Background
Metastases are responsible for the majority of deaths due to breast cancer. Metastasis is a multistage process that occurs through a sequence of steps involving dissociation of cells from the primary tumor, local invasion and migration through the basement membrane and into the circulation, extravasation into foreign organs, and finally tumor growth at distant sites [
1]. Triple-negative (negative for estrogen receptor, progesterone receptor, and human epithelial receptor 2 (HER2) gene amplification) breast cancer (TNBC) is an aggressively metastatic subtype with a disproportionately high rate of
TP53 mutation compared to other breast cancer subtypes.
The tumor suppressor protein p53 is lost or mutated in about half of all human cancers, and in tumors where this gene (
TP53) is wild-type (WT), mechanisms frequently exist to inactivate the protein. The majority of
TP53 mutations in basal-like breast cancer, an intrinsic breast cancer subtype that largely overlaps with TNBC, are insertions and deletions that result in
TP53 truncation and loss of function [
2]. p53 loss disrupts pathways that inhibit metastasis and activates pathways that promote metastasis. The pathways that are altered by p53 loss regulate multiple stages of the metastatic cascade, including the acquisition of stem cell-like properties, interactions with the extracellular matrix, adhesion and migration [
3,
4]. In addition, p53 loss disrupts cell cycle checkpoints and protects incipient tumor cells from undergoing apoptosis or entering senescence, which in turn, creates opportunities for tumor evolution and metastatic progression [
5‐
7]. Furthermore, some
TP53 mutants confer additional functions that promote metastasis [
8,
9]. Thus, the metastatic potential of tumors can be enhanced by loss of p53 or by expression of gain-of-function p53 mutants. However, studies conducted in vivo indicate that p53 loss alone is insufficient for metastasis [
4,
8‐
10]. Interestingly, a genomic study of treatment-naïve TNBC revealed that p53 loss or acquisition of somatic mutations does not always emerge as a founding event [
11], suggesting that disruption of p53 function also can influence late stages of tumor development. The presence of gain-of-function and loss-of-function mutations in breast cancer warrants a thorough characterization of these mutations in tumor progression. In this study, we specifically studied the contribution made by p53 deficiency to metastasis in late-stage triple-negative breast cancer.
Most of the existing preclinical breast cancer xenograft models used to study metastasis to lung or bone involve injecting human cancer cell lines that have been extensively cultured ex vivo into the tail vein or left ventricular chamber of the heart, respectively. These methodologies bypass all early steps in the metastatic cascade, including escape from the primary site and survival in circulation. By contrast, orthotopic patient-derived xenograft (PDX) models of breast cancer are derived by engrafting tumors obtained directly from patients into the mammary fat pads of immune-compromised mice. Human breast tumors have been shown to metastasize to physiologically relevant organs in these models, and as such, orthotopic PDX models enable all stages of the metastatic cascade to be studied within a more advanced biological context in vivo [
12‐
15].
We engineered paired isogenic PDX lines differing only in p53 status to develop a breast cancer metastasis model of TNBC that enabled longitudinal studies in mice [
16]. By studying the effect of p53 loss in an already-metastatic PDX line, we investigated whether p53 loss impacted late stages of tumor progression by examining various stages of the metastatic cascade over time. The contributions made by p53 silencing to breast tumor growth, escape from the mammary gland, homing and colonization of distant organs, and tumor growth at metastatic sites were investigated. In addition, gene expression profiling was conducted to identify p53 effectors that regulate metastasis.
Methods
Study approval
This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals from the National Institutes of Health (NIH) Institutional Animal Care and Use Committee (IACUC). The protocol was approved by the Committee on the Ethics of Animal Experiments of Washington University and the IACUC at MD Anderson Cancer Center. Mice were euthanized when they became moribund and when they reached defined study end points. Animals were euthanized as dictated by the Association for Assessment and Accreditation of Laboratory Animal Care International and IACUC euthanasia end points.
Establishment of PDX models of TNBC
PDX models were established according to published protocols [
17]. Briefly, 2.5 × 10
5 immortalized human mammary stromal fibroblasts, derived from a patient who underwent a reduction mammoplasty, were irradiated (400 Rads) and mixed with 2.5 × 10
5 nonirradiated cells. Fibroblasts were then mixed with 1 × 10
6 tumor cells stably expressing click beetle red luciferase (CBR-luc)/mCherry and 1/3 volume Matrigel. This mixture was injected into the fourth mammary fat pads of 3- to 4-week-old nonobese diabetic/severe combined immunodeficiency (NOD/SCID) mice (ordered from the National Cancer Institute). For metastasis studies, p53-deficient tumors were resected before they reached 1 cm in diameter (approximately 9 weeks post-engraftment) to enable time for metastasis to occur before mammary tumor burden reached the maximum allowable size. For studies monitoring mammary tumor growth, tumors were not resected and mice were euthanized when tumors reached 2 cm in diameter.
Bioluminescence imaging
Bioluminescence imaging (BLI) was performed in vivo as previously described [
18]. Briefly, after injection of 150 μg/g D-luciferin (Biosynth, Staad, Switzerland) in phosphate-buffered saline (PBS), intraperitoneal (i.p.), isoflurane-anesthetized mice were imaged with a charge-coupled device camera-based BLI system (IVIS Lumina and IVIS Spectrum; PerkinElmer, Waltham, MA, USA). Signals were displayed as photons/s/cm
2/sr. Regions of interest were defined manually using Living Image Software, and data were expressed as total photon flux (photons/second). The first images were taken 2–3 weeks after tumor implantation (PDX models), or immediately after tail vein injection of labeled cells, and weekly thereafter. To assess the appearance of metastases in the axillary lymph node, mammary tumors were covered to block signal from the primary site and allow visualization of metastases. To quantify organ distribution, D-luciferin was administered to live animals, and tissues were assessed with BLI ex vivo at necropsy.
Mice were euthanized a minimum of 19 weeks post-engraftment, and organs were subjected to BLI ex vivo. Organs exhibiting regions of bioluminescence with Gaussian distribution were counted as one nodule.
Blood processing for circulating tumor cell (CTC) analysis
Blood was extracted by cardiac puncture in the presence of heparin. Whole blood was immediately transferred to red blood cell (RBC) lysis buffer (Sigma-Aldrich, St. Louis, MO, USA; R7757), inverted 10 times, and placed on ice. Samples were centrifuged at 1500 rpm for 6 min, RBC lysis buffer was carefully removed, and an additional 7.5 ml of RBC lysis buffer was added. Samples were again centrifuged at 1500 rpm for 6 min. Pellets were resuspended in 350 μl of PBS with 0.5 % bovine serum albumin (BSA) and assessed immediately by flow cytometry. Whole blood from non-tumor-bearing mice was processed alongside tumor-bearing mice as negative controls. Whole blood from non-tumor-bearing mice was spiked with 20,000 BC3-p53 knockdown (KD) cultured cells as a positive control.
Flow cytometry for circulating tumor cells
For CTC studies, samples were analyzed on an Influx cell sorter (BD, Franklin Lakes, NJ, USA). The entire sample was flowed regardless of total cell number. Dead cells were gated out by SyTox Blue staining and forward/side scatter. Gates were set such that no cells appeared in the mCherry + gate in negative controls. CTCs were identified as cells positive for mCherry fluorescence.
Cell culture and lentiviral transduction of Washington University Breast Cancer line 3 (WU-BC3) cells
BC3-p53WT and BC3-p53KD cells were grown in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10 % fetal bovine serum (FBS) and 1 % penicillin-streptomycin. Cells were cultured at 37 °C, 5 % CO2, and 5 % O2 in humidified air. When logarithmically growing cells reached approximately 50 % confluency, they were transduced with lentivirus encoding CBR-luc and mCherry (FUW-CBR-luc-mCherry) in the presence of 1 μg/ml polybrene. Cells were selected and subsequently grown in the presence of puromycin. Lentivirus encoding BTG2 was produced by transfecting 293 T cells with the target plasmid along with the packaging vectors pDELTA8.9 and VSVG. Virus was removed from cells and passed through a 40 μm filter 72 h after transfection. To express BTG2 in BC3-p53KD, cells were transduced with pHAGE-BTG2 or control lentivirus (pHAGE-GFP-eGFP). Cells were transduced at a multiplicity of infection (MOI) of 1 in the presence of 10 μg/ml polybrene. Flow cytometry was used to isolate green fluorescent protein (GFP)-positive cells. Expression of BTG2 was assessed by quantitative polymerase chain reaction (qPCR).
Ribonucleic acid-sequencing (RNA-Seq) experiments and data analysis
Tumors were harvested when they reached 0.5 cm and were immediately snap frozen in liquid nitrogen to preserve RNA integrity and to avoid experimental manipulation that could lead to alterations in gene expression. Tumors were stored at -80 °C, thawed on ice in Trizol, and homogenized with a pedestal homogenizer. Total RNA was extracted with chloroform followed by purification using the RNEASY kit (Qiagen, Venlo, The Netherlands; 74104). Total RNA was treated with DNase (TURBO DNase, Life Technologies, Carlsbad, CA, USA; AM2238) followed by column purification (RNA Clean & Concentrator-5, Zymo Research, Irvine, CA, USA; R1015). mRNA was isolated by poly-A selection. RNA Integrity Number (RIN) was assessed on a Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA), and only samples with RIN ≥ 8 were sequenced. RNA-Seq libraries were prepared and sequencing was performed by the Genome Technology Access Center (Washington University, St. Louis, MO, USA). The RNA-Seq reads were first mapped to the mouse (NCBI Build 37.2) (GRCm37UCSC mm9) reference genome using Tophat2, allowing a maximum of two mismatches per 75 bp read end. The unmapped reads were then aligned to the human genome (NCBI Build 37.2) (GRCh37/UCSC hg19) using the same tool and parameters. Partek Genomics Suite v6.6 and its genomic feature database (RefSeq Transcripts - 2014-01-03) were used to quantify the gene-level read counts. The differential analyses for gene/isoform expression were performed using DESeq2. The log2 fold change 1.5 and p value cutoff <0.05 were used to identify differentially expressed genes. Pathway analysis was performed using the GeneGo application of the MetaCore program (Thomson Reuters, New York, NY, USA).
Real-time polymerase chain reaction (RT-PCR)
Total RNA was isolated by Trizol-chloroform extraction followed by purification using the RNEASY kit as described above. RNA was DNase-treated as described above. RIN was assessed as described above, and only samples with RIN ≥8 were used for qPCR analyses. A dynamic range test (standard curve) was performed using reverse transcription (RT) conditions to determine the range of RNA to be used in all subsequent reactions. Each 40 μl RT reaction consisted of no more than 4 μg template RNA. RNA was diluted 1:10 and reverse transcribed using the Superscript III system (Invitrogen, Carlsbad, CA, USA). The resulting cDNA was used for qPCR using the TaqMan Gene Expression Assays (Applied Biosystems, Foster City, CA, USA), and data were normalized to a multiplexed endogenous control, GAPDH. No-template and no-RT controls were run on each plate, and amplification was not observed for any samples. qPCR was performed on the ViiA 7 Real-Time PCR System (Life Technologies).
Immunohistochemistry (IHC)
Mammary tumors were fixed in 10 % neutral buffered formalin for 24–48 h. Samples were then washed three times with PBS and transferred to 70 % ethanol, then embedded in paraffin. Five micron sections were cut and baked at 65 °C for 60 min. Sections were deparaffinized by immersing in xylene (Thermo Fisher Scientific, Waltham, MA, USA) three times for 5 min each and rehydrated by immersing twice through a decreasing gradient of alcohol (100 %, 95 %, 70 %, 50 %, and distilled H20) for 2 min each. Antigen retrieval was carried out by boiling samples in rodent decloaker agent (Biocare Medical, Concord, CA, USA; RD913M) for 30 min followed by cooling at room temperature for an additional 15 min. Endogenous peroxidase activity was blocked by incubating sections in Peroxidase Blocking Reagent (Dako, Glostrup, Denmark) for 15 min at room temperature. Nonspecific interactions were blocked by incubating sections in Protein Block (Dako) for 1 h at 4 °C. Primary and secondary antibodies were diluted in Antibody Diluent (Dako). Antibodies were diluted as follows: cytokeratin 18 (CK-18) (Abcam, Cambridge, UK; Ab82254), 1:200; phospho-histone H3 (pHH3) (Sigma-Aldrich; H9908), 1:1500; cleaved caspase 3 (CC3) (Cell Signaling Technology, Danvers, MA, USA; 9661), 1:200. For CC3, the ImmPRESS Reagent Anti-Rabbit IgG kit was used (Vector Laboratories, Burlingame, CA, USA; MP-7401). For pHH3, the ImmPRESS Reagent Anti-Rat IgG kit was used (Vector Laboratories MP-7404). For CK-18, the ImmPRESS M.O.M kit (Vector Laboratories; MP-2400) was used.
Statistics
To normalize the photon flux values to the time of study end point, the photon flux and the logarithm of the photon flux in each organ were plotted versus time to euthanasia. Plots of the photon flux showed that most values increased approximately exponentially over time, and plots of the logarithm of the photon flux indicated that most values were fitted by a straight line [
19,
20]. We therefore assumed that the photon flux increased exponentially over time. Thus, the normalized photon flux was obtained by dividing the photon flux measured in each by the exponential value of its time to euthanasia. Since the sample sizes in most groups were relatively small, Wilcoxon rank sum test was used to evaluate the mean differences in the photon flux values in each organ, time to euthanasia, and time to lymph node metastasis between the groups of mice. F-test was used to account for the time effect in the analysis of variance (ANOVA) model to assess whether p53 silencing was significantly correlated with the log2-transformed CTC counts. Statistical analyses were performed using R (
http://www.r-project.org/).
Survival analysis
The univariate Cox proportional hazard model was used to assess the correlation of gene expression with patient overall survival and metastasis-free survival (Table S3 in Additional file
1), and the likelihood ratio test
p values were reported. The dataset from The Cancer Genome Atlas (TCGA) [
21] was used for determining overall survival of breast cancer patients and TNBC patients. Datasets from the San Diego cohort [
22], the NKI cohort [
23], and the Oxford cohort [
24] were used for determining metastasis-free survival. Log-rank tests were used to estimate the
p values between high and low expression groups (using top and bottom 25 % as the cutoff for grouping).
Preparation of tumors for flow cytometery for stem markers and mammosphere formation assays
For use in flow cytometry and mammosphere assays, BC3-p53WT and BC3-p53KD tumor cells not expressing CBR-luc-mCherry were implanted contralaterally into the fourth mammary gland (n = 4 tumors BC3-p53WT, n = 4 tumors BC3-p53KD). When tumors reached 0.5 cm in diameter, they were isolated from mammary glands and dissociated into single cells and organoids in collagenase and hylauronidase for 4 h at 37 °C on a rotator. Tumor cells were washed in DMEM/F12 with 5 % bovine calf serum, and red blood cells were lysed for 5 min at room temperature (RBC lysis buffer; Sigma-Aldrich). Cells were passed through a 100 μm, then 40 μm filter. Mouse stromal cells were depleted using magnetic-activated cell sorting (Miltenyi Biotec, Bergisch Gladbach, Germany; MACS mouse cell depletion resin) according to the manufacturer’s instructions. Effective depletion of mouse cells was confirmed by staining with a mouse-anti-H2kd (mouse-specific MHC class I H-2Kd haplotype) antibody conjugated to PE/Cy7 (BioLegend, San Diego, CA, USA clone SF1-1.1), followed by flow cytometry.
Flow cytometery
Following tumor digestion and mouse cell depletion, cells were immediately placed in PBS + 0.5 % BSA and stained with antibodies for flow cytometry for 20 min at 4 °C at a concentration of 5 million cells/mL. Antibodies used were: rat anti-CD44 conjugated to allophycocyanin (APC) (BD Pharmingen San Jose, CA, USA; clone IM7), mouse anti-CD24 conjugated to fluorescein isothiocyanate (FITC) (BD Pharmingen clone ML5), mouse anti-GD2 (unconjugated, BD Pharmingen clone 14.G2a), goat-anti-mouse secondary antibody conjugated to APC (BD Pharmingen). Cells were stained with Ghost Dye Violet 510 (Tonbo Biosciences, San Diego, CA, USA) for viability. After staining, cells were fixed in 1 % paraformaldehyde then analyzed by an LSRFortessa X-20 flow cytometer (BD).
Immediately following tumor digestion and mouse cell depletion, cells were plated at a concentration of 1000 cells per well on an ultra-low-binding 96-well plate (Corning, New York, NY, USA) in complete MammoCult medium (STEMCELL Technologies, Vancouver, BC, Canada), supplemented with 1 % methylcellulose to immobilize cells. Mammospheres were manually counted 12 days after plating. Primary mammosphere formation efficiency was calculated as percent of mammospheres formed divided by total number of cells plated. Primary mammospheres were collected, dissociated into single cells using TrypLE Express (Life Technologies), and replated in MammoCult containing 1 % methylcellulose to form secondary mammospheres. Secondary mammospheres were manually counted 12 days after plating.
3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assays
Cells were seeded in 24-well plates (20,000 cells per well) in complete cell culture media. Four consecutive time points were harvested post-seeding (24, 48, 72 and 96 h). Every 24 h, plates were incubated with MTT reagent (0.5 mg/ml) for 1 h at 37 degrees, 5 % CO2, and 5 % O2 in humidified air (normal cell culture conditions). After incubation with MTT reagent, media was removed from cells, and 500 μl DMSO was added per well to solubilize purple formazan crystals. The plate was then read on a Clariostar plate reader (BMG Labtech, Cary, NC, USA) at 570 nm for formazan absorbance and 690 nm for background absorbance, and these values were normalized by subtraction. Each condition was performed in triplicate.
Cell proliferation assays
Cells were seeded in 6-well plates (80,000 cells per well) in complete cell culture media. Four consecutive time points were harvested post-seeding (24, 48, 72 and 96 h). Every 24 h, cells were trypsinized, resuspended to break remaining cell clusters, loaded onto cell counting chambers, and read on a CellOMeter (Nexcelom Bioscience, Lawrence, MA, USA). Each condition was plated in biological triplicates, and each well was counted in duplicate. Each chamber was read in triplicate.
Western blotting
Samples were lysed in Mammalian Cell Lysis Buffer (50 mM Tris, pH 8.0, 2 mM dithiothreitol (DTT), 5 mM ethylenediaminetetraacetic acid (EDTA), 0.5 % NP-40, 100 mM NaCl) with added protease inhibitors (1 μM mycrocystin, 2 mM phenylmethanesulfonyl fluoride (PMSF), 1 × protease inhibitor cocktail (Sigma-Aldrich; P8340-5ML), 1 × phosphatase inhibitor cocktail (Calbiochem, San Diego, CA, USA; 524625-1SET)), and 10–30 μg total protein was loaded onto Bio-Rad (Hercules, CA USA) Criterion gels and transferred to nitrocellulose for Western blotting using standard procedures. Antibodies used for Western blotting included those recognizing BTG2 (sc-33775, Santa Cruz Biotechnology Inc., Dallas, TX, USA), and α-tubulin (2144, Cell Signaling Technologies). Protein was detected using enhanced chemiluminesence (ECL) (34080, Thermo Fisher Scientific) on a G-Box imager (Syngene, Frederick, MD, USA).
Ethics, consent, and permission
All animal experiments were authorized by the IACUC committee at MD Anderson. RNA-seq data analyzed in this manuscript are publically available at (GEO Accession number GSE76433). We confirm that this study did not involve human patients and no consent was required.
Discussion
Using a paired set of PDX models of TNBC differing only in p53 status that metastasize to similar organs as those observed in breast cancer patients, we demonstrated that p53 silencing enhanced tumor growth in both primary and metastatic sites. The BC3 lines used in this study were generated from a patient with metastatic breast cancer whose tumor was wild-type for p53. Thus, expression of wild-type p53 alone was insufficient to inhibit tumor metastasis in the patient and in PDX models derived from the patient’s tumor. However, p53 loss augmented metastatic potential by enhancing proliferation and reducing apoptosis and this, in turn, led to enhanced tumor growth. Thus, one contribution made by p53 loss to late-stage tumor progression was to promote tumor growth.
Previous studies conducted in vitro demonstrated that p53 inhibits pathways that promote tumor escape from the primary site, including ECM remodeling, connective tissue degradation, cell adhesion and EMT [
3,
4]. Indeed, differential gene expression analysis revealed that these pathways were also deregulated upon p53 silencing in BC3 cells (Table S2 in Additional file
7). Despite these expression changes, p53 silencing in BC3 tumor cells did not significantly increase the shedding of CTCs into the bloodstream of tumor-bearing mice. However, based on the methods employed to detect CTCs, we cannot rule out the possibility that silencing of p53 increased CTC aggregation, thus improving their ability to seed secondary sites [
41]. However, it seems more likely that CTC shedding was not affected by p53 status because the tumor line already possessed the pathway changes required for tumor escape.
p53 silencing resulted in the deregulation of many genes that likely contributed to enhanced tumor growth. It should be noted that in our study, gene expression changes resulting from p53 loss were identified by analyzing human breast tumor cells grown in the mammary fat pads of recipient mice. Mice were not exposed to DNA-damaging agents, although tumor cells are known to have intrinsic DNA damage. Troester et al. (2006) silenced p53 in established breast cancer cell lines and performed gene expression profiling on parental and p53-deficient cells both in the absence and presence of DNA damage [
34]. Four of the differentially expressed genes identified in the absence of exogenous DNA damage in their study were also present on our list. These included
BTG2, inositol polyphosphate-5-phosphatase (
INPP5D), lysyl oxidase (
LOX), and p21 protein (Cdc42/Rac)-activated kinase 6 (
PAK6) (Table S1 in Additional file
6). In contrast to this in vitro analysis on nontransformed basal-like cell lines, our gene expression analyses were derived from PDX tumors that were engrafted in vivo. Thus, our analyses identified many alterations not observed in other studies. Importantly, BTG2 has been identified on both platforms, suggesting that it is a critical regulator of p53 function.
BTG2 is a direct transcriptional target of p53 [
38‐
40] that regulates various cellular processes, including proliferation, differentiation, and apoptosis [
38,
42,
43]. BTG2 expression is deregulated in various cancers, consistent with its role as a tumor suppressor protein. We observed reduced expression of BTG2 upon p53 silencing, and restoration of BTG2 expression in p53-deficient cells reduced proliferation in vitro and reduced breast tumor growth at both primary and secondary sites in vivo. In addition, reduced BTG2 expression correlated with reduced overall survival as well as metastasis-free survival of breast cancer patients, consistent with previous reports [
42,
44‐
46]. Herein, we extended these findings by demonstrating that reduced BTG2 expression was associated with reduced overall survival in patients with TNBC. Thus, BTG2 functions as an important downstream effector of p53 to negatively regulate tumor progression and is a candidate prognostic biomarker for TNBC. Future experiments to examine the effect of BTG2 expression in additional PDX models will determine if BTG2 inhibits metastasis in multiple genetic backgrounds.
Although PDX models can largely recapitulate the heterogeneity of human tumors [
47], it is generally acknowledged that they do not faithfully replicate human tumor - human stromal interactions. This is because human stroma is replaced by mouse stroma as a function of time after engraftment. In addition, PDX models generated in immune-compromised mice do not account for the important contributions made by immune cells to malignant progression. The immune system may limit or promote metastasis, and it is possible that WT or mutant p53 controls metastasis by signaling through immune components. Therefore, a thorough examination of the function of p53 in an immune-competent model is also warranted. Interestingly, Ding et al. (2010) reported that PDX models established from a TNBC of the basal subtype were more representative of the patient’s metastasis than her primary breast tumor [
48]. Thus, data identifying these similarities between the metastasis and the xenograft validate the PDX model as a useful system for studying metastasis. Herein we report that human breast tumors were capable of completing all stages of the metastatic cascade in mice and metastatic lesions were observed in organs normally found in patients with metastatic breast cancer including lung, liver, bone, and brain. Thus, PDX models may provide a powerful system for delineating how heterogeneity contributes to metastatic progression and for identifying metastatic drivers that can then be functionally validated.
Competing interests
The authors state that they have no competing interests.
Authors’ contributions
EP, DPW, and HPW were responsible for study design and data interpretation. EP, JS, and GVE performed laboratory experiments under the supervision of DPW and HPW. SC resected tumors for long-term metastasis studies and provided technical knowledge for survival surgeries. YY and HCC performed statistical analyses under the supervision of HL and KAD. NM, KFD, and CS performed bioinformatics analyses of RNA-Seq data under the supervision of JRE and HL. EP, DPW, and HPW drafted the manuscript with critical feedback from GVE, YY, HCC, CS, and HL. All authors have critically read, edited and approved the final version of the manuscript.