.1% (56.1%–84.2%) | 67.4% (51.3%–80.5%) | 76.7% (61.0%–87.7%) |
10-fold Cross Validation | |||
Optimal Cutoff Point (No. of UETCs/10 mL) | 1.95 | 0.95 | 3.50 |
AUC (95% CI) | 0.84 (0.76–0.93) | 0.79 (0.68–0.89) | 0.91 (0.83–0.99) |
Sensitivity (95% CI) | 89.8% (71.5%–86.4%) | 90.0% (76.1%–100.0%) | 94.0% (85.6%–100.0%) |
Specificity (95% CI) | 71.5% (59.7%–83.3%) | 68.0% (52.9%–83.1%) | 76.5% (64.5%–88.5%) |
Training dataset and validating with a 10-fold cross-validation of the total study cohort of N = 122 (79 BC and 43 NC). . | |||
---|---|---|---|
. | NC vs. BC (N = 122) . | NC vs. G1 BC (N = 72) . | NC vs. G2 and G3 BC (N = 89) . |
Training Set | |||
Optimal Cutoff Point (No. of UETCs/10 mL) | 1.95 | 0.95 | 3.50 |
AUC (95% CI) | 0.85 (0.78–0.92) | 0.78 (0.66–0.89) | 0.91 (0.85–0.97) |
P-value | <0.0001 | <0.0001 | <0.0001 |
Sensitivity (95% CI) | 89.9% (80.5%–95.2%) | 89.7% (71.5%–97.2%) | 93.5% (81.1%–98.3%) |
Specificity (95% CI) | 72.1% (56.1%–84.2%) | 67.4% (51.3%–80.5%) | 76.7% (61.0%–87.7%) |
10-fold Cross Validation | |||
Optimal Cutoff Point (No. of UETCs/10 mL) | 1.95 | 0.95 | 3.50 |
AUC (95% CI) | 0.84 (0.76–0.93) | 0.79 (0.68–0.89) | 0.91 (0.83–0.99) |
Sensitivity (95% CI) | 89.8% (71.5%–86.4%) | 90.0% (76.1%–100.0%) | 94.0% (85.6%–100.0%) |
Specificity (95% CI) | 71.5% (59.7%–83.3%) | 68.0% (52.9%–83.1%) | 76.5% (64.5%–88.5%) |
NOTE: Grade information of 4/79 bladder cancer patients is not accessible.
Abbreviation: BC, bladder cancer.
Performance comparison of four techniques for a cohort of N = 40 (20 BC and 20 NC)
Technique . | Sensitivity . | Specificity . | ||
---|---|---|---|---|
. | G1 BC (N = 6) . | G2 and G3 BC (N = 14) . | Total BC (N = 20) . | Controls (N = 20) . |
Microfluidic Immunoassay | 83.3% (5/6) | 100.0% (14/14) | 95.0% (19/20) | 80.0% (16/20) |
Traditional Cytology | 16.6% (1/6) | 42.9% (6/14) | 35.0% (7/20) | 100.0% (20/20) |
BTA stat | 16.6% (1/6) | 85.7% (12/14) | 65.0% (13/20) | 60.0% (12/20) |
NMP22 Bladder Check | 16.6% (1/6) | 42.9% (6/14) | 35.0% (7/20) | 80.0% (16/20) |
Technique . | Sensitivity . | Specificity . | ||
---|---|---|---|---|
. | G1 BC (N = 6) . | G2 and G3 BC (N = 14) . | Total BC (N = 20) . | Controls (N = 20) . |
Microfluidic Immunoassay | 83.3% (5/6) | 100.0% (14/14) | 95.0% (19/20) | 80.0% (16/20) |
Traditional Cytology | 16.6% (1/6) | 42.9% (6/14) | 35.0% (7/20) | 100.0% (20/20) |
BTA stat | 16.6% (1/6) | 85.7% (12/14) | 65.0% (13/20) | 60.0% (12/20) |
NMP22 Bladder Check | 16.6% (1/6) | 42.9% (6/14) | 35.0% (7/20) | 80.0% (16/20) |
Abbreviation: BC, bladder cancer.
Performance Comparison of standalone tests with serial and parallel tests of N = 65 (45 BC and 20 NC)
Test . | Sensitivity (95% CI) . | Specificity (95% CI) . | Youden Index . |
---|---|---|---|
Standalone Test: Microfluidic Immunoassay | 91.1% (77.9%–97.1%) | 80.0% (55.7%–93.4%) | 71.1% |
Standalone Test: Traditional Cytology | 26.7% (15.1%–42.2%) | 100.0% (80.0%–100.0%) | 26.7% |
Serial Test: Microfluidic Immunoassay "AND" Cytology | 24.4% (13.4%–39.9%) | 100.0% (80.0%–100.0%) | 24.4% |
Parallel Test: Microfluidic Immunoassay "OR" Cytology | 93.3% (80.7%–98.2%) | 80.0% (55.7%–93.4%) | 73.3% |
Test . | Sensitivity (95% CI) . | Specificity (95% CI) . | Youden Index . |
---|---|---|---|
Standalone Test: Microfluidic Immunoassay | 91.1% (77.9%–97.1%) | 80.0% (55.7%–93.4%) | 71.1% |
Standalone Test: Traditional Cytology | 26.7% (15.1%–42.2%) | 100.0% (80.0%–100.0%) | 26.7% |
Serial Test: Microfluidic Immunoassay "AND" Cytology | 24.4% (13.4%–39.9%) | 100.0% (80.0%–100.0%) | 24.4% |
Parallel Test: Microfluidic Immunoassay "OR" Cytology | 93.3% (80.7%–98.2%) | 80.0% (55.7%–93.4%) | 73.3% |
A further analysis revealed several statistically significant correlations between the UETC count and clinical outcomes [Fig. 3A–F and numerical values (median and interquartile range) in Supplementary Table S3A]. The UETC count is significantly higher for patients with bladder cancer versus the NC group [53.3 (95% CI, 10.7–1001.9) vs. 0.0 (95% CI, 0.0–3.0) UETCs/10 mL; P < 0.0001; Fig. 3A; Supplementary Fig. S2]. Also, there is a significant increase in the UETC count for G2 and G3 bladder cancer patients versus G1 bladder cancer patients [485.7 (95% CI, 37.6–2015.0) vs. 13.3 (95% CI, 2.7–49.6) UETCs/10 mL; P < 0.0001; Fig. 3B].
Correlations between UETC count and key clinical outcomes. A, Comparing the UETC count between the NC and the collective bladder cancer (BC) groups. B, Comparing the UETC count between patients with G1 and G2/G3 bladder cancer. As expected, the UETC count increases with the increasing grade of the cancer. C, Correlating the UETC count with the tumor size. Once again, larger tumors generate much more UETCs. For the last three metric comparisons: correlating the UETC count with the tumor invasiveness (D), correlating the UETC count with the multifocality of the tumor (E), and correlating the UETC count with the tumor recurrence (F), the trend is still the same—UETCs increase with the severity of the cancer. Data are expressed as the median with IQRs. The P value refers to the results of the Mann–Whitney test. Each dot represents the UETC count of an individual.
Correlations between UETC count and key clinical outcomes. A, Comparing the UETC count between the NC and the collective bladder cancer (BC) groups. B, Comparing the UETC count between patients with G1 and G2/G3 bladder cancer. As expected, the UETC count increases with the increasing grade of the cancer. C, Correlating the UETC count with the tumor size. Once again, larger tumors generate much more UETCs. For the last three metric comparisons: correlating the UETC count with the tumor invasiveness (D), correlating the UETC count with the multifocality of the tumor (E), and correlating the UETC count with the tumor recurrence (F), the trend is still the same—UETCs increase with the severity of the cancer. Data are expressed as the median with IQRs. The P value refers to the results of the Mann–Whitney test. Each dot represents the UETC count of an individual.
A higher UETC count is observed among patients with tumor size ≥2 cm compared with those <2 cm [133.3 (95% CI, 26.6–1649.0) vs. 17.3 (95% CI, 3.0–148.0) UETCs/10 mL; P < 0.05], respectively (Fig. 3C). In contrast, no significant correlations are observed between the UETC count and other clinical outcomes such as invasiveness, multifocality, recurrence, etc. (Fig. 3D–F).
Paired comparison between microfluidic immunoassay and urinary markers
The traditional urinary cytology and other recently developed urine markers for clinical diagnosis and surveillance of bladder cancer are often hampered by their limited sensitivity (46). To examine the performance of our microfluidic immunoassay method with other techniques, we initiated a paired comparison between our microfluidic immunoassay with the traditional cytology and two commercially available, FDA-approved markers—BTA stat and NMP22 BladderChek on 20 randomly enrolled patients with bladder cancer and 20 NCs. Every urine sample was divided into 2 major aliquots for microfluidic immunoassay and urinary cytology keeping several drops for BTA and NMP22. The microfluidic immunoassay and cytology results were evaluated independently by 2 cytopathologists, and the BTA and NMP22 results by two independent technicians, all of whom have not been pre-informed of the status of the clinical diagnosis.
The direct comparison is shown in Fig. 4A with detailed information of each individual given in Supplementary Table S3B. The results for the sensitivity and specificity of the 4 urinary markers are tabulated in Table 3. In addition to a higher overall sensitivity achieved by the microfluidic immunoassay, another significant advantage of the method is that the sensitivity across tumor grades is acceptably high, albeit with a weakened performance in differentiating low-grade tumors (83.3% for G1 vs. 100% for G2 and 3), which is in stark contrast with the unacceptably low sensitivity of 16.6% for the 3 markers.
Paired comparison between microfluidic immunoassay and urinary markers. A, Comparison among microfluidic immunoassay, traditional cytology, BTA stat, NMP22 BladderChek in an example sample of 40 consisting of 20 patients with bladder cancer (BC) and 20 NCs. B, Representative images of UETCs captured by microfluidic chip; scale bar, 20 μm.
Paired comparison between microfluidic immunoassay and urinary markers. A, Comparison among microfluidic immunoassay, traditional cytology, BTA stat, NMP22 BladderChek in an example sample of 40 consisting of 20 patients with bladder cancer (BC) and 20 NCs. B, Representative images of UETCs captured by microfluidic chip; scale bar, 20 μm.
On the specificity side, both the microfluidic immunoassay and NMP22 BladderChek achieved an above average readout of 80% but still is significantly lower than the perfect specificity of the traditional cytology. Supplementary Table S3C summarizes the list of urinary markers that wrongly identified NCs with non-cancer urinary tract diseases as having bladder cancer. Not surprisingly, the list includes microfluidic immunoassay and this error is due to the well-known observation that noncancer urinary tract diseases often generate a false positive readout (3).
This preliminary study demonstrates the use of a combined CK20/CD44v6 oncoproteins to mark UETCs and the results show that the microfluidic immunoassay approach could be promising in detecting bladder cancer as the technique is more sensitive compared with other methods while maintaining a reasonably high level of specificity. Figure 4B shows the representative images of traditional cytology and microfluidic immunoassay.
Single-cell sequencing of captured UETCs
Supplementary Fig. S3 depicts the capture and recovery of a single cell from the voided urine of a patient with bladder cancer for a genomic analysis. From the harvested cell cohort of 3 patients with bladder cancer, we collected a total of 15 single cells comprising of 12 immunofluorescence-identified UETCs (DAPI+/CK20+/CD44v6+) and 3 normal urothelial cells (NUC) to serve as a sequence reference. We extracted the DNA from each captured cell for the WGA and whole-genome sequencing (∼1 × sequencing depth) to see if the 12 IF-identified UETCs possess genomic instability associated with tumor cells. The CNAs of single cells were analyzed with the low-pass sequencing strategy (20, 47).
The results showed that 11/12 (91.7%) of the IF-identified UETCs possess unstable genome and are clearly, tumor cells with CNAs, while the 3 NUCs are confirmed to be diploids. Figure 5 depicted in detail the CNA profiles of 4 IF-identified UETCs in contrast to that of a NUC. Supplementary Fig. S4 showed the paired tumor tissue from the same patient.
CNA profiles from single-cell sequencing of UETCs. The 500-kbp resolution CNV calling on a normal urothelial cell and four IF-identified UETCs. The regions of normal, gain, and loss copy numbers are indicated in black, red, and blue color, respectively.
CNA profiles from single-cell sequencing of UETCs. The 500-kbp resolution CNV calling on a normal urothelial cell and four IF-identified UETCs. The regions of normal, gain, and loss copy numbers are indicated in black, red, and blue color, respectively.
Discussion
Advantages of a chip-based microfluidic approach
For patients with bladder cancer, the components in their urine are complicated by the presence of UETCs, in addition to NUCs, red blood cells, crystals, urinary cylinders, and other impurities. As urinary cytology is based on centrifugation and smearing, all these components get trapped on a slide that not only result in a highly debris-cluttered view but also, adversely affect a pathologist's cytologic assessment. Our microfluidic chip does not suffer from this problem as the capture chamber depicts a clear and unobstructed view of the trapped cells devoid of debris build-up (Fig. 1H and I). In addition, the microfluidic technology is advantageous over batch-sampling techniques in terms of minimizing cell loss when dealing with sparse cell materials, because it limits the processing environment to nanoliter scale that makes it possible to carry out a precise handling of minute and sparse samples. Furthermore, microfluidic immunoassay is based on cell enumeration and is totally objective compared with traditional cytology, which is highly subjective and dependent on the skills and experience of a pathologist. Likewise, among commercial urinary markers, FISH in particular, requires scorers to be sufficiently skilled in carrying out a good analysis.
Multimarkers to enhance diagnostic accuracy
The 2 biomarkers CK20 and CD44v6 used in combination in our study to mark UETCs achieved an accurate diagnosis of bladder cancer (with a validated AUC of 0.84, sensitivity of 89.8% and specificity of 71.5%), because both of them have been separately employed to mark and identify bladder cancer in tumor tissues (33–45).
In a normal urothelium, CD44v6 is usually expressed in the basal layer with negativity towards the topmost surface layer of umbrella cells (42, 43) whereas, the CK20 expression is only restricted to superficial umbrella cells and therefore, absent in the basal layer (35). However, in tumor tissues, both CK20 and CD44v6 are frequently overexpressed in all urothelial layers (35, 42, 43). This unique expression pattern of CK20 and CD44v6 justifies our simultaneous use of the 2 biomarkers to realize a more robust detection of UETCs. As a result, we find significant correlations between UETC count and several key parameters (tumor grade and size) in Fig. 3.
CNA status of IF-identified UETCs
All UETCs are aneuploid with a chromosomal instability phenotype in contrast with the diploid NUC, which is due to the elevation of lagging chromosomes during anaphase and higher rates of chromosome mis-segregation in tumor cells. Though coming from the same patient, different UETCs exhibit distinct ploidy status and this implies the existence of intratumor heterogeneity. Differing from the genome-wide CNA profile of UETCs, the paired tumor tissue displays limited regions with CNAs (Supplementary Fig. S4). It is well known that bulk sequencing generates an averaged data with tumor signals masked by those coming from a complex population of cells that includes normal stromal cells. Furthermore, compounding these admixture tumor signals, many solid tumors are believed to harbor both epithelial and mesenchymal populations (48). Even so, the CNA regions in the paired tumor tissue have all been identified in the UETCs, which confirms that the IF-identified UETCs are indeed, real cancer cells sourced directly from the primary tumor in the bladder. They possess unique copy-number profiles with unstable genomes, suggesting that there are individual differences in the genome evolution of the tumorigenesis. Our microfluidic capture and recovery of intact UETCs constitutes an essential step for a downstreamed molecular analysis required in precision medicine and personalized treatment.
Clinical utility of UETCs and future prospect
Paired comparisons between microfluidic immunoassay and traditional cytology showed a more than 2-fold higher sensitivity difference. This is likely due to 2 reasons. First, the complicated urinary components and the muddled debris field in traditional cytology often interfere with a cytopathologist's judgment in contrast with a clean and uncluttered field of view in the microfluidic approach. Second, expressions of protein may precede the cytologic features. In other words, for tumor cells that have not yet displayed distinctive tumor morphological characteristics, they might already have expressed tumor related proteins. As observed, there are 20 patients with bladder cancer >1,000 UETCs/10 mL and the majority are Grade 2 and 3 patients (19/20, unknown: 1 case; Fig. 3B) with tumor size >2 cm (11/20, unknown: 6 cases; Fig. 3C). Clearly, advanced-stage patients shed a significantly larger amount of UETCs into their urine.
Similar to the proposed IF method, both the BTA stat and NMP22 BladderChek detect abnormal levels of urinary plasma H-related protein and urinary nuclear matrix protein 22, respectively, for the detection of bladder cancer. Given the enhanced predictive performance of the microfluidic immunoassay, it suggests that UETCs marked by CK20 and CD44v6 could become a promising biomarker for bladder cancer.
We also reported on a preliminary study to investigate the feasibility of a clinical utility by combining microfluidic immunoassay (high sensitivity) with cytology (high specificity). We looked at 65/122 patients who had both microfluidic immunoassay and traditional cytology test data, and computed the performance of a combined microfluidic immunoassay and cytology in either a serial ("AND") or parallel ("OR") test strategy compared with the standalone test of either microfluidic immunoassay or cytology. As expected, a serial test strategy yielded a 100% specificity (Table 4); however, its sensitivity is too low to be clinically useful. On the other hand, a parallel test strategy yielded an improved sensitivity (93%) compared with either of the standalone test but with a reduced specificity of 80% coming from the standalone microfluidic immunoassay test. In terms of the Youden index, the parallel test strategy is much higher than that of the serial test strategy. Although the parallel test strategy yielded a marginally higher sensitivity and Youden index over the standalone microfluidic immunoassay, it is reasonable to recommend adopting the parallel test strategy for possible clinical implementation. The other option is to consider using only the standalone microfluidic immunoassay approach in lieu of traditional cytology, but we do not think this is a good idea as it involves re-educating users to replace the traditional cytology with the standalone microfluidic immunoassay approach. Our clinical utility study is still preliminary as the cohort size is too small to yield a reasonable confidence interval in our specificity data.
We note that standardization methods for urinary biomarkers involving UETCs are rare and having examined several possible approaches, we are of the opinion that the use of urine quantity to normalize UETCs while not perfect, appears to be consistent with the conclusion put forth in the major study of Reid and colleagues (49).
We conclude by stating that although our microfluidic immunoassay of UETCs appears promising, we still need a large-scale clinical trial to verify our results, particularly for a clinical implementation. Because of the limited sample size and the fact that only intra-study 10-fold cross validation was used to verify the diagnostic model, the reported performance might be overly optimistic. To reflect a more objective diagnostic capability, a proper external validation derived from a new dataset (perhaps in a multicenter study) would be welcomed. However, our method should lead to an improved sensitivity detection for bladder cancer by a parallel combination of microfluidic immunoassay with traditional cytology and also, a reduced frequency of cystoscopy but we do not expect it to completely supplant the two conventional techniques for bladder cancer diagnosis.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Authors' Contributions
Conception and design: A. Chen, G. Fu, Z. Xu, Z. Tang, B. Jin, R.P.S. Han
Development of methodology: A. Chen, G. Fu, Y. Sun, X. Chen, K.H. Neoh, Z. Tang, Y. Dai, B. Jin, R.P.S. Han
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Chen, G. Fu, X. Chen, Z. Tang, Y. Dai, Q. Wang, J. Jin, B. Jin
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Chen, G. Fu, K.S. Cheng, Z. Tang, S. Chen, M. Liu, T. Huang, J. Jin, B. Jin, R.P.S. Han
Writing, review, and/or revision of the manuscript: A. Chen, G. Fu, Y. Sun, K.S. Cheng, K.H. Neoh, Z. Tang, S. Chen, M. Liu, B. Jin, R.P.S. Han
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Chen, G. Fu, Z. Xu, Y. Sun, K.H. Neoh, B. Jin, R.P.S. Han
Study supervision: B. Jin, R.P.S. Han
Acknowledgments
The authors thank Drs. Hao Pan and Yimin Wang for their advice and assistance in patient management. We also appreciate the use of facilities and technicians at The Key Laboratory of Combined Multi-Organ Transplantation, Ministry of Public Health, Hangzhou, Zhejiang, China. This project was supported by the Key Project of the Science and Technology Program of Zhejiang Province (), Grant No: 2014C03028 and Grantees: B. Jin (PI) and R.P.S. Han (Co-PI) and also, by the Special Funds for Future Industries of Shenzhen, Grant No: JSGG20160229123927512 and Grantee: Shifu Chen.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.