Abstract

   During neoplastic development, a multitude of changes in genome-encoded
   information are progressively selected to confer growth and survival
   advantages to tumor cells. microRNAs-mRNAs regulatory networks, given
   their role as a critical layer of robust gene expression control, are
   frequently altered in neoplasm. However, whether and how these gene
   perturbations impact metabolic homeostasis remains largely unresolved.

   Methods: Through targeted miRNA expression screening, we uncovered an
   oral squamous cell carcinoma (OSCC)-associated miRNAome, among which
   miR-31-5p was identified based on extent of up-regulation, functional
   impact on OSCC cell migration and invasion, and direct regulation of
   the rate-limiting enzyme in peroxisomal β-oxidation, ACOX1.

   Results: We further found that both miR-31-5p and ACOX1 underpin, in an
   antagonistic manner, the overall cellular lipidome profiles as well as
   the migratory and invasive abilities of OSCC cells. Interestingly, the
   extracellular levels of prostaglandin E2 (PGE2), a key substrate of
   ACOX1, were controlled by the miR-31-5p-ACOX1 axis, and were shown to
   positively influence the extent of cell motility in correlation with
   metastatic status. The promigratory effect of this metabolite was
   mediated by an elevation in EP1-ERK-MMP9 signaling. Of note, functional
   significance of this regulatory pathway was further corroborated by its
   clinicopathologically-correlated expression in OSCC patient specimens.

   Conclusions: Collectively, our findings outlined a model whereby
   misregulated miR-31-5p-ACOX1 axis in tumor alters lipid metabolomes,
   consequently eliciting an intracellular signaling change to enhance
   cell motility. Our clinical analysis also unveiled PGE2 as a viable
   salivary biomarker for prognosticating oral cancer progression, further
   underscoring the importance of lipid metabolism in tumorigenesis.

   Keywords: miRNA-mRNA network, ACOX1, lipid metabolism, oral squamous
   cell carcinoma, prostaglandins

Introduction

   MicroRNAs (miRNAs) are non-coding, single-stranded RNAs of 18-24
   nucleotides that serve as pivotal gene regulators in mammals and other
   multicellular organisms [41]^1^, [42]^2. miRNAs may silence gene
   expression at the post-transcriptional level through either perfect or
   imperfect binding to the 3' untranslated region (3' UTR) of messenger
   RNAs [43]^2. To date, 2,588 different mature human miRNAs have been
   identified and published in miRBase database and reportedly target more
   than 30% of the protein-coding genes. Given their extensive involvement
   in many biological processes, including proliferation, apoptosis,
   differentiation, and metabolic homeostasis, mutations or mis-expression
   of miRNAs have been associated with the progression of various types of
   human cancers [44]^1^, [45]^3.

   In an effort to map, from a systems perspective, the miRNA-mRNA
   networks underlying tumor pathogenesis, we designed and performed a
   targeted expression profiling of miRNAomes on the basis of in-house
   clinical specimens as well as a public cancer database. We focused this
   systems approach on oral cancer for the following reasons: 1) With
   approximately 300,373 new cases in year 2012, this globally prevalent
   cancer stands as a growing health concern [46]^4^, [47]^5. Despite
   progress in diagnosis and therapy strategies in the last three decades,
   the five-year survival rate in patients has not been significantly
   improved due to aggressive local invasion, distant metastasis, and even
   recurrence [48]^6. 2) A unique etiology - tobacco usage, betel quid
   chewing, alcohol consumption, infection with high-risk genotypes of
   human papillomavirus (HPV), and dietary intake - has been linked to
   oral squamous cell carcinoma (OSCC), the major subtype (>90%) of oral
   cancer [49]^7^, [50]^8. In line with this disease complexity, multiple
   dysregulated molecular networks underlie OSCC neoplastic development,
   including the epidermal growth factor receptor (EGFR), Ras, NF-κB,
   STAT, Wnt/β-catanin, TGF-β, and PI3K-AKT-mTOR signaling pathways
   [51]^9. However, the contribution of the transcriptome and/or
   metabolome alteration to OSCC incidence or progression has not yet been
   fully explored. 3) Given the anatomical sites of the disease and also
   because of the reportedly high stability of miRNAs in bodily fluids,
   cumulative studies have emphasized the potential of these small RNAs as
   non-invasive biomarkers for predictive monitoring of OSCC
   [52]^10^-[53]^12. For example, plasma and salivary levels of miR-31-5p
   are elevated in OSCC and could be abrogated after tumor resection
   [54]^13^, [55]^14.

   Our approach uncovered several aberrantly expressed miRNAs, some of
   which have been identified previously by numerous independent studies
   [56]^10^, [57]^12, thus strongly supporting the feasibility of our
   experimental paradigm. Among them, miR-21-5p is a frequently
   over-expressed oncogenic miRNA in OSCC as well as head and neck
   squamous cell carcinoma (HNSCC) samples. miR-21-5p plays crucial roles
   in cancer cell survival and resistance to anticancer agents, and its
   increased expression is related to poor prognosis and reduced survival
   rate in patients with HNSCC [58]^11^, [59]^15. miR-31-5p is another
   significantly up-regulated miRNA in oral cancer. In HNSCC, aberrantly
   elevated expression of miR-31-5p has been demonstrated to directly
   target the 3' UTR of factor-inhibiting hypoxia-inducible factor (FIH),
   an inhibitor of hypoxia-inducible factor (HIF), leading to activation
   of HIF downstream pathway under normoxic conditions [60]^11^, [61]^16.
   We then examined how this altered miRNAome could systematically impact
   the transcriptome and tumor biology. Interestingly, aside from
   oncogenic signaling and cell proliferation, an enrichment of predicted
   targets in fatty acid and peroxisome-associated metabolic pathways was
   evident, suggesting that distinct and hitherto unknown cellular changes
   may arise as a consequence of these OSCC-associated miRNA alterations.

   Peroxisomes are single membrane-enclosed organelles that can be found
   in virtually all eukaryotic cells. Previous studies have demonstrated
   an increased number of peroxisomes in tissues with active lipid
   metabolism, such as liver and adipose tissues [62]^17. The primarily
   metabolic functions of mammalian peroxisomes include β-oxidation of a
   broad range of fatty acids, particularly the very long chain fatty
   acids (VLCFAs), α-oxidation of branched chain fatty acids, biosynthesis
   of etherphospholipids and bile acids, and regulation of redox
   homeostasis [63]^17. Both the peroxisomal and mitochondrial β-oxidation
   undergo four identical steps for degradation of acyl-CoA: (1)
   desaturation of acyl-CoA; (2) hydration of trans-2-enoyl-CoA; (3)
   dehydrogenation of 3-hydroxyacyl-CoA; (4) thiolysis of β-ketoacyl-CoA
   to form acetyl-CoA and a new fatty acyl-CoA with two fewer carbons
   [64]^17^, [65]^18. However, the peroxisomal β-oxidation is distinct
   from its mitochondrial counterpart in several aspects of enzymology and
   catabolic substrates. For instance, mitochondrial β-oxidation
   contributes to the catabolism of a bulk of lipid substrates, such as
   short, medium and long chain fatty acids, whereas the VLCFAs are
   exclusively oxidized in peroxisomes. Additionally, peroxisomal
   β-oxidation system in mammals also engages in the metabolic degradation
   of 2-methyl branched chain fatty acids, bile acid intermediates,
   polyunsaturated fatty acids (PUFA), dicarboxylic fatty acids,
   xenobiotics, and eicosanoids (i.e. prostaglandins and leukotrienes)
   [66]^18^, [67]^19. Furthermore, the first β-oxidation step in
   mitochondria is catalyzed by acyl-CoA dehydrogenase (ACAD) and
   concomitantly reduces FAD+ to FADH2, which are used to produce ATP
   through the electron transfer chain, whereas the initial step of
   peroxisomal lipid metabolism is carried out by acyl-CoA oxidase (ACOX),
   also culminating in formation of FADH2. However, instead of acting as
   an ATP generating source, FADH2 in peroxisomes is further re-oxidized,
   with its electrons transferred to oxygen and producing H[2]O[2] as a
   by-product [68]^17^, [69]^18^, [70]^20, suggesting a role of
   peroxisomal lipid metabolism in redox regulation.

   There are at least three acyl-CoA oxidases encoded by the mammalian
   genomes - ACOX1 (with two alternative isoforms, a/b or I/II), ACOX2,
   ACOX3 and ACOX4 [71]^19 - which are different in tissue distribution
   and substrate specificity. Unlike other peroxisomal acyl-CoA oxidases,
   ACOX1 is commonly expressed in most tissues and is specific for the
   desaturation of straight chain CoA esters of VLCFAs, dicarboxylic
   acids, PUFAs, and prostaglandins. In contrast, the 2-methyl branched
   compounds, such as pristanoyl-CoA, cholestanoyl-CoA and CoA esters of
   other bile acids in S configuration, are preferentially oxidized by
   ACOX2 [72]^19. ACOX1 deficiency is connected with a lethal human
   disease, called pseudoneonatal adrenoleukodystrophy (P-NALD). The
   biochemical symptoms of this autosomal recessive disorder are
   accumulated VLCFAs and enlarged peroxisomes [73]^21. Patients with
   P-NALD often exhibit neurodevelopmental abnormalities, including
   hypotonia and seizures, and die early. Recently, it has become evident
   that activation of ER stress contributes to the progression of
   hepatocarcinogenesis in homozygous ACOX1 knockout mice [74]^22. While
   this implies a tumor-suppressive role for this oxidase, precise
   functional implication of ACOX1-mediated metabolism in human cancers
   remains largely elusive.

   In this study, we identified ACOX1 as a new target of miR-31-5p and
   further connected this microRNA to the maintenance of proper lipid
   metabolic profiles. Our results further revealed that this functional
   antagonism underpins the OSCC cell migratory and invasive capacity.
   Mechanistically, PGE2, one of the substrates of ACOX1, acts as a
   downstream mediator of these oncogenic changes by conferring an
   elevated promigratory ERK-MMP9 signaling. Clinicopathological
   significance of this regulatory pathway was further reinforced by
   distinct expression alterations of the molecular components in OSCC
   patient specimens, which are in correspondence with their functional
   consequences. In summary, our findings demonstrated that miR-31-5p acts
   as a tumor-promoting miRNA in OSCC and outlined a downstream regulatory
   network whereby tumor cell motility is favored by altered lipid
   metabolomes. Our studies also uncovered a potentially viable biomarker
   in salivary PGE2 for OSCC disease monitoring.

Materials and Methods

miRNA expression profiling analysis

   The expression levels of 270 miRNAs were determined using the stem-loop
   RT-PCR method. First, total RNA was extracted from normal and cancerous
   specimens of oral origin. 1 μg of total RNA from tissues was
   subsequently subjected to a reverse transcription reaction with pooled
   miRNA-specific stem-looped RT primers. The reversely transcribed cDNA
   was diluted and analyzed by SYBR-based real-time PCR on an Applied
   Biosystems 7900 Fast PCR System (Foster City, CA, USA). After global
   median normalization, the expression levels of individual miRNAs were
   expressed as 40 minus the value of the threshold cycle (Ct), defined as
   the cycle number where the SYBR green intensity reaches a threshold of
   0.2. For screening the differentially expressed miRNAs, three-way
   analysis of variance (ANOVA) was performed and the false discovery rate
   (FDR)-based multiple testing correction (q-value) was used to adjust
   the statistical confidence measures as previously described [75]^23.
   All the data processing and statistical analyses described above were
   performed with Partek Genomics Suite 6.5 (Partek Inc.; St. Louis, MO,
   USA). The data have been submitted to the NCBI Gene Expression Omnibus
   ([76]https://www.ncbi.nlm.nih.gov/geo/) under accession no.
   [77]GSE103931.

Cell culture

   Three human OSCC cell lines were used in this study: OECM-1, a gingival
   epidermoid carcinoma line from a Taiwanese patient, was maintained in
   Dulbecco's Modified Eagle Medium (DMEM), while tongue squamous cell
   lines SCC25 and SCC4 were cultured in DMEM/F12 medium. These OSCC cell
   lines were supplemented with 10% heat-inactivated fetal bovine serum
   (FBS; Biological industries; Cromwell, CT, USA), 1X non-essential amino
   acids and 1 mM sodium pyruvate. Except for OECM-1, the complete growth
   media for SCC25 and SCC4 require additional supplement of 400 ng/mL
   hydrocortisone (Sigma-Aldrich; St. Louis, MO, USA). Human gingival
   epithelial S-G cells were cultured in DMEM with 10% heat-inactivated
   FBS. All cells were incubated in fully humidified air containing 5%
   carbon dioxide at 37°C. Unless otherwise specified, all culture media
   and reagents were purchased from Thermo Fisher Scientific (Waltham, MA,
   USA).

Reagents

   Locked nucleic acids (LNA) against miR-31-5p and scramble control were
   purchased from Exiqon A/S (Vedbaek, Denmark), and the small interfering
   RNAs (siRNAs) against ACOX1 were obtained from Thermo Fisher Scientific
   (Waltham, MA, USA). Lipofectamine 2000, Lipofectamine RNAiMAX, and all
   RT reagents were purchased from Invitrogen (Carlsbad, CA, USA). PGE2
   and PGE2 monoclonal ELISA kit were purchased from Cayman Chemical (Ann
   Arbor, MI, USA). All other chemicals were purchased from Sigma-Aldrich
   (St. Louis, MO, USA).

RNA extraction and quantitative reverse transcription PCR (RT-qPCR)

   Total RNA was isolated from cells by the TRIzol reagent (Thermo Fisher
   Scientific) according to manufacturer's instructions. The concentration
   of extracted RNA was determined using NanoDrop spectrophotometer. For
   miRNA RT reaction, 1 μg of total RNA was first converted into cDNA by
   Superscript III reverse transcriptase (Invitrogen), in which
   miRNA-specific stem-looped RT primers were used. RT reaction was done
   under the following condition: 16°C for 30 min, followed by 50 cycles
   of 20°C for 30 s, 42°C for 30 s and 50°C for 1 s. The enzyme was then
   inactivated at 70°C for 10 min. For mRNA, 1 μg of total RNA was used as
   templates to synthesize first-strand cDNA using oligo-dT primers. RT
   reaction was carried out at 42°C for 90 min and then 70°C for 15 min.
   For the quantitative determination of the reversely transcribed
   products, real-time PCR was performed on an Applied Biosystems 7500
   Fast PCR System (Foster City, CA, USA) using the standard SYBR green
   method under the following condition: 95°C for 10 min, followed by 40
   cycles of 95°C for 15 s and 60°C for 1 min. For expression analysis,
   the Ct values of mRNA targets and miRNA were first normalized
   respectively to those of the eukaryotic translation elongation factor 1
   alpha 1 (EEF1A1) and U6 genes in the same samples. The resulting ∆Ct
   was further utilized to assess the relative gene expression between
   different experimental groups, with the data presented as fold change
   to control. Sequences of primers used in real-time PCR assay are listed
   in Table [78]S1.

Transient transfection and lentiviral-based infection

   For transient transfection of LNA, 10 nM LNAs directed against
   miR-31-5p or control LNA (scramble) were transfected into cultured OSCC
   cells using Lipofectamine 2000 according to the manufacturer's
   instructions. To deplete endogenous ACOX1, two siRNAs against ACOX1 or
   negative control (si-Ctrl) were transfected into cells through
   Lipofectamine RNAiMAX. For co-transfection experiments with miR-31-5p
   antisense oligomers and ACOX1 siRNA, the combination of 10 nM LNAs and
   10 nM siRNAs was used, together with Lipofectamine 2000 according to
   the manufacturer's instructions. All functional analyses were performed
   after 24 h of transfection.

   Double-stranded oligonucleotides encoding miR-31-5p or control (LacZ)
   sequence were synthesized and inserted into pcDNA-GW/EmGFP-miR vectors.
   The cloning primers utilized here are listed in Table [79]S1. Those
   vectors were further transferred into lentiviral expression plasmids
   using Gateway recombination technology according to the manufacturer's
   instructions. The resulting constructs were co-transfected with
   packaging vectors into HEK293-FT cells to generate lentiviral
   particles, which were further used to infect cells for establishing
   stable expression. All materials were purchased form Invitrogen.

Dual luciferase reporter assay

   The luciferase reporter vectors for the wild-type or mutant ACOX1 3'
   UTR containing the miR-31-5p binding site were constructed into
   pMIR-REPORT luciferase system (Thermo Fisher Scientific). The primers
   used here are shown in Table [80]S1. Mutated ACOX1 3' UTR was generated
   by replacement of the target sequence UCUUGC to AAGCUU using two-step
   mutagenesis PCR. The vectors expressing firefly luciferase reporter
   fused with wild-type or mutant ACOX1 3' UTR were co-transfected with
   miR-31-5p antisense oligomers or scramble control into OSCC cells.
   After 24 h, cells were harvested and the two distinct luciferase
   activities were measured using Dual-luciferase reporter assay according
   to the manufacturer's instructions (Promega; Madison, WI, USA). Renilla
   luciferase activity was used to normalize the extent of firefly
   luciferase activity.

Prediction and pathway analysis of miR-31-5p putative targets

   For miR-31-5p target prediction, the 3' UTR sequences of 17,330 human
   protein-coding genes were retrieved from the UCSC genome database
   ([81]https://genome-asia.ucsc.edu/index.html), and the sequence of
   mature hsa-miR-31-5p was downloaded from miRBase database
   ([82]http://www.mirbase.org/). Target prediction and context score
   analyses were performed using the targetscan_50.pl and
   targetscan_50_context_scores.pl Perl scripts were downloaded from the
   TargetScan website ([83]http://www.targetscan.org/vert_50/). A sum
   context score of <0.2 was used to filter for the highly probable
   targets. Consequently, a total of 1,213 putative targets (irrespective
   of species conservation) were recovered. These targets were further
   subjected to pathway enrichment analysis using online DAVID functional
   annotation tools ([84]https://david.ncifcrf.gov) and GeneGo MetaCore
   software (Thomson Reuters; New York, NY, USA).

Cell proliferation and colony formation assays

   For the growth curve assay, OSCC cells were seeded in 96-well plates.
   The cells were then fixed with 3.7% formaldehyde and stained with
   4',6-diamidino-2-phenylindole (DAPI) on indicated days. Cell numbers
   were determined using IN Cell Analyzer 1000 imaging system (GE
   Healthcare Life Sciences; Buckinghamshire, UK). For colony formation
   assay, OSCC cells were seeded at a low cell density of 100 cells
   (SCC25, OECM-1 and SCC4) into 6-well plates and maintained in complete
   medium for the indicated days (SCC25: 9 to 13 days; OECM-1: 9 days;
   SCC4: 13 days). The cells were then washed, fixed, and stained with
   crystal violet. Colony forming activity was determined by quantifying
   the visible colonies using UVP BioSpectrum Imaging System.

In vitro migratory and invasive assays

   For the wound healing-based migration assay, OSCC cells were seeded and
   grown in 6-well plates with complete medium to allow monolayer cell
   formation. The cells were scratched with a sterile pipette tip to
   create artificial wounds [85]^24. At indicated time points, images of
   the healed wound were photographed by ZEISS Axio Observer A1 microscope
   and the wound area was determined using ZEISS AxioVision 4.6 microscope
   software.

   The transwell migration and invasion assays were carried out using
   Boyden chambers with transwell inserts in 24-well plates (Corning;
   Corning, NY, USA). For the invasion assay, we pre-coated the transwell
   inserts with Matrigel Basement Membrane Matrix (BD Biosciences; San
   Jose, CA, USA). The indicated OSCC transfectants were resuspended in
   serum-free medium and seeded into the upper transwell inserts, with the
   supplement of 10% fetal bovine serum-containing culture media in the
   bottom chamber. After 14 h for OECM-1 and 24 h for SCC25, the numbers
   of cells migrated or invaded towards the bottom chamber were determined
   by crystal violet staining and quantified using IN Cell Developer
   Toolbox.

Western blotting

   Western blotting was conducted after separation of polypeptides using
   SDS-PAGE. Proteins on gel were transferred to PVDF membrane (Merck
   Millipore; Billerica, MA, USA). The membrane was further incubated with
   indicated primary and appropriate secondary antibodies. Antibodies
   against ACOX1 were obtained from Proteintech Group (Chicago, IL, USA).
   Phosphorylated form of ERK1/2 and ERK1 antibodies were purchased from
   Santa Cruz Biotechnology (Santa Cruz, CA, USA) and BD Biosciences,
   respectively. Antibodies targeting MMP3, MMP10 and phospho-Akt (Thr308)
   were also purchased from Santa Cruz Biotechnology. Antibodies against
   phospho-Akt (Ser473) were obtained from Cell Signaling Technology
   (Danvers, MA, USA). MMP2 and MMP9 antibodies were purchased from Abcam
   (Cambridge, Cambridgeshire, UK). An anti-GAPDH antibody (BioWorld; St.
   Louis Park, MN, USA) was used as control. HRP-conjugated secondary
   antibodies against rabbit IgG or mouse IgG (GeneTex) were incubated
   with membrane for 1 h at room temperature. Immunobands were detected
   using chemiluminescent HRP substrates (ECL; Merck Millipore) and
   captured by UVP BioSpectrum 600 Imaging System. The intensity of bands
   was quantified by Image J software.

Lipidomic analyses: UPLC-MS metabolomics, lipid bodies staining and PGE2
quantification

   For Ultra Performance Liquid Chromatography (UPLC)-MS metabolomics
   analysis, global metabolites were extracted from OSCC cell lysates
   using 80% methanol. Folch method (ddH2O : CHCl3 : Methanol = 3 : 8 : 4)
   was implemented for further lipid metabolites extraction. Subsequently,
   UPLC was performed and mass spectrometry was operated in negative-ion
   (ESI-) mode. Data analysis was accomplished using MetaboAnalyst tool
   and Partek Genomics Suite software.

   For lipid droplet staining, cultured OSCC cells were fixed on
   coverslips and incubated with a mixture of Alexa Fluor 488-conjudated
   BODIPY (Lipid droplet), Alexa Fluor 555-conjugated wheat germ
   agglutinin and DAPI for 1 h at room temperature prior to washing and
   mounting. The concentrations of total free fatty acids in plasma were
   measured by using the Non-Esterified Fatty Acids (NEFA) assay kit
   (Randox Laboratories; Crumlin, Antrim, UK), based on the manufacturer's
   instructions. PGE2 levels in saliva or conditioned media were detected
   using PGE2 monoclonal ELISA kit according to the manufacturer's
   instructions. Each subject signed an informed consent form approved by
   the Institutional Review Board of Chi-Mei Medical Center, permitting
   the use of plasma and saliva samples for this study.

Gelatin zymography

   To measure the gelatinase activity of the experimentally manipulated
   cell culture, 2 × 10^6 cells were first switched to serum-free
   DMEM/F12. After 24 h, the conditioned media were collected and
   concentrated by Amicon Ultra-15 centrifugal filter (Merck Millipore).
   Subsequently, all samples were run on 8% SDS-PAGE with 0.1% gelatin.
   Gels were washed twice with 2.5% Tritox X-100 for 30 min at room
   temperature, and then incubated with development buffer at 37°C
   overnight. Lastly, gels were stained with Coomassie blue to determine
   the MMP2 and MMP9 activity.

Statistical analysis

   All data are presented as mean ± SEM (standard error of the mean).
   Statistical analyses were carried out by using GraphPad Prism 6 (San
   Diego, CA, USA). Pearson correlation analysis was utilized to evaluate
   the correlation between two groups. Two-tailed Student's t-test was
   used to determine the statistical significance between independent
   samples. p-values less than 0.05 were considered statistically
   significant. Degrees of significance (*, p < 0.05; **, p < 0.01; ***, p
   < 0.001; ns, not significant) are depicted in the respective figure
   legends.

Results

Systematic expression profiling uncovers miR-31-5p as a highly expressed
miRNA in both OSCC and HNSCC

   To provide further insights into the physiological relevance of miRNAs
   in tumor malignancy, we first sought to compile, from a systems
   perspective, the miRNA-mRNA networks underlying OSCC pathogenesis. Our
   targeted, qRT-PCR-based expression profiling of 270 miRNAs in 49 human
   oral tissue samples (including 19 normal and 30 OSCC specimens)
   previously identified 49 differentially expressed miRNAs (p-value <
   0.01 and fold change > 2) in OSCC tissues - 11 of these were
   significantly up-regulated in OSCC tissues (Figure [86]S1A-S1C and
   Table [87]S2). To confirm our findings, we also analyzed HNSCC and OSCC
   microRNA deep sequencing datasets from The Cancer Genome Atlas (TCGA)
   database, and subsequently identified 40 markedly up-regulated miRNAs
   (p-value < 0.05 and fold change > 2) in 43 HNSCC tissues (Figure
   [88]1A) and 47 over-expressed miRNAs (p-value < 0.05 and fold change >
   4) in 28 OSCC subjects (Figure [89]S2), as compared with their paired
   normal specimens. Six of these were commonly over-expressed in both
   in-house OSCC samples and public TCGA HNSCC specimens (Figure [90]1B).
   To elucidate how this altered miRNAome could systemically impact tumor
   biology, we performed computational miRNA target prediction for these
   six miRNAs and subsequently utilized pathway analysis (Figure [91]S3
   and Table [92]S3). Interestingly, an enrichment of predicted targets in
   several KEGG pathways was found, with different types of lipid
   metabolic pathways shown as the top ranked terms, suggesting distinct
   cellular changes may arise as a consequence of these OSCC-associated
   miRNA alterations.

Figure 1.

   [93]Figure 1
   [94]Open in a new tab

   Systems profiling identifies miRNAs mis-regulated in HNSCC. (A)
   Hierarchical clustering of 43 normal/tumor paired HNSCC samples based
   on the expression levels of 40 significantly upregulated miRNAs. (B)
   Venn diagram shows the extent of overlap in differentially
   over-expressed miRNAs between in-house OSCC samples and public TCGA
   HNSCC data. The six commonly over-expressed miRNAs are also indicated.
   (C) Expression levels (in log2 reads per million, or RPM, value) of
   miR-31-5p in 42 HNSCC tumor tissues (red) and their matched normal
   tissues (blue) (upper panel), as well as in 28 OSCC samples (red) and
   their paired normal controls (blue) (lower panel). Paired samples are
   connected by lines. Corresponding box plots are also shown alongside.
   (D) Expression levels of miR-31-5p were analyzed in normal (n = 43) vs.
   HNSCC samples (Stage I, n = 25; Stage II, n = 73; Stage III, n = 70;
   Stage IV, n = 228) (upper panel), and in normal (n = 43) vs. OSCC
   samples (Stage I, n = 19; Stage II, n = 53; Stage III, n = 50; Stage
   IV, n = 142) (lower panel) according to AJCC stages. ***, p < 0.001,
   Student's t-test.

   We selected miR-31-5p as the target miRNA for further functional
   studies in this study on the basis of the following criteria: 1) Among
   the six miRNAs, miR-31-5p exhibited the greatest extent of expression
   up-regulation as well as the highest abundance in the OSCC tissues
   (Figure [95]S1D and Table [96]S2). 2) The tumor-associated elevation of
   miR-31-5p was coincidently evident in both the TCGA OSCC and HNSCC
   miRNA expression datasets (Figure [97]1C). In addition, analysis of
   clinical relevance of miR-31-5p expression with regard to disease
   stages revealed that, while miR-31-5p was significantly increased in
   the tumor tissues, its levels remained invariably high across all four
   stages of HNSCC and OSCC patient samples (Figure [98]1D), indicating
   that this altered elevation may be an early event in neoplastic
   progression of OSCC and HNSCC. 3) Pathway analyses of the presumptive
   targets, based on either KEGG (Figure [99]2A and Table [100]S4) or
   GeneGo/MetaCore (Figure [101]S4), showed enrichment in the regulation
   of lipid or fatty acid metabolism. To test whether these hypothetical
   molecular changes in metabolism are verifiable in a clinicopathological
   context, we monitored the levels of free fatty acids in the plasma of
   OSCC patients. Interestingly, we found that OSCC plasma exhibited
   higher levels of free fatty acids (Figure [102]2B), thus supporting our
   in silico functional predictions of this miRNA.

Figure 2.

   [103]Figure 2
   [104]Open in a new tab

   ACOX1, the rate-limiting enzyme in peroxisomal β-oxidation, is directly
   targeted by miR-31-5p. (A) Schematic depiction of the distribution of
   putative miR-31-5p targets in two fatty acid metabolism pathways. The
   candidate targets are highlighted in red. (B) The concentrations of
   total free fatty acids in 16 normal and 70 OSCC patient plasma
   specimens were detected by the NEFA kit as mentioned in Methods. (C)
   mRNA levels of ACOX1 in SCC25 and OECM-1 cells were measured by qRT-PCR
   from RNAs extracted from miR-31-5p-depleted (upper panel) or
   over-expressing (lower panel) cells. The data are expressed as fold
   change relative to corresponding controls. (D) Protein levels of ACOX1
   were detected by western blotting in SCC25 and OECM-1 cells treated
   with anti-miR-31-5p oligomers or scramble control (left panel), and in
   SCC25 cells infected with lentivirus containing miR-31-5p and control
   (LacZ) vector (right panel). GAPDH was used as a loading control. (E)
   Predicted duplex formation between hsa-miR-31-5p and human ACOX1 3' UTR
   (Wt). The nucleotides highlighted in bold (positions 2945 to 2951)
   indicate the seed region of miR-31-5p. The underline points out the
   mutated miR-31-5p target site in the ACOX1 3' UTR (Mut) construct.
   SCS25 and OECM-1 cells were transfected with a luciferase reporter
   construct that contains wild type (Wt) or mutated (Mut) ACOX1 3' UTR,
   together with anti-miR-31-5p or anti-scramble oligomers. Luciferase
   activity was determined 24 h after transfection. (F) Expression levels
   (in log2 normalized raw count) of ACOX1 in 41 HNSCC tumor tissues (red)
   and their matched normal tissues (blue). Paired samples are connected
   by lines, with corresponding box plots also shown. (G) Correlation
   between the expression levels of miR-31-5p and ACOX1 in 39 normal
   (blue) and tumor (red) paired HNSCC samples. Pearson's r analysis was
   performed. The bar graphs in (C) and (E) show mean ± SEM from three
   independent experiments (ns, not significant; *, p < 0.05; **, p <
   0.01; ***, p < 0.001; Student's t-test).

miR-31-5p directly targets ACOX1, the first and rate-limiting enzyme of
β-oxidation in peroxisomes

   In line with our observations, several reports in the literature have
   revealed that expression of miR-31-5p is frequently up-regulated in
   different cancer types, including HNSCC [105]^16. In order to further
   dissect the biological roles of miR-31-5p in OSCC, we first performed
   miR-31-5p expression profiling in four OSCC cell lines by stem-loop
   RT-PCR, which revealed that it was considerably elevated in all four
   OSCC cell lines compared with normal gingival epithelioid S-G cell
   lines (Figure [106]S5A). Next, to examine whether the aforementioned
   lipid metabolism-associated candidates were indeed regulated by
   miR-31-5p, we analyzed their mRNA and protein levels in miR-31-5p
   mis-expressed OSCC cells (Figure [107]S5B-S5D) using real-time PCR and
   western blotting, respectively. This expression profiling uncovered
   ACOX1, the first enzyme in peroxisomal lipid metabolism, as a possible
   target: depletion of miR-31-5p using antisense oligonucleotides
   increased both mRNA and protein levels of ACOX1 in various cultured
   OSCC cells (Figure [108]2C-[109]2D and S5E) and also in a
   dose-dependent manner (Figure [110]S5F). By contrast, lentiviral-based
   ectopic expression of miR-31-5p led to expression suppression of ACOX1
   (Figure [111]2C-[112]2D). Knockdown or over-expression of miR-31-5p did
   not seem to influence the expression levels of other predicted targets
   associated with lipid metabolism (Figure [113]S5E-S5F), suggesting
   miR-31-5p does not target these 3' UTR despite the presence of
   targeting sequences.

   The presence of a binding site for miR-31-5p in the 3' UTR of ACOX1
   suggests a direct regulation (Figure [114]2E). We tested this
   possibility by performing luciferase reporter assay on constructs
   harboring the wild-type or a binding site-mutated 3' UTR. Our results
   revealed that depleting endogenous miR-31-5p in OSCC cells elevated the
   luciferase activity of the wild type, but not the mutant ACOX1 3' UTR
   construct (Figure [115]2E), strengthening the direct targeting by
   miR-31-5p. Additionally, deep sequencing data from TCGA database
   further corroborated this regulatory relationship: the ACOX1 expression
   was down-regulated in HNSCC samples relative to their paired normal
   samples (Figure [116]2F), and further was inversely correlated in
   expression with miR-31-5p in the HNSCC normal/tumor paired specimens
   (Figure [117]2G). Collectively, these results connected the
   tumor-associated up-regulation of miR-31-5p with lower level of ACOX1
   in HNSCC clinical specimen.

miR-31-5p regulates the reprogramming of lipid metabolism in OSCC via
targeting ACOX1

   ACOX1 is known to desaturate straight chain lipid substrates, such as
   the CoA esters of long chain fatty acids, very long chain fatty acids,
   dicarboxylic acids, polyunsaturated fatty acids and prostaglandins
   [118]^18^, [119]^19. The negative regulation of ACOX1 thus signifies
   that miR-31-5p may be linked to tumor-associated lipid metabolic
   alterations. To test this hypothesis, we set out to assess various
   lipidomic features of cells in the context of miR-31-5p or ACOX1
   mis-expression. First, a mass spectrometry-based, untargeted lipid
   metabolite analysis was conducted on SCC25 cells with miR-31-5p or
   ACOX1 knockdown (Figure [120]3A). We found that several long chain
   fatty acids, including myristic acid (C14:0), palmitic acid (C16:0) and
   stearic acid (C18:0), were notably reduced upon inactivation of
   miR-31-5p in the cells (Figure [121]3B, left), whereas the levels of
   two C18 unsaturated fatty acids exhibited no significant change (Figure
   [122]S6A, top). While the UPLC-MS data showed that reduced expression
   of ACOX1 did not affect the abundance of C14, C16 and C18 free fatty
   acids (Figure [123]3B, left), certain phospholipids and sphingolipids,
   especially those derived from free fatty acids with 16- or 18-carbon
   chain, underwent significant increases in the same knockdown culture
   (Figure [124]3B, right). These alterations could hypothetically imply
   that the free fatty acids accumulated in the absence of ACOX1 might be
   rapidly utilized to synthesize new components of plasma membrane for
   cell proliferation and migration. Notably, we discovered that several
   phosphatidylinositols (PI), phosphatidylglycerol (PG) and sphingomyelin
   were dramatically repressed upon ACOX1 depletion (Figure [125]3B and
   S6A), whereas the miR-31-5p inactivation exhibited a discernibly
   reverse outcome. Together, these cellular lipidomic profile alterations
   highlighted that the miR-31-5p-ACOX1 axis may be linked to
   reprogramming of lipid metabolism pathways in OSCC (Figure
   [126]S6B-S6C).

Figure 3.

   [127]Figure 3
   [128]Open in a new tab

   The miR-31-5p-ACOX1 axis reprograms the fatty acid metabolism of OSCC
   cells. (A-B) SCC25 cells were transfected with scramble or miR-31-5p
   antisense oligomers, or treated with control or ACOX1 siRNAs (single or
   mixture of two). Protein levels of ACOX1 were determined using western
   blotting, with GAPDH as a loading control (A). The intensity of lipid
   metabolites was measured via UPLC-MS, and the levels of selected lipid
   metabolites are shown as mean ± SEM from at least three independent
   repeats (B). (C-D) Alexa Fluor 488-conjudated BODIPY (Lipid droplet)
   and Fluor 555-conjugated wheat germ agglutinin (WGA) staining in SCC25
   after inactivation of miR-31-5p or suppression of ACOX1. Scale bar, 20
   μm. Representative images (C) and the quantitative data (D) are also
   shown. (E) SCC25 cells were transfected with two ACOX1 siRNAs (si-ACOX1
   #1 or si-ACOX1 #2) or control (si-Ctrl) for 24 h. The concentrations of
   PGE2 in conditioned media were determined via ELISA assay. (F) PGE2
   detection assay was performed to assess PGE2 production in SCC25 cells
   with the indicated combination of co-transfection. The data shown in
   (D) to (F) were derived from biological triplicates and are expressed
   as mean ± SEM. (*, p < 0.05; **, p < 0.01; ***, p < 0.001)

   Second, given that certain peroxisomal metabolic enzymes are enriched
   in and functionally linked to lipid droplets [129]^17^, [130]^25, and
   also owing to the association of lipid droplet biogenesis with cancer
   progression [131]^26, we turned our attention to the distribution of
   these organelles. Lipid droplets are phospholipid monolayer-enclosed
   subcellular organelles composed of neutral lipids, and can be detected
   based on this lipid-rich characteristic. Lipid-specific BODIPY staining
   illustrated that depletion of endogenous miR-31-5p diminished the
   levels of cellular lipid droplets. By contrast, there was a prominent
   accumulation of lipid droplets in the ACOX1 knockdown cells (Figure
   [132]3C-[133]3D). These data therefore confirmed the link of
   peroxisomal β-oxidation with lipid droplets and further implied that
   miR-31-5p, by targeting ACOX1, could regulate the formation of these
   lipid bodies.

   Finally, it has been reported that PGE2-CoA is one of the substrates of
   ACOX1 [134]^27. Via transport into the extracellular space,
   intracellular PGE2 is also known to activate various pro-tumorigenesis
   signaling pathways. Therefore, we next sought to assess the PGE2 levels
   in the conditioned media (CM) of ACOX1-depleted OSCC cells. We found
   that siRNA abrogation of ACOX1 expression significantly enhanced the
   PGE2 levels in the CM (Figure [135]3E). Conversely, inhibition of
   miR-31-5p reduced the concentration of PGE2. In addition, simultaneous
   knockdown of miR-31-5p in the ACOX1-depleted cells reversed the
   induction of PGE2 production in response to ACOX1 down-regulation
   (Figure [136]3F), implying that the amount of PGE2 in the extracellular
   space is regulated by the miR-31-5p-ACOX1 axis in OSCC. Viewed
   together, these observations implicated miR-31-5p in the
   ACOX1-depednent reprogramming of lipid metabolism in OSCC.

Suppression of ACOX1 accelerates OSCC cell migration and invasion

   To further decipher the biological consequence of miR-31-5p-mediated
   functions in OSCC, we carried out several functional studies based on
   gain- and loss-of-function manipulation. Toward this end,
   mis-expression of miR-31-5p did not influence the cell proliferation
   and colony formation capacity in three independent OSCC cell lines
   (Figure [137]S7A-S7D). However, in vitro migratory assay showed that
   loss of miR-31-5p notably suppressed the migratory ability of OSCC
   cells, and that ectopic expression of miR-31-5p exhibited an adverse
   effect on cell migration (Figure [138]4A-[139]4B and S7E). These
   results were further supported by the Boyden chamber migration assay -
   miR-31-5p expression was correlated with the migratory and invasive
   extent of OSCC cells (Figure [140]4C-[141]4D and S7F-S7G).
   Collectively, these findings indicated that miR-31-5p acts positively
   on the motility of OSCC cells, a function that is independent of cell
   growth.

Figure 4.

   [142]Figure 4
   [143]Open in a new tab

   miR-31-5p and ACOX1 reciprocally modulate the OSCC migratory and
   invasive capability. (A-B) Wound healing migration assay was performed
   to assess the migratory ability of SCC25 upon knockdown or ectopic
   expression of miR-31-5p. Photographs were taken at indicated time
   points after cells were scratched, with the representative images shown
   in (A). Scale bar, 100 μm. Wound area was measured and normalized to 0
   h (B). Data are shown as mean ± SEM from at least three independent
   experiments (**, p < 0.01; Student's t-test). (C-D) Transwell migration
   and Matrigel invasion assays were carried out after knockdown or stable
   over-expression of miR-31-5p in SCC25 cells. The representative
   photomicrographs are shown (C). Scale bar, 200 μm. Average migratory
   area per field was quantified and expressed as fold change relative to
   control (D). (E-F) SCC25 cells were transfected with ACOX1 siRNAs or
   control siRNA, and subjected to wound healing migration (E) and
   transwell migratory (F) assays. The bar graphs in (D) to (F) show mean
   ± SEM from three independent experiments (*, p < 0.05; **, p < 0.01;
   ***, p < 0.001).

   We also examined whether the down-regulation of ACOX1 in tumor
   recapitulates similar oncogenic attributes. Repressing endogenous ACOX1
   mRNA and protein levels in two OSCC cell lines, SCC25 and OECM-1
   (Figure [144]S8A-S8B), facilitated the in vitro migratory capacity of
   these cells on the basis of a wound healing assay (Figure [145]4E and
   S8C); however, this ACOX1 abrogation had seemingly no effect on cell
   proliferation and growth (Figure [146]S8D-S8E). To further ascertain
   this functional role, we employed Matrigel invasion assay to monitor
   OSCC invasive capability, and subsequently observed that ACOX1
   attenuation considerably stimulated the migratory and invasive
   abilities of two different OSCC lines (Figure [147]4F and S8F). These
   data were reminiscent of the results found in OSCC cells
   over-expressing miR-31-5p, and further supported the suggested
   connection of the miR-31-5p-ACOX1 axis to OSCC malignancy.

PGE2 promotes OSCC cell migration and invasion through activating the
ERK-MMP9 cascade

   Having established that ACOX1 down-regulation underlies cell motility,
   we next speculated the contribution of lipid metabolites to these
   features. While few studies directly link the lipid metabolites to cell
   mobility, PGE2 is reportedly linked to tumor cell proliferation,
   survival and migration [148]^28. To further elucidate its role, we
   directly stimulated OSCC cells with PGE2 and examined their migratory
   capacity and invasiveness. Since the concentration of PGE2 in SCC25 CM
   was measured to be 4 nM, we consequently treated the OSCC cells with a
   higher dose (10 nM) to assess its effect. Our results revealed that
   this PGE2 treatment triggered the in vitro migratory and invasive
   activities of OSCC cells (Figure [149]5A).

Figure 5.

   [150]Figure 5
   [151]Open in a new tab

   The miR-31-5p-ACOX1 axis-mediated up-regulation of prostaglandin E2
   enhances the migration and invasiveness of OSCC cells. (A) SCC25 cells
   were treated with or without PGE2 (0.01 µM), or co-treated with PGE2
   and EP1 receptor antagonist (SC-51322; 0.02 µM), and analyzed by
   transwell migration assay. Representative photomicrographs are shown
   (left panel; scale bar, 200 μm). Average migratory area per field was
   quantified and expressed as fold change relative to control (right
   panel). (B) SCC25 cells were serum-starved and stimulated with PGE2
   (upper panel) or with the combination of PGE2 and SC-51322 (lower
   panel) for the indicated times. Protein levels of p-ERK, ERK1, p-AKT
   (Ser473), p-AKT (Thr308) and AKT were determined via western blotting,
   with GAPDH as a loading control. (C) SCC25 cells were stimulated with
   or without 0.01 µM PGE2 for 24 h. The expression levels of various MMP
   molecules were assessed by qRT-PCR (upper panel). Relative expression
   of MMP9 in SCC25 cells was determined upon stimulation of PGE2 alone or
   together with SC-51322 (lower panel). (D) Incubation of SCC25 cells
   with conditioned media from ACOX1-depleted or control cells for 24 h,
   followed by assessment of their migratory and invasive activities.
   Quantitative results in (A), (C) and (D) correspond to mean ± SEM from
   three independent experiments. (E) The concentrations of PGE2 in 50
   normal and 50 OSCC salivary specimens were determined by ELISA (left
   panel). The receiver operating characteristic (ROC) curve was utilized
   to distinguish OSCC from non-tumor groups (middle panel). For tumor
   state-associated alterations, the abundance of PGE2 in saliva from
   normal (n = 50) and OSCC samples sub-grouped according to AJCC stages
   (Stage I, n = 12; Stage II, n = 8; Stage III, n = 10; Stage IV, n = 20)
   was further plotted (right panel). Statistical analysis in this figure
   was done by Student's t-test (ns, not significant; *, p < 0.05; **, p <
   0.01; ***, p < 0.001).

   Cumulative evidence has suggested a role of PGE2 in tumor cell
   motility: PGE2 is known to promote colorectal cancer (CRC) cell
   migration and invasion through the β-arrestin-SRC-EGFR-PI3K-Akt
   cascade, while its role in pancreatic cancer cells is via
   ERK-ETS1-matrix metalloproteinase 2 (MMP2) signaling [152]^29^,
   [153]^30. We then set out to delineate the PGE2-dependent signaling
   associated with OSCC cell migration. Subsequent immunoblotting analyses
   showed a rapid but transient activation of ERK: phospho-ERK signal was
   detectable at 5 min following PGE2 stimulation but restored to basal
   level at 30 min (Figure [154]5B, top). In contrast, levels of
   phospho-Akt (Thr308 or Ser473) were either invariable or undetectable,
   respectively (Figure [155]5B, top), indicating that AKT-associated
   signaling was not activated in PGE2-treated OSCC cells. Next, given
   that PGE2 up-regulates the expression of certain matrix
   metalloproteinases in immunocytes and cancer cells [156]^30^-[157]^32,
   we subsequently examined the mRNA expression levels of various MMP
   molecules, such as MMP1, MMP2, MMP3, MMP9 and MMP10, in PGE2-stimulated
   OSCC cells. We discovered that PGE2 treatment markedly triggered an
   increase in MMP9 relative to control but not the other MMPs, at both
   the mRNA (Figure [158]5C, top) and protein (Figure [159]S9) levels.
   Next, given that stimulation of SCC4 cells with PGE2 induces the
   expression of one of its cognate receptors (EP1) but not the other
   three receptor subtypes [160]^33, we assessed the role of EP1 in PGE2
   downstream signaling by subjecting cells to an EP1 receptor-specific
   antagonist (SC-51322). Our results showed that treatment with SC-51322
   significantly antagonized the PGE2-induced cell migration and invasion
   (Figure [161]5A), ERK activation (Figure [162]5B, bottom) and MMP9
   expression (Figure [163]5C, bottom and S9). Together, these
   observations were thus in line with the scenario that PGE2 promotes
   OSCC cell migration through an EP1-ERK-MMP9 signaling cascade.

   To further substantiate this notion, we next performed migratory assay
   on OSCC cells cultured with the CM derived from ACOX1-silenced cells.
   Interestingly, addition of this CM evidently promoted migration and
   invasiveness of OSCC cells (Figure [164]5D), implying extracellular
   medium conditioned by ACOX1 down-regulation may accelerate OSCC cell
   migration. Owing to this possible oncogenic implication of the
   extra-tumor environment, we next addressed the possible clinical
   relevance of PGE2 secretion by quantifying the amounts of PGE2 in
   saliva specimens from OSCC patients and comparing them to healthy
   controls. Based on ELISA measurements, the salivary PGE2 levels were
   notably increased in OSCC vs. normal samples (Figure [165]5E, left).
   Furthermore, the ROC analysis validated that this salivary secretion
   may be a promising biomarker - AUC = 0.776, sensitivity = 0.70, and
   specificity = 0.68 using 0.566 ng/mL as the cut-off concentration for
   differentiating OSCC from non-tumorous group (p-value < 0.001) (Figure
   [166]5E, middle). When accounting for disease stages, we further found
   a slight but significant elevation of PGE2 levels during OSCC
   progression, and its production was maximal in stage III OSCC patients,
   characterized by local metastasis (Figure [167]5E, right).
   Collectively, these results illustrated the potential value of secreted
   PGE2 as a powerful biomarker for diagnosing OSCC progression.

miR-31-5p-ACOX1 axis enhances OSCC cell migration via up-regulating ERK-MMP9
signaling

   Upon establishing the connection between PGE2 and ERK-MMP9 signaling,
   we next aimed to corroborate the implication of miR-31-5p-ACOX1 axis in
   this functional regard. To this end, inhibiting the expression of
   miR-31-5p in OSCC cells significantly attenuated the level of ERK1/2
   phosphorylation (Figure [168]S10A-S10B). By contrast, siRNA-mediated
   ACOX1 suppression in two OSCC cell lines activated the ERK pathway
   (Figure [169]S10A-S10B). In terms of MMP expression, our data showed
   that ectopic expression of miR-31-5p significantly elevated the
   expression of MMP9 in OSCC cells (Figure [170]S10C, left), while
   depletion of endogenous miR-31-5p conversely lowered MMP9 levels
   (Figure [171]S10C, middle). Moreover, in SCC25 cells harboring
   ACOX1-specific siRNAs, we detected a discernible rise in MMP9
   expression (Figure [172]S10C, right).

   To further substantiate the functional antagonism between miR-31-5p and
   ACOX1, we next carried out co-knockdown of miR-31-5p and ACOX1 in SCC25
   cells. Consistent with earlier observations, we first found that
   miR-31-5p inhibition led to reduction in the levels and activities of
   MMP9 (Figure [173]6A-[174]6B) and activated ERK (Figure [175]6C), as
   well as a lessened migration capability (Figure [176]6D), in comparison
   with control cells. In contrast, loss of ACOX1 had an opposite effect
   on these functional attributes. Simultaneous knockdown of ACOX1 in the
   miR-31-5p-depleted cells antagonized the alterations exerted by single
   knockdown of miR-31-5p: the down-regulation of phospho-ERK and MMP9 was
   reversed upon co-treatment of ACOX1 siRNAs (Figure [177]6A-[178]6C),
   and the reduced migration and invasiveness potential as a consequence
   of miR-31-5p knockdown was rescued by ACOX1 inactivation (Figure
   [179]6D). Considered together, these results pinpointed miR-31-5p-ACOX1
   axis as the key upstream step in the promigratory ERK-MMP9 signaling
   pathway.

Figure 6.

   [180]Figure 6
   [181]Open in a new tab

   ACOX1 interferes with the promigratory ERK-MMP9 signaling in OSCC
   cells. (A) Expression levels of miR-31-5p, ACOX1 and MMP9 in SCC25
   transfected with the indicated combination of antisense oligomers
   (miR-31-5p) and siRNAs (ACOX1). (B) Gelatinase zymograms of lysates
   from SCC25 cells with the indicated co-transfection conditions. Silver
   staining was used as a loading control. (C) Immunoblotting of the
   indicated proteins for cells shown in (A). (D) SCC25 cells with the
   indicated transfection combinations were examined for transwell
   migration and Matrigel invasion activities. Average migratory area per
   field was determined and expressed as fold change relative to control.
   Scale bar, 200 μm. The bar graphs in (A) and (D) show mean ± SEM from
   three independent experiments (*, p < 0.05; **, p < 0.01; ***, p <
   0.001; ns, not significant). (E) Correlation between the expression
   levels of miR-31-5p and MMP9 (top) and between those of ACOX1 and MMP9
   (bottom) in 39 normal (blue) and tumor (red) paired HNSCC samples, as
   determined by Pearson's r analysis. (F) Schematic illustration of the
   mechanistic model for the role of miR-31-5p-ACOX1 axis in oral squamous
   cell carcinoma.

   To further strengthen the clinical relevance of this regulatory
   pathway, we analyzed the HNSCC microRNA and mRNA deep sequencing data
   archived in TCGA. The expression levels of miR-31-5p showed a positive
   correlation with those of MMP9 (Figure [182]6E, top), whereas ACOX1
   levels were negatively associated with MMP9 in the HNSCC specimens
   (Figure [183]6E, bottom). Moreover, we sought to elucidate the clinical
   significance of the levels of ACOX1 and MMP9 in patients with
   metastatic HNSCC. To this end, we analyzed two independent datasets,
   including the online Oncomine microarray and TCGA deep sequencing
   datasets, and discovered that the ACOX1 expression was distinctly
   down-regulated in HNSCC samples as compared with normal samples in both
   datasets, irrespective of the extent of local metastasis (Figure
   [184]S10D-S10E, top). Interestingly, we further discovered that the
   expression levels of ACOX1 were progressively decreased in HNSCC
   patients with higher-grade lymph node metastasis, in line with our
   observations that ACOX1 silencing raised the migratory and invasive
   potential of OSCC cells. Concordantly, the MMP9 levels were
   dramatically elevated in HNSCC relative to normal specimens, and also
   exhibited a rise in correspondence to lymph node invasion (Figure
   [185]S10D-S10E, bottom). In light of our data, we propose a model
   whereby the elevated expression of miR-31-5p in OSCC directly
   contributes to the down-regulation of ACOX1, consequently leading to
   accumulation of un-degradable lipid metabolites. In this reprogrammed
   lipidome, PGE2 represents a key determinant in OSCC cell migration and
   invasion and exerts its pro-tumorigenic role through activating
   EP1-ERK-MMP9 signaling (Figure [186]6F).

Discussion

   While the critical roles of miRNAs in carcinogenesis have been
   demonstrated in the literature, a systems characterization of
   miRNA-mRNA networks in oral cancer pathogenesis has not been
   sufficiently established. A detailed investigation of the
   OSCC-associated miRNAs and their regulated networks may provide
   important mechanistic insights into the progression and treatment of
   this disease. In this regard, our present work advanced the
   understanding of OSCC biology by presenting the following lines of
   findings. 1) We catalogued an OSCC-associated altered miRNAome and
   further uncovered distinct cellular pathways potentially targeted by
   this miRNAs-mediated regulatory network. 2) The putative link of these
   regulatory molecules to lipid metabolism was substantiated by the
   delineation of a miR-31-5p-ACOX1 axis, which was further demonstrated
   to closely modulate cellular lipidomes as well as the migratory and
   invasive capacity of cancer cells. 3) PGE2, one of the key substrates
   of the ACOX1-mediated peroxisomal β-oxidation, acted as the key
   mediator of miR-31-5p function by augmenting the ERK-MMP9 signaling.
   Importantly, clinicopathological relevance of PGE2 was further
   corroborated by the observed elevation in saliva, which is in
   association with local invasion of OSCC and also indicative of its
   value as a promising disease biomarker. Viewed together, our findings
   have outlined a miR-31-5p-driven, tumor-promoting regulatory network
   that has considerable implications in the lipid metabolomic alterations
   associated with OSCC tumor biology.

Elevated miR-31-5p is a contributor to progression of advanced OSCC

   In line with our observations, previous studies have reported that
   miR-31-5p is frequently up-regulated in multiple cancer types,
   including HNSCC [187]^16, colorectal cancer [188]^34 and lung cancer
   [189]^35. In contrast, its down-regulation in tumor has also been
   documented based on expression reduction in gastric cancer [190]^36 and
   prostate cancer [191]^37, and homozygous deletion of miR-31-5p host
   gene (MIR31HG) in mesothelioma [192]^38. Intriguingly, in esophageal
   cancer, miR-31-5p is distinctly altered between the two major subtypes
   of this malignancy, with its expression up-regulated in esophageal
   squamous cell carcinoma (ESCC) [193]^39 but down-regulated in
   esophageal adenocarcinoma (EAC) [194]^40. In line with its frequent
   up-regulation, miR-31-5p is oncogenic in several squamous cell-derived
   malignancies, including OSCC, ESCC and skin squamous cell carcinoma.
   However, the findings that miR-31-5p is significantly over-expressed in
   lung adenocarcinoma and colorectal cancer (more than 90% of cases are
   adenocarcinoma) is seemingly contradictory to the expression profiles
   observed in EAC [195]^35^, [196]^41^, [197]^42. Viewed together, these
   findings imply a complex, tumor-dependent regulation and possibly dual
   role of this miRNA in tumor malignancy. To this end, several studies
   have mechanistically dissected this tumor-associated dysregulation of
   miR-31-5p. For instance, the expression levels of both miR-31-5p and
   MIR31HG are diminished in triple-negative breast cancer (TNBC) cell
   lines through promoter hypermethylation [198]^43. In addition, the
   promoter region of MIR31HG contains multiple potential binding sites
   for C/EBP-β, which exerts positive regulation towards cigarette
   smoke-induced MIR-31HG and miR-31-5p transcription [199]^44. In OSCC,
   it has further been shown that miR-31-5p elevation is mediated through
   the EGFR-AKT-C/EBP-β regulatory axis [200]^45.

   Despite its up-regulation in OSCC, the clinical relevance of miR-31-5p
   has not been characterized in depth. For patients exhibiting oral
   potentially malignant disorder (OPMD) features, an elevated miR-31-5p
   expression is more likely to result in malignant progression [201]^46.
   Furthermore, recent work by Hung et al. has implicated miR-31-5p in the
   early stage of oral carcinogenesis by showing that miR-31-5p cooperates
   with hTERT to immortalize normal oral keratinocytes [202]^47. These
   results are then consistent with our findings that enhanced miR-31-5p
   expression occurs in early stage of OSCC progression, which are often
   characterized by fast proliferation. However, our functional analyses
   revealed that miR-31-5p stimulates OSCC migratory and invasive
   capabilities without altering the rate of cell growth (Figure
   [203]4A-[204]4D and S7). It is noteworthy that a progressive
   up-regulation of miR-31-5p is associated with advanced
   clinicopathological stages in lung adenocarcinoma, and that higher
   miR-31-5p levels in patients with lung adenocarcinoma also correlate
   with lymph node metastasis [205]^48 and poor five-year overall survival
   [206]^35. Recently, work by Lu et al. has found that miR-31-5p promotes
   the oncogenicity and stemness of HNSCC via directly targeting the 3'
   UTR of AT-rich interacting domain 1A (ARID1A), leading to
   transactivation of pluripotency genes such as Nanog, OCT4 and Sox2
   [207]^49. In line with this pro-tumorigenic regulatory network, HNSCC
   tumors carrying an elevated expression of miR-31-5p, together with a
   higher expression of pluripotency genes and lower level of ARID1A,
   display poorer survival and prognosis. Collectively, these observations
   provide support for multiple roles of miR-31-5p at distinct stages of
   carcinogenesis; it may function in early steps of OSCC development as a
   driver to promote cell immortalization and transformation, whereas its
   invariably increased level in late aggressive stages may contribute to
   other oncogenic attributes, such as migration and invasion.

The connection between miRNA and tumor-associated lipid metabolism

   The importance of miRNAs in lipid homeostasis and tumor-related
   lipidomic alteration has recently emerged. In this capacity,
   liver-specific miR-122-5p is reportedly associated with hepatic
   cholesterol and lipid metabolism, as miR-122-5p inactivation triggers
   down-regulation of plasma cholesterol and triglycerides in high-fat
   diet mice [208]^50. Mechanistically, miR-122-5p is known to mediate
   translational inhibition of the cholesterol 7α-hydroxylase (CYP7A1),
   the first and rate-limiting enzyme in cholesterol-derived bile acid
   synthesis. Concordantly, treatment of miR-122-5p inhibitors in human
   HepG2 cells leads to bile acids elevation, further supporting the role
   of miR-122-5p in CYP7A1-dependent plasma cholesterol concentrations
   [209]^50^, [210]^51. miR-122-5p homozygous knockout mice are also shown
   to manifest steatohepatitis and fibrosis features, as well as
   spontaneous hepatocellular carcinoma (HCC) [211]^52. Our present study
   has strengthened this emerging lipidomic role of miRNAs by uncovering a
   novel regulator with critical implication in tumor-associated lipid
   metabolomes. We further showed that miR-31-5p targets a key enzyme in
   peroxisome and that this miR-31-5p-ACOX1 axis may constitute an
   oncogenic switch in cancer progression.

Tumor-associated lipid alteration

   Given the critical roles of lipids in many cellular processes,
   dysregulation of lipid metabolism has been conceivably linked to cancer
   [212]^53. An important finding of our current study is the
   identification of PGE2 as key mediator of the promigratory
   miR-31-5p-ACOX1 pathway and also as a promising salivary biomarker for
   monitoring OSCC progression. Interestingly, a recent salivary
   metabolomics study on OSCC patients has unveiled a panel of five
   metabolites - lactic acid, γ-aminobutyric acid, valine, phenylalanine
   and n-eicosanoic acid - as altered metabolites in the OSCC vs.
   non-cancerous samples. Among them, the concentrations of lactic acid
   and n-eicosanoic acid are elevated but those of the other three showed
   significant reduction [213]^54. While these findings underscore the
   significance of metabolic reprogramming in malignancy, the functional
   implications of these metabolites and their tumorigenic contributions
   remain presently unclear. Aside from PGE2, our own UPLC-MS-based
   lipidomic analyses also uncovered several altered metabolites in the
   ACOX1-depleted cells - elevated levels of phosphatidylethanolamine
   (PE), phosphatidylcholine (PC), ceramide, glucosylceramide and
   lactosylceramide, in contrast to the decline of certain
   phosphatidylinositols (Figure [214]S6C) - raising the possibility that
   additional lipid metabolites may be in effect. Notably, several recent
   studies have also revealed the link of lipidomic changes in cellular
   transformation. Work by Lv et al. highlighted a rise in serum levels of
   palmitic acid (C16:0), stearic acid (C18:0) and linoleic acid (C18:2)
   in patients with breast cancer, and further identified three types of
   free fatty acid (C16:0, C18:2 and total free fatty acids) as novel
   biomarkers for breast cancer detection [215]^55. Similarly, Cheng et
   al. reported higher levels of phosphatidylethanolamine and
   phosphatidylcholine in breast cancer patients vs. normal groups
   [216]^56. Together with our findings, these reports therefore attest to
   the scenario that excess fatty acids are utilized to synthesize
   phospholipids during carcinogenesis. How this reprogrammed lipidomes
   might affect oncogenic signaling and ultimately impact tumor biology
   remains an unresolved issue and warrants further investigation.

Balanced PGE2 synthesis and degradation is pivotal in OSCC oncogenesis

   Prostaglandins (PGs) are biologically active lipids and have been
   linked to tissue homoeostasis, inflammation and cancer development.
   Given these widespread and important functions, levels of PGs are
   dynamically controlled by a myriad of enzymes and pathways. The
   biosynthesis of PGs is successively carried out by three types of
   enzymes: cytoplasmic phospholipase A2 (cPLA2), cyclooxygenases (COXs)
   and a group of PG synthases [217]^28^, [218]^57. First, arachidonic
   acids (AA) are liberated from membrane phospholipids via cPLA2. Two COX
   proteins, COX-1 and COX-2, catalyze the subsequent conversion of free
   AAs to PGG2 and then PGH2. Finally, specific PG synthases are
   responsible for metabolizing the PGH2 to their corresponding
   prostaglandins. For instance, PGE synthase (PGES) mediates the
   production of PGE2, the predominant product derived from AA. Once
   synthesized, the PGs are exported to the extracellular space by
   specific multidrug resistance-associated proteins (MRPs) and
   subsequently exert their physiological functions via interacting with
   their cognate receptors. Conversely, the metabolic breakdown of PGs is
   initiated through a two-step process: their uptake is catalyzed by PG
   transporter (PGT) [219]^57^, [220]^58, which is followed by, in the
   case of PGE2, rapid inactivation to 15-keto-PGE2 by
   15-hydroxyprostaglandin dehydrogenase (15-PGDH) and continuously to 13,
   14-dihydro-15-keto-PGE2 by prostaglandin reductase (PTGR2) [221]^59.
   Furthermore, these inactivated PGE2 are further degraded through two
   cycles of peroxisomal β-oxidation to tetranor-PGE metabolites
   (tetranor-PGEMs), the major urinary metabolites found in mammals
   [222]^60^, [223]^61. Multiple studies have ascribed ACOX1 as the
   primary oxidase for this process, by showing that synthetic PGE2-CoA is
   oxidized exclusively by rat liver ACOX1 [224]^27^, [225]^62.

   In line with the pivotal role of PGE2 in tumor progression as revealed
   by our study, recent reports have ascribed tumorigenic relevance to
   various components of the PGE2-asscoiated pathway. To this end, the
   upstream synthetic enzymes, COX-2 and PGES, are reportedly up-regulated
   in many human malignancies, including HNSCC [226]^63. Moreover,
   treatment with COX antagonists is known to significantly inhibit cell
   invasion and lessen the expression and activity of MMP2 and MMP9 in
   HNSCC cells [227]^64. In terms of the downstream signaling of PGE2, for
   which there are four cognate receptors (EP1-EP4), only EP1 is notably
   up-regulated in HNSCC compared with paired non-tumoral mucosa [228]^63,
   which possibly could be attributed to PGE2 induction [229]^33. EP4
   receptor might also contribute to metastatic behavior, as treatment
   with EP4 antagonists significantly inhibits breast cancer metastasis
   [230]^65. While the functional selectivity of distinct EP receptors in
   various cancer types remains a currently unresolved issue, blockade of
   the EP receptors and/or the PGE2 biosynthetic pathway by corresponding
   antagonists may serve as a new treatment strategy for PGE2-positive
   tumors.

Acknowledgments