Pan‐cancer analysis of genome‐wide methylation profiling discover type‐specific markers targeting circulating free DNA for the detection of colorectal cancer

Dear Editor, Colorectal cancer (CRC) is one of the most commonly diagnosed cancers and cancer-related causes of death worldwide.1 Early diagnosis is critical to provide curable treatment and improving survival rates for CRC patients.2 Circulating-free DNA (cfDNA) carrying cancer-specific methylation signature is a promising specific marker for cancer diagnosis.3 However, previous studies were based either on high-throughput sequencing or did not consider the specificity of the methylation pattern for different cancers. We aim to discover and validate a noninvasive method with type-specific DNA methylation patterns for the diagnosis of CRC. The workflow is illustrated in Figure 1 (details of the sample sources and analysis process can be found in the Supporting information). Briefly, we first compared the differentially methylated CpG sites (DMCs) between 395 CRC and 45 adjacent normal tissues from the cancer genome atlas (TCGA) dataset, and a total of 37,132 DMCs were selected based on |Δβ| > 0.20 and FDR < 0.05. We further filtered out 27,063 CpG sites due to potential noise of DNA methylation (average beta > 0.1 or < 0.9) in 1,246 samples of white blood cells (WBCs) of healthy individuals from two Gene Expression Omnibus (GEO) datasets. Finally, with the same filtering criteria as WBCs, 15 CRC-specifically hypermethylated CpG sites with average methylation levels less than 0.1 in 8,629 tissue samples from 28 other cancer types in the TCGA dataset were retained, and no CRC-specific hypomethylated CpG sites met the criteria and were retained. The heatmap showed that the 15 CpG sites (located on B3GALNT1, C6orf97, FAM72A, FAM72B, LIFR, OSMR, ZNF264, and ZNF543) well-distinguished CRC from adjacent normal tissues (Figure 2A), WBCs (Figure 2B), and 28 other types of cancer (Figure 2C) in TCGA, as well as 23 other types of cancer in GEO (Supporting information). In addition, DNA methylation has a potential role in regulating gene expression, and CpG sites, including cg14786398, were negatively correlated with their corresponding gene expression (r = −0.691, p < 0.001, Figure 2D, Supporting information). The area under curve (AUC) of the 15 CRC-specific CpG sites ranged from 0.643 to 0.903, similar to the frequently reported CpG sites of SEPT9 (0.846 to 0.967). Compared with the SEPT9, our markers have higher CRC specificity, and the misclassification rate of the 15 CpG sites in 28 other types of tumor tissues ranged from 0% to 20%, while the misclassification rates of 11 CpG sites in SEPT9 ranged from 0% to 92% (Supporting information). Furthermore, the methylation status of selected CRC-specific markers was evaluated by MethylTarget sequencing (Genesky) in CRC tissue (N = 227), adjacent normal tissue (N = 24), WBC (N = 52) and cfDNA (N = 14) samples from CRC patients and healthy controls. The candidate CpG sites of ZNF543 were significantly hypermethylated in tissues and cfDNA from CRC compared to normal tissues and cfDNA from healthy controls and were unmethylated in WBCs from both CRC and healthy controls (Figure 2E). Candidate CpG sites of B3GALNT1, C6orf97, LIFR, and ZNF264 were also hypermethylated in CRC tissues but unmethylated in normal tissues and WBCs (Supporting information). We successfully designed primers and probes for six CRC-specific markers (FAM72A, FAM72B, LIFR, OSMR, ZNF264 and ZNF543) covering 10 of 15 CpGs for further testing the methylation level of cfDNA with ddPCR-based assays (Figure 3A; Supporting information). The established three multiplex ddPCR assays (mddPCR, Assay 1, 2, 3) have superior detection sensitivity compared with traditional multiplex MethyLight (mqPCR). mqPCR assay 1 detected one methylated allele in a background of 125 unmethylated alleles (Figure 3B; limit of quantification (LOQ) = 0.8%, R2 = 0.957). In contrast, the LOQ of Assay 1 was 25-fold lower than that of mqPCR assay 1 (Figure 3C,D, Supporting information). A total of 370 blood samples were collected from 195 patients with CRC, 6 patients with hyperplastic polyps, 22 patients with advanced adenomas (AAs), 103 healthy controls, and 44 non-CRC patients with benign or malignant tumours of breast or lung. The AUCs of the three mddPCR assays for distinguishing CRC patients from healthy controls were 0.767, 0.847 and 0.771 (Figure 4A–C), respectively. Samples with detected methylated molecules were judged as positive, with a sensitivity of 57.9% for Assay 1, 71.8% for Assay 2 and 57.9% for Assay 3. The corresponding specificities were 92.2%, 94.2% and 95.1%, respectively. Furthermore, elevated methylated molecule copies were also detected in patients with AA, with positive rates of 31.8%, 22.7% and 31.8% of the three mddPCR assays. Next, we combined three mddPCR assays to evaluate the combined diagnostic performance. The AUCs of the four combination panels were 0.884, 0.821, 0.870 and 0.892, respectively (Figure 4D). Based on the optimal cutoff values, the sensitivities of the four panels were 81.5%, 70.3%, 77.9% and 84.1%, and the corresponding specificities were 89.3%, 88.3%, 90.3% and 85.4%, respectively (Figure 4D). Furthermore, the AUCs for diagnosis of non-CRC were relatively low, ranging from 0.541 to 0.599. Identification of the origin of cfDNA and the location of the cancer is critical for guiding clinical diagnosis. Currently, mSEPT9 is the only blood assay in the clinical setting for CRC screening, but its clinical usefulness is limited by its low sensitivity in early-stage CRC.4 Moreover, mSEPT9 is not a CRC-specific marker because of overlapping aberrant methylation across multiple cancers.5, 6 Recent evidence suggests that the use of tissue-specific methylation signatures will allow for tracing tissue of origin in cfDNA.7, 8 In marker discovery, we eliminated the possible confounding interference of cfDNA released by other cancer tissues or WBCs on the detection of ctDNA methylation levels in CRC. Moreover, our inhouse validation study suggested that such CRC-specific methylation patterns could be detected in tissues and cfDNA but not in WBCs. This study has several limitations. First, most of the healthy controls and non-CRC patients of our inhouse cfDNA cohort were not confirmed by colorectal endoscopy, thus, those positive results were classified as false-positives to more closely reflect the situation in the natural population in the real world, and this could have resulted in an underestimation of diagnostic performance of our mddPCR assays to a certain extent. Second, we collected a small subset of cfDNA samples from patients with AA and other non-CRC diseases, suggesting that our arrays need to be further optimized and validated in studies including more participants in the future. In conclusion, we identified a panel of CRC-specific methylation patterns by pan-cancer analysis and developed three cfDNA multiplex ddPCR assays. Our findings suggested that cfDNA methylation assays have the potential to detect early-stage CRC and its advanced precursors. However, the diagnostic performance of these arrays requires more validation before clinical implementation. This work was supported by the National Natural Science Foundation of China (Grant numbers 82073643, 81773503, and 81473055), the Heilongjiang Provincial Natural Science Foundation of China (Grant numbers ZD2021H001) and Heilongjiang Province Applied Technology Research and Development (No. GA20C016). The authors declare no conflicts of interest. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.