Molecular mechanisms of pathogenesis in hepatocellular carcinoma revealed by RNA-sequencing

Abstract

The present study aimed to explore the underlying molecular mechanisms of hepatocellular carcinoma (HCC). RNA-sequencing profiles GSM629264 and GSM629265, from the GSE25599 data set, were downloaded from the Gene Expression Omnibus database and processed by quality evaluation. GSM629264 and GSM629265 were from HCC and adjacent non-cancerous tissues, respectively. TopHat software was used for alignment analysis, followed by the detection of novel splicing sites. In addition, the Cufflinks software package was used to analyze gene expressions, and the Cuffdiff program was used to screen for differently expressed genes (DEGs) and differentially expressed splicing variants. Gene ontology functional enrichment and Kyoto Encyclopedia of Genes and Genomes pathway enrichment analyses of DEGs were also performed. Transcription factors (TFs) and microRNAs (miRNAs) that regulate DEGs were identified, and a protein-protein interaction (PPI) network was constructed. The hub node in the PPI network was obtained, and the TFs and miRNAs that regulated the hub node were further predicted. The quality of the sequencing data met the standards for analysis, and the clean reads were ~65%. Most sequencing reads mapped into coding sequence exons (CDS_exons), whereas other reads mapped into exon 3′ untranslated regions (UTR_Exons), 5′UTR_Exons and Introns. Upregulated and downregulated DEGs between HCC and adjacent non-cancerous tissues were screened. Genes of differentially expressed splicing variants were identified, including vesicle-associated membrane protein 4, phosphatidylinositol glycan anchor biosynthesis class C, protein disulfide isomerase family A member 4 and growth arrest specific 5. Screened DEGs were enriched in the complement pathway. In the PPI network, ubiquitin C (UBC) was the hub node. UBC was predicted to be regulated by several TFs, including specificity protein 1 (SP1), FBJ murine osteosarcoma viral oncogene homolog (FOS), proto-oncogene c-JUN (JUN), FOS-like antigen 2 (FOSL2) and SWI/SNF-related, matrix-associated, actin-dependent regulator of chromatin, subfamily A, member 4 (SMARCA4), and several miRNAs, including miR-30 and miR-181. Results from the present study demonstrated that UBC, SP1, FOS, JUN, FOSL2, SMARCA4, miR-30 and miR-181 may participate in the development of HCC.

Introduction

Hepatocellular carcinoma (HCC), also called malignant hepatoma, often presents with several symptoms, including bloating (resulting from fluid in the abdomen), loss of appetite, easy bruising (owing to blood clotting abnormalities) and feeling tired (1). The main causes of HCC consist of cirrhosis, which is commonly induced by alcoholism and viral hepatitis (2). In addition, chronic hepatitis B/C infection may induce the immune system to attack liver cells, further accelerating the development of HCC (3). Liver transplantation is a good option for patients with HCC; however, patients may wait for a long period of time before a suitable donor is identified (4). The 5-year overall survival rate of patients with HCC remains low, and the disease is difficult to overcome (5). It is imperative to investigate the molecular mechanisms of HCC to develop new methods for the prevention and therapy of HCC.

An increasing number of studies on the genes, miRNAs and the underlying molecular mechanisms of HCC have greatly advanced understanding in this field. Tissue factor pathway inhibitor 2 (TFPI2) has been revealed to be a crucial tumor suppressor gene that regulates tumor growth and metastasis and is silenced in HCC (6). In male patients, a phenomenon has been identified that indicates that the androgen pathway may be able to activate microRNA (miRNA) miR-216a in a ligand-dependent manner; a process that may also be enhanced by hepatitis B virus X protein (7). A previous study demonstrated that paired box 5 is a functional tumor suppressor in HCC and activates p53 and p21 signaling (8). Although previous studies have provided important insights concerning the molecular mechanisms of HCC, the understanding of HCC is still lacking.

The present study aimed to explore the underlying molecular mechanisms of the genes and miRNAs involved in HCC, to advance our understanding of this process and to improve clinical treatments. The RNA-sequencing (RNA-seq) data set GSE25599 was downloaded from the Gene Expression Omnibus (GEO) database (http://www.ncbi.nlm.nih.gov/geo) and used for alignment analysis, followed by subsequent correlational analysis, screening for differently expressed genes (DEGs) and differentially expressed splicing variants, and cluster analysis of the identified DEGs. In addition, gene ontology (GO) function and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway-enrichment analyses were conducted for the DEGs, followed by predictions of transcription factor (TF) and miRNA target genes, and the construction of a protein-protein interaction (PPI) network.

Materials and methods

Sequence data

RNA-seq data set GSE25599 was acquired from the GEO database (9), and single-end sequencing was performed using the GPL9052 Illumina Genome Analyzer platform (Illumina, Inc., San Diego, CA, USA). Included in the GSE25599 data set were the following two profiles: GSM629264 (3 runs: SRR074999, SRR075000 and SRR075001) and GSM629265 (3 runs: SRR075002, SRR07500 and SRR075004), which were derived from HCC tissue and adjacent non-cancerous tissue, respectively.

Quality evaluation of sequencing data

The quality of the sequencing data was determined using statistical approaches, as described below, to analyze base distribution and quality fluctuation of each circle of sequencing reads.

Detection of base quality

The Phred Quality (Q) scores are related to sequencing error rate and are also affected by various other factors, including sequencer type, sequencing reagents and samples. Q scores were calculated using the formula: Q=−10log₂E, where, E is the sequencing error rate.

Detection of guanine-cytosine (GC) content distribution

RNA-seq analysis was conducted following the principle of random fragmentation (10). To ensure that the sequencing depth was relatively homogeneous, the GC content should be equivalent for each sequencing cycle.

Filtering of sequencing data

Analytical accuracy was improved by filtering the ‘dirty’ raw reads with the following steps: i) Remove adaptor sequences from reads; ii) remove reads with >20% ‘N’ bases (where N is any base A, T, C or G); ii) remove low-quality reads; that is, whole reads with Q-score ≤20 in which >50% of bases are N. By using this method, clean reads were obtained and used for subsequent analyses.

Sequence alignment

TopHat version 2.0.11 (http://ccb.jhu.edu/software/tophat/index.shtml) was used to align clean reads to the human reference genome assembly GRCh37 release 66 (Homo_sapiens.GRCh37.66), which was obtained from the Ensembl Genome Browser database (ENSG00000169857; http://www.ensembl.org/index.html). TopHat runtime parameters were set as follows: ‘-G’=Homo_sapiens.GRCh37.66.gtf;‘-segment-length’=20;‘-read-realign-edit-dist’=0 and ‘-no-coverage-search’. The remaining parameters were set to default. Alignment to the reference genome was subsequently performed to identify the origin of the sequence read.

Assessment of gene expression

Detection of gene expression levels

Gene expression values were analyzed by Cufflinks software version 1.2.1 (http://cufflinks.cbcb.umd.edu). Based on the sequence alignment results of among the different groups, the reads per kilobase of transcript per million mapped reads (RPKM) value was calculated to assess the expression quantity using the following formula:

$$\text{RPKM}=\frac{\text{total exon reads}}{\text{mapped reads (millions) x exon length (kb)}}$$

Correlational analysis

The correlation of gene expression levels between samples was calculated by Pearson's correlation coefficient (r). The calculation of r between two variables, × and y, is defined as the covariance of the two variables divided by the product of their standard deviations (11):

$$\mathrm{r}=\frac{{\sum }_{\mathrm{i}}({\mathrm{x}}_{\mathrm{i}}–\overline{\mathrm{x}})({\mathrm{y}}_{\mathrm{i}}–\overline{\mathrm{y}})}{\sqrt{{\sum }_{\mathrm{i}}{({\mathrm{x}}_{\mathrm{i}}–\overline{\mathrm{x}})}^2}\sqrt{\sqrt{{\sum }_{\mathrm{i}}{({\mathrm{y}}_{\mathrm{i}}–\overline{\mathrm{y}})}^2}}}$$

If r was close to 1, it indicated that the expression pattern between the two samples exhibited high similarity.

Screening of DEGs

The Cuffdiff program was used to calculate gene expression values and to further estimate the alternatively spliced transcripts of the fragment based on its length. DEGs were screened in HCC tissues compared with adjacent non-cancerous tissues using the following thresholds: |log₂(FC)| >1, where FC is fold change; P-value <0.01; q-value (an adjusted P-value) <0.01, and fragments per kilobase of transcript per million mapped reads (FPKM) >4; in addition, DEGs identified from repetitive sequences were removed.

Screening of differentially expressed splicing variants

As with the aforementioned screening of DEGS, differentially expressed splicing variants were screened using Cuffdiff with the thresholds of P-value <0.05 and q-value <0.05.

Cluster analysis of DEGs

Distance calculations were applied for cluster analysis of DEGs using the ‘hcluster’ algorithm, as previously described (12). Relationships between samples and genes were evaluated by Spearman's and Pearson's correlation coefficient analyses.

GO and KEGG pathway enrichment analysis

The Database for Annotation, Visualization and Integrated Discovery (DAVID version 6.7; http://david.ncifcrf.gov) is an online bioinformatics tool used to extract biological meaning from a large number of genes (13). DAVID was used to analyze GO function and KEGG pathway enrichment of the screened DEGs. The Benjamini-Hochberg multiple testing procedure, which controls for false discovery rate, was applied for the correction of P-values. On the basis of the hypergeometric distribution principle, GO and KEGG terms that have P-value <0.05 and q-value <0.05 were considered to indicate a statistically significant difference.

Prediction of target genes of TFs and miRNAs

The screened DEGs were mapped to gene sets in the Molecular Signatures Database (version 3.1, http://www.broadinstitute.org/gsea/msigdb/index.jsp) using the online software tool WebGestalt version 2.0 (http://webgestalt.org). TFs and miRNAs regulating DEGs were obtained with the hypergeometric statistical method (14). The identified TFs and miRNAs, and their target genes were used to construct a miRNA-target gene network and a TF-target gene network, which were visualized with the open-source Cytoscape software (version 2.8.3, http://www.cytoscape.org).

PPI network analysis

Interaction relationships among DEGs were obtained by mapping DEGs using the BioGRID (version 3.2.115) protein network database plugin to construct a PPI network that was visualized with Cytoscape. The PPI network comprises nodes, which represent a protein, and links, which represent each pairwise protein interaction. The degree of a node corresponds to the number of interactions a protein has, and a node with a high degree was defined as a ‘hub node’ in the network. TFs and miRNAs that regulated a hub node were further predicted, and used to construct a regulatory network of the hub node.

Results

Quality evaluation of sequencing data

Calculation of the Q scores revealed that the quality of the sequencing data met the standards for further analysis. In addition, the GC content distribution was homogeneous, and the N-base content was within an acceptable range. Table I indicates that the sequencing data was well filtered for further research.

Table I.

Quality evaluation chart for sequencing data.

Sample	Raw readsa	Clean readsb	Clean basesc	sQ20 (%)d	GC (%)e	Duplication (%)f
SRR074999	21,944,622	14,292,579	571M	94.87	45.31	40.13
SRR075000	21,328,051	13,254,517	530M	94.04	45.53	38.28
SRR075001	21,532,717	13,304,142	532M	93.95	45.46	38.98
SRR075002	20,950,756	13,615,048	544M	94.81	45.36	45.40
SRR075003	21,959,501	13,835,204	553M	94.17	45.37	45.11
SRR075004	22,011,164	13,372,744	534M	93.80	45.01	42.09

^aOriginal reads transformed from original sequencing images.

^bReads filtered from raw reads.

^cTotal number of bases that were filtered.

^dPercentage of clean bases with sQ ≥20 in all clean bases.

^ePercentage of GC content in the sequence.

^fPercentage of repeated reads in whole reads. GC, guanine and cytosine; M, Megabase.

Alignment analysis

Sequencing reads mapped into different regions of the human reference genome GRCh37.66. Most of sequencing reads mapped into coding sequence exons (CDS_exons), whereas other reads mapped into different regions, such as 3′ untranslated region (3′UTR_Exons), 5′UTR_Exons and Introns. However, few novel splicing sites were obtained in the present study, which may be due to limited sequencing length.

Analysis of gene expression level

A favorable repetition in the duplicates was observed (repeatability, R²>0.94), whereas in different groups the R² values were much lower (R²<0.9). Top 10 DEGs between HCC and adjacent non-cancerous tissues were screened (Table II). Upregulated DEGs with high FC values included: α-fetoprotein, thrombospondin 4, neurotensin, preferentially expressed antigen in melanoma and sulfotransferase family 1C member 2. Downregulated DEGs with high FC values included: C-type lectin domain family 4 member M, alcohol dehydrogenase 4, 4-hydroxyphenylpyruvic acid dioxygenase and synaptotagmin 9. Genes with differentially expressed splicing variants were also screened (Table III) and included vesicle-associated membrane protein 4 (VAMP4), phosphatidylinositol glycan anchor biosynthesis class C, protein disulfide isomerase family A member 4 and growth arrest specific 5.

Table II.

Top 10 genes list - differentially expressed genes.

Gene	Locus	Control	Case	Log2(FC)	Regulation
AFP	4:74296854–74321891	10.953	6573.840	9.230	Up
THBS4	5:79287133–79379477	0.177	77.892	8.780	Up
NTS	12:86268072–86276767	0.866	315.600	8.509	Up
PRAME	22:22890122–2290900	0.0255	4.051	7.311	Up
SULT1C2	2:108905094–108926371	0.465	69.951	7.234	Up
PEG10	7:94285636–94299007	1.427	213.887	7.227	Up
NQO1	16:69740898–69760854	1.418	206.101	7.183	Up
AGR2	7:16831434–16873057	0.997	113.695	6.834	Up
GPC3	X:132669772–133119922	5.825	566.919	6.605	Up
NLRP1	17:5402747–5487832	3.232	289.985	6.488	Up
CLEC4M	19:7804878–7834490	271.072	1.483	−7.514	Down
ADH4	4:100010007–100274184	612.703	3.122	−7.617	Down
HPD	12:122277432–122326517	833.459	4.560	−7.514	Down
RP11-7M8.2.1	12:122277432–122326517	186.722	1.029	−7.503	Down
SYT9	11:7260098–7490273	4.716	0.027	−7.450	Down
CYP2E1	10:135192694–135383462	4030.920	22.780	−7.467	Down
CTD-2195M18.1.1	5:6582248–6588612	27.817	0.159	−7.455	Down
CPS1	2:211342405–211543831	2827.010	17.092	−7.367	Down
GLYAT	11:58476536–58499447	150.288	0.955	−7.298	Down
HSD11B1	1:209834708–209908295	520.703	3.393	−7.262	Down

Paired t-test (P-value) was used to identify the splicing variants that were differentially expressed between HCC and adjacent non-cancerous tissues. HCC, hepatocellular carcinoma; FC, fold change.

Table III.

Top 10 genes list - genes with significant differentially expressed splicing variants.

Gene	Locus	Sqrt (JS)	P-value	q-value	Significant
VAMP4	1:171669299–171711387	0.328	5.05×10−3	4.79×10−2	Yes
PIGC	1:171810620–172437971	0.198	2.50×10−4	3.93×10−3	Yes
PDIA4	7:148700153–148725733	0.833	5.00×10−5	9.03×10−4	Yes
GAS5	1:173831289–173866494	0.244	5.00×10−5	9.03×10−4	Yes
RARRES2	7:150035407–150038763	0.287	5.00×10−5	9.03×10−4	Yes
ABCF2	7:150904922–150924316	0.160	1.00×10−4	1.72×10−3	Yes
MRPS14	1:174968299–174992561	0.174	5.00×10−5	9.03×10−4	Yes
SLC7A2	8:17354596–17428082	0.436	5.00×10−5	9.03×10−4	Yes
INTS10	8:19674650–19709594	0.494	4.40×10−3	4.34×10−2	Yes

Paired t-test (P-value) was used to identify the splicing variants that were differentially expressed between HCC and adjacent non-cancerous tissues. HCC, hepatocellular carcinoma; Sqrt, square root. JS, JavaScript.

GO and KEGG enrichment analysis

The screened DEGs were enriched in various GO terms: Upregulated DEGs enriched in GO terms, including membrane-enclosed lumen, organelle lumen and intracellular organelle lumen, whereas downregulated DEGs enriched in GO terms, such as extracellular region part, oxidation reduction and response to wounding (Table IV). In addition, KEGG pathway enrichment analysis revealed that these upregulated DEGs also enriched in cell cycle, DNA replication and glutathione metabolism pathways, whereas the downregulated DEGs enriched in complement and coagulation cascades, fatty acid metabolism and PPAR signaling pathways (Table IV).

Table IV.

Top 5 significant GO terms.

Category	GO ID	Term	Count	Ratio	P-value	q-value	Regulation
GOTERM_CC_FAT	GO:0031974	Membrane-enclosed lumen	201	17.12	2.13×10−13	1.18×10−10	Up
GOTERM_CC_FAT	GO:0043233	Organelle lumen	197	16.78	4.37×10−13	1.21×10−10	Up
GOTERM_CC_FAT	GO:0070013	Intracellular organelle lumen	193	16.44	6.82×10−13	1.26×10−10	Up
GOTERM_CC_FAT	GO:0005829	Cytosol	151	12.86	1.70×10−11	2.35×10−9	Up
GOTERM_CC_FAT	GO:0031981	Nuclear lumen	159	13.54	6.46×10−11	7.13×10−9	Up
GOTERM_CC_FAT	GO:0044421	Extracellular region part	158	11.84	6.91×10−21	3.08×10−18	Down
GOTERM_BP_FAT	GO:0055114	Oxidation reduction	117	8.77	6.75×10−20	2.36×10−16	Down
GOTERM_BP_FAT	GO:0009611	Response to wounding	102	7.65	6.01×10−19	1.05×10−15	Down
GOTERM_CC_FAT	GO:0005615	Extracellular space	120	9.00	5.67×10−18	1.26×10−15	Down
GOTERM_CC_FAT	GO:0005576	Extracellular region	251	18.82	3.79×10−16	4.95×10−14	Down

The Fisher's exact test was used to calculate statistical significance (P-values) of enriched annotation terms. The q value is the Benjamini-Hochberg adjusted P-value. GO, gene ontology.

Prediction of TF and miRNA target genes

The predicted TF-target gene network (Fig. 1A) demonstrated that TFs, including specificity protein 1 (SP1), nuclear factor of activated T cells and forkhead box, may be able to regulate large numbers of target genes. For example, SP1 was revealed to target various genes, including cingulin (CGN), reversion-inducing cysteine-rich protein with Kazal motifs, protein kinase AMP-activated non-catalytic subunit β2 and chromosome segregation 1-like. In the predicted miRNA-target gene network (Fig. 1B), miRNAs, including miR-506, miR-17-5P and miR-15A, were also demonstrated to regulate several target genes. For example, target genes of miR-506 included CGN, ankyrin repeat domain 27 and 3′-phosphoadenosine 5′-phosphosulfate synthase 2.

Figure 1.

Regulatory networks of miRNAs, TFs and target genes. (A) Predicted TF-target gene network. (B) Predicted miRNA-target gene network. The red nodes represent the miRNAs or TFs, and the blue nodes represent their target genes. A link represents an interaction between a TF or miRNA and its target gene, whereas the size of a node corresponds to the number of interactions that a TF or miRNA has. miRNA, microRNA; TF, transcription factor.

PPI network analysis

A PPI network was constructed for the identified DEGs (Fig. 2), which placed ubiquitin C (UBC) as the hub node with the highest degree (degree=1,465). A regulatory network of the UBC gene was also constructed (Fig. 3), which predicted that UBC was regulated by a number of TFs that have also been identified as DEGs, including SP1, proto-oncogene c-Jun (JUN), FBJ murine osteosarcoma viral oncogene homolog (FOS), FOS-like antigen 2 (FOSL2) and SWI/SNF related matrix-associated actin-dependent regulator of chromatin, subfamily a member 4 (SMARCA4). A number of miRNAs were also predicted to regulate UBC expression, including miR-30, miR-181 and miR-106a (Fig. 3).

Figure 2.

Protein-protein interaction network with ubiquitin C as the hub node. In the network, a node represents a protein and a link represents each pairwise protein interaction. The red shaded nodes represent upregulated genes, and the blue shaded nodes represent downregulated genes. The size of a node corresponds to its degree (that is, the number of interactions one protein has). Green-framed nodes are those with a degree >50.

Figure 3.

Regulatory network of UBC. The UBC gene may be regulated by transcription factors, represented by the square nodes, and microRNAs, represented by the diamond nodes. Green square framed nodes were also identified as DEGs, with the red shaded nodes representing the upregulated DEGs, and the blue shaded nodes representing the downregulated DEGs. UBC, ubiquitin C; DEGs, differently expressed genes.

Discussion

HCC is a highly prevalent and lethal disease, which has proven to be difficult to treat (15). In the present study, the GSE25599 RNA-seq data set was downloaded from the GEO repository to investigate the molecular mechanisms of HCC. RNA-seq reads mainly mapped into CDS_Exons. Upregulated and downregulated DEGs between HCC and adjacent non-cancerous tissues were screened. In addition, several genes with significant differentially expressed splicing variants, such as VAMP4, were obtained. The screened DEGs mainly enriched in GO terms, such as membrane-enclosed lumen, organelle lumen and extracellular region part, and KEGG pathways, such as cell cycle and fatty acid metabolism pathways. Similarly, previous studies have demonstrated that cell cycle and fatty acid metabolism are closely associated with HCC pathogenesis (16–18).

Ubiquitination is associated with several biological processes, including cell cycle regulation, DNA repair, protein degradation and kinase modification (19). Ubiquitination of the disheveled protein has previously been reported to serve an important role in the development of several types of cancers, including liver cancer (20–22). In the present study, UBC exhibited the highest degree and was identified as the hub node of the constructed PPI network and the regulatory network, which suggested that UBC may serve an important role in HCC pathogenesis via interactions with a large number of genes. It may be a crucial target gene for HCC treatment. In addition, a previous study demonstrated that during the regulatory progress, ubiquitin-conjugating enzyme 9 (UBC9) is inhibited by miR-30 (23). Similarly, the present study predicted UBC to be regulated by miR-30, suggesting that UBC might be regulated by miR-30 in HCC.

The results from the present study suggested that the role of UBC in HCC may be regulated by a number of TFs, such as SP1, FOS, FOSL2, JUN and SMARCA4. SP1 has previously been demonstrated to bind to the GC-rich motifs of many gene promoters (24), including UBC. It has also been reported to be involved in cell differentiation, cell growth and apoptosis (25), and an interaction between UBC and SP1 has been reported (26). FOS and FOSL2 are two members of FOS gene family, which encode leucine zipper proteins that may be able to dimerize with member proteins of the JUN family to form the adaptor protein 1 (AP1) TF complex (27). AP1 has been demonstrated to modulate liver cancer initiation (28). A previous study reported that the expression of FOS is elevated in human hepatoma compared with adjacent tissues (29), suggesting that JUN, FOS and FOSL2 may be involved in the molecular mechanisms of HCC pathogenesis. In addition, the present study predicted that FOS and miR-181a regulated UBC in HCC. A previous study demonstrated that miR-181a may repress the inflammatory response in dendritic cells by targeting FOS (30); therefore, it has been inferred that FOS and miR-181a may also affect HCC-related inflammatory response. SMARCA4 is a member of the SWI/SNF family of chromatin-remodeling complexes, which exhibit helicase and ATPase activity, and regulate the transcription of several genes (31). SMARCA4 has been demonstrated to regulate the expression of CD44 by binding to breast cancer 1, and to promote cell proliferation though Notch-dependent proliferation signals (32). It has also been reported to be a tumor suppressor gene, and its loss promotes the development of small cell carcinoma of the ovary (33); SMARCA4 mutations have been identified in HCC (34). Results of the present study suggested that downregulated FOSL and SMARCA4 expression may participate in the development of HCC.

In conclusion, UBC may serve a crucial role in HCC pathogenesis, a role that may be regulated by SP1, FOS, JUN, FOSL2 and SMARCA4, which may be promising target genes for HCC treatment. A number of miRNAs, including miR-30 and miR-181, may also participate in the development of HCC; however, these results need to be investigated through biochemical studies.

Table V.

Top 5 significant KEGG pathways.

Category	KEGG ID	Term	Count	Ratio	P-value	q-value	Regulation
KEGG_PATHWAY	hsa04110:	Cell cycle	28	2.385	7.10×10−7	1.16×10−4	Up
KEGG_PATHWAY	hsa03030	DNA replication	14	1.193	1.41×10−6	1.15×10−4	Up
KEGG_PATHWAY	hsa00480	Glutathione metabolism	13	1.107	3.36×10−4	1.81×10−2	Up
KEGG_PATHWAY	hsa00970	Aminoacyl-tRNA biosynthesis	11	0.937	9.14×10−4	3.66×10−2	Up
KEGG_PATHWAY	hsa04610	Complement and coagulation cascades	32	2.399	1.87×10−14	3.43×10−12	Down
KEGG_PATHWAY	hsa00071	Fatty acid metabolism	20	1.499	7.52×10−10	6.92×10−8	Down
KEGG_PATHWAY	hsa03320	PPAR signaling pathway	25	1.874	9.89×10−9	6.07×10−7	Down
KEGG_PATHWAY	hsa00982	Drug metabolism	21	1.574	7.00×10−7	3.22×10−5	Down
KEGG_PATHWAY	hsa00830	Retinol metabolism	19	1.424	1.48×10−6	5.45×10−5	Down

The Fisher's exact test was used to calculate statistical significance (P-values) of enriched annotation terms. The q value is the Benjamini-Hochberg adjusted P-value. KEGG, Kyoto Encyclopedia of Genes and Genomes.