Although 5-Hydroxymethylcytosine signatures in cell-free DNAs provide information about tumors, the co-5-Hydroxymethylcytosine subnetworks/modules, their biological pathways, and the associations of biological processes in cfDNAs related to 5-hmC were not thoroughly investigated. 5-hydroxymethylated modules may be a cancer signature, and may be useful as clinical diagnostic biomarkers with liquid biopsy for a broad range of cancers [18, 19].
We reconstructed the 5-hydroxymethylated network and clustered it to detect the GI-cancer-related modules. The reproducibility of GI cancer-related modules was assessed [8, 11]. The biological pathways of significant modules were detected and visualized. To develop a predictive model for distinguishing GI patients from healthy individuals, the GS method and elastic net classifier were used for feature selection and model specification, respectively. The cross-validation, heatmap, and AUC were used for model validation.
cfDNAs are detectable using liquid biopsy, and could be used in the clinic to provide more precise therapies in the early stages [2, 20, 21]. Stratifying GI patients and healthy individuals using a limited number of genes would be useful and cost-effective, especially as a noninvasive method. In this study, the GreenYellow module was found to be GI-related; its potential in stratifying patients and healthy individuals was computationally validated using the elastic net model and hierarchical clustering (Fig. 3a, b). After experimental validation, it could be used in the clinic for cancer prediction.
The biological pathways and processes of three GI cancer-related modules, and their associations, were assessed in this study (Figs. 1, 2). The associations between cell-cycle arrest and apoptosis were independently investigated in several cancer-related studies [22,23,24]. In our study, a direct association between these metastasis-prone biological processes was detected for aberrant 5-hmC genes in GI cancers (Fig. 2). Close direct association may indicate concurrent activation of cell-cycle and apoptosis processes through 5-hmC aberrations in GI cancers.
In contrast, there is a close inverse association between cell-cycle-related biological processes and extracellular structure organization/cell adhesion-related biological processes (Fig. 2). In studies conducted by Walker et al. and Pickup et al., several aspects of extracellular matrix (ECM) dysregulation leading to cancer progression and metastasis were individually explored [25, 26]. It was reported that the ECM concentration increases and the ECM stiffens during tumor development. The stiffened ECM increases cell mobility and reduces the expression of genes that typically inhibit cell-cycle progression and proliferation [25, 26]. In our study, we detected an inverse association between ECM dysregulation and cell-cycle activity for the Black and GreenYellow hydroxymethylated modules (Fig. 2), which may be consistent with results reported by Walker et al. and Pickup et al. [25, 26]. However, they did not study cfDNAs. The LUX gene family secreted by primary tumor cells is responsible for ECM stiffness and total ECM concentration [25]. Some members of the LUX family were detected in our study, including lux, lux1, and lux3. They exhibited a low level of 5-hmC. An inverse association between the high volume of ECM stiffness and reduced cell cycle inhibition may be due to aberrant 5-hmC in primary cancer cells detected in cfDNAs.
Many previous studies have investigated the role of 5-hmC in the regulation of gene expression [4, 27,28,29]. Due to a lack of gene expression values, we were unable to assess the correlation between dysregulation of gene expression and aberrant 5-hmC values for detected genes in significant modules in this study. However, we assessed the gene expression changes of 11 final genes using TCGA data. Genes such as samd11 and mthfd2l were reported as diagnostic 5-hmC signatures in previous studies [2, 30]. These biomarkers were detected in the GreenYellow and Black modules. Although there is no experimental evaluation in our study, on the basis of previous studies, the new genes detected in these modules may be new 5-hmC-related biomarkers for GI cancers.
The other significant finding in this study is a close inverse association between the Black and Tan modules, which indicate apoptosis and ECM organization and pathways, respectively. Signaling properties of the ECM such as three-dimensional organization of cells and ECM structures have significant impacts on cellular processes such as proliferation and apoptosis [31]. Moreover, the DNA damage response pathway, which is significant in the Tan module, is the vital determinant of the plasticity of the cellular genome and depends on signaling pathways that regulate apoptosis. Cell adhesion to the ECM, as a part of ECM organization, regulates several of these pathways [31, 32]. Less cell–ECM contact leads to more apoptosis. This finding is consistent with the results from two 5-hmC-related modules (Figs. 1, 2). We concluded that activation of ECM and apoptosis pathways may be related to aberrant 5-hmC values. Our findings for 5-hmC pathways were not experimentally validated; however, we computationally validated the reducibility in another dataset. Accordingly, some of the associations found in our study were reported in previous studies on 5-hmC in GI cancers, including the inverse associations between cell-cycle and ECM organization, and between apoptosis and ECM organization [27, 28]. However, some of the associations in GI cancers have not been reported, and are good targets for future research. Although we assessed the associations between biological pathways of GI-related modules, the module activities were not assessed in each GI-related cancer. Cubuk et al. assessed the module activities by integrating gene expression into biological pathways [33].
As cfDNAs mirror the primary characteristics of every patient, they may be an option for precision medicine in the clinic [21, 34]. In the future, checking the 5-hmC profile may be practical in precision medicine for GI cancers with aberrations, such as a patient’s specific 5-hmC. With 5-hmC-related modules and their associations, predictive models based on 5-hmC may be highly practical for cfDNAs as they are detectable using a noninvasive method. Such predictive models could be used in the clinic to distinguish patients from healthy individuals.
The studies by Song et al. and Li et al. [4, 18] were performed to detect cancer biomarkers using statistical methods including the fold change and t-test. In this study, we used a network-based method to detect GI-related 5-hmC modules, their associations, and 5-hmC biomarkers for GI cancers.