At present, colonoscopy and pathological biopsy are still the gold standards for the diagnosis of colorectal cancer. However, the disadvantages of colonoscopy are also obvious. First, colonoscopy is highly dependent on the experience of the operator. Second, complex preoperative preparation is not suitable for mass screening. Third, patients have poor experience with this invasive test. Methylated sDNA analysis is considered a promising method for large-scale CRC screening. To date, more than 30 methylated sDNA biomarkers have been developed, but only a few biomarkers have met the clinical need for diagnostic efficiency to some extent[27]. For example, methylated BMP3 combined with NDRG4 mutant KRAS was approved by the US. The FDA has a sensitivity of 92.3% and a specificity of 86.6% for CRC diagnosis[5], and the sensitivity and specificity of SDC2, which are approved by the China FDA, are 83.8%-89.1% and 98.0%, respectively[11, 28]. Due to its suboptimal sensitivity and specificity and prolonged detection time, this method has yet to replace conventional screening methods. As a result, further development of a new panel/model with improved sensitivity and specificity is essential[29].
CRC methylation patterns exhibit heterogeneity, and a common classification approach in many studies involves categorizing CRC into hypermethylated and hypomethylated subtypes based on the CIMP[15, 30]. In our study, we screened a specific gene set for CRC diagnosis using the CIMP concept. Due to the limited size of our in-house dataset, we implemented more rigorous screening criteria, resulting in all 6 PTs exhibiting hypermethylation irrespective of their CIMP phenotype. To obtain stage-insensitive CpGs, we retained only those DMCs that did not differ significantly between eCRC and aCRC patients. Similarly, age-, sex-related and collinear CpG sites were excluded. As a result, we identified a DMG set that included many previously unreported CRC methylation markers.
We aimed to enhance the sensitivity of sDNA for diagnosing CRC, with the goal of identifying the most minimal marker for CRC detection. However, our study highlighted the challenges in achieving a significant breakthrough with single sDNA. Similar to several previously reported methylated sDNA markers[10, 31], the rates of positivity for B3GAT2, VIM, and SFRP2 in feces were notably lower than those observed in PTs. There may be several underlying factors that can account for our analysis. First, fecal samples contain fewer tumor cells[32], necessitating an experimental system with high sensitivity. Second, the composition of feces, including DNA from bacteria, along with residues from various animal and plant foods, is complex. Third, dietary structure, digestive function, and intestinal condition can also impact the outcomes of the test[33].
Therefore, by combining PRDM12 with previously reported genes in our DMG set, namely, FOXE1 and SDC2, we established a 3-target model, CDM, that exhibited exceptional performance in detecting eCRC, as validated in both the training and independent validation cohorts. The sensitivity and specificity of CDM for eCRC patients were 94.52% and 97.21%, respectively, and those for CRC patients and patients with AADs were 91.89% and 95.21%, respectively, both of which are better than those of the kits reported thus far. Furthermore, in the verification group, we added a variety of diseases, such as gastrointestinal malignancies, benign diseases, and extralimentary tumors, as interfering diseases, and the results showed that only STAD may interfere with the diagnosis of CRC by CDM. Therefore, in clinical diagnosis and treatment, attention should be given to the identification of upper and lower gastrointestinal symptoms.
The progression from atypical hyperplastic colorectal polyps to malignant colorectal adenocarcinoma involves the accumulation of genetic and epigenetic alterations[34–35], resulting in epigenetic reprogramming of human colonocytes. Different types of cancers display unique methylation patterns, and certain genes, including PRDM12, exhibit similar patterns across multiple cancer types[36–37]. Considering the potential role of PRDM12 in the occurrence of CRC, we explored its function. PRDM12 belongs to the PRDI-BF1 (positive regulatory domain I-binding factor 1) homologous domain (PRDM)-containing protein family, which is a subfamily of Kruppel-like zinc finger proteins[38]. This family plays a crucial role in regulating cancer development[39]. PRDM12, which is restrictively expressed in normal human tissues, shows elevated expression in eCRC and consistently maintains a lower expression level, suggesting its potential widespread involvement in cancer development. PRDM12 is known to play a role in the development of sensory neurons [40]; however, the role of PRDM12 in cancer has been poorly studied.
PRDM12 is an important transcriptional regulator capable of regulating neural differentiation and formation in combination with solid tumors and hematological malignancies [38, 41]. AG Reid[42] et al. reported that 15% of patients with chronic myeloid leukemia have deletion of the PRDM12 gene, which is related to rapid progression and short-term survival. Pancancer analysis suggested that PRDM12 expression is poorly correlated with patient prognosis and immunotherapy outcomes. Subsequently, we stratified CRC patients into high-expression and low-expression groups, revealing that the DEGs were primarily enriched in the cell cycle pathway. Protein interaction analysis demonstrated a close association between PRDM12 and EHMT2 (G9a), while DMFold2 predicted its involvement in various cellular processes. The findings of Yang et al[41]. In P19 embryonic carcinoma cells, PRDM12 recruits G9a to methylate lysine 9 of histone H3 (H3K9me), resulting in a decrease in chromatin structure and an increase in the proportion of G1-phase cells. However, this alteration hinders cell cycle progression, suggesting that PRDM12 may function as a tumor suppressor gene.
The present study has certain limitations. First, 25% of participants in the validation set were symptomatic patients with CRC, which could lead to an overestimation of the sensitivity of the PRDM12 test. Second, the sensitivity of PRDM12 in detecting STAD is nearly 65%, which may result in an increase in the number of endoscopies performed by individuals during physical examinations. However, from a different perspective, this could broaden the application range of PRDM12. Third, controlling the quality of specimens proves to be challenging. Patients collect their own specimens, and the use of regular toilets instead of squatting toilets in most wards contributes to difficulties in sampling and increases the possibility of specimens being soaked in water.
In summary, CIMP-based genome-wide methylation profiles provide valuable epigenetic information that can be used to develop novel methylation markers. The three-target diagnostic model constructed in this study further improved the diagnostic efficiency of methylated sDNA for eCRC.