Introduction
Breast cancer has the highest incidence and second highest mortality rate among cancers in women worldwide [
1]. The median age of diagnosis for breast cancer is 61 years and the majority of patients are postmenopausal at time of diagnosis. Endocrine therapy has been the mainstay for patients with hormone receptor-positive breast cancer (approximately 75 %) since the inception of the steroid receptor modulator, tamoxifen, in the latter part of the 20th century. One of the major advances in breast cancer therapy in the past decade has been the widespread endorsement of aromatase inhibitors (AIs) as first-line therapy for postmenopausal patients with hormone receptor-positive breast cancer. AIs work by abrogating the activity of the enzyme aromatase (Cyp 19), which converts circulating androgens into estrogen within different body compartments including mammary adipose tissue [
2‐
4]. Clinical trials into the efficacy and side-effects of these drugs in comparison to tamoxifen have heralded them as the main choice of adjuvant endocrine therapy for postmenopausal breast cancer [
5]. Despite the huge advances in the treatment and management of breast cancer the development of drug resistance remains an unresolved problem. For hormone receptor-positive breast cancer, drug resistance occurs in approximately 25 % of cases, which accounts for approximately 50,000 breast cancer recurrences/annum in the US alone [
6]. In some studies the development of resistance to endocrine therapy is estimated to be as high as 30−60 % [
7]. This may be due in part to adaptive hypersensitivity of the intact estrogen receptor (ER) [
5], selective co-activator enhancement [
8], or a shift to growth-factor pathway-dependent cell growth which is well-known to contribute to endocrine treatment failure [
9,
10]. Approximately 75 % of breast cancers also express androgen receptor, which includes a proportion of triple-negative tumours [
11,
12]. With regards to the role of androgen signalling and breast cancer survival there is conflicting evidence. In ER-positive tumours it is reported that androgen receptor (AR) expression is beneficial and it is suggested that it may compete for ER binding sites on the DNA, thereby blocking estrogen-stimulated transcription of pro-proliferative genes [
13]. Other reports however claim that high circulating levels of androgens are associated with increased breast cancer risk in both the premenopausal and postmenopausal setting [
14] and that elevated levels of AR are pro-metastatic [
15]. More recent studies have highlighted the tumour promotional effects of the AR particularly in the event of estrogen disruption [
16‐
20].
Previous work from this laboratory investigating endocrine resistance in breast cancer has shown the developmental transcription factor HOXC11 to be a strong predictor of metastasis and poor disease-free survival (DFS), independent of receptor status, tumour size, nodal status and grade [
21]. Homeobox genes encode a family of 39 proteins of which HOXC11 is a member. These proteins act as developmental transcription factors involved in growth and differentiation [
22,
23]. They play essential roles in body patterning and spatial identity, hence they are akin to a form of cellular global positioning system [
24]. Maintenance of
HOX gene expression patterns are under complex epigenetic regulation. The homeobox transcription factors are known to be regulated by steroids during embryogenesis; however, there is a growing body of evidence to suggest that these genes are also architects of steroidal regulation in endocrine tumours [
21,
25‐
27]. Studies by Norris et al. [
28] demonstrate that HOXB13 can interact with the AR to alternately suppress or activate AR-responsive genes in a promoter-dependent manner. Moreover, recent studies into the upregulation of the HOXC locus in prostate cancer have demonstrated that its ability to modulate androgen signalling is due to the abrogation of coactivator recruitment to direct androgen target genes [
25]. Thus, there is rapidly accumulating evidence to suggest that HOX genes and in particular HOXC genes are key players in modulating steroid signalling in endocrine tumours.
To further our understanding of HOXC11 and the role it plays in the development of endocrine resistance and steroidal adaptability we undertook an RNA-seq experiment to identify HOXC11 target genes in resistant breast cancer. We identified 1,919 genes, and conducted motif mapping to identify potential direct target genes of the transcription factor HOXC11. Analysis of the target genes identified a common novel motif with significant similarity to an AR response element. From these studies we identified prosaposin (PSAP) as a HOXC11 regulated gene. PSAP is a purported AR activator associated with metastatic potential in a number of neoplasms [
29‐
31]. This study supports the hypothesis that expression of HOXC11 and the subsequent secretion of PSAP can expedite endocrine resistance to aromatase inhibitor therapy via tumour promotional activation of the AR.
Methods
Cell culture
Endocrine-sensitive MCF-7 (American Type Culture Collection (ATCC) Virginia, USA ) and tamoxifen-resistant LY2 cells (kind gift from R. Clarke, Georgetown, DC, USA) were grown as previously described [
21]. MCF7-derived AI-sensitive cells (Aro) were developed in house. MCF7 Aro-derived letrozole-resistant cells (LetR) and anastrozole-resistant cells (AnaR) were created by long-term treatment of Aro with letrozole (Novartis, Basel, Switzerland) or anastrozole (AstraZeneca, Macclesfield, UK) [
21]. LY2, LetR and AnaR cells were utilised to model endocrine resistance developed through long-term adaptation to hormonal therapy. MDA-MB-453, SKBR3 and LNCaP cells were acquired from ATCC and maintained as recommended. Cells were maintained in steroid-depleted medium for 72 hours before treatment with hormones. All cells were incubated at 37 °C under 5 % CO
2 in a humidified incubator. All in-house cells were authenticated and are routinely verified as endocrine resistant.
siRNA transfection
HOXC11 was silenced by transient transfection using an experimentally verified pool of siRNA (Flexitube, Qiagen, Manchester, UK) as previously described [
21]. All transfections were carried out using Lipofectamine 2000 transfection reagent according to manufacturer’s instructions (Invitrogen, Thermo Fisher, Warrington, UK) and a non-targeting siRNA negative control (Ambion, Thermo Fisher, Warrington, UK) was used as a control for all siRNA experiments.
To assess the global effects of HOXC11 in endocrine-resistant breast cancer cells we performed RNA-seq on LY2 cells, which were transfected with either siRNA targeting HOXC11 (siRNA-HOXC11) or a scrambled negative control siRNA (scrambled) in the presence of tamoxifen (10
−8 M). Knockdown was verified by Taq-man quantitative reverse transcription PCR (qRT-PCR) prior to library preparation. RNA (10 μg) was extracted using an Oligotex mRNA kit (Qiagen) as per manufacturer’s instructions (n = 4). RNA was reverse transcribed followed by mRNA library preparation and sequencing based on a protocol outlined by Wilhelm
et al. [
32]. Sequencing was performed on an Illumina Genome Analyzer II (GAII) (54 million reads per sample on average). Four independent biological libraries were prepared for each sample to facilitate the detection of expression and estimation of variance. Multiplexing was achieved using barcoding adapters designed in house. After quality checks two of the replicates per sample group were then subjected to further downstream analysis.
Short reads of 36 bp in length were aligned to human reference genome (UCSC hg19, Ensembl GRCh37 release 64) using the splice-aware aligner Tophat [
33] allowing up to 50 multiple mapping locations and no more than 2 mismatches across each read. Differential expression genes (DEGs) were detected by using the Cufflinks/Cuffdiff program (v v1.0.2) [
34].
Estrogen response element (ERE) and androgen response element (ARE) motif analysis
The AR and ER position frequency matrices (PFM) were downloaded from the JASPAR database [
35]. The PFM was then converted to positive weight matrices (PWM) using the MEME suite programme [
36]. A 400-bp-sized window surrounding starting sites of HOXC11 target genes was selected for motif searching. FIMO [
37], a MEME suite programme, was used to identify the AR/ER motifs near the HOXC11 target gene start site with a
p value significance cutoff set at 0.001. Genes having at least one significant hit were kept, and for genes with multiple significant hits, only the best one is reported.
HOXC11 motif analysis
The global mapping of the HOXC11 motif was performed using the TFfind programme [
38]. The HOXC11 motif frequency matrix (PFM) and representing logo were identified from the UniPROBE database [
39]. The first bp and the last 3 bp of this motif were deemed to have low information content and removed from further consideration, resulting in a full motif of 12 bp (consensus sequence: AANGTCGTAAAA) being employed for binding site discovery. Mapping was carried out in the promoter region (5000 bp upstream of TSSs) of all annotated human genes (UCSC hg19, Ensembl GRCh37 release 64). Resultant hits that met the predefined cutoff (≥0.9; 1 denotes a perfect match) were kept and associated with corresponding gene symbols based on the UCSC kgXref table. Overlaps between the DEGs from the siRNA-HOXC11 RNA-seq data and those containing HOXC11 motif in their promoters were reported.
Chromatin immunoprecipitation
LY2 cells were treated with tamoxifen (10
−8 M), estrogen (10
−8 M) or vehicle. LetR cells were treated with androstenedione (Andro, 10
−7 M), estrogen (10
−8 M) or vehicle. Chromatin immunoprecipitation (ChIP) was performed as previously described [
21]. Mouse anti-HOXC11 (6 μg) (Santa Cruz Biotechnology (SCBT), Texas, USA) was added to the supernatant fraction and incubated overnight at 4 °C with rotation. Proteins were un-crosslinked, and primers were used to amplify the DNA −581 to −116 of the PSAP proximal promoter that harbors a HOXC11 binding site and a hormone response element (HRE). PSAP proximal promoter primers were forward: CCCGCTACTACAATGGGCTA, and reverse: GGGGAGGAGTGAGGAAGAAC. Distal non-promoter control primers were forward: TGGTGAGGTTGTATCCACGA, and reverse: CCACTCATGCAATGACCGTA.
PSAP ELISA
A commercially available PSAP ELISA (Cusabio, Stratech, Suffolk, UK) was used to assess serum levels of secreted PSAP in conditioned medium from breast cancer cell lines
in vitro and also in blood serum from consenting patients with breast cancer (n = 34) (see “
Clinical Samples” for more detail).
Western blotting
Protein from breast cancer cell lines were lysed, electrophoresed, and immunoblotted with a rabbit anti-human AR (1:1000 dilution) (sc-816, SCBT), β-actin loading control (Sigma Aldrich, UK) and a corresponding horseradish peroxidase-conjugated secondary antibody (Dako, Den).
HOXC11 transfection
A vector construct pCMV.SPORT-HOXC11 (Life technologies, Thermo Fisher, UK) was used to transiently overexpress HOXC11 in MCF7 cells. MCF7 were seeded at 2.5 × 105 cells in a 6-well plate and incubated overnight. Lipofectamine 2000 (Life Technologies) was used to transfect the cells with pCMV.SPORT-HOXC11 versus empty vector (2 μg) according to the manufacturer’s instructions.
TransAM AR assay
The TransAM assay (Active Motif, California, USA) was modified for use with an AR rabbit anti-human antibody (sc-816, SCBT). Optimal nuclear lysate and antibody concentrations were determined using steroid-dependent LNCaP prostate cancer cell nuclear lysate as a positive control (Additional file
1: Figures S1A and B). LetR breast cancer cells were cultured in the presence of metribolone, a synthetic androgen (R1881, Sigma Aldrich) (10 nmol/L) or recombinant human PSAP (rhPSAP) (10 ng/ml) for 1 hour. Cells were pelleted and nuclear protein isolated using a NE-PER kit (Pierce, Thermo Fisher). The assay was then performed according to manufacturer’s instructions using LetR nuclear cell lysate (20 μl) and AR antibody at a 1:250 dilution (Additional file
1: Figure S1B). On completion of the assay the plate was read at an absorbance of 450 nm (0.1sec) using a Victor2 plate reader (Perkin Elmer, Dublin, Ire).
Treatment of cells in vitro with rhPSAP
Lyophilized rhPSAP protein (Abnova, Taiwan) was reconstituted in storage buffer (50 mM Tris-HCI, 10 mM reduced Glutathione, pH = 8.0). LetR cells were steroid-depleted for 72 hours prior to the addition of recombinant PSAP protein (10 ng/ml). Cells were cultured for a further hour (TransAM) or for 24 hours (protein analysis) under standard conditions, before the cell monolayer was washed with PBS, trypsinised and the cells pelleted for further analysis.
Nuclear translocation assay
Uncoated glass microscope slides (BDH Laboratory supplies, UK) were cleaned with 100 % ethanol, air-dried and placed in sterile 6-well tissue culture dishes. MCF7 and LetR cells were seeded at 6 × 102 onto a coverslip in steroid-depleted media for 72 hours prior to treatment. For individual treatments, rhPSAP (10 ng/ml) and enzalutamide (Enza, 10 μM, Selleckchem, Stratatech) were added to the cells for 3 hours. For dual rhPSAP and Enza treatment, cells were pretreated with Enza (10 μM) for 2 hours and then co-treated with rhPSAP (10 ng/ml) in the presence of Enza for a further hour (3 hours total treatment time). Cells were washed in PBS and fixed in methanol for 10 minutes and permeabilised with 0.1 % triton X-100. Samples were then blocked with 10 % goat serum for 1 hour at room temperature (RT) and incubated with an antibody against AR (N20, sc-815 1:50; SCBT) in 10 % human serum for 90 minutes at RT. Samples were incubated with secondary antibody anti-rabbit Alexa Fluor 488 (1:200) (Thermo Fisher) in 10 % human serum for 1 hour at RT. The nuclei were stained with 4′,6-diamidino-2-phenylindole (1 μg/ml) for 1 minute. Samples washed with dH2O and coverslips containing treated samples were mounted onto slides using fluorescent mounting media (Dako). Cells were visualised by fluorescent microscopy using the CellSens Olympus software. The nuclear translocation of AR (the intensity of AR signal within the nucleus) was quantified in a minimum of 50 cells using ImageJ software.
In vitro cell migration assay
Uncoated glass microscope slides were cleaned with 100 % ethanol, air-dried and placed in sterile 6-well tissue culture dishes. Cells were plated (LetR 3 × 105 cells/2ml growth medium per well, MCF7 2.5 × 105 cells/2ml growth medium per well) and allowed to adhere for 24 hours at 37 °C under standard conditions. Cell monolayers were wounded using the pointed edge of a 20−200-μl yellow pipette tip to score laterally and through the longest length of the cell monolayer. All slides were wounded aseptically at the same time. Cells were treated with 0.1, 1.0, and 10.0 ng/ml rhPSAP and vehicle (Tris–HCl 5 nM). Five images were captured along the length of the scratch at 0, 6 and 24 hours. Five measurements were made within each image using CellSens Entry 1.8 software. Each treatment group consisted of three slides and the entire experiment was repeated in triplicate.
Transwell invasion assay
Twenty-four-well Biocoat matrigel invasion chambers (Corning, USA) were prepared as per manufacturer’s instructions: 5 × 104 cells/500 μl (either MCF7 or LetR) were seeded in serum-free medium into the upper chamber of the insert, and 10 % serum medium was added to the lower chamber to act as a chemoattractant. rhPSAP (10 ng/ml) was added to the upper chambers versus Tris–HCl (5 nM) control. Transwell plates were incubated for 48 hours under normal conditions, 37 °C, 5 % CO2. If cells had been transfected with siRNA the experiment was extended to 60 hours. Inserts were then washed with PBS (×2) and cells in the upper chamber removed using a cotton tip (Johnston & Johnston, Ire); cells remaining in the underside of the insert were then fixed in methanol (10 minutes), and 0.5 % crystal violet (Cruinn, Ire) was added to the lower chamber to visualize cells that had invaded across the membrane. Inserts were washed in dH2O (×2). Representative images (×4 magnification) (CellSens Olympus software) were then evaluated using Image J software to quantify the results. A minimum of five fields of view were evaluated per well.
MTS assay
LetR cells were steroid-depleted for 72 hours and then seeded into a 24-well plate prior to the addition of bicalutamide (Bica) (1 μM) (Sigma Aldrich) or vehicle (dimethyl sulfoxide DMSO; 0.01 %) to regular growth medium. MTS reagent (Sigma Aldrich) was added after 2, 3 and 4 days respectively and the resultant colorimetric outputs analyzed by measuring the absorbance at 490nm using a spectrophotometer (Perkin Elmer).
LetR cells were steroid depleted for 72 hours prior to being seeded into 6-well plates at a density of 5 × 102 cells per well. The cells were treated with vehicle (DMSO; 0.01 %) or Bica (1 μM) and androstenedione (100 nmol/L), and each plate was incubated over 4, 8, or 12 days. Media containing treatments were changed every 48 hours up until the end of each time point. After each time point was reached, cells were washed with PBS, fixed with methanol-glacial acetic acid and stained with 0.5 % crystal violet solution (Cruinn).
Clinical samples
Preoperative blood serum was collected from patients undergoing surgery for the resection of a clinically diagnosed primary breast tumour: 7-ml blood samples were taken in non-heparinised tubes, allowed to clot at RT for approximately 30 minutes and separated in a cooled (4 °C) centrifuge at 2,000 g for 10 minutes. Serum was stored as 0.5-ml aliquots/cryovials and placed in a freezer at −80 °C within 2 hours of being obtained. All patients provided written consent and are currently enrolled on a clinical trial (ICORG – 09–07), ethical approval was sought and granted by the appropriate Research (Medical Ethics) Committees at Beaumont and Waterford Regional Hospital (Ire). Clinical pathology data including receptor status, tumour grade, nodal status, and endocrine therapy have been collated. The median follow-up period for the cohort was 35 months.
HOXC11 and PSAP mRNA correlation in clinical datasets
The Cancer Genome Atlas (TCGA) breast cancer RNA-seq dataset was evaluated to determine correlation between PSAP and HOXC11 mRNA in a supplementary clinical cohort. The TCGA data were classified based on breast cancer subtypes (luminal A, luminal B, human epidermal growth factor receptor (Her)2 over-expressing and basal) and each subtype was assessed for correlation (Spearman) between PSAP and HOXC11. In addition, samples with high AR expression (upper quartile) were also selected and the Spearman’s correlation between HOXC11 and PSAP calculated for this subgroup. A hazard ratio (HR) curve was also generated for PSAP based on these data.
BreastMark [
40] is an algorithm that enables the identification of subsets of gene transcript/miRNAs that are associated with disease progression in breast cancer and its subtypes. High levels of PSAP and/or AR mRNA were evaluated in endocrine-treated datasets.
Statistical analysis
Graphpad Prism was used for the majority of statistical analysis. Univariate analysis was conducted using Fisher exact test for categorical variables. A p value of less than 0.05 was considered to be significant. Significance for quantitative data was evaluated by repeated measures one-way analysis of variance (ANOVA) for treatments over time and the unpaired, two-tailed Student t test was used to compare means. Spearman’s rank coefficient was used to determine correlation between variables.
Data accession code
RNA-seq data are available from the Gene Expression Omnibus (GEO) database (GSE71139).
Discussion
HOX genes have been implicated in the development of haematological and solid tumour malignancies [
42], with many studies focusing on their potential role in endocrine cancers [
27,
28,
43].
HOX genes play vital roles in body mapping during development and posterior
HOX genes in particular are under very tight regulation by estrogen [
44]. It is therefore of interest to understand how the abnormal rebooting of posterior
HOXC genes in mammary epithelial tumour cells can potentiate endocrine-resistant cancer and the development of metastasis [
21,
27,
45]. In this current study we wanted to further elucidate the role of HOXC11 with regard to endocrine resistance and steroidal adaptability in breast cancer. RNA-seq experiments identified 1,919 DEGs when HOXC11 was silenced in endocrine-resistant breast cancer. Analysis of genes harbouring an HRE in the proximal promoter resulted in a novel motif with significant sequence similarity to AR and GR. Filtering of the data using motif mapping identified 29 putative direct HOXC11 target genes including PSAP which is a known AR activator [
46]. HOXC11 recruitment to the DNA was found to be highly responsive to the steroid environment. We observed that HOXC11 recruitment to the proximal
PSAP promoter is impaired by treatment with estrogen in endocrine-resistant cell lines. This is likely due to the fact that a number of posterior
HOXC genes are estrogen-repressed [
27], the exact mechanism is unknown but it could be postulated to be due to chromatin remodelling [
47]. It is therefore notable that there was no significant alteration in HOXC11 recruitment in the AI-resistant cells treated with androstenedione; suggesting alternate patterns of HOXC11 recruitment can be dictated by the steroid microenvironment.
The upregulation of PSAP gene expression by HOXC11 in endocrine-resistant breast cancer cells was of interest primarily due to its association with tamoxifen resistance [
29], but also because of its potential as an AR activator. Previous studies by Koochekpour
et al. demonstrated PSAP to cause ligand-independent activation of AR in prostate cancer via activation of the PI3K pathway [
46]. To this end we decided to explore whether or not PSAP can also activate AR in AI-resistant breast cancer cells in the presence of unconverted endogenous androgens. AR transcription factor activity assays demonstrated successful activation of AR by treatment with rhPSAP, which was of similar magnitude to R1881 exposure. This was further validated by nuclear translocation experiments demonstrating that AR activation by PSAP in resistant cells could be attenuated by treatment with an anti-AR drug. The majority of research articles investigating the role of AR and/or androgens in breast cancer have concluded that male sex hormones and receptors have an inhibitory impact on breast cancer cell growth [
13,
48,
49]. However, if we consider the uniquely androgenic steroid environment that arises from treatment with an AI it should be queried whether this assumption is still relevant in the event of recurrence with this class of drug.
PSAP had previously been reported to stimulate ER-positive endocrine-sensitive breast cancer cell growth [
50], however, in the resistant setting, no significant impact on cell growth was observed (Additional file
9: Figure S5). The focus of functional studies was shifted to evaluate the impact of PSAP on cell migration. Treatment of cells with rhPSAP demonstrated that only endocrine-resistant breast cancer cells respond to the pro-migratory and pro-invasive effects of the protein. Further experiments indicated that HOXC11 upregulation is required for the pro-invasive impact of PSAP to manifest fully, suggesting that HOXC11 expression is key to development of the aggressive phenotype. Increased responsiveness in resistant cells may be due to a number of unexplored factors including expression of cell surface receptors capable of binding PSAP [
51] or activation of specific signalling pathways [
30]. Indeed, PSAP does appear to induce activation of p-AKT in breast cancer cells
in vitro (Additional file
10: Figure S6a and b). This leads to the appealing possibility of identifying new hallmarks of resistance to endocrine therapy and the elucidation of novel drug targets. The clinical potential of PSAP is highlighted by the significantly elevated levels of the protein detected in serum from breast cancer patients who experienced disease recurrence on endocrine therapy.
HOXC11 and PSAP mRNA levels are strongly correlated in a primary breast cancer cohort (
r
s = 0.7692, n = 51). Further analysis of TCGA datasets also demonstrate a correlation between HOXC11 and PSAP transcript levels, specifically in luminal B breast cancer in which AR is elevated (
r
s = 0.46). HOXC11 over-expression in MCF7 cells results in a significant increase in nuclear AR suggesting that HOXC11 upregulation may be a major determinant of the adaptive process. In our AI-resistant breast cancer model we have shown that PSAP is capable of upregulating and activating AR in the context of high HOXC11 expression. Collectively these data suggest that the tumour promotional activation of AR may be an adaptive response primarily in cancers exhibiting dysregulated estrogen signalling. Importantly these findings open up the possibility of utilizing anti-androgens in the treatment of a specific subtype of breast cancer expressing high levels of AR and PSAP, which may not exhibit sustained response to endocrine therapy. Studies by other groups [
16] have shown AR to be co-opted as a transcription factor in place of ER in the apocrine subtype; in our model we have also observed AR recruitment to some, but not all, ER targets evaluated (unpublished observations). Further investigations into the AR transcriptome in AI-resistant breast cancer will help elucidate which processes are being impacted. We would purport from these observations in AI-treated breast cancer (in which estrogen signalling has been dysregulated) that tumour cells have the potential to adapt and utilize bioavailable steroids such as those of adrenal origin: 75 % of breast tumours are positive for AR and so it is of interest to evaluate how these cells adapt to the preponderance of circulating androgens that occurs during prolonged treatment with AI therapy. Evidence that this may be a mechanism of resistance is emerging from clinical data suggesting that levels of androgens, and more specifically androstenedione, are increased in breast cancers refractory to AI therapy [
52]. Such disruption of normal hormonal homeostasis of the breast will result in perturbation of normal mammary epithelial maintenance [
53] and this is what we hypothesize may be contributing to the development of endocrine resistance.
Competing interests
The authors declare there have been no competing interests.
Authors’ contributions
AA carried out validation experiments and molecular assays, collated patient data, participated in study design, analysed results, and assisted in manuscript preparation. LC conducted ChIP assays, ELISAs, TransAM assay, transwell assay, and nuclear translocation assay, analysed results, and assisted in manuscript preparation. YH carried out RNA-seq bioinformatic analysis, TCGA data analysis, and assisted in the study design and manuscript preparation. DMcC participated in study design, sample collation and library preparation for RNA-seq. POG was involved in study design, bioinformatic analysis and revising the manuscript critically for important intellectual content, AH collected patient samples, participated in data analysis and study design, and revising the manuscript critically for important intellectual content. LY contributed to study design, data analysis, and manuscript preparation. MMcI contributed to study design, patient data collation, RNA-seq sampling, data analysis, transwell assay, TransAM assay optimization, and manuscript preparation. All authors read and approved the final version of the manuscript.