Tumour-microenvironmental blood flow determines a metabolomic signature identifying lysophospholipids and resolvin D as biomarkers in endometrial cancer patients

Purpose We aimed to study the potential influence of tumour blood flow –obtained from dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI)- in the metabolomic profiles of endometrial tumours. Methods Liquid chromatography coupled to mass spectrometry established the metabolomic profile of endometrial cancer lesions exhibiting high (n=12) or low (n=14) tumour blood flow at DCE-MRI. Univariate and multivariate statistics (ortho-PLS-DA, a random forest (RF) classifier and hierarchical clustering) and receiver operating characteristic (ROC) curves were used to establish a panel for potentially discriminating tumours with high versus low blood flow. Results Tumour blood flow is associated with specific metabolomic signatures. Ortho-PLS-DA and RF classifier resulted in well-defined clusters with an out-of-bag error lower than 8%. We found 28 statistically significant molecules (False Discovery Rate corrected p<0.05). Based on exact mass, retention time and isotopic distribution we identified 9 molecules including resolvin D and specific lysophospholipids associated with blood flow, and hence with a potentially regulatory role relevant in endometrial cancer. Conclusions Tumour flow parameters at DCE-MRI quantifying vascular tumour characteristics are reflected in corresponding metabolomics signatures and highlight disease mechanisms that may be targetable by novel therapies.


INTRODUCTION
Endometrial cancer (EC) is the most common gynaecologic malignancy in the high-income countries, and its incidence has increased by 21% since 2008 [1]. Presently, the treatment and prognosis are influenced by a subset of parameters based on molecular markers and the microscopic appearance of the tumours. Although approximately 75% of patients are primary curable by surgery and adjuvant treatment, 15-20% of these tumours will recur, with an aggressive and therapy-resistant behavior [2]. Several events (including genetic alterations and tumour microenvironment factors) have been reported to be activated or co-existing in aggressive endometrial cancer [3].
Angiogenesis is an essential mechanism for tumour growth, invasion and metastatic spread in many cancers types [4]. Moreover, tumour-associated vessels are frequently structurally and functionally abnormal, exhibiting delayed maturation and increased permeability [5].
In EC, increased microvascular proliferation correlates with a more aggressive phenotype [6][7][8]. Furthermore, high microvascular proliferation is associated with lower tumour blood flow (F b ) based on dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI), and low tumour F b is observed in EC patients with reduced recurrence / progression-free survival [6,9]. Altogether, is consistent with the hypothesis that vascular proliferation with coexisting disorganized angiogenesis promotes tumour progression and metastatic spread in EC [6].
Cancer was primarily considered a proliferative disorder; however, altered cell metabolism has emerged as a "hallmark of cancer", since growing tumour cells rewire their metabolic program to meet the biosynthetic and bioenergetic demands of constant cell growth and phenotype changes. Metabolomics analysis provides direct evidences of biochemical context-dependent changes that can be correlated with the pathologic state of a cell, tissue or organ [10]. Therefore, metabolomics has been applied to reveal new mechanisms in cancer development and has identified specific biomarkers relevant for diagnosis and prognosis in various cancers.
The present study explores the different metabolomic tumour signatures potentially characterizing EC lesions exhibiting high versus low tumour blood flow at DCE-MRI. The identification of high versus low F b metabolic biomarkers may highlight preoperative functional tumour characteristics with potential relevance for preoperative risk assessment and for identification of EC patients with expected poor outcome.

RESULTS
The main objective of this work was to determine metabolomics differences based on tumour F b using a non-targeted metabolomics approach. We employed multivariate statistics to study changes in whole sample metabolome both in unsupervised-principal component (PCA) and hierarchical clustering analysis-and in supervised approaches-partial least discriminate analysis (PLS-DA), ortho-PLS-DA and Random Forest (RF) [11].
Patients had a mean age of 68 years and 20/26 (77%) of tumours had endometrioid histology (Table  1). In total 17/26 (66%) were classified as early stage (FIGO 1: Federation International of Gynecology and Obstetrics), and received no adjuvant therapy in accordance with national guidelines. 4/6 (67%) of non endometrioid histological subtype patients received adjuvant chemotherapy.
In the present work PCA analysis was chosen as an unsupervised method ( Figure 1). This analysis showed that lesions with high tumour F b clustered better than lesions exhibiting low tumour blood flow. Once we discarded the presence of outliers and checked the technical quality of the technique we applied ortho-PLSDA as a specific tool for clustering ( Figure 2A). This analysis resulted in two well-defined clusters (Q2=0.656 and R2Y =0.985). RF analysis reinforced these results showing an out of bag error of 0.07 ( Figure 2B). Because of endometrioid and non-endometrioid tumours are different EC subtypes we also performed statistics excluding non-endometrioid EC patients obtaining similar results (Supplementary Figure  1A-1C). Specifically multivariate statistics show a good culsterization according to high and low tumour F b (ortho-PLSDA parameters: Q2 =0.553 and R2Y =0.979, random forest out of bag error = 0.1).
The analysis revealed 28 statistically significant molecules ( Table 2). The annotation process (using exact mass, retention time and isotopic distribution) disclosed several lipids belonging to phospholipid, glycerolypid and sphinogolipid families. Specifically, four lysophosphatidylcholines were up-regulated in tumours with high F b . Moreover, resolvin D was identified as a potential tumour F b biomarker. The ROC curve of resolvin D (AUC= 0.855, specificity= 0.9 and sensitivity = 0.7) confirmed the robustness of this metabolite to predict high tumour F b ( Figure 3A). To further study the predictive value of the metabolites found, we applied ROC curves using groups of metabolites (5, 10, 15, 25, 50 and 100) with lowest p-value obtained from T-Test between EC tumours with high versus low F b observing a higher sensitivity and specificity when only the statistical significant metabolites (T-Test, Benjamini Hochberg False Discovery Rate, p<0.05) with a potential identity were used (Supplementary Figure 2). Using these metabolites we could reach a higher predictive value than using only Resolvin D1. Finally, we applied hierarchical clustering analyses using the 25 most statistically significant metabolites and an almost perfect separation between groups was observed ( Figure 3B). www.impactjournals.com/oncotarget

Advanced
DCE-MRI techniques enable visualization and quantification of functional tumour characteristics in vivo. Previous studies revealed that in EC, low tumour F b accompanied by increased capillary leakage and high microvascular proliferation is associated with reduced recurrence/progression free survival [6,9]. These associations are somewhat puzzling, but may be related to the fact that tumoural angiogenesis generates vessels that in spite of being abundant, are heterogenous and rough making them altogether dysfunctional [12]. Further, the negative impact of low tumour F b and disorganized angiogenesis have suggested tumour hypoxia promotes tumour progression and metastatic spread in endometrial carcinomas [13].
Based on these premises we designed an experiment to evaluate potential changes in tissue metabolome according to whether the tumours preoperatively exhibited high-or low F b .
Previous studies have suggested a specific metabolomics signature of EC compared to normal tissue [14], but the present results indicate that there is a specific metabolome associated with tumour F b , suggesting important changes in tumour metabolism which could modulate EC progression and angiogenesis related resistance to treatment [15]. Among all statistically different molecules found, we detected lipid species belonging to different families including phospholipids and, specifically, lysophospholipids (LPL). LPL were first discovered as constituents of cell membranes; however, today we know that they can also act as signalling molecules, playing important roles in central processes in cancer: stimulating angiogenesis, inhibiting apoptosis or modulating immune response [16]. In our study we identified several LPL species increased in high F b EC patient tissues. Specifically, in this work we describe, for the first time, the potential role of lysophophatydilcholine (18:1), lysophosphatydilserine (14:0), lysophosphatydilinositol (20:0) and lysophosphatydilcholine (20:3) in EC tumour progression. These results reinforce the idea of LPL metabolism as a disease modifier in EC progression. Furthermore, this study has revealed that tissue resolvin D could be a good predictor of high tumour F b (generally correlated to better patient survival). Resolvin D is known to be involved in the resolution of inflammation processes and shows potent effects in protecting against neovascularization and hypoxia in some pathological diseases [17]. Thus, our findings   suggest that loss of the protective and inflammatoryresolution effect induced by tumour LPL and Resolvin D coexists with vascular proliferation leading to disorganized angiogenesis and hypoxia, relevant mechanisms in tumour progression and metastatic spread in EC [13]. All in all, the present work delves into the different signalling molecules involved in tumour progression.
These achievements can serve to better understand EC metabolism helping to develop, in a near future, novel and more personalized therapies. Moreover, our results support the promising role of DCE-MRI as preoperative noninvasive imaging tool depicting tumour vascular features relevant for the corresponding metabolomic tumour signature that may be targetable in EC therapy.

Patients and tissue samples
Endometrial tissue from hysterectomy specimens was collected from 26 patients surgically treated for endometrial carcinoma. All patients were diagnosed and treated at the same university hospital (Haukeland University Hospital, Bergen, Norway) written informed consent was obtained from all patients for the collection of imaging data and specimens for biomarker studies included in an institutional review board-approved protocol (Rek Vest 2009/2315).

Dynamic contrast enhanced (DCE)-MRI protocol and blood flow (F b ) analysis
Preoperative pelvic magnetic resonance imaging (MRI) was performed in all patients on a 1.5 Tesla MRI scanner (Siemens Avanto running Syngo v. B17, Germany) using a six-channel body coil and a standardized imaging protocol [6,9] 20 mg butylscopolamine bromide (Buscopan; Boehringer, Germany) was administered intravenously prior to scanning in order to reduce motion artefacts. Contrast agent was administrated intravenously (Dotarem, Guerbet: 0.1 mmol gadolinium/kg, 3 ml/s injection speed). Pelvic sagittal and axial oblique (perpendicular to the long axis of the uterus) T2-weighted images were acquired together with axial oblique T1weighted gradient-echo images prior to contrast agent administration. Pelvic DCE-MRI was obtained for 12 axial oblique slices applying a 3D spoiled gradient echo (FLASH) sequence (echo time/repetition time (TE/ TR) = 1.05/2.64 ms, flip angle (FA) = 12°, matrix = 256x256, field of view (FOV) = 300 x 300 mm 2 , slice thickness = 5 mm, number of averages (NA) = 1). The temporal resolution was 2.49 s. and sequential images were acquired from 30 s before i.v. administration of contrast medium to 6.3 min after contrast injection. At 2 min post contrast medium administration, a pause of 33 s was utilized to acquire axial oblique T1-weighted gradient-echo contrast-enhanced images.
Regions of interests (ROIs) were drawn within the endometrial tumour by a radiologist (ISH with > 7 years of experience with pelvic MRI) who was blinded for tumour stage, histological diagnosis and patient outcome. Care was taken to avoid including necrotic and haemorrhagic areas of the tumours in the ROIs. The dynamic image series were analysed using an in-house implementation of the adiabatic approximation model of Johnson and Wilson [18]. Representative diagnostic MRI images and parametric DCE-MRI maps illustrating endometrial tumours with high and low blood flow (F b ) are shown in Figure 4.

Metabolomic analyses
Tissue samples (50-70 mg) were homogenized using 20 volumes of cold methanol as previously described [14]. Metabolite extracts were subjected to mass spectrometry using an HPLC 1290 series coupled to an ESI-Q-TOF MS/MS 6520 (Agilent Technologies, Santa Clara, CA, USA) as previously described [14]. Multivariate statistic analyses were performed using Metaboanalyst platform [19]. The preliminary identification of differential metabolites (Student's T-Test, Benjamini Hochberg False Discovery Rate, p<0.05) was performed using the PCDL database from Agilent (Agilent Technologies, Barcelona, Spain), which accounts retention times in a standardized chromatographic system, exact mass and isotope distribution as an orthogonal searchable parameters to complement accurate mass data (AMRT approach) according to previously published works [14]. The version of the PCDL database used had retention times and accurate mass data for 679 compounds. To complete the identification process we searched for unidentified metabolites in Metlin Database (https://metlin.scripps.edu/ index.php) which includes accurate masses and MS/MS spectrum for 961.829 molecules.

Ethical approval
All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional research committee and with the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards. Written informed consent was obtained from all patients for the collection of imaging data and specimens for biomarker studies; included in an institutional review boardapproved protocol (Rek Vest 2009/2315). This article does not contain any studies with animals performed by any of the authors.