Multifactor dimensionality reduction method identifies novel SNP interactions in the WNT protein interaction networks that are associated with recurrence risk in colorectal cancer

Abstract

Background:

Interactions among genetic variants are rarely studied but may explain a part of the variability in patient outcomes.

Objectives:

In this study, we aimed to identify 1 to 3 way interactions among SNPs from five Wnt protein interaction networks that predict the 5-year recurrence risk in a cohort of stage I-III colorectal cancer patients.

Methods:

423 patients recruited to the Newfoundland Familial Colorectal Cancer Registry were included. Five Wnt family member proteins (Wnt1, Wnt2, Wnt5a, Wnt5b, and Wnt11) were selected. The BioGRID database was used to identify the proteins interacting with each of these proteins. Genotypes of the SNPs located in the interaction network genes were retrieved from a genome-wide SNP genotype data previously obtained in the patient cohort. The GMDR 0.9 program was utilized to examine 1-, 2-, and 3-SNP interactions using a 5-fold cross validation step. Top GMDR 0.9 models were assessed by permutation testing and, if significant, prognostic associations were verified by multivariable logistic regression models.

Results:

GMDR 0.9 has identified novel 1, 2, and 3-way SNP interactions associated with 5-year recurrence risk in colorectal cancer. Nine of these interactions were multi loci interactions (2-way or 3-way). Identified interaction models were able to distinguish patients based on their 5-year recurrence-free status in multivariable regression models. The significance of interactions was the highest in the 3-SNP models. Several of the identified SNPs were eQTLs, indicating potential biological roles of the genes they were associated with in colorectal cancer recurrence.

Conclusions:

We identified novel interacting genetic variants that associate with 5-year recurrence risk in colorectal cancer. A significant portion of the genes identified were previously linked to colorectal cancer pathogenesis or progression. These variants and genes are of interest for future functional and prognostic studies. Our results provide further evidence for the utility of GMDR models in identifying novel prognostic biomarkers and the biological importance of the Wnt pathways in colorectal cancer.

Background

One of the most common cancers in the world is colorectal cancer (1). This disease includes the cancers of the colon and rectum, has a number of identified genetic and environmental/life-style risk factors, shows geographic difference in incidence and survival rates, and is overall characterized by moderate to low survival rates (1–5). While there are a number of disease and patient related factors that help prognosis (6, 7), identifying additional factors is needed to improve the precision of prognosis (8). Personalized/Precision Medicine approaches can improve prognosis by focusing on new biomarkers. Germline (i.e. not tumor) genetic variations, such as SNPs, are candidate biomarkers, as they are relatively stable, abundant, and show variability among individuals (9). They also have the potential to help identify the biological bases of human conditions and phenotypes. As such, there has been an emphasis on examining SNP–outcome associations in cancers, including in colorectal cancer.

An important aspect of cancer genetics that these analyses miss is that potential interactions among SNPs may be associated with patient outcomes. For example, most of the studies in colorectal cancer have so far focused on individual SNPs’ associations with outcome risk/survival times (10–15). However, it is possible that a SNP’s genotypes may not be associated with the outcome on its own, but they may when they exist together with another SNP’s genotypes (16). Examining interactions can be challenging, however, as the number of variables examined increases, so does the need for computational resources. Methodologies (such as, Multifactor Dimensionality Reduction [MDR]) that address this challenge and tools that utilize these methods (such as GMDR 0.9) have been developed to examine interactions in a relatively feasible way (16–18). We and others have previously shown the utility of MDR and examining interactions among genetic variables in colorectal cancer (19–23). As these studies showed, examining interactions is a promising research area and can reveal new biomarkers that can predict patient outcomes.

Wnt genes have important roles in normal cellular functions as well as colorectal cancer development and its progression (24–26). Therefore, they are excellent biological candidates to examine in colorectal cancer. In this study, our aim was to use a Generalized MDR tool (GMDR 0.9 (18)) to examine the interactions among the germline variables of five Wnt protein interaction networks in relation to 5-year recurrence-free survival status in a cohort of stage I-III colorectal cancer patients.

Data and methods

Ethics statement

This study was conducted with ethics approval by the Health Research Ethics Authority of Newfoundland and Labrador (HREB #2018.051; #2009.106). This study was a secondary use of data study, hence, HREB waived the requirement for patient consent.

Clinical and genetic patient data

The baseline features of the patient cohort are shown in Table 1. The patients included in this study were recruited to the Newfoundland Familial Colorectal Cancer Registry (NFCCR) between 1999-2003, and were followed up until 2018 (11, 27, 28). Clinical and pathological data were collected by the NFCCR using medical records, tumor registry data, and other resources as described in other publications (11, 27, 28). Genetic data was a part of the previously obtained genomewide SNP genotype data (29). PLINK (1.7) (30) was used to manage the SNP genotype data used in this study. The 5-year local or distant recurrence-free survival (RMFS) status was the response variable. All patients were unrelated to each other and of Caucasian background (29).

Identification of Wnt interactome networks using the BioGRID database

We focused on five WNT genes in non-canonical/β-catenin independent pathway: Wnt1, Wnt2, Wnt5a, Wnt5b, and Wnt11 (25). We utilized the BioGRID database (31) to retrieve the interaction networks for each of these genes. Table S1 shows the genes in each protein interaction network after implementing quality control measures [described in detail in Curtis et al. (23)]. We then used the SNP genotype data and PLINK to retrieve the SNPs located in these genes using the following criteria: Minor Allele Frequency (MAF) >= 0.05; Hardy-Weinberg Equilibrium (HWE) > 0.0001; and missing genotype data = 0%. Pruning was performed by PLINK to remove the SNPs in high-LD, as explained in Curtis et al. (23). As a result we ended up with 6 (in Wnt1), 18 (in Wnt2), 53 (in Wnt5a), 4 (in Wnt5b), and 20 (in Wnt11) genes respectively (Supplementary Table S1), and around 1,000 SNPs in these Wnt interaction networks (Supplementary Table S2).

GMDR 0.9 runs, permutation testing, and statistical analyses

GMDR 0.9 (18, 32) was previously downloaded from the UAB Department of Biostatistics Section on Statistical Genetics website. All analyses were done as described in Curtis et al. (23). In brief, we have conducted 1-way interactions (examining the interactions among the three potential genotypes of single SNPs); 2-way interactions (examining the interactions among the genotypes of two SNPs); and 3-way interactions (examining the interactions among the genotypes of three SNPs). Known prognostic markers, disease stage, tumor location, and adjuvant chemotherapy and radiotherapy statuses were used as variables. A 5-step cross-validation was implemented; where the dataset was partitioned into five parts, with four parts used as the training cohort and the remaining 5^th part used as the testing (i.e. validation) cohort. Each of the five partitions was designated the testing cohort once and the results for each of these datasets were compared. GMDR analysis was repeated 20 times, using different random seeds, and the MDR model that was identified most frequently and with the highest Testing Balance Accuracy (TBA) score was selected as the top model for the examined interaction. Rarely, we used the higher Cross Validation Consistency (CVC) and specificity data to break a tie. We performed 1-way analysis iteratively for each dataset, and if a model was found to be significant (i.e. a SNP with a main effect was identified in the dataset), such SNPs were removed from the dataset. This was repeated until no significant 1-way model was found (16, 33). The significance of the top model was then examined using permutation testing. Models that were significant (p < 0.001) were examined in multivariable logistic regression models, adjusting for disease stage, tumor location, and adjuvant chemotherapy and radiotherapy status. A p-value < 0.05 was considered significant in multivariable logistic regression models. Kaplan Meier curves were created for overall survival (OS) and RMFS using the long-term follow up data (28) and end point status (in OS, the end point was death from any cause). In addition, 18 patients who were censored before or at the 5 year time point and as such were excluded from the GMDR and logistic regression analyses, were included in the Kaplan Meier analyses to limit bias. Figure 1 shows the OS and RMFS curves for these patients. The 5-year OS probability was around 80% and a little bit less than that for the RMFS. R (version 3.5.3) (34) and SPSS (version 28.0.0.0 (190); 35) were used for data processing and statistical analyses.

Figure 1

Functional annotation of genes and SNPs using databases

Identified SNPs and genes were examined for their functional and biological features using literature and databases. Specifically, SNPs that were identified in 1-SNP, 2-SNP, and 3-SNP interaction analyses were checked for their functional consequences in RegulomeDB (v2.0.3) (36), which uses a ranking system to describe the regulatory potential of SNPs (ranks 1a-1f specify expression quantitative trait loci [eQTLs]). In addition, SNPs were searched in the GTEx (data release v8) (37) database to see whether they were eQTLs. The latter database provided eQTL data for colon tissues only. The dbCPCO database (38) was utilized to search whether the identified SNPs were previously reported to be associated with clinical outcomes in colorectal cancer. Information about the genes/proteins was collected from literature and the Gene Entrez database (39). The dbSNP database (40) was utilized to retrieve information on SNP annotations (e.g. intronic, missense).

Results

The clinical features of the patient cohort are summarized in Table 1.

By examining around 26,298,702 interactions (2-SNP interactions=173,425; 3-SNP interactions=26,125,277) in five Wnt interactome networks, our investigation identified 32 novel interactions associated with 5-year RMFS status in colorectal cancer.

Interactions were identified in each of the networks examined. Twenty-three 1-SNP interactions, where the genotypes of individual SNPs were identified as associated with the 5-year RMFS status in multivariable logistic regression models are shown in Supplementary Table S3. Additionally, we identified nine multi-SNP interactions (four 2-way and five 3-way interactions) that were associated with the 5-year RMFS status in the patient cohort, when adjusted for other prognostic markers in the logistic regression models (Table 2). Kaplan Meier curves for these interactions are shown in Figure 2 (for the Wnt5a network) and Supplementary Figure S1 (for the Wnt1, Wnt2, Wnt5b, and Wnt11 networks).

Figure 2

In all logistic regression models, the direction of the Odds Ratios (ORs) were consistent with the high-risk and low-risk genotype combinations identified in the top MDR models.

As observed in other studies, as the order of interactions increased (from 1-SNP to 3-SNP), the significance of the association detected increased as well. In other words, the 3-SNP interaction models were the ones that best separated patients based on their 5-year RMFS status (p-values 10^-7 to 10^-12). ORs were also higher in 3-way models compared 1-SNP and 2-SNP models (Table 2).

All SNPs identified in interaction models were common in the patient cohort (MAF ≥ 5%). Only one of the SNPs (FUCA2. rs11155297, NP_114409.2:p.Ala233Glu) was a missense variant, whereas the other variants were intronic/non-coding or 3’-UTR variants (Supplementary Table S4). Interestingly, some of the SNPs were predicted to be functional (i.e. eQTLs) based on the RegulomeDB scores or GTEx data (Table 3). Another interesting finding was that both of the SNPs in a 2-way interaction were eQTLs (in sigmoid and transverse colon tissues): FUCA2.rs11155297 (eQTL for ADAT2) and TMED7.rs10075869 (eQTL for AC010226.4). Some of the genes in the interaction networks as well as those that are associated with eQTLs had literature findings indicative of their involvement in pathogenesis of colorectal cancer or its progression (Supplementary Table S5).

Discussion

As a globally common disease with moderate to low survival rates (1–5), it is critical to identify new biomarkers that can help prognosis in colorectal cancer. One under-studied but promising research area is that of interactions, where multiple variables together relate to a phenotype, such as recurrence status in cancer patients. Here, we report our study that used GMDR 0.9 (18), a data reduction tool, to examine interactions among different SNPs from the Wnt interactome genes in a cohort of colorectal cancer patients from Newfoundland and Labrador (27). As a result, we were able to identify novel SNP interactions in the five WNT interactome sets that were predictive of 5-year RMFS status in colorectal cancer.

Our results underline the utility of MDR in identifying previously unknown interactions and potential prognostic biomarkers. As shown in Table 2 and Supplementary Table S3, GMDR 0.9, permutation testing, and multivariable logistic regression analyses identified interactions that can distinguish patients based on their 5-year RMFS risk when adjusted for prognostic covariates. Note that high and low risk patient group classifications made by GMDR 0.9 were supported by the direction of effect (i.e. ORs) in the regression models. This increases the confidence in the GMDR 0.9 risk classification system. Among the interactions identified, 2-SNP and 3-SNP interactions (n=9) are the most interesting ones, as they significantly contribute knowledge to the largely unknown multi-loci interactions in colorectal cancer. According to the dbCPCO database (38), only one of the SNPs identified in this study – HSPA5.rs12009 – was investigated in relation to colorectal cancer outcomes in a stage II-III patient cohort. It was not associated with either stage or time-to-recurrence in the study patient cohort which consisted of mixed-ethnicities (41). Of note, in our study HSPA5.rs12009 was identified in both WNT2 and WNT5A pathway analyses and is an eQTL. Prognostic associations of this and other SNPs identified in this study can be verified in additional colorectal cancer cohorts.

It is hard to predict the biological reasons behind such interactions without detailed experimental analyses. However, whether or not these SNPs are likely to have biological roles in the phenotype can be first assessed using existing information, for example, based on their genomic/genic locations, predictive tools/databases, and previously conducted large-scale experiments (such as eQTL analyses). In this regard, RegulomeDB (36) and GTEx (37) database information suggested that a number of the identified variants were eQTLs that were associated with the expression levels of genes. In four cases, these genes included the genes that the SNPs were located in. For example, WLS.rs2116046 was an eQTL for GNG12-AS1 as well as WLS in transverse colon (note that sequences of WLS and GNG12-AS1 overlap, and rs2116046 is also located in GNG12-AS1) (Table 3). An interesting case was identified in the 2-way interaction results where both of the SNPs identified turned out to be eQTLs: FUCA2.rs11155297 (eQTL for ADAT2 in both sigmoid and transverse colon) and TMED7.rs10075869 (eQTL for AC010226.4 in both sigmoid and transverse colon). Little is known about the latter gene but ADAT2 is involved in tRNA modification (42), upregulated through translational control by BRCA1 and a marker of BRCA1 depletion in cell lines and human breast tumors (43), identified as over-expressed in different solid tumors (44), and linked to malignant transformation through its roles in affecting chromatin, transcriptional processes, and apoptosis (44). Also, while neither TMED7 (transmembrane p24 trafficking protein 7) nor FUCA2 (alpha-L-fucosidase 2) genes are known to be linked to colorectal cancer, FUCA2 is associated with various cancers, tumor microenvironment and prognostic features (45). Although this is the first time ADAT2 has been linked to colorectal cancer, our data and literature information make ADAT2 (as well as FUCA2.rs11155297) interesting candidates for future studies in colorectal cancer progression and prognosis.

Additional genes are worth discussion. Supplementary Table S5 shows the information collected about the genes in the five Wnt interactome networks as well as the genes associated with the eQTL SNPs. A significant portion of the genes was already linked to colorectal cancer pathogenesis or progression/prognosis by previous studies. For example, expression levels, deletion, or biological functions of ROR2, SFRP1, LRP6, PITX2, WLS, GPC1, HCK, HPN, MKRN2, and DDX58 were associated with disease features, patient outcomes/prognosis, or invasive and other malignant features of colorectal tumors (46–57). This literature information supports our findings and can be partly attributed to the fact that the Wnt pathway is one of the most studied pathways in colorectal cancer, increasing the chances of finding literature information on genes functioning in Wnt-related biological processes. Overall, future biological studies and/or interventions can be planned for the genes and eQTLs identified by our analyses.

This study has a number of strengths and limitations. This is one of the few large-scale studies examining such a large number of interactions in colorectal cancer. Interactions identified are novel. The patient cohort is a well annotated cohort and the genes selected have been previously shown to have abnormalities/functional roles in colorectal cancer development and or progression (24–26). The 5-step cross validation and repeating the analyses 20 times helped reduce the false-positive findings, in addition to the permutation testing. Additionally, we examined SNP interactions in protein interaction networks, increasing the biological plausibility of the identified interactions. Our cohort, however, consists of only Caucasian patients, therefore, our results may not be applicable to other ethnicities/populations. The identified variants associated with the 5-year recurrence-free survival need to be verified in other patient cohorts for generalizability of our findings. The study focused on common (MAFs >=5%) SNPs from the autosomal chromosomes, and hence, missed examining the associations of rare variables and variables from sex-chromosomes. As we reported earlier, GMDR 0.9 has certain limitations, so it may have missed interactions (23). However, use of permutation testing, repeating the MDR procedure and choosing the most frequently identified MDR model for each examined interaction, and multivariable regression modeling also have limited the false-positive findings.

In conclusion, we present novel 1 to 3 way SNP interactions that predict the 5-year RMFS status in colorectal cancer. These interactions are excellent candidates for further verification in other patient cohorts. We also identified a number of genes that are biologically linked to colorectal cancer: they form an exciting set for future studies or interventions in colorectal cancer. Our results also indicate that MDR and other data reduction methods should be utilized more widely for comprehensive investigations of statistical interactions in patient prognosis. Finally, our findings also re-emphasize and strengthen the importance of Wnt protein pathways in colorectal cancer.

References

• 1

  SungHFerlayJSiegelRLLaversanneMSoerjomataramIJemalAet al. Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin (2021) 71:209–49. doi: 10.3322/caac.21660

  • CrossRef
  • Google Scholar

• 2

  BoylePLangmanJS. ABC Of colorectal cancer: Epidemiology. BMJ (2000) 321:805–8. doi: 10.1136/bmj.321.7264.805

  • CrossRef
  • Google Scholar

• 3

  JaspersonKWTuohyTMNeklasonDWBurtRW. Hereditary and familial colon cancer. Gastroenterology (2010) 138:2044–58. doi: 10.1053/j.gastro.2010.01.054

  • CrossRef
  • Google Scholar

• 4

  FerlayJSoerjomataramIDikshitREserSMathersCRebeloMet al. Cancer incidence and mortality worldwide: Sources, methods and major patterns in GLOBOCAN 2012. Int J Cancer (2015) 136:E359–86. doi: 10.1002/ijc.29210

  • CrossRef
  • Google Scholar

• 5

  HermelinkRLeitzmannMFMarkozannesGTsilidisKPukropTBergerFet al. Sedentary behavior and cancer–an umbrella review and meta-analysis. Eur J Epidemiol. (2022) 37:447–60. doi: 10.1007/s10654-022-00873-6

  • CrossRef
  • Google Scholar

• 6

  ComptonCC. Colorectal carcinoma: diagnostic, prognostic, and molecular features. Mod Pathol (2003) 16:376–88. doi: 10.1097/01.MP.0000062859.46942.93

  • CrossRef
  • Google Scholar

• 7

  ZlobecILugliA. Prognostic and predictive factors in colorectal cancer. J Clin Pathol (2008) 61:561–9. doi: 10.1136/jcp.2007.054858

  • CrossRef
  • Google Scholar

• 8

  SavasSLiuG. Genetic variations as cancer prognostic markers: review and update. Hum Mutat (2009) 30:1369–77. doi: 10.1002/humu.21078

  • CrossRef
  • Google Scholar

• 9

  1000 Genomes Project ConsortiumAbecasisGRAltshulerDAutonABrooksLDDurbinRMet al. A map of human genome variation from population-scale sequencing. Nature (2010) 467:1061–73. doi: 10.1038/nature09534

  • CrossRef
  • Google Scholar

• 10

  TheodoratouEDinFVFarringtonSMCetnarskyjRBarnetsonRAPorteousMEet al. Association between common mtDNA variants and all-cause or colorectal cancer mortality. Carcinogenesis (2010) 31:296–301. doi: 10.1093/carcin/bgp237

  • CrossRef
  • Google Scholar

• 11

  NegandhiAHydeADicksEYounghusbandBParfreyPGreenRet al. MTHFR Glu429Ala and ERCC5 His46His polymorphisms are associated with prognosis in colorectal cancer patients: Analysis of two independent cohorts from Newfoundland. PloS One (2013) 8:e61469. doi: 10.1371/journal.pone.0061469

  • CrossRef
  • Google Scholar

• 12

  PhippsAIPassarelliMNChanATHarrisonTAJeonJHutterCMet al. Common genetic variation and survival after colorectal cancer diagnosis: a genome-wide analysis. Carcinogenesis (2016) 37:87–95. doi: 10.1093/carcin/bgv161

  • CrossRef
  • Google Scholar

• 13

  YuYWerdyaniSCareyMParfreyPYilmazYESavasS. A comprehensive analysis of SNPs and CNVs identifies novel markers associated with disease outcomes in colorectal cancer. Mol Oncol (2021) 15:3329–47. doi: 10.1002/1878-0261.13067

  • CrossRef
  • Google Scholar

• 14

  LabadieJDSavasSHarrisonTABanburyBHuangYBuchananDDet al. Genome-wide association study identifies tumor anatomical site-specific risk variants for colorectal cancer survival. Sci Rep (2022) 12:127. doi: 10.1038/s41598-021-03945-x

  • CrossRef
  • Google Scholar

• 15

  QuintanilhaJCFWangJSibleyABXuWEspin-GarciaOJiangCet al. Genome-wide association studies of survival in 1520 cancer patients treated with bevacizumab-containing regimens. Int J Cancer (2022) 150:279–89. doi: 10.1002/ijc.33810

  • CrossRef
  • Google Scholar

• 16

  LeeSKwonM-SOhJMParkT. Gene-gene interaction analysis for the survival phenotype based on the cox model. Bioinformatics (2012) 28:i582–8. doi: 10.1093/bioinformatics/bts415

  • CrossRef
  • Google Scholar

• 17

  MotsingerAARitchieMD. Multifactor dimensionality reduction: An analysis strategy for modelling and detecting gene - gene interactions in human genetics and pharmacogenomics studies. Hum Genomics (2006) 2:318. doi: 10.1186/1479-7364-2-5-318

  • CrossRef
  • Google Scholar

• 18

  LouX-YChenG-BYanLMaJZZhuJElstonRCet al. A generalized combinatorial approach for detecting gene-by-Gene and gene-by-Environment interactions with application to nicotine dependence. Am J Hum Genet (2007) 80:1125–37. doi: 10.1086/518312

  • CrossRef
  • Google Scholar

• 19

  PanderJWesselsJAMGelderblomHvan der StraatenTPuntCJAGuchelaarH-J. Pharmacogenetic interaction analysis for the efficacy of systemic treatment in metastatic colorectal cancer. Ann Oncol (2011) 22:1147–53. doi: 10.1093/annonc/mdq572

  • CrossRef
  • Google Scholar

• 20

  AfzalSGusellaMJensenSAVainerBVogelUAndersenJTet al. The association of polymorphisms in 5-fluorouracil metabolism genes with outcome in adjuvant treatment of colorectal cancer. Pharmacogenomics (2011) 12:1257–67. doi: 10.2217/pgs.11.83

  • CrossRef
  • Google Scholar

• 21

  AfzalSGusellaMVainerBVogelUBAndersenJTBroedbaekKet al. Combinations of polymorphisms in genes involved in the 5-fluorouracil metabolism pathway are associated with gastrointestinal toxicity in chemotherapy-treated colorectal cancer patients. Clin Cancer Res (2011) 17:3822–9. doi: 10.1158/1078-0432.CCR-11-0304

  • CrossRef
  • Google Scholar

• 22

  SaracSBRasmussenCHAfzalSThirstrupSJensenSAColding-JørgensenMet al. Data-driven assessment of the association of polymorphisms in 5-fluorouracil metabolism genes with outcome in adjuvant treatment of colorectal cancer. Bas. Clin Pharmacol Toxicol (2012) 111:189–97. doi: 10.1111/j.1742-7843.2012.00885.x

  • CrossRef
  • Google Scholar

• 23

  CurtisAYuYCareyMParfreyPYilmazYESavasS. Examining SNP-SNP interactions and risk of clinical outcomes in colorectal cancer using multifactor dimensionality reduction based methods. Front Genet (2022) 13:902217. doi: 10.3389/fgene.2022.902217

  • CrossRef
  • Google Scholar

• 24

  de LauWBarkerNCleversH. WNT signaling in the normal intestine and colorectal cancer. Front Biosci (2007) 12:471–91. doi: 10.2741/2076

  • CrossRef
  • Google Scholar

• 25

  NieXLiuHLiuLWangY-DChenW-D. Emerging roles of wnt ligands in human colorectal cancer. Front Oncol (2020) 10:1341. doi: 10.3389/fonc.2020.01341

  • CrossRef
  • Google Scholar

• 26

  DisomaCZhouYLiSPengJXiaZ. Wnt/β-catenin signaling in colorectal cancer: Is therapeutic targeting even possible? Biochimie (2022) 195:39–53. doi: 10.1016/j.biochi.2022.01.009

  • CrossRef
  • Google Scholar

• 27

  WoodsMOYounghusbandHBParfreyPSGallingerSMcLaughlinJDicksEet al. The genetic basis of colorectal cancer in a population-based incident cohort with a high rate of familial disease. Gut (2010) 59:1369–77. doi: 10.1136/gut.2010.208462

  • CrossRef
  • Google Scholar

• 28

  YuYCareyMPollettWGreenJDicksEParfreyPet al. The long-term survival characteristics of a cohort of colorectal cancer patients and baseline variables associated with survival outcomes with or without time-varying effects. BMC Med (2019) 17:150. doi: 10.1186/s12916-019-1379-5

  • CrossRef
  • Google Scholar

• 29

  XuWXuJShestopaloffKDicksEGreenJParfreyPet al. A genome wide association study on Newfoundland colorectal cancer patients’ survival outcomes. biomark Res (2015) 3:6. doi: 10.1186/s40364-015-0031-6

  • CrossRef
  • Google Scholar

• 30

  PurcellSNealeBTodd-BrownKThomasLFerreiraMABenderDet al. PLINK: a tool set for whole-genome association and population-based linkage analyses. Am J Hum Genet (2007) 81:559–75. doi: 10.1086/519795

  • CrossRef
  • Google Scholar

• 31

  OughtredRRustJChangCBreitkreutzB-JStarkCWillemsAet al. The BioGRID database: A comprehensive biomedical resource of curated protein, genetic, and chemical interactions. Protein Sci (2021) 30:187–200. doi: 10.1002/pro.3978

  • CrossRef
  • Google Scholar

• 32

  ChenG-BXuYXuH-MLiMDZhuJLouX-Y. Practical and theoretical considerations in study design for detecting gene-gene interactions using MDR and GMDR approaches. PloS One (2011) 6:e16981. doi: 10.1371/journal.pone.0016981

  • CrossRef
  • Google Scholar

• 33

  RitchieMDHahnLWRoodiNBaileyLRDupontWDParlFFet al. Multifactor-dimensionality reduction reveals high-order interactions among estrogen-metabolism genes in sporadic breast cancer. Am J Hum Genet (2001) 69:138–47. doi: 10.1086/321276

  • CrossRef
  • Google Scholar

• 34

  Core Team, R. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing (2017).

  • Google Scholar

• 35

  IBM SPSS Statistics for Windows. (2017).

  • Google Scholar

• 36

  BoyleAPHongELHariharanMChengYSchaubMAKasowskiMet al. Annotation of functional variation in personal genomes using RegulomeDB. Genome Res (2012) 22:1790–7. doi: 10.1101/gr.137323.112

  • CrossRef
  • Google Scholar

• 37

  LonsdaleJThomasJSalvatoreMPhillipsRLoEShadSet al. The genotype-tissue expression (GTEx) project. Nat Genet (2013) 45:580–5. doi: 10.1038/ng.2653

  • CrossRef
  • Google Scholar

• 38

  SavasSYounghusbandHB. dbCPCO: a database of genetic markers tested for their predictive and prognostic value in colorectal cancer. Hum Mutat (2010) 31:901–7. doi: 10.1002/humu.21285

  • CrossRef
  • Google Scholar

• 39

  MaglottDOstellJPruittKDTatusovaT. Entrez gene: gene-centered information at NCBI. Nucleic Acids Res (2011) 39:D52–7. doi: 10.1093/nar/gkq1237

  • CrossRef
  • Google Scholar

• 40

  SherrySTWardMHKholodovMBakerJPhanLSmigielskiEMet al. dbSNP: the NCBI database of genetic variation. Nucleic Acids Res (2001) 29:308–11. doi: 10.1093/nar/29.1.308

  • CrossRef
  • Google Scholar

• 41

  WinderTBohanesPZhangWYangDPowerDGNingYet al. GRP78 promoter polymorphism rs391957 as potential predictor for clinical outcome in gastric and colorectal cancer patients. Ann Oncol (2011) 22:2431–9. doi: 10.1093/annonc/mdq771

  • CrossRef
  • Google Scholar

• 42

  SchaubMKellerW. RNA Editing by adenosine deaminases generates RNA and protein diversity. Biochimie (2002) 84:791–803. doi: 10.1016/s0300-9084(02)01446-3

  • CrossRef
  • Google Scholar

• 43

  BerthelEVincentAEberstLTorresAGDacheuxEReyCet al. Uncovering the translational regulatory activity of the tumor suppressor BRCA1. Cells (2020) 9:941. doi: 10.3390/cells9040941ac

  • CrossRef
  • Google Scholar

• 44

  CaronCLestratCMarsalSEscoffierECurtetSVirolleVet al. Functional characterization of ATAD2 as a new cancer/testis factor and a predictor of poor prognosis in breast and lung cancers. Oncogene (2010) 29:5171–81. doi: 10.1038/onc.2010.259

  • CrossRef
  • Google Scholar

• 45

  LiuQDongH-TZhaoTYaoFXuYChenBet al. Cancer-associated adipocytes release FUCA2 to promote aggressiveness in TNBC. Endocr. Relat Cancer (2022) 29:139–49. doi: 10.1530/ERC-21-0243

  • CrossRef
  • Google Scholar

• 46

  HiroseHIshiiHMimoriKTanakaFTakemasaIMizushimaTet al. The significance of PITX2 overexpression in human colorectal cancer. Ann Surg Oncol (2011) 18:3005–12. doi: 10.1245/s10434-011-1653-z

  • CrossRef
  • Google Scholar

• 47

  MeiHLianSZhangSWangWMaoQWangH. High expression of ROR2 in cancer cell correlates with unfavorable prognosis in colorectal cancer. Biochem Biophys Res Commun (2014) 453:703–9. doi: 10.1016/j.bbrc.2014.09.141

  • CrossRef
  • Google Scholar

• 48

  XuHJiangWZhuFZhuCWeiJWangJ. Expression of wntless in colorectal carcinomas is associated with invasion, metastasis, and poor survival. APMIS (2016) 124:522–8. doi: 10.1111/apm.12534

  • CrossRef
  • Google Scholar

• 49

  YaoQAnYHouWCaoY-NYaoM-FMaN-Net al. LRP6 promotes invasion and metastasis of colorectal cancer through cytoskeleton dynamics. Oncotarget (2017) 8:109632–45. doi: 10.18632/oncotarget.22759

  • CrossRef
  • Google Scholar

• 50

  GuCWangXLongTWangXZhongYMaYet al. FSTL1 interacts with VIM and promotes colorectal cancer metastasis via activating the focal adhesion signalling pathway. Cell Death Dis (2018) 9:1–14. doi: 10.1038/s41419-018-0695-6

  • CrossRef
  • Google Scholar

• 51

  RoseweirAKPowellAGMTHorstmanSLInthagardJParkJHMcMillanDCet al. Src family kinases, HCK and FGR, associate with local inflammation and tumour progression in colorectal cancer. Cell Signal (2019) 56:15–22. doi: 10.1016/j.cellsig.2019.01.007

  • CrossRef
  • Google Scholar

• 52

  ChangZLiuXZhaoWXuY. Identification and characterization of the copy number dosage-sensitive genes in colorectal cancer. Mol Ther Methods Clin Dev (2020) 18:501–10. doi: 10.1016/j.omtm.2020.06.020

  • CrossRef
  • Google Scholar

• 53

  BusuiocCCiocan-CartitaCABraicuCZanoagaORadulyLTrifMet al. Epithelial-mesenchymal transition gene signature related to prognostic in colon adenocarcinoma. J Pers. Med (2021) 11:476. doi: 10.3390/jpm11060476

  • CrossRef
  • Google Scholar

• 54

  HeYGongPWangSXuQChenJ. The significance of homeodomain transcription factor 2 in colon cancer cells. Biomed Eng. Online (2021) 20:81. doi: 10.1186/s12938-021-00912-5

  • CrossRef
  • Google Scholar

• 55

  DengYFuHHanXLiYZhaoWZhaoXet al. Activation of DDX58/RIG-I suppresses the growth of tumor cells by inhibiting STAT3/CSE signaling in colon cancer. Int J Oncol (2022) 61:1–13. doi: 10.3892/ijo.2022.5410

  • CrossRef
  • Google Scholar

• 56

  LuFChenSShiWSuXWuHLiuM. GPC1 promotes the growth and migration of colorectal cancer cells through regulating the TGF-β1/SMAD2 signaling pathway. PloS One (2022) 17:e0269094. doi: 10.1371/journal.pone.0269094

  • CrossRef
  • Google Scholar

• 57

  Zaragoza-HuescaDNieto-OlivaresAGarcía-MolinaFRicoteGMontenegroSSánchez-CánovasMet al. Implication of hepsin from primary tumor in the prognosis of colorectal cancer patients. Cancers (2022) 14:3106. doi: 10.3390/cancers14133106

  • CrossRef
  • Google Scholar