Introduction
The Long Interspersed Nuclear Element-1 (LINE-1, L1), representing a family of non-long-terminal repeat (LTR) transposable elements (TEs), constitute 17% of the human genome with ~ 500,000 copies widely distributed in the genome [
1]. Transcriptionally active L1s can propagate themselves and insert into a gene locus in the genome via the reverse transcription of the transposon [
2] (called L1 retrotransposition (LRT)). In other cases, through the activation of the L1 antisense promoter (L1-ASP), the L1s within an intronic region can also be transcribed into the adjacent exon of a gene through a splicing site to generate L1-gene chimeric transcripts (LCTs) which might affect the expression or functions of a gene [
3,
4]. The LCTs through L1-ASP activation were reported to have abnormally high expression in most cancer tissues and be associated with oncogenic activity [
5‐
7].
Increased L1 activity is associated with cancer, neural degenerative diseases and many other diseases. Therefore, the detection of LRTs and trans-splicing events is key to understand how L1s can alter gene expression or function leading to disease development and progression. Many strategies have been performed to observe LRTs at the DNA level based on the whole exome or genome sequencing and L1-targeted sequencing [
8‐
12]. While DNA sequencing may lack the resolution to capture trans-splicing events and quantify the level of L1 activities at the transcriptional level, alternative approaches have begun to place some emphasis on developing informatic tools to detect LCTs based on short-read RNA sequencing data [
13‐
15]. Because of technical limitations, such as requiring high sequencing depth for sequencing assembly or other restrictions, LCT events detected by these tools remain limited. In addition, these analytical tools rely on paired end sequencing data, making it difficult to infer from single cell sequencing data to study LCT events at a single cell level. Therefore, an analytical framework that allows for accurate detection of genome-wide LCTs at whole transcriptome or single cell level is needed in cancer biology to better interrogate the role LCTs play in oncogenesis.
Lung cancer remains among the most common cancer types which has estimated 1.6 million death each year [
16] and the 5-year survival rate of only about 18.1%, mainly due to the late diagnosis of advanced disease [
16]. Non-small cell lung cancer (NSCLC), with common subtypes as lung adenocarcinoma (LUAD) and lung squamous cell carcinoma (LUSC), account for ~ 85% of lung cancer patients [
17]. Results from whole exome sequencing (WES) of tumor samples have shown that the NSCLC corresponds with higher L1 activity among different cancer types [
18]. In our previous study, we detected 13 frequent and recurrent LCTs from RNA sequencing data of TCGA LUSC cohort by the DeFuse program, a gene fusion detection-based program with a limitation in genome-wide LCT detection [
19]. We detect one of the most prominent tumor-specific LCTs,
L1-FGGY, which is formed by transcription of L1 in the intron region into the exon 13 of FGGY through L1-ASP activation. Interestingly,
FGGY is involved with arachidonic acid metabolism and regarded as a putative tumor suppressor gene.
L1-FGGY interferes with the tumor suppressor function of
FGGY, thereby, promotes cancer cell proliferation, invasion and accelerated tumor growth corresponding with a tumor microenvironment deplete of immune cell populations to forge a cytotoxic response to tumor cells. This implicates the role of L1 in altering tumor cell metabolism thus allowing for the evasion of immunity in NSCLC [
19], thereby requiring further genome-wide LCT assessment and functional studies in tumorigenesis.
In this study, we generated a novel bioinformatic approach called, ReFuse (Retrotransposon-gene fusion estimation program), as a means to accurately detect LCTs at a genome-wide scale from both bulk and single cell RNA sequencing data of NSCLC with greater sensitivity. Our approach reveals that LCT frequently affect genes corresponding with mitochondrial biogenesis and energetics linked with overall metabolic capacities with the underlying oncogenic roles of L1-FGGY in driving metabolic reprogramming in NSCLC. Finally, we confirm that reverse transcriptase inhibitors can disrupt L1 activity that results in the metabolic programming as putative rationales for considering the treatment of patients observed with higher levels of L1 activity. Our study is the first to report a functional role of L1 activity resulting in the reprogramming of metabolism leading to changes in the tumor microenvironment leading to accelerated lung cancer progression. The intersection between LCTs and their corresponding targets in the human genome leading to increased oncogenesis could represent a promising prognosis biomarker and therapeutic target in NSCLC.
Discussion
In this study, we developed a local-alignment based bioinformatic tool to detect genome-wide LCTs with a high sensitivity in lung cancer in both bulk and single cell RNA sequencing data. We observe that LCTs are preferably involved among genes with mitochondrial function(s) and metabolic process, leading to the reprogramming of metabolism favorable for tumor cell growth and survival and subsequent progression leading to metastasis. In particular, in our comprehensive in vitro and in vivo functional studies of the LCT to reveal the highly recurrent L1-FGGY that directly influences oncogenesis by stimulating the 12-LOX pathway through the pathways that manage arachidonic acid and fatty acid metabolism which in-turn stimulates Wnt signaling and other oncogenic signaling pathways. Finally, a set of metabolic transcripts and LCT markers were determined to identify the patients with a high risk of poor outcomes who would benefit by treatments with NVR and other inhibitors of metabolism. Our investigations are initial studies using genome-wide and accurate detection of LCTs with high sensitivity using both bulk and single cell RNA sequencing data from lung cancer specimens and represents to reveal the functional role of L1 integration in metabolic programming in tumorigenicity of lung cancer. This could reflect the similar behavior of LCTs in other cancer types.
In previous study, we used the DeFuse program that was used to detect the fusion from two different gene loci, to identify a limited number of LCTs for LUSC. Considering this general approach may not detect many fusion events within a gene locus, we developed the ReFuse program, which can identify L1-chimeric transcript using a more sensitive detection methods and alignments of the genomic locations of the L1 position and LCT partner. Here, we have identified many LCTs with the majority involving splice-junction sites through L1-ASP activation that implicate intronic L1-ASP activation as a predominant mechanism of how L1 integration is involved in lung cancer tumorigenicity at the transcript level. However, due to the high similarity of short-read sequences of L1 families, there are limitations is revealing the exact origin of many L1s from short read RNA sequencing, therefore, long read sequencing (ISO-Seq) of the whole LCT could enhance the analysis as a framework and as a reference transcriptome for considering the origins of L1s to target their LCTs and further validate their frequencies and importance in human cancer.
Metabolic reprogramming has been shown as a means of cancer cells to navigate a microenvironment favorable to promote tumor occurrence and progression [
41,
42]. Specific metabolites can directly participate in the transformation processes or support the biological processes that facilitate tumor growth and progression [
43,
44] by a variety of internal and external factors. The downstream pathways of oncogenes and tumor suppressors have been reported to regulate the metabolism of cancer cells [
45]. Genomic changes may also lead to an increase or decrease in the expression of genes encoding metabolic enzymes [
46,
47]. As a widely distributed transposable element, L1 has been known as promoting tumor development by disturbing the transcription of tumor-related genes, as well as trigging genome instability, however, the roles of L1 in directing tumor cell metabolism and the surrounding microenvironment remains unclear. In this LCT study, we observed the L1s form LCT transcripts by preferably integrating with metabolic genes to affect mitochondrial function and metabolic process, which indicated a promising mechanism in cancer metabolism modulation, and even put forward a novel mechanism of L1 regulation in cancer development.
The balance of glucose and lipid metabolism is important for maintaining cell physiological homeostasis and normal biological functions [
48,
49]. Dysfunction of glucose and lipid metabolism can lead to a variety of diseases, especially in cancer [
50]. In LUSC, there is an obvious imbalance of glucose and lipid metabolism, which presents upregulation of fatty acid oxidation [
51,
52]. Targeted inhibition of fatty acid synthesis can effectively reduce tumor cell proliferation and invasion [
42], suggesting that lipid reprogramming is of great significance for maintaining the progression of LUSC. Therefore, breaking through the traditional therapy and mining and potential therapeutic targets at the metabolome level might improve the prognosis and outcome of LUSC.
FGGY is a metabolic gene which encodes carbohydrate kinase [
53].
FGGY was first reported to be related with sporadic amyotrophic lateral sclerosis [
54]. Besides that, FGGY can regulate dietary obesity in mice by regulating lipid metabolism [
55]. Here we showed that there was an abnormal fatty acid metabolism in
L1-FGGY+ LUSC. Specially,
L1-FGGY activated the downstream of AA metabolism, i.e. 12-LOX/GPR31/Wnt signaling. AA is an essential fatty acid, and its metabolites participate in a variety of physiological activities in cells, such as cell proliferation, migration, and apoptosis [
56]. AA metabolism includes three pathways: LOX, CYP450 and COX [
57]. Among them, LOX is a key enzyme in the metabolism of fatty acid, which catalyzes AA to generate HETE and other biologically active metabolites, which could affect cell metabolism and signal transduction, thereby playing an important role in cancer cells and inflammatory response [
58]. Fatty acids could promote the expansion of natural killer cells by improving energy metabolism, including enhancing the oxygen consumption rate (OCR), promoting ATP production and elevating the energy flux [
20]. PGE2, an AA metabolism, has been reported to enhance oxidative phosphorylation in macrophages [
59]. Here consistent with the analysis results from database which indicated the LCT transcripts affected mitochondrial function, we observed
L1-FGGY could promote mitochondrial oxidative phosphorylation activity, enhance membrane potential and produce more ATP levels. We noticed the increased degree of oxidative phosphorylation triggered by
L1-FGGY alone was modest, which indicated
L1-FGGY and other LCT transcripts also co-played roles in mitochondrial function alteration and thus metabolic reprogramming. Since the results from transmission electron microscopy (TEM) showed
L1-FGGY did not alter the morphology of mitochondria severely, we hypothesized that the accelerated energy metabolism and more energy production might be caused by the activated AA metabolism, but not the alteration of mitochondrial morphology.
GPR31 is a member of the G protein-coupled receptor (GPCR) family and is located in the cell membrane [
35,
60]. A large amount of evidence shows that GPCR is important for 12S-HETE-mediated signal transduction and may be involved in the progression of a variety of tumors [
35]. We found that
L1-FGGY increased the expression of GPR31 on the cell membrane, but did not affect its RNA level, suggesting
L1-FGGY may regulate the membrane expression of GPR31 through post-translational modification. It is reported in the literature that the level of GPCR membrane protein is regulated by deubiquitinating enzyme (USP) [
61]. USP regulates the ubiquitination level of the target protein GPCR to reduce GPCR degradation and increase its expression on the cell membrane [
62]. Here we discovered that
L1-FGGY down-regulated the expression of
FGGY, followed by reducing the binding of FGGY and USP24, which increased free USP24 to increase the deubiquitination level of GPR31, and eventually promoted its membrane expression and therefore activated 12-LOX/Wnt signaling.
The coordination between the immune, stromal and tumor cell populations within the microenvironment play a central role in navigating tumorigenicity and metastasis [
39]. Our study indicates that 12S-HETE, the 12-LOX metabolite, was involved in
L1-FGGY mediated AA pathway activation and immune microenvironment dysregulation. 12S-HETE has previously been identified as a mediator in inflammatory response [
63,
64]. In hepatocytes, treatment with 12S-HETE resulted in greater p65 (RELA), JNK, p38 and ERK phosphorylation and inflammatory gene expression, suggesting the proinflammatory action of 12S-HETE [
64]. While delivery of 12S-HETE to the airway of mice has been reported to attenuate allergic airway inflammation [
65]. Furthermore, 12S-HETE has been shown to promote secretion of IL-4 and IL-13, thereby polarizing macrophages to a more M2-like phenotype [
66]. In this study, we observed the
L1-FGGY directed elevation of 12S-HETE also involves proteins with functions in T cell activation and L1 activity is correlated with the down-regulation of T cell activation pathways, implicating a negative regulation of L1 in tumor immune microenvironment as an alternative mechanism for the tumor to escape from immune surveillance, and even suggesting a novel treatment by combined of targeting L1 and immunotherapy. However, more extensive functional and mechanism-based studies with immune cells are needed to elucidate the roles of L1 in innate and adaptive immune response activity associated with tumorigenicity and progression.
Here we showed L1 promoted tumor progression via coordinating its effects on multiple metabolic processes and immune activities. We not only proposed L1 as a candidate marker for cancer diagnosis and prognosis, but also suggested potential drugs to develop more effective cancer treatment strategies for patients carrying the LCT events.
Data and Method
L1 and Refseq reference
The repeat gene family annotation for hg38 was downloaded from the RepeatMasker [
67] database and repeat families annotated as “LINE/L1” were extracted for further analysis. We then selected the L1 families whose coordinates located within the upstream and downstream 50 kb of coding genes from NCBI Refseq database [
68]. The sequences of these L1 families were extracted from the human genome hg38 using bedtools getfasta [
69] based on their genome coordinates. makeblastdb of blast 2.2.26 + [
70] was used to build the reference for blast mapping. Meanwhile, all the repeat sequences from RepeatMasker were also extracted and genomeGenerate was used to build the reference for STAR mapping to remove false positive chimeric candidates.
Refseq RNA sequences was downloaded from NCBI data base. Repeatmasker 4.0.7 (
http://www.repeatmasker.org >) was used to identify and mask the repeat sequences in the RNA sequence using hmmer method. The repeat sequence identified from the RNA sequence was replaced with Ns. Two reference transcriptomes were generated with makeblastdb of blast 2.2.26 + , one with original Refseq sequence and one for masked reference. A bwa reference for the unmasked RNA sequence was also built with bwa index with default parameters [
71].
Refuse strategy
Raw RNA sequences were mapped to the Refseq transcript reference (unmasked) using bwa mem with default parameters. Unmapped reads and soft clipped reads were selected using samtools 1.9 [
72] and further aligned to L1 reference sequence using blastn in blast/2.2.26 + with default parameters. Reads with one end partially mapped to L1 family were trimmed and the unmapped end was blast against the Repeat-masked Refseq sequence using blast-short since some fragments might be short in length (Repeat-masked Refseq was used here to avoid false positive reads that mapped to different L1-repeat sequence in two ends). Then the reads that partially aligned to L1 and Refseq was identified. Considering balstn can only handle reads longer than 18 bp, some partially mapped reads might be missed. We then extend the identified reads ReadLength-1 bp along the corresponding L1 and Refseq sequencing from the junction site and constructed a candidate L1-chimeric sequence set. The candidate chimeric sequences are clustered with cd-hit [
73] with 100% identity to remove duplicates and then mapped to all repeat sequences using STAR [
74] in a gap allowed manner to remove false positive that containing repeat sequence fusions. Finally, all the unmapped reads were blast back to this candidate L1-chimeric sequence set to find more support reads. The final expression value for L1-chimeric transcript was quantified by the counting the supported reads.
Identification of differentially expressed LCT Events
Raw LCT-sample counts were normalized with the total RNAseq reads in each sample and then log transferred. Then two strategies were used to identify differentially expressed LCT events. 1) limma test [
75] was performed on the normalized expression value of LCTs in tumor samples against normal controls. The LCT events with L1 chimeric
p value < = 0.05, log2 fold change > 0 were selected as differentially expressed events. 2) For each event, the number of occurrences in cancer and control of LCTs and normal transcripts were counted and Fisher’s exact test was performed to calculate the significance p value. The LCT events with
p value < = 0.05 were selected as differentially expressed events. Finally, the union set of differentially expressed events identified in the two strategies were selected as the differentially expressed events.
Correlation analysis and survival analysis
The “M” methylation value was calculated as described in [
76] and the methylation value of each sample was calculated as the median methylation M value of all sites in the sample and the correlation with LCT expression was tested using Pearson correlation test. A sample level copy number was quantified by the sum of absolute Gistic2 copy number value of all the genes in each sample. The correlation of sample level copy number and the LCT expression was tested using Pearson correlation test. A sample level LCT activity was calculated as the sum of LCT supported reads and normalized by the total RNAseq read number of each sample.
All LCT activity related genes were identified by correlating the overall LCT expression level and individual gene expression levels using a Pearson correlation test. LCT correlated genes were identified with p value < = 0.0001 and further divided into positive related (> 0) and negative related (< 0) based on the estimated correlation value by Pearson correlation test between sample-level LCT and gene log2 FPKM expression. The Cox model was used for survival analysis.
Immune markers and GSEA analysis
In-house markers for immune cells populations were identified based on 7 individual PBMC single cell and 5 bulk RNAseq data sets of sorted immune cell. For each cell type, their expression profile was compared against other cell types one by one with Wilcoxon rank sum test. Genes with
p value < 0.01 and mean log
2 fold change larger than 0 in all comparisons was selected as the specific markers of the cell type. Each cell type was calculated in the same way in each sample and markers identified more than 3 times was selected as the meta-marker for the cell type. A “gmt” file for GSEA analysis was then constructed using the identified meta marker list [
77]. GSEA analysis was done by “fgsea” R package with fold change for each gene calculated using limma [
75].
Single cell clustering, trajectory and differential analysis
Expression profile of raw gene counts on cell level of each sample was downloaded from Array Express. Genes expressed in less than 3 cells and Cells expressing more than 8000 genes were filtered. Cells whose mitochondrial RNA content takes larger than 15% of the total RNA were filtered. Filtered cells are further clustered using Seurat 3.1.5 [
78] with resolution 0.8 and first 10 PCA dimensions, and cell clusters were annotated based on the markers in the original paper and another single cell papers on lung cancer [
79]. Differentially expressed genes between cell types and L1versus non-L1 cells were identified using Wilcoxon rank sum test with p value less than 0.01 and log (fold change) greater than 0.25.
Apply refuse on single cell RNAseq data
Raw sequence data of the single cell study were downloaded from Array Express. The file containing RNA sequence reads (R2 for 10 × v2 and R1 for 10 × v3) of each sample was submitted to run on Refuse to identify LCT events as single end bulk RNAseq data. The barcodes of LCT supporting reads were extracted in the barcode sequencing file and clustered using CD-hit [
73] using end-to-end mode with identity score larger than 0.85 to allow for sequencing errors. Barcodes clustered together were annotated as the same cell and the barcode found in expression profiles was selected as the representative barcode of the cell cluster.
LCT affected candidate genes and drug repurposing
Genes identified to be positively correlated with overall survival related LCT activity (
p < = 0.0001 and estimate > 0) and differentially expressed by cells with L1 versus without L1 in > 3 cell types/patients (myeloid, T cell and tumor cell in 5 patients, totally 15 comparisons) were selected as candidate genes affected by LCT. Those genes were submitted to the CMAP database [
80] for drug repurposing for potential treatment on patients with high LCT activity.
The log2 RPKM gene expression in all the patients of each candidate gene was z-normalized, and an aggregate score was calculated by summing up the z score of all the 47 candidate genes for each sample. The patients were divided into 4 groups with the threshold of aggregate gene score and survival related LCT expression. By iterating the two thresholds, a high-risk group was identified having 50% fatality rate within 3 years, which could be highly related to LCT activity.
Potential treatment for LCT related patients
Raw sequence data of cancer cell line treated with Metformin and simvastatin were generated from NCBI GEO data base with accession number GSE146982, GSE141052 and GSE149566 [
81,
82]. LCT analysis was performed with ReFuse and the overall LCT levels was summarized by adding up all the junction reads supporting LCT events and normalized with the read depth in each sample. For simvastatin data sets, few LCT were found as it is not from a cancer cell line, so we quantified the L1 activity with L1 reference sequence. First, the raw sequences were mapped to Refseq using bwa mem 0.7.15 [
71] and the unmapped reads were then mapped to L1 reference using STAR 2.7.0f [
74]. The overall L1 activity was quantified by counting all the reads mapped to L1 reference and normalized by the read depth in each sample.
All the LUSC patients were obtained from Cancer Biobank of Tianjin Medical University Cancer Institute and Hospital (TJMUCH, Table S20-S21) which were treated with partial lung resection surgery at the Department of Lung Cancer of TJMUCH. No prior treatments, including chemotherapy or radiotherapy, were conducted before lung resection surgery was performed. This project was approved by the Ethics Committee of Tianjin Medical University (Approved No.: Ek2020111) and written informed consents were obtained from the patients. All experiments were performed in accordance with the principles of the Declaration of Helsinki.
Cell lines
NCI-H520 and SK-MES-1 were purchased from Cellcook Co., Ltd. with cell authentication via STR multi-amplification method. KLN205 was obtained from Chinese Academy of Medical Sciences tumor cell libraries. NCI-H520 was cultured in RPMI1640 (Gibco BRL). SK-MES-1 was cultured in MEM. KLN205 was cultured in H-DMEM. All medium contained 10% FBS and 1% penicillin/streptomycin. For 12/15-LOX inhibitor treatment experiments, ML355 and PD146176 was diluted in medium, followed by replacing the cell medium 5 h after cells seeded respectively. Cell lines were routinely evaluated for Mycoplasma contamination. All experiments were completed less than 2 months after establishing stable cell lines or thawing early-passage cells.
Mouse models
The DBA2/2NCrl mice are an inbred line and were obtained from SPF biotechnology Co. Ltd. (Beijing). The weights and tumor sizes of each mouse were monitored every 2 days. Each experimental group contained 5 mice. The tumor volume (V) of the xenograft was calculated by the formula: V = π × L × W × H/6 (L: length, W: width, H: height). For drug treatment studies, animals were subjected to treatment with either NVR or ML355 every day, or subjected with the two inhibitors simultaneously. All animal protocols were approved by the Ethics Committee for Animal Experiments of TJMUCH (Approved No.: NSFC-AE-2020101), and was performed in accordance with the Guide for the Care and Use of Laboratory Animals.
Lentivirus construction
The construction of human
L1-FGGY insertion lentivirus was performed as previously described [
19]. For mouse
L1-FGGY insertion lentivirus construction, the recombinant lentivirus with
L1-FGGY sequence was generated by co-transfection in the packaging KLN205 cells. For
GPR31/
USP24 knockdown lentivirus construction, the recombinant lentivirus with
GPR31/
USP24 shRNA sequence (constructed by Hanbio Co., Ltd.) was generated by co-transfection in H520
OV−L1−FGGY cells as previously described [
19].
RNA was extracted with TRIzol™ reagent (Life Technologies). Reverse transcription was performed with PrimeScript™ RT Master Mix (Takara) according to the manufacturer’s instructions. qPCR was performed using TB Green™ Premix Ex Taq™ (Takara) and ABI PRISM 7500 real-time PCR System (Applied Biosystems). The primers used are shown in Table S23. The relative mRNA levels were calculated as previously described [
19].
RNA library preparation, sequencing and enrichment analysis
Library preparation and sequencing steps were performed as previously described [
19]. Briefly, the libraries were sequenced on Illumina® (NEB) following manufacturer’s recommendations. The RNAseq data has been uploaded to GEO database (accession number: GSE181042 and GSE181043). Raw sequencing data was aligned to hg38 reference using STAR 2.5.3a [
74] and HTseq 0.11.2 [
83] was used to quantify the gene-sample expression profiles. Differentially expressed genes (DEG) were identified with limma voom [
84] with FDR adjusted
p value < 0.01. KEGG and GO function enrichment analysis of the interested gene sets were performed using clusterProfiler package [
85].
Quantitative proteomics
The quantitative proteomic studies were performed by Jingjie PTM BioLab Co. Ltd (Hangzhou) as previously described [
86]. Briefly, the protein extracted was digested and then the resulting peptides were desalted, reconstituted, tandem mass tag labeled, and analyzed by Liquid chromatography-tandem mass spectrometry (LC–MS/MS). Tandem mass spectra were searched against human Uniprot database (
http://www.ebi.ac.uk/uniprot/) concatenated with reverse decoy database.
AA metabolite detection was performed by Shanghai Applied Protein Technology Co., Ltd. Cells were homogenized on ice in a mixture of chloroform, methanol and water. The samples were then centrifuged and the supernatant was transferred to an LC sampling vial. The deposit was rehomogenized with methanol and supernatant was added to the same vial. After reconstituted with mobile phase, the extract as well as reference standards were analyzed with ACQUITY ultra performance liquid chromatography coupled with mass spectrometer (Waters). UPLC-MS raw data obtained with negative mode were analyzed using TargetLynx applications manager to obtain calibration equations and the quantitative concentration of each AA metabolite in the samples.
Gene expression (RNA) profiling: NanoString methodology
Gene expression analysis was conducted on the NanoString nCounter gene expression platform (NanoString Technologies) as previously described [
87]. Briefly, RNA was mixed in a 3′-biotinylated capture probe and 5′-reporter probe tagged with a fluorescent barcode. Probes and target transcripts were hybridized overnight. Hybridized samples were run on the NanoString nCounter preparation station by using a high-sensitivity protocol. The cartridge samples were scanned at maximum resolution by using the nCounter digital analyzer. GEP scores were calculated as a weighted sum of normalized expression values for the genes.
Cell proliferation assay
The cell proliferation was detected by Cell Counting Kit 8 (CCK8) proliferation assay as previously described [
19]. Briefly, cells were trypsinized and incubated with CCK8 for 4 h. Then the absorbance reading at 450 nm was taken by a microplate reader (Synergy HT).
Wound healing assay
The wound healing assay was performed as previously described [
19]. Briefly, when the seeded cells reached 80 ~ 90% confluency, we made a straight line in the cell monolayer. At 0 and 48 h, images were obtained, and the distance of the wound was measured.
Transwell invasion assay
The transwell invasion assay was performed as previously described [
19]. Briefly, we seeded cells in Matrigel and serum-free RPMI-1640. Medium supplemented with 10% FBS was placed in the lower chamber of the Transwell. After 48 h’ incubation, we fixed the cells on the membrane’s lower surface and subjected them to staining with 1% toluidine blue. After staining photographs were taken under a microscope, the number of invading cells was recorded.
Enzyme-linked immunosorbent assay (ELISA)
The levels of 5S-HETE, 12S-HETE, 15S-HETE, PGD2, PGE2, PGF2, Wnt3a and Wnt5a either in cell culture supernatants or from tissue samples were measured using commercially available ELISA kits (Abcam and Bioswamp) according to the manufacturer’s instructions.
Western blot
Proteins were electrophoresed by SDS/PAGE and blots were incubated overnight with primary antibody. The following antibodies were used: anti-GPR31 (Abcam, ab75579), anti-ATPase Na+/K+ β2 (Bioss, bs-1152R), anti-Wnt3a (Cell signaling Technology (CST), #2391), anti-Wnt5a (CST, #2530), anti-pGSK-3β (phospho Ser9, CST, #5558), anti-JNK (phospho Thr183/Tyr185, CST, #4671), anti-β-catenin (CST, #8480), anti-HA tag (CST, #5017), anti-flag tag (CST, #14,793), anti-USP24 (Proteintech, 13,126–1-AP), anti-FGGY (Abnova, ABN-H00055277), anti-β-Actin (CST, #4967), and anti-GAPDH (Abcam, ab181602). After incubated with HRP-conjugated α-rabbit or α-mouse secondary antibodies for 1 h, protein bands were detected with chemiluminescence substrate (Perkin Elmer) using the ChemiDoc Imaging System (Bio-Rad).
Immunoprecipitation
Cell lysates were harvested using lysis buffer, rotated at 4 °C and as previously described [
88]. Lysates were clarified by spinning. Protein concentrations were measured using BCA standard curves (Pierce). Flag antibody (for binding to flag-GPR31, Thermo Fisher Scientific) was added to protein lysate and rotated overnight. IP was carried out using the Invitrogen Dynabeads Protein G Immunoprecipitation Kit as directed. Lysates were next subjected to SDS-PAGE and immunoblot analysis. Each immunoprecipitation experiment was performed a minimum of twice.
Immunohistochemistry
All procedures were performed as described above [
89]. The antibodies are as follows: anti-FGGY (Abcam), anti-12-LOX (Abcam), anti-15-LOX (Abcam), anti-GPR31 (Abcam), and a biotinylated secondary goat anti-mouse IgG antibody (Santa Cruz), labeled with HRP using a DAB staining kit (Maixin Biotechnology) according to the manufacturer’s instructions. For negative controls, IgG1 was used. Positively stained cells were counted in 5 fields at 200 × magnification.
Flow cytometry
Cells were incubated with different antibodies for 30 min at 4 °C as indicated. The following antibodies were used: PerCP anti-mouse CD45, APC anti-mouse CD3, FITC anti-mouse CD4, PE anti-mouse CD8, PE anti-mouse CD11c, and FITC anti-mouse CD86. We selected isotype-matched immunoglobulin G1 antibodies (BD Biosciences) to serve as a negative control. The cells were analyzed on a BD FACS CantoTM II flow cytometer and FlowJo software (BD Biosciences).
Multispectral immunofluorescence (IF) staining
We performed multispectral IF staining as previously described [
90]. In brief, the slides were deparaffinized and rehydrated. After antigen retrieval and blocking, the primary antibody was applied and incubated overnight. Opal polymer HRP was used as the secondary antibody. The slides were washed, and tyramide signal amplification (TSA) dye (PerkinElmer) was applied. The slides were then exposed to microwaves to remove the primary and secondary antibodies, washed, and blocked again. Afterward, other primary antibodies, as well as DAPI were applied successively. Finally, slides were placed on a coverslip. Five fields at 200 × magnification was imaged and recorded, and StrataQuest Image Analysis software was used to generate a spectral library for unmixing.
Transmission electron microscopy (TEM)
Cells were fixed with 2.5% glutaraldehyde, postfixed with 0.5% osmium tetroxide and contrasted using tannic acid and uranylacetate. Specimens were dehydrated in a graded ethanol series and embedded in Polybed. Ultrathin sections were analyzed in a HT7800 transmission electron microscope.
Measurements of oxygen consumption and extracellular acidification
The rates of oxygen consumption rate (OCR) and extracellular acidification rate (ECAR) in various cell lines were measured with a Seahorse Bioscience XF-24 extracellular flux analyzer, as detailed previously [
91,
92]. Cells were seeded at a density of 1 × 10
4 cells per well on Seahorse XF-24 polystyrene tissue culture plates. Inhibitors were used at the following concentrations: Oligomycin (1.5 μM), Carbonyl cyanide 4-trifluoromethoxy-phenylhydrazone (FCCP) (0.8 μM), Antimycin A (1.5 μM) and Rotenone (3 μM).
Assessment of mitochondrial membrane potential (MMP)
JC-1 was used to measure the MMP according to the manufacturer’s instructions (Bioss). Cells in 6-well plates were incubated with JC-1 staining solution at 37 °C for 30 min and then washed with JC-1 buffer. Fluorescent signals were obtained using flow cytometry.
Measurement of intracellular ATP levels
ATP levels were measured using the Enhanced ATP Assay Kit (Beyotime Biotechnology) according to the manufacturer’s instructions. The substrate was gently mixed with reaction reagent at room temperature. The luminescence was then measured using a Beckman Coulter.
Statistical analysis
Data were statistically analyzed with SPSS 20.0 and GraphPad Prism 5.0 software following the manufacturers’ instructions. Measurement data were expressed as means ± standard deviations. We analyzed correlations between 2 datasets using Spearman’s correlation coefficient. One- and two-way analysis of variance with subsequent Bonferroni post-hoc tests was used for comparisons between 2 groups. Cumulative survival was determined via the Kaplan–Meier method. All data were normally distributed. P < 0.05 was taken to indicate a statistically significant result.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.