Cancer Biol Med 2021. doi: 10.20892/j.issn.2095-3941.2020.0131

ORIGINAL ARTICLE

Analysis of the HNF4A isoform-regulated transcriptome identifies CCL15 as a downstream target in gastric carcinogenesis

Zhen Ni1,2,*, Wenquan Lu3,*, Qi Li4, Chuan Han4, Ting Yuan5, Nina Sun6, Yongquan Shi1 1State Key Laboratory of Cancer Biology & Institute of Digestive Diseases, Xijing Hospital, Air Force Medical University of PLA, Xi’an 710032, China; 2Department of Gastroenterology, General Hospital of Western Theater Command, Chengdu 610083, China; 3Department of Gastroenterology, First Affiliated Hospital of Zhengzhou University, Zhengzhou 450052, China; 4Department of Endocrinology, General Hospital of Western Theater Command, Chengdu 610083, China; 5Department of Gastroenterology, 989 Hospital of the People’s Liberation Army, Luoyang 471003, China; 6Department of Gastroenterology, First Affiliated Hospital of Xi’an Medical College, Xi’an 710038, China

ABSTRACT Objective: Hepatocyte nuclear factor 4α (HNF4A) has been demonstrated to be an oncogene in gastric cancer (GC). However, the roles of different HNF4A isoforms derived from the 2 different promoters (P1 and P2) and the underlying mechanisms remain obscure. Methods: The expression and prognostic values of P1- and P2-HNF4A were evaluated in The Cancer Genome Atlas (TCGA) databases and GC tissues. Then, functional assays of P1- and P2-HNF4A were conducted both in vivo and in vitro. High-throughput RNA-seq was employed to profile downstream pathways in P1- and P2-HNF4A-overexpressing GC cells. The expression and regulation network of the candidate target identified by RNA-seq were characterized based on data mining and functional assays. Results: HNF4A amplification was a key characteristic of GC in TCGA databases, especially for the intestinal type and early stage. Moreover, P1-HNF4A expression was significantly higher in tumor tissues than in adjacent non-tumor tissues (P < 0.05), but no significant differences were found in P2-HNF4A expression (P > 0.05). High P1-HNF4A expression indicated poor prognoses in GC patients (P < 0.01). Furthermore, P1-HNF4A overexpression significantly promoted SGC7901 and BGC823 cell proliferation, invasion and migration in vitro (P < 0.01). Murine xenograft experiments showed that P1-HNF4A overexpression promoted tumor growth (P < 0.05). Mechanistically, RNA-seq showed that the - interactions pathway was mostly enriched in P1-HNF4A-overexpressing GC cells. Finally, (C-C motif) ligand 15 was identified as a direct target of P1-HNF4A in GC tissues. Conclusions: P1-HNF4A was the main oncogene during GC progression. The cytokine-cytokine receptor interaction pathway played a pivotal role and may be a promising therapeutic target. KEYWORDS Gastric cancer; carcinogenesis; HNF4A; CCL15; transcriptomics

Introduction deaths1. Due to a lack of specific clinical manifestations in the early stage, most GC patients are generally diagnosed at Gastric cancer (GC) remains one of the most common malig- an advanced stage with an overall 5-year survival of less than nancies, and one of the most frequent causes of cancer-related 20%, especially in China2. Therefore, increasing an under- standing of GC pathogenesis may identify novel strategies for more effective treatments. *These authors contributed equally to this work. Correspondence to: Yongquan Shi Hepatocyte nuclear factor 4α (HNF4A) has been previously 3-5 E-mail: [email protected] shown to be an oncogene in GC . Mechanistically, HNF4A ORCID ID: https://orcid.org/0000-0001-9515-7577 is a target of NF-κB6, and interacts with the Wnt pathway3 Received March 24, 2020; accepted August 21, 2020 and controls tumor metabolism required for GC progression Available at www.cancerbiomed.org 5 ©2021 Cancer Biology & Medicine. Creative Commons through (isocitrate dehydrogenase 1) IDH1 . However, the Attribution-NonCommercial 4.0 International License specific target genes and downstream signaling pathways of Cancer Biol Med Vol 18, No 2 May 2021 531

HNF4A remain unknown. In addition, a total 9 HNF4A iso- microarrays. Then, we performed transcriptome-wide map- forms (HNF4A1-9), transcribed from P1 or P2 promoters, ping of both P1- and P2-HNF4A targets in GC cell lines. have been shown to display distinct tissue expressions and Unbiased analysis revealed that the cytokine-cytokine recep- functions7,8 (Figure 1A). Although both P1- and P2-HNF4A tor interaction and PPAR signaling pathway genes were most are present in GC tissues9, it remains unclear which splice var- enriched in P1- and P2-dependent manners. We also identi- iant is the most relevant. fied chemokine (c-c motif) ligand 15 (CCL15), a member of In this study, we initially determined the expressions of the chemokine family, as a direct target gene of P1-HNF4A, both P1- and P2-HNF4A using TCGA databases and tissue which was required for GC progression.

A HNF4α

P2 1D P1 1A

A/B AF1 DBD LBD AF2F P2 P1

B

Adjacent Tumor

P1-HNF4α

CD 300 120 P1-HNF4α Negative

e Positive

al 90 200 iv 60 ent su rv 100 rc 30

Pe P = 0.0029 P1-HNF 4 α H-scor 0 0 0306090 120 150 Tumor Adjacent Survival times (month)

Figure 1 Expression and prognostic value of P1-HNF4A in gastric cancer (GC). (A) Schematic of the HNF4A gene and its isoforms (P1 and P2). The antigens recognized by isoform-specific (P1 and P2) antibodies are labeled. (B) Representative expression of P1-HNF4A in tumor and adjacent non-tumor tissues of GC patients. Scale bars = 200 μm and 50 μm. (C) H-score of P1-HNF4A in tumor and adjacent non-tumor tissues using tissue microarray and immunohistochemical analyses (*P < 0.05). (D) The prognostic value of P1-HNF4A in GC patients according to negative and positive expression (P < 0.01). 532 Ni et al. Distinct function of HNF4A isoforms in gastric carcinogenesis

Materials and methods every 2–3 days with 2 µg/mL puromycin-containing RPMI 1640 medium. Cell culture Cell proliferation assay

The human GC cell lines, AGS, MKN45, BGC823, AZ521, Cell proliferation was examined using CCK-8 kits (Thermo N87, and SGC7901, and the normal gastric epithelial cell Fisher Scientific) according to the manufacturer’s instruc- line, GES-1 (American Type Culture Collection, Manassas, tions, to determine cell growth. A total of 103 target cells VA, USA) were cultured in RPMI 1640 medium (Thermo were seeded in 96-well plates for the assays in 100 μL of Fisher Scientific, Waltham, MA, USA) at 37 °C with a humid- complete medium. Then, 10 μL of CCK-8 reagent was ified atmosphere of 5% CO supplemented with 10% fetal 2 added to the wells and incubated for 2 h. The cultures were bovine serum (Biological Industries, Kibbutz Beit Haemek, assayed each day for 4 continuous days, and the absorbance Israel) and 1% penicillin-streptomycin solution (Thermo was read at 450 nm with a reference wavelength at 650 nm Fisher Scientific). All cell line identities were verified through using a Varioskan Flash Multimode Reader (Thermo Fisher STR DNA profiling and were negative for mycoplasma Scientific). Each experiment was performed in triplicate and contamination. repeated 3 times.

Cell transfection Cell migration and invasion assays

HNF4A2 (NM_000457) and HNF4A8 (NM_175914) over- Cell migration experiments were performed in a transwell expression lentiviruses and a negative control were con- chamber covered with polycarbonate membranes (Corning, structed by GeneCopoeia (Rockville, MD, USA). Human NY, NY, USA). Cell invasion assays were performed using CCL15-shRNA lentivirus and negative controls were con- chambers coated with Matrigel matrix (BD Science, Sparks, structed by GenePharma Technologies (Shanghai, China) MD, USA). Resuspended cells (1 × 105 cells/well) were added (Table 1). Logarithmic phase cells (1 × 105) were seeded in to the upper chambers and cultured for 24 h. Migrated or 24-well culture plates. After adherence, lentiviral vectors invaded cells were stained with 0.1% Crystal Violet for 20 min were added at a final concentration of ~50 multiplicity of at room temperature. Quantification of migrated or invaded infection (MOI) with 0.5 μg/mL polybrene for 10 h at 37 °C. cells was performed using ImageJ software (National Institutes Fresh medium was replaced 10 h after transfection. The cells of Health, Bethesda, MD, USA). were transferred into 25 cm2 flasks after reaching 70%–80% confluence. Then, the cell culture medium was replaced Cell cycle assay

Table 1 The sequences of shRNAs used in this study Target cells were seeded in 6 cm plates and harvested after Gene symbols Sequences 5ʹ-3ʹ ­reaching approximately 70%–80% confluence. Then, resuspended Human CCL15-121 Sense CCUUACACAUGAAGAGGCATT cells were fixed in 75% ethanol, stained with propidium iodide according to the manufacturer’s protocol (Sigma-Aldrich, Antisense UGCCUCUUCAUGUGUAAGGTT St. Louis, MO, USA) and sorted using fluorescence-activated­ Human CCL15-222 Sense CAGCCAGAAAGCACUACAATT cell sorting (BD Sciences, San Jose, CA, USA). Data were ana- Antisense UUGUAGUGCUUUCUGGCUGTT lyzed using ModFit software (BD Sciences). Human CCL15-291 Sense AGCCCTTGAGTCCGGTGTCTT Murine xenograft model Antisense AAGACACCGGACTCAAGGGCT

Negative control Sense UUCUCCGAACGUGUCACGUTT Animal experiments were approved by the Institutional Antisense ACGUGACACGUUCGGAGAATT Animal Care and Use Committee of the Air Force Medical shRNAs, short hairpin RNA; HNF4A, hepatocyte nuclear factor 4α; University of PLA (Approval No. 2018-kq-010). Briefly, 6- CCL15, chemokine (C-C motif) ligand 15. to 8-week-old BALB/c nude female mice (Shanghai SLAC Cancer Biol Med Vol 18, No 2 May 2021 533

Laboratory Animal Co., Shanghai, China) were subcutane- Transcriptome sequencing and analysis were conducted ously implanted with 1 × 107 SGC7901 cells transfected with by OE Biotech (Shanghai, China). Raw data were processed a stable control, P1- or P2-HNF4A (n = 5). Then, the ani- using Trimmomatic (Illumina). Differentially-expressed genes mals were euthanized and the tumor volumes were calculated (DEGs) were identified using the DESeq (2012) R package using the formula (V = length × width2 × 0.5) after 2 weeks. (Bioconductor; http://www.bioconductor.org/). A P value < The tumor tissues were then harvested and processed for fur- 0.05 and fold changes > 2 or fold changes < 0.5 was set as the ther hematoxylin and eosin staining and immunohistochem- thresholds for significant differential expression. Hierarchical istry (IHC) analysis. cluster analysis of DEGs was performed to characterize patterns. (GO) enrichment and lysate preparation and Western blot Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses of DEGs were performed using R based Cells were digested with RIPA cell lysis buffer (Beyotime on the hypergeometric distribution. Biotechnology, Shanghai, China) with protease and phos- phatase inhibitor cocktails (MCE, Shanghai, China) and were Quantitative real-time PCR collected using a cell scraper. Total protein was extracted and quantified using the bicinchoninic acid method (Thermo Total RNA was extracted using the RNeasy Mini Kit Fisher Scientific). (20–30 μg) were resolved using (QIAGEN, Duesseldorf, Germany), according to the manu- 10% PAGE (Bio-Rad, Hercules, CA, USA), transferred to a facturer’s instructions. In total, 0.5 μg RNA was synthesized nitrocellulose membrane (Pall Corporation, Port Washington, into cDNA using the PrimeScript RT Reagent Kit (TaKaRa, NY, USA) at 25 V for 30 min, and blocked for 1 h in 10% Shiga, Japan) and Mir-X mRNA First-Strand Synthesis nonfat milk in 1× TBS/0.1% (v/v) Tween 20 at room tem- Kit (TaKaRa) in a 10 μL volume. Real-time PCR was per- perature. Primary antibodies (GAPDH, 1:1,000, #5174; formed on a CFX96 system using TB Green Premix Ex Taq Technology; β-actin, 1:1,000, #4970; Cell II (TaKaRa) with 2 μL cDNA and 0.8 µL primers in a final Signaling Technology; P1-HNF4A, 1:1,000, ab41898; Abcam; volume of 20 μL. The final PCR conditions were as follows: P2-HNF4A, 1:1,000, PP-H6939-00; R & D Systems; and predenaturation at 95 °C for 10 min, followed by 44 cycles CCL15, 1:1,000, ab219388; Abcam) were added and incubated at 95 °C denaturation for 10 s, 60 °C annealing for 20 s, and overnight at 4 °C. Secondary antibodies (goat anti-rabbit IgG, 72 °C extension for 10 s. The target mRNA gene was nor- HRP #7074 or goat anti-mouse IgG, HRP, 1:2,000, #7076; Cell malized to human GAPDH using the 2-ΔΔCT method. Primer Signaling Technology) were incubated at room temperature sequences are shown in Table 2. for 1 h. Signals were detected using Western Lumax Light Sirius HRP substrate reagent (ZETA-Life, San Francisco, CA, Tissue microarrays and immunohistochemistry USA). All data were normalized to human β-actin or GAPDH. The bands were scanned using a ChemiDocXRS + Imaging GC tissue microarrays, including 98 cases of gastric tumor and System (Bio-Rad). adjunct non-tumor tissues, were obtained from Outdo Biotech (HStmA180Su15; Shanghai, China). Patient characteristics are RNA-seq shown in Table 3. This study was approved by the Institutional Ethics Committee of Xijing Hospital of the Air Force Medical Total RNA was extracted using the mirVana miRNA Isolation University of PLA (Approval No. KY20183093-1). Kit (Ambion, Austin, TX, USA). RNA integrity was evaluated Sections were dried overnight at 60 °C. Antigen retrieval using the Agilent 2100 Bioanalyzer (Agilent Technologies, was performed by heating in 1× citrate for 2 min and then Santa Clara, CA, USA). The samples with RNA integrity num- cooling to room temperature. Endogenous peroxidase activ- ber (RIN) ≥ 7 were subjected to subsequent analysis. The ity was blocked using 3% H2O2 for 10 min. Slides were libraries were constructed using the TruSeq Stranded mRNA incubated with mouse anti-human antibodies against Ki67 LT Sample Prep Kit (Illumina, San Diego, CA, USA). Then, (1:1,000, ab92742), P1-HNF4A (1:100, ab41898; Abcam), these libraries were sequenced using the Illumina HiSeq X Ten P2-HNF4A (1:100, PP-H6939-00; R & D Systems), and platform, and 125 bp/150 bp paired-end reads were generated. CCL15 (1:200, ab219388; Abcam) at 4 °C overnight. The 534 Ni et al. Distinct function of HNF4A isoforms in gastric carcinogenesis

Table 2 The sequences of primers used in this study Dual-Luciferase Reporter Assay Kit (Promega) at 48 h after

Gene symbols Primers 5ʹ-3ʹ reporter transfection using Lipofectamine 2000 Transfection Reagent (Invitrogen, Carlsbad, CA, USA) in SGC7901 cells. HNF4A Forward GTTCAAGGACGTGCTGCTCCTA Firefly luciferase activity was normalized to Renilla luciferase Reverse AGGCATACTCATTGTCATCGATCTG activity, and the final data are presented as the fold induc- CCL15 Forward AGGCCCAGTTCACAAATGA tion of luciferase activity compared to that of the negative Reverse CAGTCAGCAGCAAAGTGAAAG control.

GAPDH Forward GCACCGTCAAGGCTGAGAAC Chromatin immunoprecipitation (ChIP) Reverse TGGTGAAGACGCCAGTGGA

CCL15 Promoter 1 Forward TAACCTGTGAGCCTTCGATG ChIP assays were performed using the EZ ChIP™ Kit Reverse GTGACCTATGATCACACCACTG (Millipore, Billerica, MA, USA). The cells were cross-linked

CCL15 Promoter 2 Forward GGAAGGTCTTCTAGCCTCTAATC with 1% formaldehyde for 10 min at 37 °C and quenched with 2.5 M glycine for 5 min at room temperature. DNA was immu- Reverse CACTGCAAACTCCACCTTCTG noprecipitated from the sonicated cell lysates using HNF4A CCL15 Promoter 3 Forward TAGGCTCAAGTATGTGCACTA antibody (Abcam, Cambridge, MA, USA) and subjected to Reverse GGTTTGACCTTTGGATTGGG PCR to amplify the HNF4 binding site (Table 2). The ampli- CCL15 Promoter 4 Forward GAGTAGTCTTGGAAAAGGCAAC fied fragments were analyzed on an agarose gel. A nonspecific antibody against IgG served as a negative control. Reverse GTAATACCAGTACTTTGGGAG HNF4A, hepatocyte nuclear factor 4α; CCL15, chemokine (C-C Data mining using public databases motif) ligand 15.

The mRNA expression and DNA copy number of HNF4A in sections were then counterstained with hematoxylin staining GC were analyzed within the Oncomine (http://oncomine. and dehydrated. org), GEPIA (http://gepia.cancer-pku.cn), and Ualcan (http:// The slides were scanned and viewed using a Pannoramic ualcan.path.uab.edu) databases. The GEPIA and Ualcan data- Viewer (3DHISTECH, Budapest, Hungary). The stain- bases were used to analyze candidate gene alterations that ing intensity of P1- and P2-HNF4A was semiquantitatively were identified as being enriched in RNA-seq. We further used determined using the H-score method10. Only unequivocally the LinkedOmics (http://www.linkedomics.org), cBioPortal stained nuclei were considered positive. H-scores < 50 and (http://cbioportal.org), and GEPIA databases to characterize ≥ 50 were considered as negative and positive, respectively. the relationships between HNF4A and candidate target genes. H-scores < 150 and ≥ 150 were considered low and high expressions, respectively. CCL15 staining was semiquantita- Statistical analysis tively performed according to the staining intensity and per- centage of positive cells11. IHC scores < 6 and ≥ 6 were consid- All cell culture experiments were performed in triplicate to ered low and high expressions, respectively. reduce error and ensure reproducibility. All quantitative data are expressed as the mean ± SEM. Differences between two Dual-luciferase reporter assays groups were examined using an unpaired Student’s t-test. Differences between multiple groups were compared by one- Briefly, 2,000 bp fragments of the human CCL15 promoter way analysis of variance with Dunnett’s post hoc tests. All cat- were obtained from Ensembl and predicted in the JASPAR egorical data are expressed as rates. Differences between two database. The wild-type (WT) and corresponding muta- groups were examined using the χ2 test. Statistical analysis was tional CCL15 promoter fragments covering predicted DR1 performed using SPSS 13.0 statistical software for Windows sites were PCR-amplified and cloned into the firefly luciferase (SPSS, Chicago, IL, USA) or GraphPad Prism 5.0 (GraphPad reporter plasmid, pGL3-basic vector (Promega, Madison, Software, San Diego, CA, USA). P < 0.05 was considered sta- WI, USA). Luciferase activity was then detected using a tistically significant. Cancer Biol Med Vol 18, No 2 May 2021 535

Table 3 Clinical characteristics of the patients analyzed using tissue microarrays

Characteristics n P1-HNF4A P CCL15 P Negative Positive Low High Age (years) > 0.05 > 0.05

< 60 32 20 12 16 16

≥ 60 66 27 39 27 39

Gender > 0.05 > 0.05

Male 62 30 32 29 33

Female 36 16 20 12 24

TNM stage > 0.05 > 0.05

I+II 39 17 22 20 19

III+IV 59 29 30 22 37

T stage > 0.05 > 0.05

T1+T2 14 5 9 8 6

T3+T4 84 42 42 34 50

LN metastasis > 0.05 < 0.05

No 26 10 16 16 10

Yes 72 36 36 25 47

Distant metastasis > 0.05 > 0.05

No 89 40 49 38 51

Yes 9 6 3 3 6

WHO type > 0.05 > 0.05

Adenocarcinoma 47 20 27 15 32

TA 17 5 12 4 13

MA 10 8 2 5 5

SRC 14 8 6 6 8

UC 10 4 6 1 9

TA, tubular adenocarcinoma; MA, mucus adenocarcinoma; SRC, signet ring cell carcinoma; UC, undifferentiated carcinoma.

Results than in normal tissues in TCGA Gastric, Deng Gastric, Cho Gastric, and DErrico Gastric (P < 0.01), especially for intes- The expression levels of HNF4A in TCGA tinal type adenocarcinomas (P < 0.01) (Supplementary databases Figure S1). Data in the Ualcan database also showed signif- icantly higher expression levels of HNF4A in tumor tissues Initially, the mRNA expression and DNA copy number of than in non-tumor tissues (Supplementary Figure S2A). HNF4A in GC were analyzed within the Oncomine, GEPIA, Early stages showed significantly higher HNF4A expression and Ualcan databases. Data in the Oncomine database levels than those in late stages (Supplementary Figure S2B revealed that mRNA expression and DNA copy number and S2C). Finally, data in the GEPIA database also revealed variation of HNF4A were significantly higher in GC tissues significantly higher HNF4A expression levels in tumor tissues 536 Ni et al. Distinct function of HNF4A isoforms in gastric carcinogenesis than in non-tumor tissues (Supplementary Figure S3A). showed dramatic differences (Supplementary Figure S5A). Furthermore, isoform analysis indicated that HNF4A1 and Then, we directly introduced P1-HNF4A (HNF4A2) and HNF4A2 expressions were significantly higher than HNF4A7 P2-HNF4A (HNF4A8) into SGC7901 and BGC823 cells, and HNF4A8 expressions (Supplementary Figure S3B). respectively, 2 cell lines with relatively low P1- and P2-HNF4A Together, these results showed that HNF4A amplification expressions (Supplementary Figure S5B). The CCK-8 assay was common in GC, especially in the intestinal type and indicated that P1-HNF4A overexpression promoted GC early stages. In addition, P1-HNF4A upregulation was more cell proliferation (P < 0.01) (Figure 2A). However, no sig- remarkable than P2-HNF4A upregulation. nificant difference was noted after P2-HNF4A overexpres- sion (P > 0.05) (Figure 2A). Furthermore, flow cytometry The expression levels of P1- and P2-HNF4A in analysis showed that P1-HNF4A overexpression promoted gastric tumors and adjacent tissues G1/S phase arrest (P < 0.01), and P2-HNF4A overexpres- sion showed no significant effect on the cell cycle (P > 0.05) Although high HNF4A expression has been shown in GC tis- (Figure 2B). Furthermore, P1- and P2-HNF4A promoted sues3-5, previous studies did not distinguish the functions of ERK1/2 pathway activation, with a more significant ten- different HNF4A isoforms. We therefore detected the expres- dency in P1-HNF4A-overexpressing cells (Supplementary sion levels of both P1- and P2-HNF4A in gastric tumors and Figure S5B). adjacent non-tumor tissues using tumor microarrays. The Next, transwell invasion and migration assays also showed results indicated that P1-HNF4A expression was significantly that P1-HNF4A overexpression facilitated GC cell invasion increased in tumor tissues compared with non-tumor tis- and migration (P < 0.01), but P2-HNF4A overexpression sues (P < 0.05) (Figure 1B and 1C). However, no significant showed no effect (P > 0.05) (Figure 2C and 2D). Moreover, difference was noted in P2-HNF4A expression (P > 0.05) we performed an in vivo xenograft mouse experiment using (Supplementary Figure S4A and S4B), although the H-score stably transfected SGC7901 cells. The results showed that the of P2-HNF4A was higher in non-tumor tissues than in tumor average tumor volume in the P1-HNF4A group was signifi- tissues (P > 0.05) (Supplementary Figure S4B). Next, we cantly larger than that in the control group (P < 0.05) (Figure examined P1-HNF4A expression according to cancer clini- 3A and 3B). However, no significant difference was found in cal features. As shown in Table 2, P1-HNF4A expression lev- the average tumor volume in the P2-HNF4A group compared els showed an upregulation in the I+II, T1+T2, M0, and N0 with the control (P > 0.05) (Figure 3A and 3B). In addition, stages, but no significant difference was noted (P > 0.05). tumors from P1-HNF4A overexpression mice showed more Finally, patients with positive P1-HNF4A staining showed a significant Ki67 expression than those from P2-HNF4A over- significantly shorter survival compared with negative cases expression mice (Figure 3C and 3D). (P < 0.05) (Figure 1D). However, no significant difference was Taken together, the above in vitro and in vivo results indi- noted between low and high P2-HNF4A expression patients cated that P1-HNF4A could significantly promote GC pro- (Supplementary Figure S4C). Together, these results indi- gression, and that P2-HNF4A showed no significant effect. cated that P1- and P2-HNF4A might play different roles dur- ing GC progression. P1-HNF4A expression was the main iso- RNA-seq showed distinct pathway and form in gastric carcinogenesis and predicted poor prognoses. gene enrichment in P1- and P2-HNF4A- However, no significant difference was observed in P2-HNF4A overexpressing cells expression levels. To investigate the underlying mechanisms mediating P1- P1- and P2-HNF4A showed distinct functions and P2-HNF4A functions, we performed RNA-seq using both in vitro and in vivo P1- and P2-HNF4A-overexpressing SGC7901 cells. For this purpose, we transfected SGC7901 cells with P1-HNF4A, We further examined the roles of both P1- and P2-HNF4A in P2-HNF4A and negative control lentiviruses. Increased GC cells in vitro and in vivo. First, we detected the expression expression of P1- and P2-HNF4A upon transfection was levels of P1- and P2-HNF4A in GC cell lines. Western blot confirmed by Western blot (Supplementary Figure S5B). results indicated that P1- and P2-HNF4A expression levels Then, we performed RNA-seq to examine the effects Cancer Biol Med Vol 18, No 2 May 2021 537

0.6 0.6 A SGC7901-NC ** BGC823-NC ** SGC7901-P1 BGC823-P1 SGC7901-P2 BGC823-P2 0.4 0.4

OD450 0.2 OD450 0.2

0.0 0.0 0 1 2 3 4 5 0 1 2 3 4 5 Time (day) Time (day) B NC P1-HNF4α P2-HNF4α NC HNF4α2 HNF4α8 400 500 80 n.s 400 320

400 ** 60 n.s 300 240 300 cent ** Number 40 Number BGC823 Number 200 160 200 n.s

Cell per 20

80 n.s 100 100 0 0 0 0 04080 120 160 0306090 120 150 04080 120 160 G0-G1 SG2-M Channels (FL2-A) Channels (FL2-A) Channels (FL2-A) NC HNF4α2 HNF4α8

400 80 n.s 800 n.s

600 ** ** 300 60 600 en t 400 200

Number 40 400 Number SGC7901 Number

200 n.s 100 Cell pe rc

200 20 ** 0 0 0 0 0306090 120 150 04080 120 160 0306090 120 150 G0-G1 S G2-M Channels (FL2-A) Channels (FL2-A) Channels (FL2-A) n.s. C NC P1-HNF4α P2-HNF4α 2500 ** 2000 1500 1000

Cell numbe r 500

SGC7901 0 n.s. 2000 **

1500

1000 Cell numbe r BGC823 500

0 n.s. D NC P1-HNF4α P2-HNF4α 1500 ** 1200

900

600 Cell number SGC7901 300

0 n.s. 2000 **

1500

1000 BGC823

Cell number 500

0 NC P1-HNF4α P2-HNF4α

Figure 2 The effects of both P1- and P2-HNF4A on malignant phenotypes of gastric cancer cells in vitro. (A) P1-HNF4A or P2-HNF4A overex- pression lentivirus were transfected into SGC7901 and BGC823 cells. Cell proliferation was examined using the CCK8 assay. OD450 absorbance was detected every day for a continuous 4 days (**P < 0.01, n.s., nonsignificant,n = 3). (B) The effects of P1- and P2-HNF4A overexpression on 538 Ni et al. Distinct function of HNF4A isoforms in gastric carcinogenesis

SGC7901 and BGC823 cell cycles using a flow cytometry assay (**P < 0.01; n.s., nonsignificant,n = 3). (C) The effects of both P1- and P2-HNF4A overexpression on SGC7901 and BGC823 cell migrations using transwell experiments. Migrated cells were counted using ImageJ (**P < 0.01; n.s., nonsignificant, n = 3) (Crystal Violet staining, 20×). (D) The effects of both P1- and P2-HNF4A overexpression on SGC7901 and BGC823 cell invasion using transwell experiments. Invaded cells were counted by ImageJ (**P < 0.01; n.s., nonsignificant,n = 3).

A NC P1-HNF4α P2-HNF4α B n.s. 1500

) * 3

m 1200

900

600

mor volume (m 300 Tu 0 Control P1-HNF4α P2-HNF4α

NC P1-HNF4α P2-HNF4α C

HE 50µm 50µm 50µm

200µm 200µm 200µm D

ki67 50µm 50µm 50µm

200µm 200µm 200µm

Figure 3 The roles of both P1- and P2-HNF4A on gastric cancer (GC) progression in vivo using murine xenograft experiments. (A) BALB/c nude mice underwent orthotopic implantation with SGC7901 cells stably transfected by a negative control, P1-HNF4A, and P2-HNF4A overexpression­ lentiviruses. (B) The tumors were extracted and measured. Tumor volume was calculated using the formula (V = length × width2 × 0.5). The results are expressed as the mean ± SEM (*P < 0.05, n.s., nonsignificant). (C) H&E staining of sections from orthotopic specimens. Scar bars = 200 μm and 50 μm). (D) Ki67 staining of section from orthotopic specimens by immunohistochemistry. Scale bars = 200 μm and 50 μm).

of P1- and P2-HNF4A on the SGC7901 transcriptome. changes, which could be because P1-HNF4A typically has We found that both P1- and P2-HNF4A overexpression a more potent transactivation function than P2-HNF4A. resulted in considerable changes in gene expression at the KEGG pathway analysis of the differentially regulated genes mRNA level (Supplementary Figure S6A). Using 2-fold as showed that there was a significant upregulation of genes a cutoff to designate DEGs, we found 310 and 105 genes involved in the cytokine-cytokine receptor interaction, and that were downregulated and 763 and 190 genes that were a decrease in genes involved in the calcium signaling path- upregulated in P1- and P2-HNF4A-overexpressing cells, way in P1-HNF4A cells (Figure 4A). In P2-HNF4A cells, respectively (Supplementary Figure S6B). There was an there was a significant upregulation of genes involved in the overall greater effect in P1-HNF4A than in P2-HNF4A in PPAR signaling pathway and a decrease in genes involved in terms of the number of dysregulated genes with large fold valine, leucine, and isoleucine biosynthesis (Figure 4B). Cancer Biol Med Vol 18, No 2 May 2021 539

A HNF4α2-UP vs. Control (UP):KEEG Enrichment top 20 hsa05414:Dilated cardiomyopathy hsa05412:Arrhythmogenic right ventricular cardiomyopathy hsa05410:Hypertrophic cardiomyopathy hsa05321:Inflammatory bowel disease hsa05205:Proteoglycans in cancer

hsa05202:Transcriptional misregulation in cancer Number hsa05146:Amoebiasis 10 15 hsa05143:African trypanosomiasis 20 hsa05132:Salmonella infection 25 30 hsa04979:Cholesterol metabolism 35 hsa04947:Protein digestion and absorption P-value hsa04928:Parathyroid hormone synthesis, secretion and action 0.003 hsa04913:Ovarian steroidogenesis 0.002 hsa04668:TNF signaling pathway 0.001 hsa04640:Hematopoietic cell lineage hsa04610:Complement and coagulation cascades hsa04512:ECM-receptor interaction hsa04151:PI3K-Akt signaling pathway hsa04060:Cytokine-cytokine receptor interaction hsa03320:PPAR signaling pathway

2.042.5 3.0 3.5 .0 Enrichment score

B HNF4α8-UP vs. Control (UP):KEEG Enrichment top 20 hsa05414:Dilated cardiomyopathy hsa05412:Arrhythmogenic right ventricular cardiomyopathy hsa05410:Hypertrophic cardiomyopathy hsa05323:Rheumatoid arthritis hsa05222:Small cell cancer hsa05202:Transcriptional misregulation in cancer hsa04979:Cholesterol metabolism Number 4 hsa04975:Fat digestion and absorption 6 hsa04928:Parathyroid hormone synthesis, secretion and action 8 10 hsa04921:Oxytocin signaling pathway hsa04657:IL-17 signaling pathway P-value hsa04640:Hematopoietic cell lineage 0.03 hsa04350:TGF-beta signaling pathway 0.02 0.01 hsa04216:Adrenergic signaling in cardiomyocytes hsa04152:AMPK signaling pathway hsa04512:NF-kappa B signaling pathway hsa04060:Cytokine-cytokine receptor interaction hsa04151:cAMP signaling pathway hsa04010:MAPK signaling pathway hsa03320:PPAR signaling pathway

284 6 Enrichment score

Figure 4 RNA-seq screen of target genes and signaling pathways using downstream P1- and P2-HNF4A overexpression in SGC7901 cells. (A) Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis showing enrichment of the top 20 signaling pathways in P1-HNF4A overexpression compared to the negative control in SGC7901 cells. (B) KEGG pathway analysis showing enrichment of the top 20 signaling pathways in P2-HNF4A overexpression compared to the negative control in SGC7901 cells.

CCL15 is a direct target of P1-HNF4A and is examined the expression of these genes within TCGA data- functionally required in GC bases. Data in GEPIA revealed that the expressions of CCL5, CCL15, CCL20, CSF2RA, CXCL1, CXCL8, CXCL16, CXCR4, To test the deregulated genes involved in the cytokine-cy- GDF15, IL1RN, IL2RB, IL2RG, IL22RA1, INHBA, TNFRSF1B, tokine receptor interaction pathway in GC development, we and TNFRSF14 were significantly higher in GC tissues than in 540 Ni et al. Distinct function of HNF4A isoforms in gastric carcinogenesis normal tissues (Supplementary Figure S7). Data from Ualcan (Figure 6B). Next, we constructed luciferase reporter gene revealed that the expressions of BMP2, CCL5, CCL15, CCL20, fragments covering the two DR1 sequences (wild-type and IL8, CD70, CSF2RA, CXCL1, CXCL16, CXCR4, TGFB2, GDF15, mutant) (Figure 6B). The results revealed that HNF4A2 posi- IL11, IL1R1, IL1RN, IL13RA2, IL32, IL2RB, IL2RG, IL22RA1, tively regulated the CCL15 promoter containing the two DR1 IL18R1, IL18RAP, INHBA, INHBB, INHBE, TNFRSF9, sites between ~81~95 and ~243~257 (Figure 6B and 6C). TNFRSF1B, and TNFRSF14 were significantly higher in GC Chromatin immunoprecipitation (ChIP) assays further con- tissues than in normal tissues (Supplementary Figure S8). firmed that HNF4A2 bound to the speculative sites (~81~95 Among the aforementioned gene deregulations in GC, we and ~243~257) on the CCL15 promoter in SGC7901 cells focused on , including CCL15, the role of which (Figure 6D). Furthermore, to determine the clinical relevance, remained unclear in GC progression. Then, we examined the we divided the GCs into P1-HNF4A-negative and P1-HNF4A- in vitro role of CCL15 on GC cell proliferation and invasion. positive subgroups according to the H-score. CCL15 high Initially, we detected the mRNA expression of CCL15 in the expression cases in the P1-HNF4A-positive group were found normal gastric epithelial cell line, GES-1, and the GC cell lines, to be significantly higher than those in the P1-HNF4A-negative BGC823, MKN45, AZ521, SGC7901, and AGS. Notably, the group (P < 0.01) (Figure 6E and 6F). We further determined mRNA level of CCL15 in GC cells far surpassed that in GES-1 the correlation between HNF4A and CCL15 in TCGA data- cells (Supplementary Figure S9). Then, shRNA-mediated bases. Both data in the GEPIA, cBioportal, and LinkedOmic CCL15 knockdown cells were generated in MKN45, a cell line databases revealed that HNF4A was positively correlated with with relatively high CCL15 expression (Figure 5A). Significant CCL15 (Figure 6G, 6H, and 6I). reductions in the proliferation rate were observed in shCCL15 cells compared with control cells using the CCK-8 assay (P < Discussion 0.01) (Figure 5B). Significantly reduced cell migration and invasion were also observed in shCCL15 cells compared with Accumulating evidence has revealed the vital role of HNF4A in control cells in the MKN45 cell line (P < 0.01) (Figure 5C GC initiation and progression. However, the functions of dif- and 5D). To determine whether CCL15 was also upregulated ferent HNF4A isoforms and underlying molecular mechanisms in primary GCs, we analyzed CCL15 expression using a GC in GC cell growth and invasion have not been completely clar- tissue microarray. The results indicated that CCL15 expres- ified. In this study, we demonstrated that P1-HNF4A was the sion was significantly higher in tumors than in paracancerous main isoform upregulated in GC progression. Furthermore, tissues (P < 0.01) (Supplementary Figure S10A and S10B). we revealed that the cytokine-receptor pathway was one of the Subgroup analysis indicated that patients with high CCL15 key downstream signaling pathways. We also demonstrated expression showed more lymph node metastasis than those that CCL15 was the most abundant chemokine downstream with low CCL15 expression (P < 0.05) (Table 3). Survival of P1-HNF4A in human GC, with significant prognostic value. analysis showed that patients with high CCL15 expression had These findings revealed complex tumor-promoting effects a poorer prognosis than patients with low CCL15 expression shaped by the HNF4A-CCL15 axis in human GC. (P < 0.01) (Figure 5E). We further explored the prognos- HNF4A is a highly conserved member of the nuclear recep- tic value of CCL15 in the cBioportal database. High CCL15 tor superfamily, and regulates genes involved in hepatocyte expression showed poorer prognosis in both the Firehose leg- differentiation12,13. It also plays key roles in kidney14, pan- acy and PanCancer Atlas groups (Figure 5F and 5G). In sum- creas15, and gut16 development, and regulates the expression of mary, these results supported CCL15 as an oncogene in GC genes involved in metabolism17,18, homeostasis19-21, differen- progression. tiation22, and immune response23,24. The roles and functions To check if CCL15 acted as a downstream target, we per- of HNF4A in HCC and CRC have been clearly defined25-27. formed qRT-PCR. The results indicated that both P1- and In contrast, the role of HNF4A in stomach development and P2-HNF4A could significantly upregulate CCL15 expression, GC has only preliminarily been investigated. Recent studies with P1-HNF4A being more prominent (P < 0.01) (Figure showed that HNF4A is functionally required for GC survival 6A). Then, we examined the promoters of CCL15 using by interacting with the Wnt and NF-κB pathways and regulat- JASPAR. The results showed that the promoter of CCL15 ing tricarboxylic acid (TCA) cycle metabolism3,5,6. In contrast, contained two predicted DR1 motifs (AGGTCAnAGGTCA) another recent study showed that the tumor suppressor ITLN1 Cancer Biol Med Vol 18, No 2 May 2021 541

NC shCCL15-222 NC shCCL15-291 2.0 NC ABshCCL15-222 shCCL15-291 CCL15 1.5

1.0 OD450 0.5 β-actin 0.0 MKN45 d1 234 MKN45 C NC shCCL15-222 shCCL15-291 ** 1000 * 800 600 400 Cell number Cell migratio n 200 0 ** 1200 D ** 900 600 300 Cell numbe r Cell invasion 0 NC MNK45

shCCL15-222shCCL15-291 EF 120 Altered group Low 100% Unaltered group High 90% 90 80% 70% Logrank test P-value:1.231e-3 60 60% 50% P<0.01

Percent survival 40% 30 30%

20% TCGA, Firehose legacy 0 Disease/progression-free survival 10% 0 30 60 90 120 150 0% Time (month) 0 10 20 30 40 50 60 70 80 90 100 110 120 Time (month) G Altered group 100% Unaltered group 90%

80% Logrank Test P-value:2.978e-4 70% 60% 50% 40% 30%

20% TCGA, PanCancer Atlas

Disease/progression-free survival 10% 0% 0 10 20 30 40 50 60 70 80 90 100 110 120 Time (month)

Figure 5 The effects of CCL15 on malignant phenotypes of gastric cancer (GC) cells in vitro. (A) Negative control and sh-CCL15 lentivirus were transfected in MKN45 cells and CCL15 protein expression was detected by Western blot. β-actin was used as internal control. (B) MKN45 cell viability infected with lentivirus expressing shRNAs targeting CCL15 or control shRNA was examined by CCK8, and OD450 was detected at the indicated time (**P < 0.01, n = 3). (C, D) Cell migration and invasion was performed using transwell assays. Migrated (C) or invaded (D) cells were counted by ImageJ (*P < 0.05, **P < 0.01, n = 3) (Crystal Violet staining, 20×). (E) The prognostic value of CCL15 in GC patients using tissue microarrays. (F) The prognostic value of CCL15 in GC patients of the Firehose legacy cohort using the cBioPortal database (Z-Score = 2). (G) The prognostic value of CCL15 in GC patients of the PancancerAtlas cohort using the cBioPortal database (Z-Score = 2). 542 Ni et al. Distinct function of HNF4A isoforms in gastric carcinogenesis

A Bc 1.8 SGC7901-NC 25 ** SGC7901-HNF4α 2 ** 1.5 20 1.2 ** 15 0.9 ** 10 0.6 5

Relative CCL15 mRNA 0.3 Firefly/Rellina Luciferase 0 0.0 NC P1-HNF4α P2-HNF4α CCL15 mutCCL15-1 PGL3-basic mutCCL15-mutCCL15-2 3 D NC ChIP1 ChIP2 CCL15

292 bp 4229 bp 280 bp 298 bp 209 bp 77 bp ChIP1 ChIP2

MMMM

M Input HNF4α Ig G Blan k Input HNF4α Ig G Blan k Input HNF4α Ig G Blan k Input HNF4α Ig G Blan k

2000 bp

1000 bp 750 bp 500 bp

250 bp

100 bp

P1-HNF4α CCL15 E F CCL15-Low CCL15-High 60 **

Adjacen t 50

40

30

Percent (%) 20

10 Tumo r 0 Negative Positive

GH I Pearson-Correlation: 0.4526 P = 0.00069 CCL15 vs. HNF4A P-value: 2.356e–22 R = 0.16 15 Sample size: (n = 415) Spearman: 0.26 6 (P = 2.40e–7) 10 Pearson: 0.50 10 4 (P = 2.44e–24) 5

5 2 Log2(CCL15 TPM) 0 CCL1 5

0 –5 0

0 246 8 RNA Expression, RSEM: HNF4α (log2 ) 08246 10 12 14 Log2(HNF4A TPM) –2 0 2 4 6 8 10 12 14 HNF4A RNA Expression, RSEM: CCL15 (log2)

Figure 6 CCL15 is a direct target of P1-HNF4A in gastric cancer. (A) Relative CCL15 mRNA was examined by qRT-PCR in MKN45 cells trans- fected with the control, P1-HNF4A, and P2-HNF4A overexpression lentiviruses. GAPDH was used as internal control (**P < 0.01, n = 3). (B) CCL15 promoter fragment (~2,000 bp) from Ensemble was predicted using JASPAR. Reporter genes containing the classical DR1 motifs or mutational sequences were constructed. (C) Negative control (NC) and the HNF4A2 overexpression plasmid were transiently transfected with CCL15 promoter reporter constructs with the DR1 motif or mutation for 24 h, and luciferase activity was assayed thereafter. The promoter activity was expressed as fold-induction (means ± SEM) compared to that of the NC, n = 3. (**P < 0.01). (D) A chromatin immunoprecipitation Cancer Biol Med Vol 18, No 2 May 2021 543

assay was performed to show the direct binding of HNF4A2 to the CCL15 promoter. M, Marker. (E) Paraffin sections of human gastric cancer tissues and adjacent non-tumor tissues were immunostained with anti-P1-HNF4A and anti-CCL15 antibodies. H&E staining. Scale bars = 200 μm and 50 μm. (F) The percent of CCL15-low and CCL15-high expression cases in P1-HNF4A negative and positive tissues, respectively according staining score (**P < 0.01). (G) The correlation of CCL15 and HNF4A in the GEPIA database. (H) The correlation of CCL15 and HNF4A in the cBioPortal database. (I) The correlation of CCL15 and HNF4A in the LinkedOmics database. promoted HNF4A expression by repressing NF-κB in GC28, this study, we demonstrated that P1- and P2-HNF4A exhib- and high HNF4A expression showed improved survival out- ited significantly different functions during GC progression. comes28. Similar inconsistent results have also been observed However, unlike in HCC36 and CRC37, P1-HNF4A was the in CRC27,29. These paradoxical conclusions indicated that the main oncogene, while P2-HNF4A expression was irrelevant to functions of HNF4A were characterized with time and spatial GC. Previous studies have demonstrated that P1-HNF4A plays distribution characteristics. The expression of HNF4A might key roles in both intestinal epithelium and liver cell develop- be dynamically regulated during GC initiation and progres- ment and differentiation16,23, so it is possible that tissue spe- sion. Our preliminary results found that HNF4A was upregu- cific transcription factors might suppress tumorigenesis. In lated in GC, especially in the early stage, which supported the contrast, ectopic expression of lineage-survival oncogenes oncogenic roles of HNF4A. may promote cancer by reactivating early developmental pro- A total of nine HNF4A splice variants have been generated grams. These results were consistent with the high expression via two alternative promoters (P1 and P2 promoters) with of P1-HNF4A in intestinal-type GC4, so we proposed that the tissue- and cell-specific expression patterns30,31. P1-driven function of HNF4A was tissue-specific, which might account HNF4A includes HNF4A1-6 isoforms and P2-driven HNF4A for the distinct roles of P1- and P2-HNF4A in different can- includes HNF4A7-9 isoforms respectively30. P1-driven cer development. Furthermore, integration of RNA-seq data HNF4A is expressed in the adult liver32 and kidney7, whereas suggests that this functional difference is due to differential P2-driven HNF4A is expressed in the fetal liver, the adult expression of certain target genes, with HNF4A2 upregulat- stomach7 and pancreas30. Both P1- and P2-HNF4A are ing genes involved in cell proliferation and inflammation, and expressed in the adult colon and intestine, but with distinct HNF4A8 upregulating genes involved in metabolism. In addi- location distributions30. The unique expression patterns of tion, P1-HNF4A typically has a more potent transactivation P1- and P2-driven HNF4A suggested distinct functional roles function than P2-HNF4A, which displays a similar pattern as of different HNF4A isoforms under physiological condi- in CRC. tions. Furthermore, previous research has demonstrated that Recent evidence suggests the involvement of HNF4A HNF4A isoforms showed significant metabolic differences in the pathophysiology of inflammatory diseases, such as and varied susceptibilities to colitis-associated colon cancer inflammatory bowel diseases24,38,39. HNF4A has also been in exon-swap mice33,34. In gastric carcinogenesis, both P1- shown to interact with the IL-6/STAT3 and NF-κB path- and P2-HNF4A were positive in gastric precancerous lesions ways6,25. Additionally, HNF4A was involved in IL-1β signal and tumor tissues9,35, which suggests that the HNF4A gene transduction through the regulation of IL-1R6. These results may exhibit oncogenic activity in GC. However, the relative indicated that inflammation regulation might be a poten- contributions of the different HNF4A isoforms (P1- and tial molecular mechanism involving the functional roles P2-HNF4A) have not been determined. of HNF4A. In this study, we identified CCL15 as a direct P1-HNF4A acts as a tumor suppressor both in HCC and P1-HNF4A target gene that promoted GC cell proliferation CRC, inhibiting cell proliferation and the inflammation path- and invasion. CCL15 has been shown to play a critical role way25,34. In contrast, P2-HNF4A promotes colitis and coli- in tumor progression in various cancer types, including liver tis-associated colon cancer development34. Previous studies and colorectal cancers40-42. However, the role of CCL15 in showed that stomach tissues were positive for only P2-HNF4A7. GC remains unclear. Accumulating evidence has suggested In contrast, Moore et al.19 indicated that HNF4A, especially that CCL15 may have a crucial role in the progression of P1-HNF4A, regulated gastric epithelium homeostasis and was tumor cells via the CC chemokine receptor42,43. Previous necessary for the maintenance of ZC secretory architecture. In studies have preliminarily indicated that CCL15 expression 544 Ni et al. Distinct function of HNF4A isoforms in gastric carcinogenesis is higher in gastric tumor tissues than in normal tissues44,45. Availability of data and materials In human CRC cells, loss of SMAD4 leads to the upregu- lation of CCL15 expression40,42,43. Additionally, in human The RNA-seq datasets generated and analyzed during the liver cancer cells, both inflammatory factors and epigenet- current study are available in the NCBI SRA repository ics regulate CCL15 expression41. Mechanistically, CCL15 (PRJNA612930). could promote tumor progression through the only recep- tor, CCR1, which is expressed by both immune cells such as monocytes41, lymphocytes46, neutrophils40, eosinophils47 References and tumor cells41,48. Our results revealed that CCL15 expres- sion was higher in GC tissues than in normal tissues, and 1. Bray F, Ferlay J, Soerjomataram I, Siegel RL, Torre LA, Jemal A. predicted a poor prognosis. CCL15 knockdown significantly Global cancer statistics 2018: GLOBOCAN estimates of incidence repressed GC cell proliferation, which indicated that CCL15 and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2018; 68: 394-424. might promote GC cell proliferation in an autocrine man- 2. Strong VE, Wu AW, Selby LV, Gonen M, Hsu M, Song KY, et al. ner (Supplementary Figure S11). Whether CCL15 promotes Differences in gastric cancer survival between the U.S. and China. J GC progression through recruitment of CCR1 immune cells Surg Oncol. 2015; 112: 31-7. in a paracrine manner requires further exploration. More 3. Chang HR, Nam S, Kook MC, Kim KT, Liu X, Yao H, et al. importantly, our results indicated the positive correlation of HNF4alpha is a therapeutic target that links AMPK to WNT P1-HNF4A with CCL15 in a subset of GC patients. Thus, signalling in early-stage gastric cancer. Gut. 2016; 65: 19-32. 4. Chia NY, Deng N, Das K, Huang D, Hu L, Zhu Y, et al. Regulatory aberrant P1-HNF4A expression in certain GC patients might crosstalk between lineage-survival oncogenes KLF5, GATA4 and provide a possible clue for the development of an effective GATA6 cooperatively promotes gastric cancer development. Gut. therapy by CCL15 signaling pathway blockade. 2015; 64: 707-19. 5. Xu C, Ooi WF, Qamra A, Tan J, Chua BY, Ho S, et al. HNF4alpha Conclusions pathway mapping identifies wild-type IDH1 as a targetable metabolic node in gastric cancer. Gut. 2020; 69: 231-42. 6. Ma L, Zeng J, Guo Q, Liang X, Shen L, Li S, et al. Mutual In summary, our study preliminarily showed the biologi- amplification of HNF4alpha and IL-1R1 composes an cal and clinical significance of different HNF4A isoforms inflammatory circuit in Helicobacter pylori associated gastric in GC. Compared to P2-HNF4A, P1-HNF4A might be the carcinogenesis. Oncotarget. 2016; 7: 11349-63. main driver oncogene in GC progression. Mechanistically, 7. Dean S, Tang JI, Seckl JR, Nyirenda MJ. Developmental and tissue- P1-HNF4A directly regulated the cytokine-receptor pathway, specific regulation of hepatocyte nuclear factor 4-alpha (HNF4- thereby promoting tumor growth and progression. CCL15 alpha) isoforms in rodents. Gene Expr. 2010; 14: 337-44. 8. Tanaka T, Jiang S, Hotta H, Takano K, Iwanari H, Sumi K, et al. could be regarded as a biomarker to guide therapy in GC Dysregulated expression of P1 and P2 promoter-driven hepatocyte patients with a high expression of P1-HNF4A, and pharma- nuclear factor-4alpha in the pathogenesis of human cancer. J ceutical intervention of the CCL15 signaling pathway may Pathol. 2006; 208: 662-72. provide a promising strategy to improve the prognoses of 9. Takano K, Hasegawa G, Jiang S, Kurosaki I, Hatakeyama K, Iwanari GC patients. H, et al. Immunohistochemical staining for P1 and P2 promoter- driven hepatocyte nuclear factor-4alpha may complement mucin Grant support phenotype of differentiated-type early gastric carcinoma. Pathol Int. 2009; 59: 462-70. 10. Zhang Q, Agoston AT, Pham TH, Zhang W, Zhang X, Huo X, et al. This work was supported by the National Natural Science Acidic bile salts induce epithelial to mesenchymal transition via Foundation of China (Grant No. 81873554) and Shaanxi VEGF signaling in non-neoplastic Barrett’s cells. Gastroenterology. Foundation for Innovation Team of Science and Technology 2019; 156: 130-44. (Grant No. 2018TD-003). 11. Liu X, Chen B, You W, Xue S, Qin H, Jiang H. The membrane bile acid receptor TGR5 drives cell growth and migration via activation of the JAK2/STAT3 signaling pathway in non-small cell lung cancer. Conflict of interest statement Cancer Lett. 2018; 412: 194-207. 12. Colletti M, Cicchini C, Conigliaro A, Santangelo L, Alonzi No potential conflicts of interest are disclosed. T, Pasquini E, et al. Convergence of Wnt signaling on the Cancer Biol Med Vol 18, No 2 May 2021 545

HNF4alpha-driven transcription in controlling liver zonation. 27. Vuong LM, Chellappa K, Dhahbi JM, Deans JR, Fang B, Bolotin Gastroenterology. 2009; 137: 660-72. E, et al. Differential effects of hepatocyte nuclear factor 4alpha 13. Bolotin E, Liao H, Ta TC, Yang C, Hwang-Verslues W, Evans isoforms on tumor growth and T-Cell Factor 4/AP-1 interactions JR, et al. Integrated approach for the identification of human in human colorectal cancer cells. Mol Cell Biol. 2015; 35: 3471-90. hepatocyte nuclear factor 4alpha target genes using protein binding 28. Li D, Zhao X, Xiao Y, Mei H, Pu J, Xiang X, et al. Intelectin 1 microarrays. Hepatology. 2010; 51: 642-53. suppresses tumor progression and is associated with improved 14. Marable SS, Chung E, Adam M, Potter SS, Park JS. Hnf4a deletion survival in gastric cancer. Oncotarget. 2015; 6: 16168-82. in the mouse kidney phenocopies Fanconi renotubular syndrome. 29. Darsigny M, Babeu JP, Seidman EG, Gendron FP, Levy E, Carrier JCI Insight. 2018; 3: e97497. J, et al. Hepatocyte nuclear factor-4alpha promotes gut neoplasia 15. Jennings RE, Berry AA, Gerrard DT, Wearne SJ, Strutt J, in mice and protects against the production of reactive oxygen Withey S, et al. Laser capture and deep sequencing reveals the species. Cancer Res. 2010; 70: 9423-33. transcriptomic programmes regulating the onset of pancreas and 30. Ko HL, Zhuo Z, Ren EC. HNF4alpha combinatorial isoform liver differentiation in human embryos. Stem Cell Rep. 2017; 9: heterodimers activate distinct gene targets that differ from their 1387-94. corresponding homodimers. Cell Rep. 2019; 26: 2549-57. 16. Chen L, Toke NH, Luo S, Vasoya RP, Fullem RL, Parthasarathy A, 31. Guo S, Lu H. Novel mechanisms of regulation of the expression et al. A reinforcing HNF4-SMAD4 feed-forward module stabilizes and transcriptional activity of hepatocyte nuclear factor 4alpha. J enterocyte identity. Nat Genet. 2019; 51: 777-85. Cell Biochem. 2019; 120: 519-32. 17. Armour SM, Remsberg JR, Damle M, Sidoli S, Ho WY, Li Z, et al. 32. Li K, Zhang H, Wang Y, Wang Y, Feng M. Differential expression An HDAC3-PROX1 corepressor module acts on HNF4alpha to of HNF4alpha isoforms in liver stem cells and hepatocytes. J Cell control hepatic triglycerides. Nat Commun. 2017; 8: 549. Biochem. 2006; 99: 558-64. 18. Xu Y, Zalzala M, Xu J, Li Y, Yin L, Zhang Y. A metabolic stress- 33. Briancon N, Weiss MC. In vivo role of the HNF4alpha AF-1 inducible miR-34a-HNF4alpha pathway regulates lipid and activation domain revealed by exon swapping. EMBO J. 2006; 25: lipoprotein metabolism. Nat Commun. 2015; 6: 7466. 1253-62. 19. Moore BD, Khurana SS, Huh WJ, Mills JC. Hepatocyte nuclear 34. Chellappa K, Deol P, Evans JR, Vuong LM, Chen G, Briancon N, factor 4alpha is required for cell differentiation and homeostasis in et al. Opposing roles of nuclear receptor HNF4alpha isoforms in the adult mouse gastric epithelium. Am J Physiol Gastrointest Liver colitis and colitis-associated colon cancer. eLife. 2016; 5: e10903. Physiol. 2016; 311: 267-75. 35. Kojima K, Kishimoto T, Nagai Y, Tanizawa T, Nakatani Y, Miyazaki 20. San RA, Aronson BE, Krasinski SD, Shivdasani RA, Verzi MP. M, et al. The expression of hepatocyte nuclear factor-4alpha, a Transcription factors GATA4 and HNF4A control distinct aspects developmental regulator of visceral endoderm, correlates with the of intestinal homeostasis in conjunction with transcription factor intestinal phenotype of gastric adenocarcinomas. Pathology. 2006; CDX2. J Biol Chem. 2015; 290: 1850-60. 38: 548-54. 21. Cattin AL, Le Beyec J, Barreau F, Saint-Just S, Houllier A, Gonzalez 36. Fekry B, Ribas-Latre A, Baumgartner C, Deans JR, Kwok C, Patel FJ, et al. Hepatocyte nuclear factor 4alpha, a key factor for P, et al. Incompatibility of the circadian protein BMAL1 and homeostasis, cell architecture, and barrier function of the adult HNF4alpha in hepatocellular carcinoma. Nat Commun. 2018; 9: intestinal epithelium. Mol Cell Biol. 2009; 29: 6294-308. 4349. 22. Farkas AE, Hilgarth RS, Capaldo CT, Gerner-Smidt C, Powell 37. Chellappa K, Jankova L, Schnabl JM, Pan S, Brelivet Y, Fung CL, DR, Vertino PM, et al. HNF4alpha regulates claudin-7 protein et al. Src tyrosine kinase phosphorylation of nuclear receptor expression during intestinal epithelial differentiation. Am J Pathol. HNF4alpha correlates with isoform-specific loss of HNF4alpha 2015; 185: 2206-18. in human colon cancer. Proc Natl Acad Sci U S A. 2012; 109: 23. Lau HH, Ng N, Loo L, Jasmen JB, Teo A. The molecular functions 2302-7. of hepatocyte nuclear factors – In and beyond the liver. J Hepatol. 38. Chahar S, Gandhi V, Yu S, Desai K, Cowper-Sal-lari R, Kim Y, 2018; 68: 1033-48. et al. Chromatin profiling reveals regulatory network shifts and a 24. Babeu JP, Boudreau F. Hepatocyte nuclear factor 4-alpha protective role for hepatocyte nuclear factor 4alpha during colitis. involvement in liver and intestinal inflammatory networks. World J Mol Cell Biol. 2014; 34: 3291-304. Gastroenterol. 2014; 20: 22-30. 39. Meddens CA, Harakalova M, van den Dungen NA, Foroughi AH, 25. Hatziapostolou M, Polytarchou C, Aggelidou E, Drakaki Hijma HJ, Cuppen EP, et al. Systematic analysis of chromatin A, Poultsides GA, Jaeger SA, et al. An HNF4alpha-miRNA interactions at disease associated loci links novel candidate genes to inflammatory feedback circuit regulates hepatocellular oncogenesis. inflammatory bowel disease. Genome Biol. 2016; 17: 247. Cell. 2011; 147: 1233-47. 40. Yamamoto T, Kawada K, Itatani Y, Inamoto S, Okamura R, 26. Yin C, Wang PQ, Xu WP, Yang Y, Zhang Q, Ning BF, et al. Iwamoto M, et al. Loss of SMAD4 promotes lung metastasis of Hepatocyte nuclear factor-4alpha reverses malignancy of colorectal cancer by accumulation of CCR1+ tumor-associated hepatocellular carcinoma through regulating miR-134 in the through CCL15-CCR1 axis. Clin Cancer Res. 2017; 23: DLK1-DIO3 region. Hepatology. 2013; 58: 1964-76. 833-44. 546 Ni et al. Distinct function of HNF4A isoforms in gastric carcinogenesis

41. Liu LZ, Zhang Z, Zheng BH, Shi Y, Duan M, Ma LJ, et al. CCL15 45. Raja UM, Gopal G, Shirley S, Ramakrishnan AS, Rajkumar T. recruits suppressive to facilitate immune escape and Immunohistochemical expression and localization of / disease progression in hepatocellular carcinoma. Hepatology. 2019; chemokines/growth factors in gastric cancer. Cytokine. 2017; 89: 82-90. 69: 143-59. 46. Dyer DP, Medina-Ruiz L, Bartolini R, Schuette F, Hughes CE, Pallas 42. Liang Q, Tang C, Tang M, Zhang Q, Gao Y, Ge Z. TRIM47 is K, et al. redundancy and specificity are context up-regulated in colorectal cancer, promoting ubiquitination and dependent. Immunity. 2019; 50: 378-89. degradation of SMAD4. J Exp Clin Cancer Res. 2019; 38: 159. 47. Fulkerson PC, Schollaert KL, Bouffi C, Rothenberg ME. IL-5 43. Itatani Y, Kawada K, Fujishita T, Kakizaki F, Hirai H, triggers a cooperative cytokine network that promotes eosinophil Matsumoto T, et al. Loss of SMAD4 from colorectal cancer cells precursor maturation. J Immunol. 2014; 193: 4043-52. promotes CCL15 expression to recruit CCR1+ myeloid cells 48. Li BH, Garstka MA, Li ZF. Chemokines and their receptors and facilitate liver metastasis. Gastroenterology. 2013; 145: promoting the recruitment of myeloid-derived suppressor cells into 1064-75. the tumor. Mol Immunol. 2020; 117: 201-15. 44. Camargo MC, Song M, Shimazu T, Charvat H, Yamaji T, Sawada N, et al. Circulating inflammation markers and risk of gastric Cite this article as: Ni Z, Lu W, Li Q, Han C, Yuan T, Sun N, et al. Analysis of the and esophageal cancers: a case-cohort study within the Japan HNF4A isoform-regulated transcriptome identifies CCL15 as a downstream public health center-based prospective study. Cancer Epidemiol target in gastric carcinogenesis. Cancer Biol Med. 2021; 18: 530-546. Biomarkers Prev. 2019; 28: 829-32. doi: 10.20892/j.issn.2095-3941.2020.0131 Cancer Biol Med Vol 18, No 2 May 2021 1

Supplementary materials

HNF4α copy number in TCGA Gastric HNF4α copy number in Deng Gastric 2.0 0. No value (2) 1.0 0. No value (3) 1. Blood (236) 1. Gastric tissue (98) 1.5 2. Stomach (94) 2. Diffuse (36) s 3. Diffuse (45) 0.5 3. Adenocarcinoma (63) 1.0 4. Adenocarcinoma (173) 4. Gastric cancer (42)

5. Intestinal type (41) 5. Intestinal type (38) 0.5 6. Papillary (3) 0.0 6. Mixed (11) 7. Tubular (31) log2 copy number units 0.0 log2 copy number unit 8. Mucinous (12)

–0.5 –0.5 0 1 2345 6078 123456

P = 4.06E-7 P = 6.19E-23 P = 7.22E-8 P = 0.105 P = 8.51E-7 P = 0.002 P = 2.74E-4 P = 4.59E-7 P = 3.44E-5 P = 3.73E-5 P = 0.007 Fold = 1.104 Fold = 1.191 Fold = 1.279 Fold = 1.127 Fold = 1.283 Fold = 1.129 Fold = 1.040 Fold = 1.086 Fold = 1.084 Fold = 1.104 Fold = 1.125

HNF4α expression in Cho Gastric HNF4α expression in DErrico Gastric 1.0 2.5 0. No value (19) 0. No value (33)

1. Diffuse (31) 2.0 1. Diffuse (6) 2. Adenocarcinoma (4) 2. Intestinal type (26) 1.5 0.5 3. Intestinal type (20) 3. Mixed (4) 4. Mixed (10) 1.0

5. Stromal tumor (6) 0.5 0.0 0.0

–0.5 log2 median-centered intensit y log2 median-centered intensity –0.5 –1.0 0123 45 0 123 P = 2.47E-4 P = 0.029 P = 0.002 P = 0.001 P = 0.157 P = 2.43E-5 P = 0.136 Fold = 1.120 Fold = 1.145 Fold = 1.094 Fold = 1.144 Fold = 1.239 Fold = 2.518 Fold = 2.120

HNF4α expression in Cui Gastric HNF4α expression in Wang Gastric 4.5 0.5 0. No value (80) 1. Gastric mucosa (12) 4.0 0.0 3.5 1. Gastric cancer (80) 2. Gastric tissue (3) y 3.0 –0.5 2.5 3. Gastric cancer (12) 2.0 –1.0 1.5 –1.5 1.0 0.5 –2.0 0.0 –0.5 –2.5 –1.0 –3.0 –1.5 –2.0 –3.5

log2 median-centered intensit –2.5 –4.0 –3.0 log2 median-centered intensity –3.5 –4.5 1 2 123 P = 0.037 P = 0.039 Fold = 1.264 Fold = 1.584

Figure S1 The expression of HNF4A in the Oncomine database including TCGA Gastric, Deng Gastric, Cho Gastric, DErrico Gastric, Cui Gastric, and Wang Gastric cohorts. 2 Ni et al. Distinct function of HNF4A isoforms in gastric carcinogenesis

AB** 200 ** 250 ** ** ** 150 200

150 100 **

100 50

50 Transcript per millio n Transcript per millio n 0 0

–50 –50 Normal Primary tumor Normal Grade 1 Grade 2 Grade 3 (n = 34) (n = 415) (n = 34) (n = 12) (n = 148) (n = 246)

TCGA samples TCGA samples

CD** ** 200 ** ** 250 ** 150 200 ** ** 100 150

100 50 50 Transcript per millio n

0 Transcript per millio n 0

–50 –50 Normal Stage 1 Stage 2 Stage 3 Stage 4

(n = 34) (n = 18) (n = 123) (n = 169) (n = 41) Normal = 34) (n AdenoNOS (n = 155) (n = 69) = 12) = 76) AdenoDiffuse IntAdenoNOS(n = 73) n = 20) TCGA samples (n ( n (n = 7) IntAdenoTubular ( AdenoSignetRing IntAdenoPapillary IntAdenoMucinous TCGA samples

Figure S2 The expression of HNF4A in the Ualcan database according to tumor (A), grade (B), stage (C), and tumor type (D) (**P < 0.01). Cancer Biol Med Vol 18, No 2 May 2021 3

A F value = 0.101 Pr(>F) = 0.96 8 8

6 6

4 4

2 2

0 0

STAD Stage IStage II Stage III Stage IV (num(T) = 408; num(n) = 211)

B 7.7

STAD

0.0 8 8 1 2 6 5 5 5 4 (HNF4A–002) (HNF4A–001) (HNF4A–003) (HNF4A–006) (HNF4A–004) (HNF4A–005) (HNF4A–008) (HNF4A–007) (HNF4A–201) ENST00000316099. ENST00000316673. ENST00000372920. ENST00000415691. ENST00000443598. ENST00000457232. ENST00000609262. ENST00000609795. ENST00000619550.

STAD 8 8 1 2 6 5 5 5 4 (HNF4A–002) (HNF4A–001) (HNF4A–003) (HNF4A–006) (HNF4A–004) (HNF4A–005) (HNF4A–008) (HNF4A–007) (HNF4A–201) ENST00000316099. ENST00000316673. ENST00000372920. ENST00000415691. ENST00000443598. ENST00000457232. ENST00000609262. ENST00000609795. ENST00000619550.

Figure S3 The expression of HNF4A in the GEPIA database. (A) The expression of HNF4A in tumor and non-tumor cells (*P < 0.05). (B) The expression of different HNF4A isoforms in stomach adenocarcinomas. 4 Ni et al. Distinct function of HNF4A isoforms in gastric carcinogenesis

A

Adjacent Tumor

P2-HNF4α

B C 300 n.s. 150 P2-HNF4α Low High e l 100

200 vi va

cent sur 50 100 r Pe P2-HNF4α H-scor 0 0 0306090 120 150 Tumor Adjacent Survival time (month)

Figure S4 Expression and prognostic value of P2-HNF4A in gastric cancer (GC). (A) Representative pictures of the expression of P2-HNF4A in tumor and adjacent non-tumor tissues of GC patients. Scale bars = 200 μm and 50 μm. (B) H-score of P2-HNF4A in tumor and adjacent non-tumor tissues using tissue microarray using immunohistochemical analysis (n.s., nonsignificant). (C) The prognostic value of P2-HNF4A in GC patients according to low and high expressions (P > 0.05). Cancer Biol Med Vol 18, No 2 May 2021 5

A HNF4a2_UP-vs.-control HNF4a2_UP-vs.-HNF4a8_UP A (1127) (1073) MKN45 AZ521 AGS GES- 1 BGC823 N8 7 SGC7901 669

P1-HNF4α 305 245

82 P2-HNF4α 71 77 β-actin

65 B SGC7901 BGC823 HNF4a8_UP-vs.-control NC P1 P2 NC P1 P2 (295)

P1-HNF4α B Statistic differently expressed gene 816 800 763

P2-HNF4α

600

Type pERK1/2 400 Up Down 310 311

p-p38 200 190

Differently expressed gene number 105

p-JNK 0

HNF4α2-UP vs. HNF4a8-UP HNF4α2-UP vs. control HNF4α8-UP vs. control

GAPDH Figure S6 RNA-seq screen of target genes downstream of P1- and P2-HNF4A overexpression in SGC7901 cells. (A) A Venn diagram showing the overlap between genes dysregulated by overexpres- Figure S5 The effects of P1- and P2-HNF4A on mitogen-activated sion of P1-HNF4A or P2-HNF4A. (B) Statistics of differently-ex- protein kinase (MAPK) signaling activation. (A) The relative protein pressed gene numbers between negative control, P1-HNF4A or expression levels of P1- and P2-HNF4A in normal gastric epithe- P2-HNF4A overexpression SGC7901 cells. lial cell line (GES-1) and gastric cancer cell lines (SGC7901, N87, AGS, AZ521, MKN45 and BGC823) were detected by Western blot. β-actin was used as internal control. (B) The protein expression of MAPK in SGC7901 and BGC823 after P1- and P2-HNF4A overex- pression was detected by Western blot. GAPDH was used as an internal control. 6 Ni et al. Distinct function of HNF4A isoforms in gastric carcinogenesis

8 7 CCL5 CCL15 CCL20 CSF2RA

8 10 6

6 5 8 6 4 4 6 4 3 4 2 2 2 2 1

0 0 0 0

CXCL1 CXCL8 CXCL16 CXCR4 10 10 8 8

8 8 6 6 6 6

4 4 4 4

2 2 2 2

0 0 0 0

12 10 GDF15 IL1RN IL2RB 10 IL2RG 6 10 8 8 5 8 6 4 6 6 3 4 4 4 2 2 2 2 1

0 0 0 0

IL22RA1 INHBA TNFRSF1B TNFRSF14 6 3.5

6 3.0 5 6

2.5 4 4 4 2.0 3 1.5 2 2 2 1.0

1 0.5

0 0 0 0.0 T = 408 n = 211 T = 408 n = 211 T = 408 n = 211 T = 408 n = 211

Figure S7 The expressions of CCL5, CCL15, CCL20, CSF2RA, CXCL1, CXCL8, CXCL16, CXCR4, GDF15, IL1RN, IL2RB, IL2RG, IL22RA1, INHBA, TNFRSF1B, and TNFRSF14 in stomach adenocarcinomas (STAD) in the GEPIA database (*P < 0.05). Cancer Biol Med Vol 18, No 2 May 2021 7

CXCR4 CXCL16 IL18 CXCL1 250 ** 500 350 400 ** ** 300 ** 200 400 300 250 150 300 200 200 100 200 150 100 100 50 100 50 Transcript per millio n Transcript per millio n Transcript per millio n Transcript per millio n 0 0 0 0 –50 –100 –50 –100

CCL20 CCL15 8 CD70 400 ** 60 ** 250 CCL5 ** ** 50 6 300 200 40 150 4 200 30 100 2 100 20 50 10 Transcript per millio n Transcript per millio n Transcript per millio n

0 0 Transcript per millio n 0 0

–2 –100 –10 –50

12.5 TGFB2 INHBE 10 TNFRSF9 100 TNFRSF1B 0.6 ** ** ** ** 8 80 10 0.5 0.4 6 60 7.5 0.3 4 40 5 0.2 2 20 2.5 0.1 Transcript per millio n Transcript per millio n Transcript per millio n 0 Transcript per millio n 0 0 0

–2 –20 –2.5 –0.1

IL32 70 IL12RA1 IL18RAP 5 IL13RA2 800 3 ** ** 60 ** ** 2.5 4 600 50 2 3 40 400 1.5 30 2 1 200 20 1 10 0.5 Transcript per millio n Transcript per millio n 0 Transcript per millio n Transcript per millio n 0 0 0

–200 –10 –0.5 –1

50 IL2RB 250 IL1RN 100 IL1R1 IL2RG ** 500 ** ** ** 40 200 80 400 30 150 60 300 20 100 40 200 10 50 20 100 Transcript per millio n 0 Transcript per millio n 0 Transcript per millio n 0 Transcript per millio n 0 –10 –50 –20 –100

BMP2 INHBB 25 CSF2RA 60 15 ** 8 IL18R1 ** ** ** 20 50 12.5 6 40 10 15 30 7.5 4 10 20 5 2 5 10 2.5 Transcript per millio n Transcript per millio n Transcript per millio n Transcript per millio n 0 0 0 0

–5 –10 –2.5 –2

TNFRSF14 500 GDF15 150 100 INHBA 12.5 IL11 ** ** ** ** 400 125 80 10 100 300 60 7.5 75 200 40 5 50 100 20 2.5 25 Transcript per millio n Transcript per millio n Transcript per millio n Transcript per millio n 0 0 0 0

–100 –25 –20 –2.5 Normal Primary tumor Normal Primary tumor Normal Primary tumor Normal Primary tumor (n = 34) (n = 415) (n = 34) (n = 415) (n = 34) (n = 415) (n = 34) (n = 415)

Figure S8 The expressions of BMP2, CCL5, CCL15, CCL20, IL8, CD70, CSF2RA, CXCL1, CXCL16, CXCR4, TGFB2, GDF15, IL11, IL1R1, IL1RN, IL13RA2, IL32, IL2RB, IL2RG, IL22RA1, IL18R1, IL18RAP, INHBA, INHBB, INHBE, TNFRSF9, TNFRSF1B, and TNFRSF14 in stomach adenocarcino- mas (STAD) in the Ualcan database (**P < 0.01). 8 Ni et al. Distinct function of HNF4A isoforms in gastric carcinogenesis

200 150 100 GES-1 50 SGC7901 N87 L15 mRNA AGS

CC 4 AZ521 3

lative MKN45 2 Re BGC823 1 0

Figure S9 The relative mRNA expression levels of CCL15 in a nor- mal gastric epithelial cell line (GES-1) and gastric cancer cell lines (SGC7901, N87, AGS, AZ521, MKN45, and BGC823) was detected by qRT-PCR (n = 3). GAPDH was used as an internal control.

A B 150 High CCL15 Low CCL15 120 ) Adjacent 90 ∗∗

60 Percent (%

30 Tumo r 0 Tumor Adjacent

Figure S10 The expression of CCL15 in a gastric cancer tissue microarray. (A) The expression of CCL15 was examined in tumor and non-­ tumor adjacent tissues by immunohistochemical (IHC) analysis using a gastric cancer tissue microarray. Scale bars = 200 μm and 50 μm. (B) CCL15 staining in tumor and adjacent non-tumor tissues was semiquantitatively calculated according to the staining intensity and ­percentage of positive cells (**P < 0.01). IHC scores < 6 and ≥ 6 were considered low and high expressions, respectively.

P2-HNF4α

Epithelial cells

Paracrine

P1-HNF4α Tumor cells CCL15 Immune cells

Cytokines Progression Crosstalk

Figure S11 Schematic model of the proposed theory in this study.