Integrated Multi-Omics Analysis of Colorectal Cancer Reveals Mir-20a as a Regulator of Fatty Acid Metabolism in Consensus Molecular Subtype 3

microRNAs (miRNAs) serve important roles in metabolism. The consensus molecular subtype (CMS) 3 of colorectal cancer (CRC) is characterized by activated fatty acid (FA) metabolism. We aimed to identify essential miRNAs of CMS3-CRC and analyze the regulatory role in the FA metabolism. The RobustRankAggreg method by integrating multi-omics data including genome, epigenome, transcriptome and interactome, was applied to lter out functional genes (Fgenes). The backward derivation approach based on Fgenes and miRNA-gene interactions was further applied to identify functional miRNAs (Fmirs). Nine human CRC cell lines with different CMSs were investigated. RT-qPCR, western blotting and immunouorescence were performed to examine the effect of miR-20a on FA synthesis and Wnt/β-catenin signaling. The effect of miR-20a on cell proliferation and metastasis were studied by clone-formation, EdU assay, wound healing and transwell assay. We identied 12 Fmirs by integrating multi-omics features in CMS3-CRC. These Fmirs exhibited signicantly enriched CRC driver miRNAs and signicant impacts on CMS3-CRC cell growth. Beyond the ndings, miR-20a was signicantly correlated with Wnt/β-catenin signaling and participated in FA metabolism subpathway. In vitro assays combined with bioinformatics analyses demonstrated that elevated miR-20a up-regulated FA synthesis enzymes FASN, ACAC and ACLY via Wnt/β-catenin signaling, and nally promoted proliferative and migration of CMS3-CRC cells.

expression values on the log2 scale. For RNAseq data, we used log2 transformed FPKM matrix after adding 1 to avoid unde ned values.

Classi cation of colorectal cancer cell lines
The gene expression data of 57 large intestine cell lines were download from the Cancer Cell Line Encyclopedia (CCLE) database (https://portals.broadinstitute.org/ccle). The CMScaller algorithm was used to assign CMSs to CRC cell lines (https://github.com/Lothelab/CMScaller). CMS cell lines were also supplemented with the reported results [15].
PPI and TF-gene interactions in the context of CMS3-CRC were ltered based on their expression correlation measured by the Pearson correlation analysis. We only retained the signi cantly co-expressed interactions with the threshold of adjusted P < 0.05 and |ρ|>0.3. "ρ" represents the Pearson correlation coe cient. miRNA-mediated gene regulation miRNA-gene interactions were all experimentally validated and collected from public databases, including TarBase[18], miRTarBase [19], and miRecords [20]. miRNA-gene regulations in the context of CMS3-CRC were ltered based on their expression correlation measured by the Pearson correlation analysis with the threshold of P < 0.05 and |ρ|>0.3.
We note that these target genes of miRNAs were derived from TCGA data. To avoid biased results, we took the target genes as a gene signature for each miRNA and performed an evaluation step (sigQC) to ensure that the signature used has suitable properties across different datasets. Here, we present radar plots summarizing the gene signature quality control metrics implemented by the R package sigQC [21].

Identi cation of functional miRNAs
We identi ed Fmirs by rstly identifying Fgenes. Based on the individual-level differentially expressed genes calculated by RankComp algorithm [22]. We identi ed genes with signi cantly upregulated or downregulated patterns and de ned them as CMS-related genes. Then, we proposed a computational method to prioritize CMS-related genes and identify Fgenes in CMS3-CRC by integrating multi-omics features [13]. Brie y, for each CMS-related gene in CMS3-CRC, we calculated the scores of seven features that characterized the genomic and transcriptomic activity, regulation by co-expressed TFs, and central roles in PPIN of gene regulators. The aggregated score for each gene was further calculated by integrating the seven separate ranks in CMS3-CRC, using the robust rank aggregation method in R package RobustRankAggreg [23]. Then, we utilized the most effective feature and identi ed top-ranked 100 genes as the Fgenes.
For miRNA, we directly analyzed the 1,881 miRNAs measured by the TCGA-COAD and TCGA-READ project. We identi ed Fmirs using a backward derivation approach. Firstly, we identi ed signi cantly coexpressed miRNA and genes for CMS3-CRC among the collected miRNA-gene interactions. Then, miRNAs whose target genes were signi cantly intersected with Fgenes were de ned as Fmirs. The hypergeometric test was used to calculate the signi cance as follows: Where N is the number of target genes that overlap with all detected genes in the expression pro les, F is the number of target genes that overlap with Fgenes, T i is the number of target genes of miRNA i, and F i is the number of target genes for miRNA i that overlap with Fgenes.
Manually curation of CRC driver miRNAs A panel of oncogenic and tumor-suppressive miRNAs was collected from the public databases miRCancer [24] and OncomiRDB [25]. Brie y, the microRNA-cancer association in miRCancer was extracted using the text mining algorithm against PubMed based on 75 rules. The entries of CRC-related miRNA regulations in OncomiRDB were manually curated from abstracts with direct experimental evidence. Finally, we retained miRNAs associated with CRC and collected 250 CRC-related miRNAs.

Subpathway analysis
The R package "Subpathway-GMir" was performed to identify the metabolism subpathways mediated by miRNAs [26]. We mapped the Fmirs and signi cantly co-expressed target genes as nodes into the reconstructed metabolism pathway graphs obtained from the package. Then, we identi ed miRNAmediated FA metabolism subpathways based on the "lenient distance" similarity method. It evaluated the signi cance of candidate subpathways using the hypergeometric method. The miRNA-target interactions for CMS3-CRC were collected as described above.
All media for the human cell lines were supplemented with 10% FBS and 1% penicillin/streptomycin antibiotics. SW48, SW948 and SW480 cell lines were cultured in 100% air at 37 °C. Other cell lines were cultured under 95% air-5% CO 2 at 37 °C. All cell lines were validated by STR DNA nger-printing.
Experiments were carried out within 6 months after acquisition of the cell lines. In addition, mycoplasma contamination was ruled out using a PCR-based method.
HT29 and LS174T cells were transfected with miR-20a mimic and negative control (NC) or miR-20a inhibitor and inhibitor NC (inNC) respectively using Lipofectamine 2000 (Invitrogen, USA). After 48 h transfection, these cells were harvested for further experiments.

Western Blotting
The cells were washed with ice-cold PBS and lysed with RIPA buffer containing protease inhibitors. After mixed with SDS sample buffer, the proteins were heated to 99 °C for 10 min, separated on 10% SDSpolyacrylamide gels, transferred to PVDF mem-branes, and probed with antibodies against β-catenin (1:1000, CST, USA) and β-actin (1:3000, Proteintech, China) at 4 °C overnight and blots were visualized using gel imaging systems (Bio-Rad). ECL was used to visualize the immunoblot signals.

Real Time-quantitative PCR (RT-qPCR)
Total RNA of HT29 and LS174T cells were extracted using total RNA extraction kit (Thermo Fisher Scienti c, Waltham, USA). 1 μg RNA was used for reverse transcription into complementary DNA (cDNA) using the PrimeScript RT-PCR Kit (Takara, Tokyo, Japan). Real-time polymerase chain reaction (RT-PCR) was conducted using the SYBR Premix Ex-Taq II kit (Takara, Tokyo, Japan) on the Quant Studio 3 Real-Time PCR System (Applied Biosystems, CA, USA). The fold changes were determined using the 2 -△△CT method. All primers used were as follows: Quantitative-FASN, forward 5'-AAGGACCTGTCTAGGTTTGATGC-3', reverse 5'-TGGCTTCATAGGTGACTTCCA-3', Quantitative-ACACA, Clone-formation assay, EdU assay, Transwell assay and Wound-healing assay For the clone formation assay, HT29 and LS174T cell lines were transfected with mimic/NC (inhibitor/inNC) on day 1. Digest the cells the next day and seed 200 cells into a six-well plate, followed by the addition of 2 ml of 10% FBS DMEM for 14 days. After 14 days, the cells were xed in 4% paraformaldehyde and stained with crystal violet for 30 min at room temperature. Colonies consisting of > 50 cells were counted. For EdU assay, cells were plated in 12-well plates and performed transfection. After 48h, 5-ethynyl-2′-deoxyuridine (EdU) assay (BeyoClick™ EdU-488 Kit, Beyotime, Shanghai, China) were performed to analyze the cell proliferation. The cells were incubated with 10μM EdU solution for 2 h and xed with 4% paraformaldehyde. And the cells were washed with PBS for 3 times and 0.5% TritonX-100 once. Then, the cells were stained with 100 μL 1 × DAPI solution. After washed with 100μL PBS for 3 times, images were obtained from uorescence microscope for further calculation of proliferation rates.
Transwell assays were conducted using an 8 μm pore Transwell lter (Corning, US). HT29 or LS174T cells were seeded in the upper chamber with serum-free medium. And medium with 10% FBS was added to the lower chamber. After incubation for 48 h, the Transwell lter was washed, xed with 4% paraformaldehyde and in turn stained with 0.1% crystal violet staining solution.
For wound-healing assay, HT29 or LS174T cells were seeded on six-well plates. Then, a wound across the wall was introduced. After washing with PBS, the plate was incubated with serum-free medium for 24 h. Cell migration was observed and photographed by microscopy.

Immunofluorescence analysis
The cells were seeded on poly-lysine-coated coverslips for 24 h, xed with 4% paraformaldehyde for 15 min, and permeabilized in 1% Triton X-100 for 10 min. The cells were then blocked with 5% BSA in PBS for 30 min at room temperature and incubated in antibodies overnight at 4°C. The coverslips were washed three times with PBS buffer, and the secondary antibody was applied for 1 h at 37°C. The coverslips were then mounted onto glass slides using an antifade mounting medium (Beyotime, China). Cells were then examined under a laser scanning confocal microscope (Zeiss LSM 510 Meta).

Identi cation of functional miRNAs in CMS3-CRC tumors
In order to nd out the functional miRNAs (Fmirs), we initially identi ed Fgenes in CMS3-CRC by integrating multi-omics features. We rstly summarized CMS3-reglated genes that were signi cantly frequently up-or down-regulated in CMS3-CRC tumors from the individual-level genes identi ed by RankComp algorithm (Cumulative binomial model, adjusted P < 0.05). Fgenes were de ned as the topranked 100 CMS3-related genes with highest aggregated multi-omics scores (Additional le 2: Table S2). For CMS3-CRC cell line HT29, Fgenes showed signi cantly lower dependency scores than other genes using RNAi data ( Figure 1A), indicating their critical roles for cancer cell growth. Besides, Fgenes were signi cantly enriched with tumor driver genes identi ed by TUSON explorer [27] (Additional le 3: Figure  S1A), and can separate CMS3-CRC tumors from normal samples (Additional le 3: Figure S1A). All results suggested that Fgenes were involved in the essential function of CMS3-CRC tumors.
Then, we identi ed signi cantly correlated miRNA-gene interactions in CMS3-CRC using paired miRNA and mRNA expression data from TCGA (Pearson correlation analysis, P < 0.05 and |ρ|>0.3). Taking the signi cant interacting genes as a gene set signature for each miRNA, we found that CRC datasets are comparable in quality for the application of the gene signature using sigQC ( Figure 1B, taken miR-20a as an example). Thus, we identi ed Fmirs based on the identi ed miRNA-gene interactions using TCGA data.
The target genes of 12 miRNAs were enriched with Fgenes and these miRNAs were de ned as Fmirs (Hypergeometric test, P < 0.05), 11 of which were signi cantly deregulated in CMS3-CRC compared to normal samples. Hierarchical clustering analysis found that Fmirs can separate CMS3-CRC tumors from normal samples ( Figure 1C). Moreover, Fmirs were signi cantly enriched with driver miRNA of CRC collected from miRCancer and OncomiRDB databases (Hypergeometric test, P < 0.05, red in Figure 1C). Besides, targets of several Fmirs such as miR-363, miR-20a and miR-148a, were signi cantly overrepresented at the top of gene list in ascending order in the RNAi and CRISPR screened HT29 cell lines (GSEA, P < 0.05, Figure 1D), indicating their critical roles for cancer cell growth. miR-20a promoted FA synthesis metabolism in CMS3-CRC cells Accumulating evidence suggests that alterations in lipid metabolism represented by FA metabolism, contribute to overall metabolic reprogramming in cancer cells [28,29]. CMS3-CRC had signi cantly higher enrichment score (ES) in FA metabolism pathway than the other subtypes ( Figure 2A). Then, we identi ed FA metabolism subpathway mediated by Fmirs in CMS3-CRC. The subpathway contained 34 consecutive differential molecules, including six miRNAs and 28 genes ( Figure 2B). For the six miRNAs, all except miR-363 showed signi cantly higher expression in CMS3-CRC tumors compared with normal samples. Particularly, miR-148a and miR-20a were also signi cantly over-expressed in CMS3-CRC than at least two CMSs ( Figure 2C). miR-148a is associated with lipid metabolism in the liver [30], and its target Fgene MMP7 participants in esophageal cancer metabolism [31]. The target Fgene FOXQ1 of miR-20a can regulate liver metabolism [32]. Thus, miR-148a and miR-20a may be essential for FA metabolism in CMS3-CRC and were selected for further analysis.
Using CMScaller algorithm, we assigned CMSs to CRC cell lines. The expression of miR-148a and miR-20a was measured by RT-qPCR in the four CMSs cell lines and human colon epithelial cell line ( Figure  2D). It was demonstrated that miR-20a was upregulated in CMS3-CRC cell lines HT29 and LS174T, when compared with colon epithelial cell line NCM460, which is consistent with the results identi ed in tissues.
Moreover, miR-20a was also upregulated in CMS3-CRC cell lines in comparison with the other subtypes. The results were not observed for miR-148a (Additional le 4: Figure S2). To explore whether the miR-20a was involved in abnormal FA metabolism in CMS3-CRC tumors, we rst investigated miRWalk and TargetScan database, and found potential complementary base pairing between miR-20a and key FA metabolism enzymes including FASN, ACAC and ACLY. The three genes were responsible for de novo FA synthesis and over-expressed in CMS3-CRC.
The HT29 and LS174T cells were transfected with miR-20a mimic, miR-20a inhibitor or NC to analyze multiple FA synthesis related genes, including FASN, ACAC and ACLY. We observed that transfected miR-20a mimic into HT29 and LS174T cell lines upregulated the expression of miR-20a ( Figure 3A). Simultaneously, the expression of FASN, ACAC and ACLY were increased in response to miR-20a mimic versus NC ( Figure 3B). The subsequent loss-of-function experiments revealed that HT29 and LS174T cells transfected with miR-20a inhibitor had diminished miR-20a, FASN, ACAC and ACLY expressions compared with inNC groups (Figure 3C-D). These results indicated that upregulation of miR-20a promoted FA synthesis in CMS3-CRC tumors. miR-20a promotes FA synthesis via Wnt signaling pathway Activation of the Wnt signaling pathway led to lipid accumulation in cells [33]. The target Fgene FOXQ1 of miR-20a, which can regulate liver metabolism [32], is a marker for activation of Wnt signaling in solid tumors [34]. Thus, we inferred that miR-20a promoted FA synthesis via Wnt signaling pathway in CMS3-CRC tumors. Using ssGSEA, we calculated the ES of Wnt gene sets listed in MSigDB. The expression of miR-20a was signi cantly positively correlated with canonical Wnt signaling pathway (Pearson correlation analysis, P = 0.0108, ρ = 0.2766), but negatively correlated with non-canonical Wnt signaling pathway (Pearson correlation analysis, P = 0.0176, ρ = -0.2585). The canonical Wnt signaling pathway causes an accumulation of β-catenin in the cytoplasm and induce a cellular response. Therefore, we sought to determine whether Wnt participated in miR-20a induced FA synthesis through exploring the effects on β-catenin expression.
After transfecting miR-20a mimic, inhibitor, and paired NC, the expression of β-catenin was detected by western blotting (WB) and immuno uorescence (IF) in HT29 and LS174T cell lines. As expected, WB analysis showed that miR-20a mimic-treated cells expressed higher β-catenin compared with NC groups ( Figure 4A), whereas lower β-catenin was detected in miR-20a inhibitor-treated cells ( Figure 4B). We obtained similar results in IF staining assays ( Figure 4C). miR-20a mimic-treated HT29 cells showed signi cantly higher β-catenin positive staining counts than the NC group. In contrast, miR-20a inhibitortreated HT29 cells showed a reversed phenotype. The results suggest that miR-20a promotes Wnt/βcatenin level in CMS3-CRC cell lines. Previous studies have reported that the activated Wnt signaling pathway can up-regulate the activity of FA synthesis in tumor. Therefore, we used LiCl (10mmol/L) to treat HT29 and LS174T cells to activate Wnt signaling pathway and analyze the expression of FA-related genes. The results showed that activating the Wnt signaling pathway can signi cantly promote the upregulation of FA metabolism genes (FASN, ACAC and ACLY) in CMS3-CRC cell lines ( Figure 4D).

Elevated miR-20a promoted proliferation and migration of CMS3-CRC
Combined with the report that elevated FA synthesis promotes cancer cell proliferation and migration, we took on experiments to investigate miR-20a function during CRC tumorigenesis. GSEA was run to explore the potential role of miR-20a in CMS3 and showed that miR-20a played an important role in promoting cell cycle and DNA replicant (adjusted P < 0.05, Figure 5A). Besides, expression of miR-20a was signi cantly correlated with the proliferation marker MKI67 (Pearson correlation analysis, P = 1.72E-03, ρ = 0.34, Figure 5B). The effects of miR-20a on cell proliferation of HT29 and LS174T cells were examined using clone-formation assay. The assay showed that elevated miR-20a expression promoted the clone formation ability of HT29 cells compared with NC groups. In contrast, miR-20a inhibitor-treated HT29 cells showed a reversed phenotype. We repeated these experiments in LS174T cells and received coherent outcomes ( Figure 5C). EdU assay showed that miR-20a promoted proliferation of HT29 and LS174T cells compared to NC groups ( Figure 5D). While inhibition of miR-20a will reduce cell proliferation, suggesting that miR-20a had a critical role in CMS3-CRC tumor proliferation.
We performed wound healing assay and transwell assay to determine the effect of miR-20a on the migration ability of HT29 cell line. In the wound healing assay, cell motility was monitored at designated time points after scratches. The miR-20a overexpressed cells migrated toward the wound more quickly than the NC groups ( Figure 6A). In contrast, miR-20a inhibitor-treated HT29 cells showed a reversed phenotype ( Figure 6B). We repeated these experiments in LS174T cells and received coherent outcomes, suggesting that miR-20a had a unanimous pivotal role in CMS3-CRC oncogenesis, that is, inducing tumor migration in vitro ( Figure 6C-D). A cell migration assay in a transwell system was performed to further assess the effect of miR-20a on HT29 cell migration. Our results showed that transfection with miR-20a mimic in HT29 cells effectively increased the migratory abilities ( Figure 6E). In contrast, the migratory abilities exerted reversed trend in miR-20a inhibitor-treated HT29 cells.

Discussion
In recent years, the CMSs, which are emerging as critical factor for prognosis and treatment of CRC, have attracted wide attention. We have identi ed a core set of FGRs for each CMS by integrating multi-omics data. FGRs played critical roles in regulating immune microenvironment of CMSs. miRNAs are also gene regulators serving an important role in tumorigenic processes. Thus, we identi ed 12 Fmirs associated with CMS3-CRC based on Fgenes integrating from multi-omics features. Fmirs played a critical role in CMS3-CRC cell growth and FA metabolism pathway.
Consistent with our nding, the Fmir miR-20a has been documented to be highly expressed in CRC tissues versus normal mucosal tissues [35], and this upregulation in CRC has also been con rmed in serum and plasma [36]. Beyond these ndings, we demonstrated that miR-20a promoted FA synthesis in CMS3-CRC. Activation of the Wnt signaling pathway led to lipid accumulation in cells [33]. We further found that miR-20a upregulated Wnt/β-catenin signaling by systematically coupling bioinformatics analyses and in vitro experiments. β-catenin is a major component of cell-cell adhesion structures and functions as a controller of cell migration, colony formation and stem cell properties through translocation into nucleus [37,38]. Aberrant β-catenin accumulation in the cytoplasm usually translocates to the nucleus and was associated with tumor relapse and metastasis in breast cancer patients [39]. The underlying mechanism needs to be further explored. Finally, we found that proliferation and migration abilities of the miR-20a overexpressed cells were signi cantly promoted, whereas the proliferation and migration abilities were lower in the miR-20a inhibitor groups.

Conclusions
In conclusion, miR-20a could promote the proliferative and migration in CMS3-CRC via regulating fatty acid metabolism and Wnt signaling pathway, highlighting novel mechanisms associated with CMS3-CRC cancerogenesis.

Declarations
Ethics approval and consent to participate Not applicable.

Consent for publication
Not applicable.

Availability of data and materials
The datasets analyzed during the current study are available in the databases TCGA and GEO.

Competing interests
The authors declare that they have no competing interests.