Journal of Clinical Trials
Open Access

Identification of Methylation Driven Biomarkers for Diagnosis and Prognosis in Colorectal Cancer by Integrative Analysis of TCGA, GTEx, and GEO Database

Lichao Cao^1,2, Ying Ba³, Jin Yang^1,2* and Hezi Zhang^{3* *}Correspondence: Jin Yang, Department of Provincial Key Laboratory of Biotechnology of Shaanxi Province, Northwest University, Xi’an, China, Email: Hezi Zhang, Department of Gene Technology, Shenzhen Nucleus Gene Technology Co., Ltd., Shenzhen, China, Email:

Author info »

Introduction

CRC is one of the most prevalent malignant tumors and the second leading cause of mortality worldwide [1]. Despite the progress made in diagnosis and therapy, CRC patients usually develop recurrence and metastasis, leading to dramatic decreases in the 5-year survival rate [2]. Therefore, there is an urgent need to improve the diagnosis, treatment, and prognosis for patients with CRC.

Molecular characterization has great potential for improving understanding of tumor development and is extensively utilized to predict tumor diagnosis, treatment, and prognosis [3,4]. Using bioinformatics methods and machine learning, various types of biomarkers have been found to be related to the diagnosis and prognosis of tumors, including microRNAs [5,6], long non-coding RNAs [7,8], DEGs [9], DNA methylation [10,11] and others.

DNA methylation as an important regulator of gene expression has been researched in various cancers, such as endometrial cancer [12], prostate cancer [13], and hepatocellular carcinoma [14]. Although the mechanism of DNA methylation is not fully understood, it is assumed that DNA methylation could affect the binding of the transcription factors to their DNA targets sites, and then alter the expression of downstream genes [15-17], which may promote oncogenesis [15,18]. MethylMix, a R package that can identify MDGs through integrative analysis of DNA methylation and gene expression data from normal samples and tumor samples [19,20], was used to investigate the relationship between MDGs and the prognosis of CRC patients [21,22]. However, no studies have reported whether MDGs could be used as diagnostic indicators for CRC.

The public databases, especially TCGA and GEO, provide convenient access to systematic collections of sequencing data with detailed clinical and pathological information which have been applied in malignant tumor research. Meanwhile, the GTEx database contains a large number of gene expression samples of normal human patients that have been integrated with TCGA or GEO in various studies [23-25].

Based on existing literature data, we carried out studies to identify a set of MDGs using DNA methylation and gene expression data from TCGA, GEO, and the GTEx, and construct an independent diagnosis model using RF, SVM, and LR algorithms. Additionally, the MDGs, the presumed potential prognostic indicators, were analyzed using univariate and multivariate Cox regression analyses. Our findings may suggest that the potential methylation driven biomarkers could prompt more individualized diagnoses and therapies for CRC.

Materials and Methods

TCGA and GTEx data acquisition

The mRNA expression (RNA-seq data from Illumina platform) and DNA methylation data (Illumina Human Methylation 450) of CRC from TCGA and the mRNA expression of normal samples from the GTEx were downloaded from the UCSC Xena platform (https://xenabrowser.net/datapages/). Subsequently, the samples from TCGA were divided into the normal group and the tumor group. 667 samples (51 normal and 616 tumor) were selected for DEG analysis and MDG analysis was performed on 433 samples (45 normal and 388 tumor), detailed information is shown in Table 1. We selected 307 normal samples (123 females and 184 males) of mRNA expression data of colon tissue from the GTEx to provide supplementary data for normal samples from TCGA.

Table 1: Clinical information of TCGA and GEO dataset.

GEO data acquisition

We downloaded gene expression data by array (GSE21815, GSE28000, GSE39582, and GSE44076) and methylation data by genome tilling array (GSE48684, GSE53051, GSE77718, and GSE101764) from the GEO database (https://www.ncbi.nlm.nih.gov/geo/), and the sample size of each dataset can be found in Figure 1. Among them, GSE39582 has detailed clinical and phenotypic information, with a total of 566 tumor and 19 normal samples in Table 1.

Figure 1: Clinical information of TCGA and GEO dataset.

Screening DEGs, DMGs and MDGs

We filtered out the genes with a read count of less than 10 in more than 50% of all the samples. The DEGs were identified by comparing the tumor group with the normal groups using the limma R package [26], with the cutoff criteria defined as | log 2 Fold Change (FC)| >0.585 and adjusted P-value <0.001. In addition, the probe with the highest absolute value of delta Beta and adjusted P-value less than 0.001 was selected as the representative of gene methylation level using the ChAMP R package [27], and |deltaBeta| >0.3 was used to screen DMGs.

Subsequently, the tumor methylation matrix, the normal methylation matrix, and the tumor expression matrix with overlapping DEGs and DMGs were constructed to identify MDGs using the R package MethylMix [19,20]. Significant methylation events were filtered using the correlation coefficient <-0.3 and P-value <0.05 as the cutoff criteria, and the correlation of the MDGs was visualized by the corrplot R package. The unpaired t test statistically analyzed the differences of the MDGs between the tumor samples and the normal samples.

Validation of MDGs using GEO datasets

To validate the MDGs, the gene expression profiling datasets (GSE21815, GSE28000, GSE39582, and GSE44076) were separately analyzed by the online software GEO2R (http://www.ncbi.nlm.nih.gov/geo/geo2r/), and the expression value of each MDG was obtained. The methylation value of each MDG in the gene methylation profiling datasets (GSE48684, GSE53051, GSE77718, GSE101764) was calculated using the ChAMP R package.

Identification and validation of diagnosis biomarkers

The mRNA expression datasets from TCGA and the GTEx were integrated and randomly divided into a training dataset and a validation dataset with a 3:1 ratio in tumor samples and normal samples; Principal Component Analysis (PCA) investigated the distribution of the combined samples of TCGA and the GTEx. To choose the optimal algorithm to build the diagnosis model, three different algorithms were used to evaluate the accuracy of the 10 MDGs as an independent predictor with default parameters and calculate the P-value of each MDG. The chosen algorithm was used to further evaluate the combined indicator for diagnosis of CRC. The GSE39582 dataset, containing 7 of the MDGs, was analyzed as a validation dataset in combination with training datasets from TCGA and the GTEx, the Receiver Operating Characteristic (ROC) curve was utilized to depict the sensitivity and specificity of the diagnosis models using independent MDGs or the combined MDGs panel.

Identification and evaluation of prognosis biomarkers

The mRNA expression data with OS information was selected to determine the prognostic biomarkers. The univariable Cox proportional hazards regression model was produced using the Survival R package, with the P-value cutoff of 0.05. Subsequently, Kaplan-Meier survival and multivariate Cox regression analyses were performed to determine independent prognostic factors; the coefficient of the MDGs was obtained from the multivariate Cox results. The risk score based on the signature of each CRC patient was calculated using the following formula:

MDGs risk score=ΣCox coefficient of gene χi ×Scale expression value of gene χi

To further evaluate the predictive efficiency of the constructed MDGs risk score model, we used the ROC curve to reflect the sensitivity and specificity of survival prediction and quantified the Area Under the Curve (AUC) using the survival ROC R package. The optimal cutoff risk score was designated at the turning point of the ROC curve, where the difference between true positive and false positive is the most significant. The patients above the cutoff value were in the high-risk group, while the patients below it was in the low-risk group. In addition, Kaplan-Meier curves were plotted to distinguish the two groups using the survminer R package.

Construction and validation of the nomogram model

To improve the accuracy of the prognostic model, we constructed a nomogram that visualized the prognostic value of different patients’ characteristics by integrating risk score and clinical information, including age, gender, and tumor stage. This analysis was performed using the rms R package to plot calibration curves to evaluate the predicted probabilities in comparison with the ideal predictive line. In addition, a forest plot based on univariable Cox analysis illustrated the relationship between clinical information and OS. Moreover, the Concordance index (C-index) indicated the predictive accuracy of the nomogram.

Results

Identification of MDGs as candidate diagnostic and prognostic biomarkers

The data generation and analysis workflow are shown in Figure 1. After merging the expression datasets and methylation datasets of Colon Adenocarcinoma (COAD) and Rectum Adenocarcinoma (READ) from TCGA, and expression datasets from TCGA and the GTEx, respectively, PCA showed that the normal samples and the tumor samples were well separated. The samples of READ and COAD, TCGA, and the GTEx were randomly distributed, which indicated that the merged data sets could be used for further analysis (Figure S1A and S1B). Expression and DNA methylation data from TCGA were separately analyzed to screen DEGs and DMGs by comparing normal samples and tumor samples. As a result, 522 DEGs (266 upregulated and 256 downregulated) were identified (Figure 2A and Table S1), and the detailed information about DMGs was shown in Table S2.

Figure 2: Identification of candidate biomarkers. (A) Volcano plot on DEGs. The green dots represent downregulated genes, while the red dots represent upregulated genes. (B) The correlations between each candidate biomarker. The color of the number in the grid represents the strength of correlation between the corresponding genes. (C) Statistical differences of the DMGs between normal and tumor groups were compared through the unpaired t test. The values of P are labelled above each boxplot with asterisks. (*P<0.05, ** P<0.01, *** P<0.001, **** P<0.0001).

Subsequently, R package MethylMix was used for identifying MDGs with a filter criterion set as Cor <-0.3, | deltaBeta | >0.3 and adjusted P-value <0.001, and 10 MDGs were identified as candidate diagnostic and prognostic biomarkers. Pearson’s correlation test statistically analyzed the correlations between methylation degree and gene expression of the MDGs using the cor.test function of the R language (https://www.r-project.org/) (Figure 3) and found that the methylation degree of those 10 MDGs was negatively correlated with their expression. The higher the methylation degree was, the lower the gene expression, and the detailed information of the MDGs is shown in Table 2. The correlations between each candidate biomarker, which may help us choose the optimal independent gene set for the diagnosis of CRC, were investigated using the corrplot R package (Figure 2B). In addition, the unpaired t test was conducted to quantify the difference of each candidate biomarker between the normal samples and the tumor samples; significant difference was found in all the candidate biomarkers (Figure 2C).

Table 2: The information of MDGs.

Figure 3: Correlation between the expression and methylation degree of the ten MDGs. X-axis represents the degree of methylation and Y-axis represents the gene expression level.

Moreover, we verified the gene differences and methylation differences of these MDGs in other datasets (4 expression GEO datasets and 4 methylation GEO datasets) and found that the trend was consistent. The significant differences of gene expression and methylation of these MDGs were verified in at least one other dataset in Table 3. The DEGs of each GEO dataset is shown in Figure S1C-S1F.

Table 3: Statistical information of expression and methylation values in MDGs.

Evaluation of the MDGs as candidate diagnostic biomarkers

We used three different algorithms to evaluate the performance of MDGs in establishing prognosis models. Among them, SVM algorithms performed the worst (Figure S2); RF and LR algorithms had similar results, which can be seen in Figure 4 and Figure S3, respectively. However, when using the LR algorithm to construct prediction models with joint MDGs, the algorithm does not aggregate; using the RF algorithm, the calculated P-value of each MDG is less than 0.0001, while in the other two algorithms, the P-value of SPTBN5 is greater than 0.0001 (Table 4). Therefore, the RF algorithm was selected to further build the diagnostic model of the joint MDGs. When using MDGs to build prediction models with the RF algorithm, 9 out of 10 MDGs revealed excellent performance as independent diagnostic predictors, where the Area Under the Curve (AUC) of CLDN1 is 0.988, EPHX4 is 0.971, TCN1 is 0.966, ARHGAP20 is 0.921, LY6G6D is 0.886, FAM150A is 0.869, KCNJ12 is 0.857, KRT7 is 0.860, and STK33 is 0.842 (Figure 4 A-4I). Additionally, the performances of the combined MDGs were further assessed in samples from TCGA-GTEx and GEO and the outcome of the AUC was 1.0 and 0.996, respectively (Figure 4 K and 4L). The importance of each predictor for the combined predictive model is shown in Table 5. The larger the number of Mean Decrease Accuracy and Mean Decrease Gini, the more important the predictor for the model. It was found that the numbers of Mean Decrease Accuracy and Mean Decrease Gini were positively correlated with the AUC of the predictive model of the predictor. Moreover, unsupervised hierarchical clustering of the combined panels with ten makers or seven makers indicated that the constructed models could accurately distinguish CRC patients from normal controls (Figure 5A-5D). Collectively, the 10 MDGs could be candidate biomarkers for diagnosis of CRC samples and 9 of them performed well as independent indicators.

Table 4: The performance of the MDGs as independent indicator using RF, LR and SVM algorithm.

Table 5: The importance of diagnostic biomarkers.

Figure 4: ROC of the diagnostic prediction model with candidate biomarkers using RF algorithm. (D, A-J) independent signature. (E, K, L) combined DMGs panel. (K) ten MDGs, (L) seven MDGs.

Figure 5: Unsupervised hierarchical clustering of the MDGs for use in the diagnostic prediction model in the training (A) and validation datasets (B) from TCGA and GTEx. (C, D) Unsupervised hierarchical clustering of seven DEGs for use in the diagnostic prediction model in the training dataset from TCGA and GTEx (C) and validation dataset from GEO (D).

Construction and evaluation of MDGs related prognostic model in CRC patients

Univariate Cox regression analysis revealed that 2 of the 10 MDGs were independent prognostic indicators of OS, as the hazard ratio of STK33 was 1.09 (95% CI: 1.01-1.17, P=0.021) and EPHX4 was 0.91 (CI: 0.83-0.99, P=0.032) (Figure 6A). Subsequently, Kaplan- Meier analyses and log-rank tests using STK33 and EPHX4 as independent prognostic indicators indicated that patients with high-risk scores suffered poor OS, with the P-value of 0.008 and 0.047, respectively (Figure 6B and 6C). We calculated the risk score of each CRC patient using the formula as follows: (1.09 × STK33)+(0.91 × EPHX4); then CRC patients (n=635) were categorized into the high-risk score group or the low-risk score group, according to the optimal cut-off value of the risk score obtained from the survminer R package. The results also indicated that high-risk score patients had a worse OS rate than low-risk score patients (P<0.0001) (Figure 6D). The prognostic accuracy of the risk score model was investigated as a continuous variable (Figure 6E). The AUC of the prognostic model for OS was 0.569 at 3 years, 0.633 at 4 years, and 0.626 at 5 years.

Figure 6: Identification and assessment of prognostic biomarkers. (A) Univariable Cox proportional hazards regression analysis of the relation between the candidate MDGs and OS status. (B, C) Kaplan-Meier curves indicate that two MDGs significantly related OS (B) STK33, (C) EPHX4. (D) Kaplan-Meier curve shows that OS in the low score group was significantly higher than that in the high score group. (E) Time-dependent ROC curve analysis of the prognostic biomarkers.

Construction and validation of the nomogram model

In the nomogram the score for each variable can be found on the point scale, then used to estimate the probability of survival at 1, 3, and 5 years by calculating the total score (Figure 7A). The forest plot shows that patient characteristics, including age (>60), tumor stage (III and IV), and risk score are associated with OS (P-value<0.001) (Figure 7B).

To validate the nomogram’s performance, we plotted the calibration curves and observed that the predictive curves were close to the ideal curve (Figure 7C-7E), which indicates good performance. Furthermore, the predictive accuracy of this nomogram (C-index: 0.72) was higher than the risk score model (C-index: 0.57).

Figure 7: Construction and validation of a nomogram. (A) Nomogram to predict the probability of OS in 1, 3 and 5 years for CRC. (B) Forest plots showing the associations between patient's characteristics and OS. (C-E) Calibration plot of the nomogram to predict the probability of OS at 1, 3 and 5 years.

Discussion

Many previous studies have reported that DMGs and DEGs can be used as prognostic and diagnostic biomarkers for CRC [11,28-30]. A previous study also reported that MDGs can assist with determining the prognosis of CRC patients [21]. However, no studies researched the relationship between MDGs and the diagnoses of CRC patients. In this study, we investigated the interplay between DMGs and DEGs and chose the genes with negative and high correlation, namely MDGs, as candidate predictors for the prognosis and diagnosis of CRC. We identified 10 MDGs that could be considered as potential diagnostic and prognostic predictors for CRC patients: CLDN1, EPHX4, TCN1, LY6G6D, FAM150A, KRT7, STK33, ARHGAP20, KCNJ12 and SPTBN5. Among them, STK33 and EPHX4 were correlated with both diagnosis and prognosis of colorectal cancer, while the others only performed well as diagnostic indicators of colorectal cancer. Consistent with our findings, various studies have demonstrated that STK33 is overexpressed in hypopharyngeal squamous cell carcinoma [31], hepatocellular carcinoma [32], and human large cell lung cancer [33]. STK33 hypermethylation may be a biomarker for the diagnosis, prognosis, and suitable treatment of CRC [34]. Previous studies demonstrated that EPHX4 was significantly upregulated in rectal cancer [35] and pseudomyxoma peritonei [36]. However, the pathological role and clinical significance of EPHX4 for CRC have rarely been reported. As a member of the claudin family and tight junction-related proteins, CLDN1 was demonstrated to be associated with dysfunction or abnormal expression in various tumors [37,38]. In addition, a previous study revealed that aberrant expression of CLDN1 regulated the AMPK/ STAT1/ULK1 signaling pathways, leading to the promotion of proliferation and metastasis in esophageal squamous cancer [39]. CLDN1 was experimentally demonstrated to be remarkably upregulated in CRC patients and could be considered as a methylated diagnostic biomarker in CRC patients and normal control groups [40]. A growing number of studies have verified that the overexpression of TCN1 is associated with tumor invasion and metastasis in CRC [41,42] and experimentally demonstrated that TCN1 was significantly overexpressed in colon cancer tissues compared with normal controls at the mRNA and protein level, and could be considered as potential prognostic biomarker of colon cancer [43]. A previous study showed that KRT7 plays a significant role in tumor metastasis and is considered as a prognostic biomarker and potential target for therapeutic prevention of metastasis [44]. In addition, KRT7 was down-regulated and hypermethylated in CRC tissues compared with adjacent normal tissues and may lead to the occurrence of CRC [45]. Y6G6D belongs to a cluster of leukocyte antigen-6 genes, which was conspicuously overexpressed (around 15-fold) and considered a promising biomarker of immunotherapy for microsatellite stable CRC [46]. FAM150A can Activate Lymphoma Kinase (ALK) by binding to its extracellular domain [47,48]; ALK has been used as an effective biomarker in various human cancers, such as neuroblastoma and non-small cell lung cancer [49]. Additionally, the DNA methylation status of FAM150A was indicated to be a diagnostic and prognostic indicator for clear cell Renal Cell Carcinoma (ccRCC) using the high-performance liquid chromatography method [50]. However, no studies found that FAM150A was associated with CRC. To the best of our knowledge, there is still a lack of research about the relationship between ARHGAP20, KCNJ12, and SPTBN5 and oncogenesis, which may represent novel predictive biomarkers.

The ROC curves were conducted to evaluate the diagnostic performance of each MDG as an independent indicator, and the results showed that the MDGs had relatively high diagnostic values for CRC, except SPTBN5 (set cutoff of AUC as 0.7). The AUC of SPTBN5 as a diagnostic indicator was 0.679 (P-value=3.19E-07) using the RF algorithm, 0.596 (P-value=0.000786) using the LR algorithm, and 0.5 (P-value=1) using the SVM algorithm. A previous study integrated and analyzed the expression data of CRC from the GEO and TCGA databases and constructed the diagnostic model using 10 hub genes; the AUC of the 10 genes were 0.900, 0.927, 0.869, 0.863, 0.980, 0.682, 0.903, 0.790, 0.995, and 0.989 for CCL19, CXCL1, CXCL5, CXCL11, CXCL12, GNG4, INSL5, NMU, PYY, and SST [51]. The results showed that the performance of the two models is similar, and our model integrates the GTEx dataset, which has a relatively large sample size and has been verified with the GEO dataset. The Kaplan-Meier curve of the 10 MDGs showed that STK33 and EPHX4 were significantly related to the prognosis of CRC patients. The AUC of the prognostic model was 0.569 for 3 years, 0.633 for 4 years, and 0.626 for 5 years. A previous study established an immune-related prognostic model for CRC using 9 genes, the AUC of which is 0.627 for 3 years, 0.632 for 4 years, and 0.630 for 5 years [52]. Although our prognostic model is relatively poor, it was built based on fewer genes. Furthermore, we constructed a nomogram based on the multivariable Cox regression coefficients of risk score, age, gender, and tumor stage to further validate our findings. The C-index of the nomogram was significantly higher than the C-index of the risk score model, which was remarkably associated with OS, suggesting that the two MDGs could be used as prognostic indicators.

Conclusion

We identified 10 MDGs that could be used as potential biomarkers for CRC, of which 9 performed well as independent diagnostic predictors and 2 could be used as prognostic indicators.

Our findings suggest that the identified MDGs could be biomarkers for diagnosis and prognosis of CRC.

Supporting Information

Additional file 1: Table S1-Detailed information of DEGs.

Additional file 2: Table S2-Detailed information of DMGs

Data Availability Statement

The data that support the findings of this study are openly available in TCGA and GTEx from the UCSC Xena (https://xenabrowser. net/datapages/) and GEO with accession numbers GSE21815, GSE28000, GSE39582, GSE44076, GSE48684, GSE53051, GSE77718, and GSE101764.

Acknowledgement

This work was supported by a grant from the Key Science and Technology Program of Shaanxi Province (2019ZDLSF02-05) and National Natural Scientific Foundation of China (81974378 and 82003115).

References

1. Bray F, Ferlay J, Soerjomataram I, Siegel RL, Torre LA, Jemal A. Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2018;68(6):394-424.

   [Cross ref] [Google scholar] [Pubmed]

2. Siegel RL, Miller KD, Goding Sauer A, Fedewa SA, Butterly LF, Anderson JC, et al. Colorectal cancer statistics, 2020. CA Cancer J Clin. 2020;70(3):145-164.

   [Cross ref] [Google scholar] [Pubmed]

3. Wang T, Birsoy K, Hughes NW, Krupczak KM, Post Y, Wei JJ, et al. Identification and characterization of essential genes in the human genome. Science. 2015;350(6264):1096-101.

   [Cross ref] [Google scholar] [Pubmed]

4. Dexheimer GM, Alves J, Reckziegel L, Lazzaretti G, Abujamra AL. DNA methylation events as markers for diagnosis and management of acute myeloid leukemia and myelodysplastic syndrome. Dis markers. 2017; 5472893.

   [Cross ref] [Google scholar] [Pubmed]

5. Chen W, Gao C, Liu Y, Wen Y, Hong X, Huang Z. Bioinformatics analysis of prognostic miRNA signature and potential critical genes in colon cancer. Front Genet. 2020;11:478.

   [Cross ref] [Google scholar] [Pubmed]

6. Zhou XG, Huang XL, Liang SY, Tang SM, Wu SK, Huang TT, et al. Identifying miRNA and gene modules of colon cancer associated with pathological stage by weighted gene co-expression network analysis. Onco Targets Ther. 2018;11:2815.

   [Cross ref] [Google scholar] [Pubmed]

7. Lin Y, Pan X, Chen Z, Lin S, Chen S. Identification of an immune-related nine-lncRNA signature predictive of overall survival in colon cancer. Front Genet. 2020;11:318.

   [Cross ref] [Google scholar] [Pubmed]

8. Jin L, Li C, Liu T, Wang L. A potential prognostic prediction model of colon adenocarcinoma with recurrence based on prognostic lncRNA signatures. Hum Genomics. 2020;14(1):1-1.

   [Cross ref] [Google scholar] [Pubmed]

9. Yang Z, Chen Y, Wu D, Min Z, Quan Y. Analysis of risk factors for colon cancer progression. Onco Targets Ther. 2019;12:3991. 

   [Cross ref] [Google scholar] [Pubmed]

10. Hao X, Luo H, Krawczyk M, Wei W, Wang W, Wang J, et al. DNA methylation markers for diagnosis and prognosis of common cancers. Proc Natl Acad Sci. 2017;114(28):7414-7419.

    [Cross ref] [Google scholar] [Pubmed]

11. Yang C, Zhang Y, Xu X, Li W. Molecular subtypes based on DNA methylation predict prognosis in colon adenocarcinoma patients. Aging (Albany NY). 2019;11(24):11880-11892.

    [Cross ref] [Google scholar] [Pubmed]

12. Ying J, Xu T, Wang Q, Ye J, Lyu J. Exploration of DNA methylation markers for diagnosis and prognosis of patients with endometrial cancer. Epigenetics. 2018;13(5):490-504.

    [Cross ref] [Google scholar] [Pubmed]

13. Lesche R, Payne S, Model F, Weiss G, Sledziewski A. DNA methylation markers for diagnosis and prognosis of prostate cancer. Tumor Biology. 2007;28:51-51.

    [Google scholar] [Pubmed]

14. Xu RH, Wei W, Krawczyk M, Wang W, Luo H, Flagg K, et al. Circulating tumour DNA methylation markers for diagnosis and prognosis of hepatocellular carcinoma. Nat Mate. 2017;16(11):1155-1161.

    [Cross ref] [Google scholar] [Pubmed]

15. Herman JG, Baylin SB. Gene silencing in cancer in association with promoter hypermethylation. N Engl J Med. 2003;349(21):2042-2054.

    [Cross ref] [Google scholar] [Pubmed]

16. Vaissière T, Sawan C, Herceg Z. Epigenetic interplay between histone modifications and DNA methylation in gene silencing. Mutat Res. 2008;659(2):40-48. 

    [Cross ref] [Google scholar] [Pubmed]

17. Yassi M, Davodly ES, Shariatpanahi AM, Heidari M, Dayyani M, Heravi-Moussavi A, et al. DMRFusion: A differentially methylated region detection tool based on the ranked fusion method. Genomics. 2018;110(6):366-374.

    [Cross ref] [Google scholar] [Pubmed]

18. Egger G, Liang G, Aparicio A, Jones PA. Epigenetics in human disease and prospects for epigenetic therapy. Nature. 2004;429(6990):457-463.

    [Cross ref] [Google scholar] [Pubmed]

19. Cedoz PL, Prunello M, Brennan K, Gevaert O. MethylMix 2.0: An R package for identifying DNA methylation genes. Bioinformatics. 2018;34(17):3044-3046.

    [Cross ref] [Google scholar] [Pubmed]

20. Gevaert O. MethylMix: An R package for identifying DNA methylation-driven genes. Bioinformatics. 2015;31(11):1839-1841.

    [Cross ref] [Google scholar] [Pubmed]

21. Wang X, Zhang D, Zhang C, Sun Y. Identification of epigenetic methylation-driven signature and risk loci associated with survival for colon cancer. Ann Transl Med. 2020;8(6).

    [Cross ref] [Google scholar] [Pubmed]

22. Dai QX, Liao YH, Deng XH, Xiao XL, Zhang L, Zhou L. A novel epigenetic signature to predict recurrence-free survival in patients with colon cancer. Clin Chim Acta. 2020;508:54-60.

    [Cross ref] [Google scholar] [Pubmed]

23. Cheng S, Jiang Z, Xiao J, Guo H, Wang Z, Wang Y. The prognostic value of six survival-related genes in bladder cancer. Cell Death Discov. 2020;6(1):1-1.

    [Cross ref] [Google scholar] [Pubmed]

24. Wang Y, Li J, Shao C, Tang X, Du Y, Xu T, et al. Systematic profiling of diagnostic and prognostic value of autophagy-related genes for sarcoma patients. BMC cancer. 2021;21(1):1.

    [Cross ref] [Google scholar] [Pubmed]

25. Xia WX, Yu Q, Li GH, Liu YW, Xiao FH, Yang LQ, et al. Identification of four hub genes associated with adrenocortical carcinoma progression by WGCNA. PeerJ. 2019;7:e6555.

    [Cross ref] [Google scholar] [Pubmed]

26. Ritchie ME, Phipson B, Wu DI, Hu Y, Law CW, Shi W, et al. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res. 2015;43(7):e47.

    [Cross ref] [Google scholar] [Pubmed]

27. Tian Y, Morris TJ, Webster AP, Yang Z, Beck S, Feber A, et al. CHAMP: Updated methylation analysis pipeline for illumina beadchips. Bioinformatics. 2017;33:3982-3984.

    [Cross ref] [Google scholar] [Pub Med]

28. Pan F, Chen T, Sun X, Li K, Jiang X, Försti A, et al. Prognosis prediction of colorectal cancer using gene expression profiles. Front Oncol. 2019;9:252.

    [Cross ref] [Google scholar] [Pub Med]

29. Wen S, He L, Zhong Z, Mi H, Liu F. Prognostic model of colorectal cancer constructed by eight immune-related genes. Front Mol Biosci. 2020;7:604252.

    [Cross ref] [Google scholar] [Pub Med]

30. Zhao QQ, Jiang C, Gao Q, Zhang YY, Wang G, Chen XP, et al. Gene expression and methylation profiles identified CXCL3 and CXCL8 as key genes for diagnosis and prognosis of colon adenocarcinoma. J Cell Physiol. 2020; 235: 4902-4912.

    [Cross ref] [Google scholar] [Pub Med]

31. Huang L, Chen C, Zhang G, Ju Y, Zhang J, Wang H, et al. STK33 overexpression in hypopharyngeal squamous cell carcinoma: Possible role in tumorigenesis. BMC Cancer. 2014;15:13.

    [Cross ref] [Google scholar] [Pub Med]

32. Yang T, Song B, Zhang J, Yang GS, Zhang H, Yu WF, et al. STK33 promotes hepatocellular carcinoma through binding to c-Myc. Gut. 2016;65:124-133.

    [Cross ref] [Google scholar] [Pub Med]

33. Wang P, Cheng H, Wu J, Yan A, Zhang L. STK33 plays an important positive role in the development of human large cell lung cancers with variable metastatic potential. Acta Biochim Biophys Sin. 2015;47:214-223.

    [Cross ref] [Google scholar] [Pub Med]

34. Yin MD, Ma SP, Liu F, Chen YZ. Role of serine/threonine kinase 33 methylation in colorectal cancer and its clinical significance. Oncol Lett. 2018;15:2153-2160.

    [Cross ref] [Google scholar] [Pub Med]

35. Flebbe H, Hamdan FH, Kari V, Kitz J, Gaedcke J, Ghadimi BM, et al. epigenome mapping identifies tumor-specific gene expression in primary rectal cancer. Cancers. 2019;11:214-223.

    [Cross ref] [Google scholar] [Pub Med]

36. Roberts DL, O'Dwyer ST, Stern PL, Renehan AG. Global gene expression in pseudomyxoma peritonei, with parallel development of two immortalized cell lines. Oncotarget. 2015;6:10786-10800.

    [Cross ref] [Google scholar] [Pub Med]

37. Martinez C, Rodino-Janeiro BK, Lobo B, Stanifer ML, Klaus B, Granzow M, et al. MiR-16 and miR-125b are involved in barrier function dysregulation through the modulation of claudin-2 and cingulin expression in the jejunum in IBS with diarrhoea. Gut. 2017; 66:1537-1538.

    [Cross ref] [Google scholar] [Pub Med]

38. Ashikari D, Takayama K, Obinata D, Takahashi S, Inoue S. CLDN8, an androgen-regulated gene, promotes prostate cancer cell proliferation and migration. Cancer Science. 2017;108:1386-1393.

    [Cross ref] [Google scholar] [Pub Med]

39. Wu J, Gao F, Xu T, Li J, Hu Z, Wang C, et al. CLDN1 induces autophagy to promote proliferation and metastasis of esophageal squamous carcinoma through AMPK/STAT1/ULK1 signaling. J Cell Physiol. 2020;235:2245-2259.

    [Cross ref] [Google scholar] [Pub Med]

40. Kerachian MA, Javadmanesh A, Azghandi M, Shariatpanahi AM, Yassi M, Shams Davodly E, et al. Crosstalk between DNA methylation and gene expression in colorectal cancer, a potential plasma biomarker for tracing this tumor. Scientific Reports. 2020;10:1537-1538.

    [Cross ref] [Google scholar] [Pub Med]

41. Feodorova Y, Tashkova D, Koev I, Todorov A, Kostov G. Novel insights into transcriptional dysregulation in colorectal cancer. Neoplasma. 2018; 65: 415-424.

    [Cross ref] [Google scholar] [Pub Med]

42. Li M, Zhao LM, Li SL, Li J, Gao B. Differentially expressed lncRNAs and mRNAs identified by NGS analysis in colorectal cancer patients. Cancer Med. 2018;7:4650-4664.

    [Cross ref] [Google scholar] [Pub Med]

43. Liu GJ, Wang YJ, Yue M, Zhao LM, Guo YD, Liu YP, et al. High expression of TCN1 is a negative prognostic biomarker and can predict neoadjuvant chemosensitivity of colon cancer. Scientific Reports. 2020;10.

    [Cross ref] [Google scholar] [Pub Med]

44. Jiang L, Tolani B, Yeh CC, Fan Y, Reza JA, Horvai AE, et al. Differential gene expression identifies KRT7 and MUC1 as potential metastasis-specific targets in sarcoma. Cancer Manag Res. 2019;11: 8209-8218.

    [Cross ref] [Google scholar] [Pub Med]

45. Wang JY, Wang CL, Wang XM, Liu FJ. Comprehensive analysis of microRNA/mRNA signature in colon adenocarcinoma. Eur Rev Med Pharmacol Sci. 2017;21:2114-2129.

    [Cross ref] [Google scholar] [Pub Med]

46. Giordano G, Parcesepe P, D'Andrea MR, Coppola L, Di Raimo T, Manfrin E, et al. JAK/Stat5-mediated subtype-specific lymphocyte antigen 6 complex, locus G6D (LY6G6D) expression drives mismatch repair proficient colorectal cancer. J Exp Clin Cancer Res. 2019;38: 28.

    [Cross ref] [Google scholar] [Pub Med]

47. Guan J, Umapathy G, Yamazaki Y, Wolfstetter G, Mendoza P, Pfeifer K, et al. FAM150A and FAM150B are activating ligands for anaplastic lymphoma kinase. Elife. 2015;4:e09811.

    [Cross ref] [Google scholar] [Pub Med]

48. Janostiak R, Malvi P, Wajapeyee N. Anaplastic lymphoma kinase confers resistance to braf kinase inhibitors in melanoma. I Science. 2019;16:453-467.

    [Cross ref] [Google scholar] [Pub Med]

49. Hallberg B, Palmer RH. Mechanistic insight into ALK receptor tyrosine kinase in human cancer biology. Nat Rev Cancer. 2013;13:685-700.

    [Cross ref] [Google scholar] [Pub Med]

50. Yotani T, Yamada Y, Arai E, Tian Y, Gotoh M, Komiyama M, et al. Novel method for DNA methylation analysis using high-performance liquid chromatography and its clinical application. Cancer Sci. 2018;109:1690-1700.

    [Cross ref] [Google scholar] [Pub Med]

51. Chen L, Lu D, Sun K, Xu Y, Hu P, Li X, et al. Identification of biomarkers associated with diagnosis and prognosis of colorectal cancer patients based on integrated bioinformatics analysis. Gene. 2019;692:119-125.

    [Cross ref] [Google scholar] [Pub Med]

52. Zhao XJ, Liu JZ, Liu SZ, Yang FF, Chen EF. construction and validation of an immune-related prognostic model based on TP53 status in colorectal cancer. Cancers. 2019;11.

    [Cross ref] [Google scholar] [Pub Med]