Risk SNP-Mediated Enhancer–Promoter Interaction Drives Colorectal Cancer through Both FADS2 and AP002754.2

Colorectal cancer is a leading cancer-related cause of death worldwide (1). In China, colorectal cancer is the third most commonly diagnosed cancer and the fifth leading cause of cancer-related deaths, causing a major public health burden (2, 3). Convincing evidence is noted that some environmental exposures, including alcohol intake, tobacco smoking, overnutrition, and obesity could contribute to the risk of colorectal cancer (4). In addition, inherited susceptibility is also a major component of colorectal cancer predisposition, with an estimated 12%–35% risk attributed to genetic factors (5, 6).

To date, genome-wide association studies (GWAS) have identified more than 100 colorectal cancer risk loci (7–22). Approximately 90% of the SNPs identified by GWAS are located in noncoding regions (23). It is a formidable challenge for decoding the function of these noncoding variants and identifying their target genes, which are critical for translating GWAS findings into clinically therapeutic targets. The traditional annotation of GWAS hits usually focuses on the nearest or most biologically plausible gene candidate, which may not be the true targets for some variants. Interestingly, recent studies have found that some noncoding SNPs within potential regulatory elements regulate the expression of distal genes by long-range genome interactions (24, 25), which makes us speculate whether there are complex regulatory mechanisms between noncoding SNPs and gene expression. The expression quantitative trait loci (eQTL) approach, which evaluates the association of a variant genotype with both local and distal gene expression levels, is used to pinpoint the candidate target genes of noncoding variants for complex traits, including colorectal cancer, using multilevel information provided in the Cancer Genome Atlas (TCGA; ref. 26) and the Genotype-Tissue Expression project (27). However, most eQTL-based strategies are based on data derived from mRNA expression, while ignoring the noncoding RNA expression, which has been proposed to carry out diverse functions in cancers (28).

In this study, we first conducted a comprehensive eQTL analysis to evaluate the associations between variants and candidate mRNAs or long noncoding RNAs (lncRNA) expression in TCGA colorectal cancer samples. Next, through integrating a large-scale population consisting of 6,024 cases and 10,022 controls and a series of functional experiments, we demonstrated that a risk SNP located in the 11q12.2 region, acts as an allele-specific enhancer to provoke FADS2 and lncRNA AP002754.2 expression through long-range enhancer–promoter interactions, thus resulting in an increased risk of colorectal cancer. Mechanistically, AP002754.2 exerts a transcriptional activator that promotes FADS2 expression, which functions as a potential oncogene facilitating colorectal cancer cell proliferation and xenograft growth by increasing the metabolism of oncogenic molecule prostaglandin E2 (PGE2). Our findings represent a functional model in which a noncoding variant can facilitate long-range chromatin interactions to regulate expression of multiple genes including mRNAs and lncRNAs in a multithreading manner, which provide a more comprehensive picture of gene expression determinants in colorectal cancer as well as novel insights into the underlying biology of colorectal cancer risk loci.

All data for colon adenocarcinoma (COAD) and rectum adenocarcinoma (READ) including genotypes, mRNA expression, noncoding RNA expression, copy number variation, CpG methylation levels, and clinical parameters (age, sex, and tumor stage) were downloaded from the TCGA portal. To increase the power for eQTL discovery, we imputed variants for all colorectal cancer samples using IMPUTE2 with 1000 Genomes Phase 3 as reference (Supplementary Fig. S1; refs. 27, 29). Following criteria were used to exclude SNPs in the imputation process: (i) imputation confidence score, INFO (an index of imputation quality and its value takes between 0 and 1, where value near 1 indicates a SNP has been imputed with high certainty) < 0.4; (ii) minor allele frequency (MAF) < 5%; (iii) SNP missing rate ≥ 5% for best-guessed genotypes at posterior probability ≥0.9; (iv) Hardy–Weinberg equilibrium P value < 1 × 10⁻⁶; (v) mapping to locations on sex chromosome. We downloaded the tag SNPs (N = 103; P ≤ 5 × 10⁻⁸; Supplementary Table S1) located in colorectal cancer GWAS loci from the National Human Genome Research Institute GWAS Catalog and derived their linkage disequilibrium (LD) SNPs (r² ≥ 0.2, European) from the Haploreg database, respectively. Finally, a total of 8,716 SNPs were included for further integrative eQTL analysis.

We conducted an integrative eQTL analysis to evaluate the associations between those variant genotypes and the expression levels of genes (1 Mb distance to the candidate SNP) and adjusted for the effect of copy number variation, CpG methylation levels, population structures (principal components), and clinical parameters (age, sex, and tumor stage) on gene expression levels. The details of the principal components' calculation can be seen in our previous study (30). A FDR P < 0.05 was considered statistically significant. We prioritized the LD block (r² ≥ 0.5) with the most statistically significant eQTL as a candidate for further analysis.

We next performed a functional annotation for eQTLs in the LD block by using multiple bioinformatic tools, including the Haploreg, ANNOVAR, rSNPBase, RegulomeDB, and CistromeDB, which integrate multiple histone modification chromatin immunoprecipitation (ChIP)-seq peaks, transcription factor (TF) ChIP-seq peaks, and DNase hypersensitive site data. Finally, we selected the functional variant with the highest potential in the LD block (r² ≥ 0.5) for further population and experimental validation. The flowchart of the procedures is shown in Supplementary Fig. S1.

A two-stage case–control study was conducted to evaluate the associations between eQTLs and colorectal cancer risk. The phase I contained 1,524 patients with colorectal cancer and 1,522 cancer-free controls recruited from the cancer hospital of Chinese Academy of Medical Sciences in Beijing, China. The phase II consisting of 4,500 cases and 8,500 cancer-free controls were recruited from Tongji Hospital of Huazhong University of Science and Technology (HUST, Wuhan, China). All controls were cancer-free individuals selected from a community nutritional survey in the same region during the same period when patients were recruited and matched to the cases by gender and age (± 5 years; ref. 31). At recruitment, peripheral blood samples and demographic characteristics including age, gender, and smoking and drinking status (Supplementary Table S2) were obtained from the medical records of these individuals. Written informed consent was obtained from each subject, the studies were conducted in accordance with recognized ethical guidelines, and this study was approved by the Chinese Academy of Medical Sciences Cancer Institute and the Institutional Review Board of Tongji Medical College, HUST (Wuhan, China).

Genomic DNA was extracted from peripheral blood samples using the Relax Gene Blood DNA System Kit (Tiangen, China) according to the protocol. SNPs were genotyped using the TaqMan SNP Genotyping system in both stages. Quality control was implemented as described previously (31).

HCT116, SNU-C1, HT115, LoVo, SW480, and CoLo205 cells were obtained from the China Center for Type Culture Collection (Wuhan, China) in June 2017 and were cultured in DMEM supplemented with 10% FBS (Gibco) and 1% antibiotics (100 U/mL penicillin and 0.1 mg/mL streptomycin) at 37°C in a humidified atmosphere of 5% CO₂. All cell lines used in this study were authenticated by short tandem repeat profiling (Applied Biosystems) and tested for the absence of Mycoplasma contamination (MycoAlert); the latest date of test was June 1, 2018.

A total of 1,000-bp DNA fragments surrounding the SNP rs174575 C or G allele were subcloned into the pGL3-promoter vector (Promega). The full-length cDNAs of E2F1, FADS2, and AP002754.2 (ENST00000542121.1) were subcloned into the pcDNA3.1 (+) vector (Invitrogen), respectively. All plasmids were synthesized by Genewiz Biological Technology.

For lentivirus production and transfection, the full-length cDNA of FADS2 and AP002754.2 were subcloned into pLVX-PGK-Puro vector, respectively. Here, the pLVX-PGK-Puro empty vector was used as control. Lentivirus was produced in 293T cells by transfecting target plasmids with X-tremeGENE9 Transfection Reagent (Roche). The Lenti-XTM concentrator was used to concentrate lentivirus and puromycin (2 mg/mL) was chosen for antibiotic selection. The transfection effect was measured by qRT-PCR (Supplementary Fig. S8).

For RNA interference, the siRNA oligonucleotides targeting E2F1, FADS2, AP002754.2 and the nontargeting siRNA control were purchased from RiboBio and transfected using Lipofectamine RNAiMAX (Invitrogen). The siRNA sequences are shown in Supplementary Table S3 and the knockdown effect was determined by qRT-PCR (Supplementary Fig. S8).

Female BALB/c nude mice, ages 4–5 weeks, purchased from the Vital River Laboratory Animal Technology, were allowed to acclimate to local conditions for 1 week and maintained under a 12-hour dark/12-hour light cycle with food and water provided ad libitum. For the xenograft tumor growth assay, nude mice (5 per group) were injected subcutaneously in the rear flank with 0.1 mL of cell suspension containing 2 × 10⁶ cells. When a tumor was palpable, it was measured every 5 days and its volume was calculated according to the following formula volume = 0.5 × length × width².

To evaluate the efficacy of PGE2 on colorectal cancer tumor growth in vivo, we established cohorts of mice-bearing tumor xenografts driven by HCT116 and LoVo cells. After 5 days, the mice were grouped and administered with vehicle or PGE2 (catalog no. 14010, Cayman Chemical) via intraperitoneal injection (300 μg, once daily for 6 weeks), respectively. PGE2 was dissolved in 100% ethanol and diluted to desired concentration with PBS immediately before the injection (32). Tumor volumes were measured every week and calculated using the following formula 0.5 × length × width². All experimental procedures were performed in accordance with the relevant institutional and national guidelines and approved by the Institutional Animal Care and Use Committee of Huazhong University of Science and Technology.

The goodness-of-fit χ² test was used to assess Hardy-Weinberg equilibrium for the genotype distribution of the candidate SNP in the controls. A two-sided Student t test was used to estimate the differences in age between cases and controls, while the differences in gender, age group, and smoking and drinking status were analyzed using Pearson χ² test. For the association analysis, unconditional multivariate logistic regression was employed to estimate ORs and 95% confidence intervals (CI) for the association between candidate SNPs and the colorectal cancer risk, with adjustments for gender, age group, and smoking and drinking status. Multiple genetic models, such as allelic, dominant, recessive and additive genetic model, were applied to assess the genetic susceptibility of variants to colorectal cancer, respectively. P values <0.05 were considered statistically significant. The paired Student t test was used to test gene expression differences between tumor tissues and matched normal tissues. Gene coexpression was tested using Spearman correlation. For functional assays, in the relevant figures, the figure legends denote the statistical details of experiments including the statistical tests used, the numbers of replicates, and the data presentation type. All statistical analyses were performed using the R software (3.30) or SPSS software (21.0).

The data during the current study are available from the corresponding author on reasonable request. Experimental details are available in the Supplementary Methods.

As shown in Fig. 1, the genotype, gene expression, copy number, and methylation data of a total of 328 patients were included for further integrative eQTL analysis. Adjusting for the effects of multiple confounders on gene expression levels, we totally identified 1,923 variants with statistically significant eQTLs in TCGA colorectal cancer tumor tissues (FDR P < 0.05; Fig. 1A; Supplementary Fig. S1). We prioritized the LD block (r² ≥ 0.5) within the most significant eQTL as a candidate for further analysis. The variant rs1535 was identified as the tag SNP in this region from a previous GWAS study (22; Supplementary Fig. S2). However, causal variants in this region have not been investigated. Therefore, we performed a functional annotation for these eQTLs in the LD block (r² ≥ 0.5) using multiple bioinformatic tools, including the Haploreg, ANNOVAR, rSNPBase, RegulomeDB, and CistromeDB, which integrate multiple histone modification and TFs ChIP-seq peaks and DNase hypersensitive site data. Variant with the highest potential to be functional in this LD block was selected as candidate causal one. We found that the SNP rs174575 located in the first intron of FADS2 is enriched within active histone modification peaks (H3K4me1, H3K4me3, and H3K27ac) and open chromatin accessibility (DNase-seq and ATAC-seq peaks; Fig. 1B). Moreover, this variant has statistically significant eQTLs with both FADS2 and lncRNA AP002754.2 expression levels in TCGA colorectal cancer samples (Fig. 1C), which are in line with the results from our own colorectal cancer samples (Fig. 1D) and the characteristics of own colorectal cancer tissues are presented in the Supplementary Table S4. Notably, rs174575 is in LD with the tag SNP rs1535 (r² = 0.67). Hence, here we selected the SNP rs174575 as the functional candidate for further interpretation.

To investigate the association between the SNP rs174575 (MAF is 0.255, 0.166, 0.210, and 0.362 in the European, East Asian, African, and Admixed American populations, respectively, from the 1000 Genomes Phase 3 browser) and colorectal cancer risk, we conducted a two-stage case–control study, consisting of 6,024 cases and 10,022 controls. The demographic characteristics of the study subjects are detailed in Supplementary Table S2. As shown in Table 1, rs174575 confers a significant genetic predisposition to colorectal cancer in both stages, with adjustment for gender, age group, and smoking and drinking status and its MAF is 0.10 in our population (Chinese population). Furthermore, we combined the results from the phase I and phase II, and found that the rs174575[GG] genotype is still associated with an increased risk of colorectal cancer with an OR of 1.26 (95% CI = 1.17–1.36, P = 2.57 × 10⁻⁹ in the additive model). Consistently, statistically significant associations are observed in other genetic models. In addition, we also measured the association between the candidate variant rs174575 and the risk of colorectal cancer, with adjustment for age group and gender and interestingly found that the statistically significant associations are still observed (Supplementary Table S5).

Having demonstrated that rs174575 is associated with colorectal cancer risk and the expression of FADS2 and AP002754.2, we next sought to elucidate the underlying mechanisms. The allele-specific activity of SNPs in regulatory regions might be due to the different binding affinity of TFs. We first conducted TF motif analysis by using multiple databases, including Cistrome and JASPAR, and identified E2F1 motif as a candidate factor that specifically binds to the rs174575[G] allele (Fig. 2A). Moreover, ChIP-seq data from the ENCODE database also indicated that E2F1 maps within the region surrounding SNP rs174575 in colorectal cancer LoVo cells (Fig. 1B).

In addition, we experimentally validated the binding of E2F1 to this region by electrophoretic mobility shift assays (EMSA) and ChIP-qPCR assays. We found that the rs174575[G] allele preferentially binds to nuclear extracts compared with the rs174575[C] allele in colorectal cancer LoVo and HCT116 cells (Fig. 2B). Furthermore, ChIP-qPCR assays in three cell lines with different rs174575 genotypes (HCT116[CG], SNU-C1[CG], and HT115[CC]) also showed that a stronger E2F1 binding is enriched in the rs174575 region in HCT116 and SNU-C1 cells with the G allele than in HT115 cells lacking this allele (Fig. 2C). Intriguingly, these binding signals are significantly attenuated when E2F1 is knocked down. These findings suggested that E2F1 preferentially binds to the risk allele of rs74575 in an allele-specific manner.

E2F1, as a classical TF, plays a crucial role in colorectal cancer tumorigenesis (33). E2F1 is significantly overexpressed in tumor tissues compared with their normal tissues in multiple datasets, such as TCGA, Hong, Notteman, and our own colorectal cancer tissues (Fig. 2D and E; Fig. S3A). In addition, we also detected the baseline levels of E2F1 in HCT116 and LoVo cell line, revealing that the E2F1 is expressed in both cells (Supplementary Fig. S3B). Furthermore, we tested the effect of E2F1 on colorectal cancer cell phenotypes and found that the overexpression of E2F1 is also found to facilitate cell proliferation, whereas the depletion of E2F1 can inhibit cell proliferation in both cells (Fig. 2F; Supplementary Fig. S3C and S3D). This result is consistent with the data from genome-wide CRISPR/Cas9-based loss-of-function screens of CCK-81 and SW403 colorectal cancer cells (34), revealing that E2F1 is essential for cell viability (Fig. 2G and H). Moreover, the expression of E2F1 is positively correlated with the expression of FADS2 and AP002754.2 in both our patients with colorectal cancer and TCGA colorectal cancer samples (Fig. 3A and 3B). Furthermore, when we overexpressed E2F1, expression in three cell lines with different genotypes of rs172575, FADS2, and AP002754.2 expression was increased accordingly in HCT116 and SNU-C1 cells with the rs174575[G] allele in a dose-dependent manner (Fig. 3C and D), but not in HT115 cells lacking this allele (Fig. 3E), which partly indicated the allele-specific regulatory effects of E2F1 on these two target genes. Taken together, these data demonstrated that rs174575 modulates E2F1 binding to affect FADS2 and AP002754.2 expression.

To determine how the risk SNP rs174575 affects the expression of FADS2 and AP002754.2, we first performed luciferase reporter assays and found that the construct containing the rs174575[G] allele exhibits higher enhancer activity than that containing the rs174575[C] allele (Fig. 3F and G). In addition, when we overexpressed E2F1 in HCT116 and LoVo cells at an increasing dose, the differences in luciferase activity between the risk and nonrisk alleles of the SNP are enhanced in a dose-dependent manner (Fig. 3H; Supplementary Fig. S4A). In contrast, the differences between both alleles of rs174575 are significantly attenuated when E2F1 is knocked down (Fig. 3I; Supplementary Fig. S4B), suggesting that rs174575 has an allele-specific enhancer activity for target genes that is modulated by E2F1.

In addition, to further validate the allele-specific enhancer activity for risk SNP rs174575 on target genes FADS2 and AP002754.2, we thus inserted the enhancer-rs174575 into the downstream of the FADS2 or AP002754.2 promoter, respectively, in an enhancer report assay (Fig. 4A). The results showed that the rs174575[G] allele has a higher enhancer activity for the FADS2 or AP002754.2 than the rs174575[C] allele or promoter vector, respectively (Fig. 4B and C). Likewise, the effects are provoked when E2F1 is overexpressed in HCT116 and LoVo cell lines, respectively (Fig. 4D and E; Supplementary Fig. S4C and S4D), indicating that rs174575 acts as an allele-specific enhancer mediated by E2F1 to directly regulate the transcriptional activity of FADS2 and AP002754.2.

The variant rs174575 is located 41.5 Kb downstream of the FADS2 transcriptional start site (TSS) and 178.8 kb upstream of AP002754.2 TSS, respectively. To examine whether there is a direct chromatin interaction between enhancer-rs174575 and FADS2 or AP002754.2 promoter, we first performed allele-specific chromosome conformation capture (3C) assays in multiple colorectal cancer cell lines with different genotypes of the risk SNP (HCT116[CG], SNU-C1[CG], HT115[CC], LoVo[CC], SW480[CC], and CoLo205[CC] cell lines). As illustrated in Fig. 4F, when anchored at the enhancer region containing rs174575, both the FADS2 and AP002754.2 promoters show a stronger interaction with the risk SNP region than any of the other neighboring NcoI sites tested. Notably, the interaction frequency is more statistically significant in cell lines carrying the rs174575[G] allele (HCT116 and SNU-C1 cells) than in other cell lines lacking the risk allele (HT115, LoVo, SW480, and CoLo205 cells, Fig. 4F), suggesting that risk SNP rs174575 can establish allele-specific long-range chromatin loops with FADS2 and AP002754.2 promoters. Moreover, consistent with the above 3C results, the rs174575 enhancer was found to form strong chromatin interactions with the promoters of both FADS2 and AP002754.2 in the rs174575-centered 0.5 Mb window from Hi-C data of small bowel tissue and HCT116 colorectal cancer cells (35, 36; Supplementary Fig. S5A and S5B). Finally, the genotype-specific ChIP-qPCR results also showed that the binding peaks of E2F1 overlap not only the region containing rs174575, but also the FADS2 and AP002754.2 promoters. Consistently, the binding of E2F1 to these three regions are more statistically significant in HCT116 and SNU-C1 cells carrying rs174575[G] allele than in HT115 cells lacking this allele (Fig. 4G). Interestingly, the binding signals are substantially attenuated when E2F1 is knocked down (Fig. 4G), further supporting the conclusion that E2F1 is closely involved in the allele-specific chromatin interaction between rs174575-enhancer and FADS2 or AP002754.2 promoter. Altogether, we demonstrated that the risk allele of SNP rs174575 acts as an allele-specific enhancer to provoke the expression of FADS2 and AP002754.2 by E2F1 mediating long-range enhancer–promoter interactions.

Several lncRNAs have been reported to possess enhancer-like function in regulating the expression of their neighboring protein-coding genes in human cells (37, 38). Interestingly, we also observed a positive correlation between the expression of AP002754.2 and FADS2 in both TCGA cohort and our own patients with colorectal cancer (Fig. 5A). To extend our findings and to determine whether the regulation of neighboring protein-coding gene FADS2 is mediated by lncRNA AP002754.2, we overexpressed and knocked down the expression of AP002754.2 in colorectal cancer cells, and then measured the alteration of FADS2 expression levels. The results showed that the overexpression of AP002754.2 dramatically increases FADS2 expression levels, in a dose-dependent manner (Fig. 5B). Unsurprisingly, the knockdown of AP002754.2 results in a statistically significant reduction for the expression levels of FADS2 (Fig. 5C). Collectively, these results reveal that AP002754.2 can exert a transcriptional activator function regulating the expression of the neighboring protein-coding gene FADS2.

To investigate the potential role of FADS2 and AP002754.2 in colorectal cancer pathogenesis, we first compared the expression of these two genes in tumor and adjacent normal tissues from multiple databases, including our own patients with colorectal cancer, TCGA samples, Oncomine database, and GEO datasets. The results revealed that FADS2 and AP002754.2 are significantly overexpressed in tumor tissues than in normal tissues in our colorectal cancer samples (Fig. 5D and E), which are consistent with the data from other databases (Fig. 5D; Supplementary Fig. S6A). These results provided evidences that the upregulation of FADS2 and AP002754.2 expression may be correlated with the development of colorectal cancer.

We next examined the effect of these two genes on cell phenotypes. The results showed that the overexpression of FADS2 and AP002754.2 substantially increase the cell proliferation rate of HCT116 and LoVo cells, respectively (Fig. 5F; Supplementary Fig. S6B) and whereas the knockdown of FADS2 or AP002754.2 remarkably reduces this effect (Fig. 5G; Supplementary Fig. S6C). Expectedly, the results are similar in the colony formation assays (Fig. 5H and I; Supplementary Fig. S6D and S6E). Interestingly, these two genes display a synergistic effect facilitating colorectal cancer cell proliferation (Fig. 5F–I; Supplementary Fig. S6B–S6E). In accordance with these results, the essential role of FADS2 in cell proliferation is also verified in SW403 and MDST8 colorectal cancer cells (Fig. 5J) from the data of genome-wide CRISPR/Cas9-based loss-of-function screening (34). Furthermore, to assess the potential role of FADS2 and AP002754.2 in vivo, we overexpressed these genes in HCT116 and LoVo cells, respectively, and then injected these cells into the flank of nude mice. The results showed the growth of xenograft tumors with FADS2 or AP002754.2 overexpression is substantially increased, compared with that of control tumors. Interestingly, we found that the group with both FADS2 and AP002754.2 overexpression presents a largest tumor volume among all tested groups (Fig. 6A and B). Altogether, these findings further supported that FADS2 and AP002754.2 can function as potential oncogenes involved in colorectal cancer tumorigenesis.

In light of AP002754.2, which acts its role by modulating the FADS2 expression, here we focused our functional analysis on FADS2. FADS2 encodes delta-6 fatty acid desaturase, which is the key enzyme in the biosynthesis of polyunsaturated fatty acids, including arachidonic acid (39, 40). Interestingly, arachidonic acid is the precursor of PGE2, which has been reported to promote cell growth and metastasis in tumors (41, 42). Therefore, we overexpressed and knocked down the expression of FADS2 in colorectal cancer cell lines and then measured the levels of PGE2 released into the culture medium by ELISA. The results showed that overexpression of FADS2 significantly increases PGE2 levels in colorectal cancer cell lines, whereas the knockdown of FADS2 results in a statistically significant reduction in PGE2 levels (Fig. 6C). Moreover, in consistent with results in colorectal cancer cells, there has also been a statistically significant correlation between FADS2 and PGE2 in serum of 81 colorectal cancer individuals (Supplementary Fig. S7A). Individuals carrying rs174575[G] allele have a higher both FADS2 and PGE2 levels compared with other genotypes (Fig. 1D; Supplementary Fig. S7B).

Furthermore, we tested the effect of PGE2 on colorectal cancer cell phenotypes and found that the treatment of PDE2 results in an increase in colorectal cancer cell proliferation, compared with that of controls (Fig. 6D; Supplementary Fig. S7C). Moreover, to evaluate the efficacy of PGE2 on colorectal cancer tumor growth in vivo, we established cohorts of mice-bearing tumor xenografts driven by colorectal cancer cells. After 5 days, the mice were grouped and administered with vehicle or PGE2 via intraperitoneal injection (300 μg) once daily, respectively (Fig. 6E). In accordance with results found in cell lines, we also observed that the treatment of PGE2 could significantly increase tumor growth, compared with the treatment of vehicle (Fig. 6F and G). The animal experiments described in Fig. 6 were conducted in accordance with relevant institutional and national guidelines and approved by the Institutional Animal Care and Use Committee of Huazhong University of Science and Technology. Taken together, these findings raised the possibility that FADS2 can facilitate colorectal cancer development by promoting the metabolism of PGE2.

So far, large genome-mapping consortia and GWASs have identified more than 100 genomic loci associated with colorectal cancer susceptibility, but only a handful have been conclusively ascribed to a specific causal variant with direct insight into the underlying disease biology. Identifying the precise gene targets for these noncoding variants and interrogating the biological mechanisms underlying associations remain formidable challenges. In this study, through multidisciplinary strategies comprised of an integrative eQTL-based approach, a large-scale population study and a series of biochemical experiments, we identified a functional variant rs174575 at 11q12.2 that acts as an allele-specific enhancer to distally regulate FADS2 and lncRNA AP002754.2 expression by long-range enhancer–promoter interactions, thus resulting in an increased risk of colorectal cancer. Mechanistically, AP002754.2 further exerts a transcriptional activator that promotes FADS2 expression. Furthermore, FADS2 and AP002754.2 display a synergistic effect facilitating colorectal cancer cell proliferation by increasing the metabolism of PGE2, which is closely involved in colorectal cancer tumorigenesis.

The discovery of target genes and their potentially oncogenic network of GWAS-identified colorectal cancer loci by eQTL is an important approach to dissect the etiology of cancer and to develop new strategies for cancer prevention and treatment, especially based on tumor tissues, which might have unique advantages. Ongen and colleagues revealed that colorectal cancer tumor-specific eQTLs are more enriched for low GWAS P values than eQTLs in normal tissues (43). Although the effects of genetic and epigenetic factors, such as methylation and somatic copy number alterations, on gene expression levels may be challenges in most eQTL studies based on tumors (13, 44, 45), a systematically practical solution was leveraged to detect the target genes adjusting for these confounders in our study. In addition, we combined GWAS within TCGA to investigate the germline determinants of gene expression in colorectal cancer tumors, not only focusing on the mRNA expression, but also addressing the lncRNA expression, which have been implicated in various physiologic and pathologic processes of diseases (28). Here, we performed an integrative eQTL-based analysis in colorectal cancer tumor tissues and found that a variant rs174575 at 11q12.2 is associated with the expression of FADS2 and AP002754.2 with the strongest statistical significance.

In addition, our study provides key mechanistic insight by which a functional noncoding variant may affect the expression of multiple target genes, in a multithreading manner, which is consistent with a previous observation (46). Enhancer–promoter interaction, one of the transcriptional regulation mechanisms, has been illustrated to be a general feature of gene control and essential for tumor pathologic activation (47). In the present study, we found that there have significant interactions between rs174575-enhancer region and the promoter regions of both FADS2 and AP002754.2 in an allele-specific manner, suggesting that the risk allele of rs174575 can facilitate long-range enhancer–promoter interactions to promote FADS2 and AP002754.2 expression. This finding is in line with the luciferase reporter assays and eQTL results of rs174575 with these two target genes, revealing that rs174575[GG] genotypes has higher transcriptional activity or expression levels for FADS2 and AP002754.2 than that of other genotypes. Notably, a looped genomic interaction is mediated by some TFs, which facilitate the folding of the 3D genome and bring distal regulatory elements and promoters into proximity. E2F1 is an important TF, that recruits the RNA polymerase II cofactor to mediate long-range loop formation, which is closely involved in cell growth, differentiation, and apoptosis of tumors (48, 49). Here we interestingly found that E2F1 is overexpressed in colorectal cancer tumor tissues derived from multiple databases, and binds to the rs174575[G] allele to bring the rs174575 enhancer element to the promoters of FADS2 and AP002754.2, thereby elevating the expression of these two target genes. Taken together, these results elucidated that the risk SNP rs174575, as an allele-specific enhancer, facilitates long-range enhancer–promoter interactions mediated by E2F1 to regulate expression of multiple genes, which provides functional evidence to support our population finding that rs174575[G] is significantly associated with the increased risk of colorectal cancer.

Our study implicates FADS2 and AP002754.2 as the target genes for the risk SNP rs174575 located at locus 11q12.2. FADS2 encodes delta-6 fatty acid desaturase, which is the key enzyme in the biosynthesis of polyunsaturated fatty acids (PUFA). Of these PUFAs, arachidonic acid is the precursor of PGE2, which has been reported to facilitate colorectal cancer tumorigenesis via the G_s-axin–β-catenin axis (41) and also validated to promote colorectal cancer cell proliferation in vitro and in vivo in this study. Furthermore, we revealed that FADS2 is overexpressed in colorectal cancer tumor tissues compared with their paired normal tissues in multiple independent cohorts and is more highly expressed in advanced stages of colorectal cancer. Mechanistically, FADS2 overexpression could provoke colorectal cancer cell proliferation and xenograft tumor growth by facilitating the metabolism of the oncogenic molecule PGE2, whereas a reduction in FADS2 expression significantly attenuated this effect. In terms of another target AP002754.2, we also found that it is overexpressed in tumor tissues and advanced stages of colorectal cancer, and functions as a potential oncogene promoting colorectal cancer cell proliferation and tumor growth. In addition, we interestingly observed that AP002754.2 further exerts a transcriptional activator that promotes FADS2 expression, which partly accounts for this finding that these two genes display a synergistic effect facilitating colorectal cancer cell proliferation. Our finding is in accordance with recent studies (24, 37, 38), supporting an unanticipated role for a class of lncRNAs as critical regulators modulating the expression of their neighboring protein-coding genes in human cells. Future investigations are warranted to elucidate the precise mechanisms between these two target genes. Collectively, these findings provide strong evidence that FADS2 and AP002754.2 paly crucial roles in colorectal cancer tumorigenesis.

Our study also has limitations. First, our risk models can be further improved. Many environmental and lifestyle factors such as diet, obesity, physical activity, nonsteroidal anti-inflammatory drugs, postmenopausal hormone use, smoking, and alcohol have been shown associated with the colorectal cancer risk (50) and should be adjusted for further improving risk determination. Second, eQTLs performed in both tumor tissue and normal tissue were not systematically interrogated, especially in normal–tumor matched samples. Finally, advanced in vivo experiments, such as patient-derived xenograft or patient-derived organoid models, are encouraged to investigate the effect of target genes on colorectal cancer development.

In conclusion, through integrating an eQTL-based approach, epidemiologic studies and a series of biochemical experiments, our findings provide an oncogenic regulatory circuit by which a functional noncoding variant rs174575 facilitates long-range enhancer–promoter interactions mediated by E2F1 to promote the expression of FADS2 and AP002754.2, which reciprocally exerts a transcriptional activator promoting FADS2 expression. Multipronged approaches comprised of in vitro and in vivo experimentation utilized in this study aimed to provide evidences in depth to explain the population association between the functional SNP rs174575 and colorectal cancer susceptibility.

No potential conflicts of interest were disclosed.

Conception and design: X. Miao

Development of methodology: J. Tian, J. Lou, M. Rao

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Tian, J. Lou, M. Rao, Z. Lu, Y. Zhu, D. Zou, X. Peng, H. Wang, M. Zhang, S. Niu, Y. Li

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Tian, Y. Cai

Writing, review, and/or revision of the manuscript: J. Tian, J. Chang, X. Miao

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): R. Zhong, X. Miao

Study supervision: J. Chang, X. Miao

We are grateful to all the study participants, research staff, and students who participated in this work. This work was supported by the National Science Fund for Distinguished Young Scholars (NSFC-81925032), the National Key Research and Development Plan Program (2016YFC1302702, 2016YFC1302703), the National Program for Support of Top-Notch Young Professionals, and the Program for HUST Academic Frontier Youth Team for X. Miao, and the Fundamental Research Funds for the Central Universities, HUST (2018JYCXJJ013 to J. Tian).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.