Study population
The study cohort consisted of patients with histologically confirmed urothelial carcinoma and healthy donors. The cancer cohort included specimens that were purchased from Geneticist Inc. (Glendale, CA, USA), and samples collected at the Department of Urology, USC Norris Comprehensive Cancer Center (Los Angeles, CA, USA), in accordance with an institutional review board approved protocol. The presence of BC, as well as the tumor type and grade, was confirmed by cystoscopy and histology from resections or biopsies. No exclusion criteria were applied for BC patients. Samples from healthy donors were collected from consenting volunteers over 21 years old. “Healthy” status was based on self-reporting and defined as no history of any type of tumor. No follow-up information was collected for both cohorts.
Urine sample collection, stabilization, and shipment
Urine samples were collected and stabilized using the Bladder CARE Urine Collection Kit (Pangea Laboratory). No specific guidelines were established for the timing of urine sample collection. Samples purchased from Geneticist Inc. (Glendale, CA, USA) were analyzed with Multistix 10SG Urinalysis Test Strip (Siemens) prior to stabilization. Stabilized urine samples can be kept at room temperature for up to one month without DNA degradation or loss [42] and were mailed at ambient temperature to Pangea Laboratory for Bladder CARE analysis. A detailed description of the Bladder CARE workflow is presented in Fig. 1.
Clinical procedures and Bladder CARE test
Stabilized urine was processed and analyzed at Pangea Laboratory. After isolation using Quick-DNA™ Urine Kit (Zymo Research, D3061), urine DNA samples were quantified with Femto™ Human DNA Quantification Kit (Zymo Research, E2005). As low as 5 ng of urine DNA was analyzed in duplicate with Bladder CARE.
Briefly, the test measures the methylation level of three BC-specific biomarkers and two internal control loci (the last informing about the input DNA amount used in the test and the efficiency of the digestion step for each sample tested) using methylation-sensitive restriction enzymes coupled with qPCR [39]. Positive and negative Bladder CARE control samples were also analyzed in parallel with clinical samples in order to confirm the validity of the test. A detailed description of the method is reported elsewhere [39, 40].
Calculation of Bladder CARE test results (Bladder CARE Index—BCI)
Bladder CARE results are expressed as the Bladder CARE Index (BCI). BCI values are calculated by integrating the methylation level of the three BC biomarkers and the two internal controls in a proprietary algorithm developed by Pangea Laboratory [39, 40].
Based on BCI values, samples are categorized as ‘Negative,’ ‘High-Risk,’ and ‘Positive.’ Specifically, samples with BCI < 2.5 are considered Negative for the presence of BC, while samples with BCI between 2.5 and 5, and > 5 are classified as High-Risk and Positive, respectively ([40], Fig. 1).
Statistical analyses
Results from a previous pilot study (57 healthy subjects and 51 BC patients; unpublished) were used to define each group’s mean BCI values with 95% CIs for the control and the cancer cohorts (control mean of 1.3 with a 95% CI of 0.994 to 1.61 and cancer mean of 213.7 with a 95% CI of 86.7 to 341). We conducted a power analysis using the n_2t_unequal() function from the dvmisc package in R, which allows group-specific variances (the cancer group has a much higher variance than control). Using the group-specific variances from the pilot data, an alpha threshold of 0.05, and a difference of means of 200 BCI units, we established that we would need 45 samples per group to achieve 80% power in a comparison of control and cancer groups. In our study we included 77 BC patients and 136 healthy subjects. Using the power_2t_unequal() function we calculated that a sample size of 77 per group would result in 96.2% power at alpha of 0.05 to detect a difference in means of 200 BCI units.
Sensitivity, specificity, PPV and NPV of Bladder CARE were calculated based on the number of true-positive, true-negative, false-positive, and false-negative cases. All the samples classified as Negative by Bladder CARE (BCI < 2.5) were considered negative in our study, while all the samples having a BCI > 2.5 (High-Risk and Positive Bladder CARE results) were considered positive.
The statistical significance of differences in BCI values was determined using Student’s two-tailed t tests. Box plots of the cohorts by BCI values and the receiver operating characteristic (ROC) curve for the BCI classification (Fig. 2) were generated in python 3.7.2 using numpy 1.16.2 and pandas 0.24.2 packages for data processing, custom code to generate the true-positive rates and false-positive rates for the ROC curve sliding threshold, scikit-learn 0.20.3 to calculate the ROC area under the curve (AUC), and seaborn 0.9.0 with matplotlib 3.0.2 for visualization.
To provide an estimated probability of BC occurrence based on the BCI (Fig. 3, Additional file 1: Figure S1), a logistic regression model was developed utilizing BCI values and their corresponding clinical sample information (cystoscopy results and self-report). Logistic regression modeling was evaluated using a stratified k-fold cross-validation approach, where the dataset was divided into five similarly sized subsets containing a ratio of outcomes representative of the total dataset. Model selection and generation was then repeated five times (training with four of the subsets and testing with one), rotating the subset to be tested. Diagnostics measurements were generated for each fold, with the resulting model being chosen where the measurements most closely matched the mean of the cross-validation diagnostics. Calculations were performed in python 3.7.2 using the numpy 1.16.2 and pandas 0.24.2 packages for data processing, statsmodels 0.9.0 for model generation, scikit-learn 0.20.3 for additional validation, and seaborn 0.9.0 with matplotlib 3.0.2 for visualization.
Comparison between MSRE-qPCR and MS-qPCR
Six standard DNA samples (800 ng each) were generated by combining different proportions of untreated and artificially methylated blood DNA (produced using M.SssI CpG Methylase; Zymo Research, E2011). Specifically, samples 1 to 6 contained 100%, 33.3%, 11.1%, 3.7%, 1.23%, and 0.41%, of artificially methylated DNA. In total, 400 ng of each sample was bisulfite-treated using EZ DNA Methylation-Lightning Kit (Zymo Research, D5031), while the remaining 400 ng was digested using 10U of the methylation-sensitive restriction enzyme HpaII (New England BioLabs, R0171S) and purified with DNA Clean and Concentrator-5 (Zymo Research, D4013). Samples were then amplified by qPCR (CFX96 Touch qPCR System, Bio-Rad) using primers designed for bisulfite-converted methylated DNA (MS-qPCR) and genomic untreated DNA (MSRE-qPCR). Primers (sequences available upon request) have similar efficiencies and were designed on the same region of the CpG island of the human MGMT gene, which is known to be unmethylated in blood samples collected from healthy donors (Additional file 2: Table S1). Both amplicons have similar length and were amplified in triplicates in 20 µl of reaction containing ZymoTaq qPCR Premix (Zymo Research, E2055), 0.4 µM of each primer, and the equivalent of 100 ng of input DNA prior to bisulfite conversion or digestion. The program used to generate both amplicons has an initial 10-min denaturation step at 95 °C followed by 40 cycles of denaturation at 97 °C for 20 s and annealing/extension at 58 °C for one minute. Data were analyzed using Bio-Rad CFX Maestro Software (Bio-Rad).
Methylation of the human MGMT target region in untreated and artificially methylated blood DNA was determined by MSRE-qPCR according to the protocol described above. Results are represented in Additional file 2: Table S1.
LOD and linearity of Bladder CARE
A set of 12 spike-in samples containing different amounts of LD583 (bladder carcinoma cell line [41]) DNA in a background of blood DNA isolated from a healthy donor were generated. Specifically, samples 1 to 12 contain 100%, 33.3%, 11.1%, 3.7%, 1.23%, 0.41%, 0.14%, 0.046%, 0.015%, 0.005%, 0.0017%, and 0% of LD583 DNA, respectively (Additional file 2: Table S2). In total, 2 µg of each sample was submitted to Pangea Laboratory and analyzed in triplicates (500 ng each) with Bladder CARE. The number of cancer cells originally used to generate the standards was calculated considering that a single human genomic DNA molecule weighs 3.59 pg [43]. For each sample, standard deviation and standard error were calculated based on the BCI values obtained from each of the three technical replicates. Student’s two-tailed t test was performed in order to determine the significance of BCI changes between consecutive samples (Additional file 2: Table S2).