Table 2 Examples of biomarker discovery studies using DNA methylation genome-wide approaches.
From: From concept to clinic: a roadmap for DNA methylation biomarkers in liquid biopsies
Study | Cancer type | Goal | Genome-wide platform | Approach for candidate biomarkers identification | Number of candidate biomarkers | Final biomarkers and performance in liquid biopsies | Detection method | Test |
|---|---|---|---|---|---|---|---|---|
SOURCE MATERIAL: NON-LIQUID BIOPSY* | ||||||||
Oh et al. [60] | Bladder cancer | (Early) Detection | Agilent 244 K microarrays | Identification of DMPs in tumors vs adjacent non-tumor tissue. | 9 most hypermethylated probes | PENK Sensitivity 76%, Specificity 92% | qPCR from urine DNA | Commercially available as EarlyTect BCD (Promis Diagnostics) |
Lange et al. [65] | CRC | Detection | Illumina 27 K and 450 K arrays | Identification of markers with high methylation across all colorectal tumors while harboring low methylation in peripheral blood leukocytes, adjacent-normal colonic tissue samples and other cancer types. | 10 CpG sites | THBD and C9orf50, performing better as a panel AUC (plasma) = 0.80; AUC (serum) = 0.83 | MethyLight and digital MethyLight from plasma and serum cfDNA | NA |
Lin et al. [61] | CRC | Detection | Illumina 450 K arrays | Identification of DMPs in tumors vs adjacent non-tumor tissue. | 5 most hypermethylated probes | AGBL4, FLI1 and TWIST1. Sensitivity for any one hypermethylation marker across all stages >90%; Stage I-III: Sensitivity of any two or three markers 53–58 and 28–39%, respectively; Stage IV: Sensitivity of any two or three markers 73 and 56%, respectively | qMSP from plasma cfDNA | NA |
Uehiro et al. [141] | Breast cancer | Detection | Illumina 450 K arrays | Identification of markers with low methylation in the non-breast cancer samples, and of breast cancer subtype dominant markers. | 140 | Best detection model included four methylation markers (RASGRF1, CPXM1, HOXA10, and DACH1) and two parameters (cfDNA concentration and the mean of 12 candidate methylation markers) AUC = 0.92; Sensitivity: 91%; Specificity: 83% | ddMSP from plasma cfDNA | NA |
Xu et al. [62] | Hepatocellular carcinoma | Detection | Illumina 450 K arrays | Identification of DMPs in tumors tissues vs blood leukocytes from healthy individuals. | 1000 most DMPs | 10 biomarker panel AUC = 0.97 | Targeted BS-seq from plasma cfDNA | Commercially available as HelioLiver (Helio Genomics), where the panel is analysed in combination with other methylation markers, serum protein markers and clinical variables. |
Ooki et al. [77] | Non–small cell lung cancer | (Early) Detection | Illumina 450 K arrays | Identification of DMRs in the proximity of the TSS using bumphunter, combined with a candidate gene approach to prioritize methylated biomarkers with known biological function in lung cancer. | 30 | 6 gene panel (CDO1, HOXA9, AJAP1, PTGDR, UNCX and MARCH11). In serum: Sensitivity 72%; Specificity: 71%. Similar detection accuracy was observed in pleural effusion and ascites. | qMSP from serum, pleural effusion and ascites DNA | NA |
Feber et al. [91] | Bladder cancer | Detection | Illumina 450 K arrays | Identification of probes with no or very low methylation in normal urothelium, blood and non-cancer urine samples and high methylation in bladder cancer. | 432 CpG loci | 150 CpG loci (“UroMark”). Sensitivity 98%; Specificity 97%; NPV 97% | Targeted BS-seq from urinary sediment DNA | NA |
Li et al. [80] | Liver cancer | Detection | Illumina 450 K arrays | Identification of clusters of at least 3 CpGs with high methylation variation across all normal and cancer types, as described in the precursor model “CancerLocator” [142] | CpG clusters (unknown number) | “CancerDetector” (probabilistic model) Sensitivity: 95%; Specificity: 100% | Low coverage WGBS from plasma cfDNA | NA |
Liu, Toung et al. [63] | Multiple cancers | Detection/ Classification | Illumina 450 K arrays | Identification of hypermethylated probes in tumors vs normal tissues; removal of sites with high methylation in normal plasma and sites with low coverage | 9223 hypermethylated probes | Methylation score based on the 9223 hypermethylated probes. Detection: Sensitivity 84%; Specificity 100% Classification: Specificity 79% | Targeted BS-seq from plasma cfDNA | NA |
Jensen et al. [143] | CRC | Detection | Illumina 450 K arrays | Prioritization of CpG sites highly methylated in CRCs, low methylated in peripheral blood leukocytes and minimally methylated in other cancers and normal colorectal mucosa. | 50 CpG sites | 3 biomarker panel (“TRiMeth”) Sensitivity 85%; Specificity 99% | ddMSP from plasma cfDNA | NA |
Cristall et al. [81] | (Metastatic) triple-negative breast cancer | Detection | Illumina 450 K arrays | Identification of triple-negative breast cancer-specific hypermethylated regions with at least 2 methylated CpG residues within 300 bp of each other. Combination with seven additional regions previously observed to be prognostic relevant for triple-negative breast cancer. | 71 hypermethylated regions | 53 amplicons from 47 regions (“mDETECT”). In serum: AUC = 0.97; Sensitivity 93%; Specificity 100% In plasma: AUC = 0.92; Sensitivity 76%; Specificity 100% | Multiplexed NSG from serum or plasma cfDNA | NA |
Kandimalla et al. [78] | Multiple GI cancers | Detection/ Classification | Illumina 450 K arrays | Identification of DMRs between individual GI cancers and adjacent normal, and across GI cancers. | 67,832 tissue-specific DMRs | Three distinct DMR panels: 1) cancers-specific detection panels (AUC = 0.90-0.98); 2) pan-GI detection panel (AUC of 0.88; and 3) tissue of origin prediction panel EpiPanGI Dx (accuracy of 0.85–0.95 for most cancers) | Targeted BS-seq in plasma | NA |
Liang et al. [64] | Multiple cancers | Detection | Illumina 450 K arrays | Identification of DMPs in tumors vs normal tissue samples, with low methylation in white blood cells. Combination with CpG sites showed to be associated with common cancers in previous studies. | 80,672 CpG sites | 2473 co-methylation blocks in early-stage lung cancer: Sensitivity 52–81%; Specificity: 96% | Machine learning algorithms based on WGBS results from plasma cfDNA | CE-marked OverC Multi Cancer Detection Blood Test |
Gouda et al. [66] | (Metastatic) CRC | Detection | Illumina 450 K arrays | Identification of hypermethylated CpG sites with low methylation in samples from healthy individuals and a methylation frequency of more than 50% in tumors | 32 hypermethylated CpGs | 32 CpG sites Sensitivity 85%; Specificity 92% | BS-seq from plasma cfDNA | NA |
Manoochehri et al. [144] | Triple-negative breast cancer | Detection | Illumina 450 K and EPIC arrays | Identification of DMRs in tumors vs normal tissue | 23 DMRs | 3 biomarker panel (SPAG6, LINC10606 and TBCD/ZNF750) AUC = 0.78 | ddPCR from plasma cfDNA | NA |
Chen et al. [133] | CRC | (Early) Detection/ Surveillance | Microarray, WGBS and RRBS | Identification of DMPs in a variety of cancer types. | 11,787 CpG sites spanning 595 genomic regions | “PanSeer” assay, interrogating 10,613 CpG sites across 477 genomic regions Detection: Sensitivity 88%; Specificity 95% Pre-diagnostic: Sensitivity 91%; Specificity 95% | Logistic regression classifier based on semi-targeted PCR results (training set) from plasma cfDNA | Commercially available as ColonAiQ (Breakthrough Genomics), developed later in [132] for specific detection and monitoring of CRC using a 6 marker panel. |
Pharo et al. [70] | Bladder cancer | Detection | RRBS | Identification of DMRs in bladder cancer cell lines vs other urological cancer cell lines, by sliding window. | 32 DMRs | 8 biomarker panel (“BladMetrix”) Sensitivity 92%; Specificity 93%; NPV 98% | qMSP from urine DNA | NA |
SOURCE MATERIAL: LIQUID BIOPSY | ||||||||
|---|---|---|---|---|---|---|---|---|
Hlady et al. [47] | Hepatocellular carcinoma | (Early) Detection | Illumina 450 K arrays | Identification of DMPs in hepatocellular carcinoma vs cirrhosis plasma samples, followed by lasso linear regression and stepwise manual recursive partitioning. | 23 DMPs | 5-marker panel (cg04645914, cg06215569, cg23663760, cg13781744 and cg07610777) AUC = 0.956 | Bisulfite pyrosequencing in plasma cfDNA | NA |
Sabedot et al. [67] | Glioma | Detection | Illumina EPIC arrays | Identification of DMPs in gliomas vs non-glioma central nervous system conditions and healthy controls. Selection of the DMPs that matched the methylation patterns of glioma tissue. | The 5000 most DMPs | Machine learning-based classification model glioma-epigenetic liquid biopsy score (GeLB) Sensitivity: 100%; Specificity: 98% | Array-based from serum cfDNA | NA |
Gallardo-Gómez et al. [68] | CRC | Detection | Illumina EPIC arrays | Differential analysis of advanced neoplasia vs no neoplasia, from results obtained from a sample pooling strategy. Performance of a statistical biomarker prioritization (SES algorithm combined with classification models) | 26 probes | Pyrosequencing from serum cfDNA | NA | |
Stone et al. [83] | Esophageal adenocarcinoma | Detection | Illumina EPIC arrays | Creation of modules of gene-based methylation probes using weighted gene co-expression network analysis.; Determination of module significance to disease and gene importance to module. | 7 probes | Combination of 7 probes, age and sex: Sensitivity 88%; Specificity 31% | Array-based from salivary DNA | NA |
Shen et al. [72] | Several tumor types | Detection/Classification | cfMeDIP–seq | Identification of DMRs in tumors vs normal samples using a window approach | Tumor type specific (proof-of-concept). | Robust performance in cancer detection and classification across several tumor types. | MeDIP–seq from plasma cfDNA | NA |
Nuzzo et al. [79] | Renal cell cancinoma | Detection | cfMeDIP–seq | Identification of DMRs in cancer vs controls and urothelial bladder cancer by a window approach | 300 DMRs, including the 150 most hypermethylated and the 150 most hypomethylated regions | 300 DMRs Plasma: AUC = 0.99 Urine: AUC = 0.86 | MeDIP–seq from urine and plasma cfDNA | NA |
Van Paemel et al. [145] | Multiple pediatric solid tumors | Classification | cf-RRBS | Identification of clusters containing at least three CpGs covered on the Illumina HM450K array. | 14,103 clusters | Performance not provided (proof-of-concept). | Classifier applied to cfDNA from plasma and cerebrospinal fluid | NA |
Chan et al. [71] | Hepatocellular carcinoma | Detection | WGBS | Comparison of mean methylation densities in region “bins” between cancer cases and healthy controls. | Genome-wide hypomethylation (no specific biomarkers identified) | Genome-wide hypomethylation. At high depth: Sensitivity 74%; Specificity 94%; At lower depth: Sensitivity 68%; Specificity 94% | WGBS from plasma cfDNA | NA |
Liu et al. [49] | Multiple cancers | Detection/Prediction of tissue of origin | WGBS | Not detailed. | >100,000 informative methylation regions (found in [48]) | Panel of >100 000 informative methylation regions. Detection: Sensitivity (early cancer) 45-67%; Specificity: 99% Prediction of tumor of origin: Sensitivity 96%; Specificity: 93% | Machine learning classifier based on BS-seq results from plasma cfDNA | Commercially available Galleri (Grail), later validated in [50] for early detection of several cancers. |
