Non-invasive diagnosis and surveillance of bladder cancer with driver and passenger DNA methylation in a prospective cohort study

Abstract

Background: State-of-art non-invasive diagnosis processes for bladder cancer (BLCA) harbour shortcomings such as low sensitivity and specificity, unable to distinguish between high- (HG) and low-grade (LG) tumours, as well as inability to differentiate muscle-invasive bladder cancer (MIBC) and non-muscle-invasive bladder cancer (NMIBC). This study investigates a comprehensive characterization of the entire DNA methylation (DNAm) landscape of BLCA to determine the relevant biomarkers for the non-invasive diagnosis of BLCA.

Methods: A total of 304 samples from 224 donors were enrolled in this multi-centre, prospective cohort study. BLCA-specific DNAm signature discovery was carried out with genome-wide bisulfite sequencing in 32 tumour tissues and 12 normal urine samples. A targeted sequencing assay for BLCA-specific DNAm signatures was developed to categorize tumour tissue against normal urine, or MIBC against NMIBC. Independent validation was performed with targeted sequencing of 259 urine samples in a double-blinded manner to determine the clinical diagnosis and prognosis value of DNAm-based classification models. Functions of genomic region harbouring BLCA-specific DNAm signature were validated with biological assays. Concordances of pathology to urine tumour DNA (circulating tumour DNA [ctDNA]) methylation, genomic mutations or other state-of-the-art diagnosis methods were measured.

Results: Genome-wide DNAm profile could accurately classify LG tumour from HG tumour (LG NMIBC vs. HG NMIBC: p = .038; LG NMIBC vs. HG MIBC, p = .00032; HG NMIBC vs. HG MIBC: p = .82; Student's t-test). Overall, the DNAm profile distinguishes MIBC from NMIBC and normal urine. Targeted-sequencing-based DNAm signature classifiers accurately classify LG NMIBC tissues from HG MIBC and could detect tumours in urine at a limit of detection of less than .5%. In tumour tissues, DNAm accurately classifies pathology, thus outperforming genomic mutation or RNA expression profiles. In the independent validation cohort, pre-surgery urine ctDNA methylation outperforms fluorescence in situ hybridization (FISH) assay to detect HG BLCA (n = 54) with 100% sensitivity (95% CI: 82.5%-100%) and LG BLCA (n = 26) with 62% sensitivity (95% CI: 51.3%-72.7%), both at 100% specificity (non-BLCA: n = 72; 95% CI: 84.1%-100%). Pre-surgery urine ctDNA methylation signature correlates with pathology and predicts recurrence and metastasis. Post-surgery urine ctDNA methylation (n = 61) accurately predicts recurrence-free survival within 180 days, with 100% accuracy.

Conclusion: With the discovery of BLCA-specific DNAm signatures, targeted sequencing of ctDNA methylation outperforms FISH and DNA mutation to detect tumours, predict recurrence and make prognoses.

Keywords: bladder cancer; diagnosis and prognosis; methylation; non-invasive screening; prospective cohort study; urine tumour DNA.

FIGURE 1

Flow chart of the study.

FIGURE 2

Classification of bladder cancer (BLCA) differentially methylated region (DMR). (A) Cancer‐associated DMR could result from four different scenarios: (1) inherited cell‐type‐specific DNA methylation (DNAm) signature from the clonal ancestor of cancer, which is a normal cell (Type I, T1DMR); (2) de novo DNAm shift during oncogenesis (Type II, T2DMR); (3) inherited cell‐type‐specific DNAm signature from a non‐cancer cell which is absent in normal tissue (Type III, T3DMR), such as immune cells; and (4) de novo DNAm shift accompanied oncogenesis in a non‐cancer cell type (Type IV, T4DMR), such as reprogrammed fibroblast. Considering only the cancer‐cell‐derived DNAm, two types of cancer‐cell‐associated DMR might present: the T1DMR, which is present in the ancestral cell of cancer, and T2DMR, which underwent tumour‐specific DNAm change. (B) T1DMR does not contain tumour‐specific haplotype (TSH) and shows low haplotype diversity. Chromatin accessibility on T1DMR does not change between ancestral cells and cancer. But T1TSH prevalence is highly correlated with tumour fraction in tissues. In contrast, T2DMR exhibits that high haplotype diversity has TSH and shows chromatin accessibility change. Furthermore, T2TSH prevalence is linked to tumour grade or clinical stage. (C) Association of haplotype from T1DMR/T2DMR with pathological traits. (D) Examples of four T2DMR that display oncogenesis‐associated de novo methylation/demethylation, as revealed by gradually changing TSH prevalence in tumour samples.

FIGURE 3

Selective driver and tissue‐of‐origin DNA methylation signature outperform tumour genomic mutation in classifying bladder cancer (BLCA) tumour tissues. (A) DNA methylation and mutation profiling on resected non‐muscle‐invasive bladder cancer (NMIBC) or muscle‐invasive bladder cancer (MIBC) tumour tissues showing haplotype (rows) prevalence in pathologically defined tumour samples (columns). Haplotype prevalence is Z‐scaled. Pathological classifications (grade, T‐stage, invasiveness [invasive] and muscle‐invasion) and DNA mutations of known BLCA‐associated oncogenes and tumour suppressors are revealed in the heat map. DNA methylation haplotype prevalence strongly correlates with pathological grade and invasiveness in tumour tissues. (B) DNA methylation‐based BLCAS classifier score predicts tumour grade in MIBC and NMIBC tissues. (C) The presence of mutations or pathology features, or DNA methylation class (luminal or basal, defined by haplotype prevalence), in NMIBC and MIBC.

FIGURE 4

Urine DNA methylation signal non‐invasively detects bladder cancer (BLCA). (A) Experiment design: DNA methylation assays were performed on pre‐surgery urine and validated with resected pathology classification. (B) Cancer‐specific methylation score (cancer methylation score) of individual samples. Samples are grouped/coloured by their class (normal, LG and HG). The individual pathological type of each sample is denoted as the shape of a dot. (C) Receiver‐operating curve and area‐under curve for DNA methylation signature to classify high‐grade (HG), low‐grade (LG) or all cancer samples from benign bladder disease and normal donors. (D and E) Sensitivity and specificity of urine sedimentary cell FISH assay (FISH) or DNA methylation assay (urine cancer score [UCAS]) for LG and HG cancer.

FIGURE 5

Pre‐surgery urine differentially methylated region (DMR) DNA methylation signal non‐invasively classifies bladder cancer (BLCA) and predicts progression‐free survival. Experiment design: pre‐surgery urine DNA methylation signal, or pathological features of resected samples on the first trans‐urethral resection of bladder tumour (TURBT) sample, was used to predict progression‐free survival (A) or metastasis‐free survival (B). Only DNA methylation scores negativeness, but none of the pathological features were significantly associated with a progression‐free survival benefit. DNA methylation, WHO grade and tumour invasiveness, but not muscle invasion, are associated with metastasis‐free survival. In other words, pre‐surgery, BLCA DNA methylation signature negative urine is from benign, slow‐growing cancer, which is very unlikely to develop disease recurrence, and pre‐surgery BLCA DNA methylation basal class are associated with tumours with metastasis potential.

FIGURE 6

Post‐surgery urine DNA differentially methylated region (DMR) methylation signal detects minimal residual disease and predicts recurrence. (A) Paired pre‐first trans‐urethral resection of bladder tumour (TURBT) urine and pre‐second surgery urine are compared for the same individual. Although the pre‐first TURBT urine is 100% positive (13/13) in these patients, second TURBT/radical cystectomy (RC) only finds residual tumour in 100% of individuals (4/4) who show positive cancer methylation score. (B) Post‐first TURBT urine DNA cancer methylation signature (high‐grade or residual disease: high risk; low‐grade and no‐residual‐disease: low risk) stratifies patients into risk groups that are strongly associated with recurrence. A 100% (30/30) of low‐risk patients do not show recurrence within 180 days of the first TURBT, whereas >90% (30/31) of high‐risk patients show recurrence within 180 days.