Abstract
Purpose: Endometrial cancer (EC) diagnosis relies on the observation of tumor cells in endometrial biopsies obtained by aspiration (i.e., uterine aspirates), but it is associated with 22% undiagnosed patients and up to 50% of incorrectly assigned EC histotype and grade. We aimed to identify biomarker signatures in the fluid fraction of these biopsies to overcome these limitations.
Experimental Design: The levels of 52 proteins were measured in the fluid fraction of uterine aspirates from 116 patients by LC-PRM, the latest generation of targeted mass-spectrometry acquisition. A logistic regression model was used to assess the power of protein panels to differentiate between EC and non-EC patients and between EC histologic subtypes. The robustness of the panels was assessed by the "leave-one-out" cross-validation procedure performed within the same cohort of patients and an independent cohort of 38 patients.
Results: The levels of 28 proteins were significantly higher in patients with EC (n = 69) compared with controls (n = 47). The combination of MMP9 and KPYM exhibited 94% sensitivity and 87% specificity for detecting EC cases. This panel perfectly complemented the standard diagnosis, achieving 100% of correct diagnosis in this dataset. Nine proteins were significantly increased in endometrioid EC (n = 49) compared with serous EC (n = 20). The combination of CTNB1, XPO2, and CAPG achieved 95% sensitivity and 96% specificity for the discrimination of these subtypes.
Conclusions: We developed two uterine aspirate-based signatures to diagnose EC and classify tumors in the most prevalent histologic subtypes. This will improve diagnosis and assist in the prediction of the optimal surgical treatment. Clin Cancer Res; 23(21); 6458–67. ©2017 AACR.
This article is featured in Highlights of This Issue, p. 6379
Translational Relevance
Endometrial cancer (EC) is the fourth most common cancer in women, and its incidence and mortality rates are increasing. Current diagnosis is based on the histologic examination of cells contained in uterine aspirates, also named pipelle biopsies. However, this procedure results in about 22% of undiagnosed patients and up to 50% of incorrectly assigned EC histotype and grade. We aimed to identify protein biomarkers in the fluid of uterine aspirates to overcome these drawbacks. We established a 2-protein panel that can accurately detect EC among women entering the diagnostic process (AUC of 0.96). Moreover, we described a 3-protein panel that allows for the accurate discrimination of EC histologic types (AUC of 0.99). In combination with the histopathologic examination, these panels are expected to preclude the need of more invasive diagnostic samplings and help to predict the optimal surgical treatment for patients with EC.
Introduction
Endometrial cancer (EC) is a tumor originating in the inner layer of the uterus, the endometrium, and constitutes the fourth most common cancer in women in developed countries. In the United States, about 61,380 new cases and 10,920 deaths are estimated for 2017, and its incidence and mortality rates are increasing annually worldwide (1).
Abnormal uterine bleeding (AUB) is one of the most common symptoms for gynecologic consultations (2). It affects up to 11% of postmenopausal women (3) and up to 30% of women during their reproductive years (4). Although only 5% to 10% of women with AUB will have an EC (5), women with this symptom are alerted to consult a specialist, as AUB occurs in 90% of EC cases. This represents that a large number of women with benign disorders presenting AUB need to undergo a diagnostic process to rule out EC. This process consists of a pelvic examination and a transvaginal ultrasonography, followed by the histopathologic examination of an endometrial biopsy, which is preferably obtained by a minimally invasive aspiration from the uterine cavity using a cornier pipelle (i.e., uterine aspirate or pipelle biopsy) (5, 6). Diagnosis is achieved by the observation of abnormal cells in the uterine aspirate, and presents high sensitivity to detect EC (7). However, high failure rates with an average of 22% of histologically inadequate specimens have been reported (8). In those cases, a more invasive testing, that is, dilatation and curettage or hysteroscopy, must be performed, with the added risk of anesthesia, infection and perforation, and higher health care costs (9).
In addition, the biopsies should provide information about tumor histology and tumor grade to help in the risk stratification of the patients with EC and guide the surgical staging procedure. Unfortunately, the limited number of cells available for examination and the high interobserver variability in the pathologic interpretation result in 40% to 50% of discordances in EC histotype and grade between biopsies and final hysterectomy specimens (10, 11). Therefore, the identification of sensitive, specific, and reproducible biomarkers that improve diagnosis and preoperative assessment of the histologic type and grade of EC tumors is crucial to appropriately manage patients with EC and decrease mortality and morbidity associated with this disease.
To date, many studies have been conducted to identify EC protein biomarkers, mainly in tissue and serum samples (12–14), but none of them have been translated into clinical utility. In a previous study, we demonstrated that the fluid fraction of uterine aspirates are minimally invasive samples with an important value for the screening of EC protein biomarkers (15). Following this investigation, here we evaluated by liquid chromatography with mass spectrometry detection using parallel reaction monitoring acquisition method (LC-PRM) the diagnostic performance of 52 proteins in uterine aspirate samples obtained from 116 patients covering the broad clinical heterogeneity of EC cases and benign pathologies with AUB. In addition, the potential of those proteins to differentiate between histological EC types was assessed. This study permitted the development of protein panels that achieve excellent performance to diagnose EC and to discriminate between the two main EC histological subtypes.
Materials and Methods
Patients and sample description
A total of 116 women were recruited in the Vall Hebron University Hospital (Barcelona, Spain), the Hospital Universitari Arnau de Vilanova (Lleida, Spain), and the University Medical Center Freiburg (Freiburg, Germany) from 2012 to 2015. Informed consent forms, approved by the Ethical Committees of each Hospital, were signed by all patients. All women entered the EC diagnostic process due to EC suspicion, that is, presentation of AUB and/or a thickness of the endometrium higher than 4 mm for postmenopausal women (16) and 8 mm for premenopausal women based on the results of a transvaginal ultrasonography (17). From the 116 women, 69 were diagnosed with EC, including 49 endometrioid EC (EEC) and 20 non-endometrioid serous ECs (SECs). The remaining 47 women were women without EC with normal endometrium or diagnosed with benign disorders. Clinicopathologic features of the patients are described in Supplementary Table S1.
Uterine aspirates were collected by aspiration with a Cornier Pipelle (Eurogine Ref. 03040200) and transferred to 1.5-mL microtubes. Phosphate buffer saline was added in a 1:1 (v/v) ratio and centrifuged at 2,500 × g for 20 minutes in order to separate the fluid fraction from the cellular fraction. The fluids of uterine aspirates, ranging volumes from 100 μL to 1 mL, were kept at −80°C until used.
Sample preparation for LC-PRM analysis
The sample preparation for the LC-PRM analysis was performed as described previously (15). Briefly, fluid fractions from uterine aspirates were sonicated, and 50 μL of each sample was depleted of albumin and immunoglobulin G proteins using the Albumin and IgG depletion spin trap kit (GE Healthcare). Total protein concentration was estimated by Bradford assay. Then, all 116 samples were divided in two equal aliquots of 12.5 μg to perform the subsequent steps and analyses with technical duplicates. Samples were denaturated, reduced, and alkylated prior to a two-step sequential proteolysis using first Lys-C endoproteinase MS grade (Thermo Fisher Scientific) at a protease/total protein amount ratio of 1/150 (w/w) for 4 hours at 37°C, and second, trypsin (Promega) at a protease/total protein amount ratio of 1/50 (w/w) at 37°C overnight. A mixture of the stable isotope-labeled synthetic peptides (Thermo Fisher Scientific, crude quality) was spiked in each sample (C terminal arginine 13C6, 15N4, Δm = 10 Da, C terminal lysine 13C6, 15N2, Δm = 8 Da or when it was not applicable with a heavy leucine 13C6, 15N1, Δm = 7 Da or phenylalanine 13C9, 15N1, Δm = 10 Da). Finally, samples were purified by solid phase extraction (Sep Pak tC18, 50 mg, Waters), vacuum dried, and resuspended in 0.1% formic acid prior LC-PRM analysis.
LC-PRM setup
The separation of the peptides was performed on a Dionex Ultimate 3000 RSLC chromatography system operated in column-switching mode. The equivalent of 250 ng of digested sample was injected and loaded onto a trap column (75 μm × 2 cm, C18 pepmap 100, 3 μm) using a mobile phase of 0.05% trifluoroacetic acid and 1% acetonitrile in water at a flow rate of 5 μL/minute. Peptides were further eluted onto the analytical column (75 μm × 15 cm, C18 pepmap 100, 2 μm) at 300 nL/minute by a linear gradient starting from 2 % solvent A to 35% solvent B in 48 minutes. The solvent A was 0.1% formic acid in water and the solvent B was 0.1% formic acid in acetonitrile.
The PRM analysis was performed on a Q Exactive plus mass spectrometer (Thermo Fisher Scientific). The MS cycle consisted of a full MS1 scan performed at a resolving power of 70,000 (at 200 m/z) followed by time scheduled targeted MS2 scans, with a normalized collision energy of 20, acquired at a resolving power of 35,000 (at 200 m/z). The quadrupole isolation window of precursor ions was set to 1 m/z unit for the MS2 events and the duration of the time scheduled windows for each pair of endogenous and isotopically labeled peptides was set to 2 minutes.
PRM data processing
The PRM data were processed as described by Martinez and colleagues (15). Briefly, the areas of extracted ion chromatograms (XIC) of the five most intense fragment ions of each precursor (i.e., PRM transition) were extracted using the Skyline program (v3.1; McCoss Lab, University of Washington, USA). The identity of the peptides was confirmed using a spectral matching approach based on the cosine of the spectral contrast angle (cos θ) calculated between the peak areas of the five transitions of the reference (a PRM acquisition of the synthetic peptides mix without biological matrix) and the areas of the corresponding transitions for the endogenous and heavy peptides in the clinical samples. Peptide detection and identification were accepted if both the cos θ of the endogenous and the isotope labeled version of a peptide were higher than 0.98 (18). MS measurements below the limits of quantification generated scores below 0.98, and in such cases the area values were replaced by an estimation of the background.
Statistical analysis
The light/heavy area ratio of each peptide was extracted from Skyline, and the average between duplicates was calculated. The statistical analysis was performed in SPSS (v20.0; IBM) and Graph Pad Prism (v.6.0; GraphPad Software). Comparison of the levels of the monitored peptides between groups of patients was performed using the nonparametric Mann–Whitney U test, as the data did not follow a normal distribution. P values were adjusted for multiple comparisons using the Benjamini–Hochberg FDR method (19). Adjusted P values lower than 0.05 were considered statistically significant. Receiver-operating characteristic (ROC) analysis was used to assess the specificity and sensitivity of the biomarkers and the area under the ROC curve (AUC) was estimated for each individual protein.
Development of the classifiers
A logistic regression model was adjusted to the data in order to assess the power of the different combinations of proteins to classify samples in two clinical categories. ROC curves were generated for each of these regression models; the AUC, and the sensitivity and specificity at the "optimal" cutoff point for discrimination between groups were obtained. The optimal cutoff corresponded to the threshold that maximized the distance to the identity (diagonal) line. The optimality criterion was: max (sensitivities + specificities). AUCs 95% confidence intervals (CIs) were computed with the Delong's method (20). The 95% CIs of the sensitivity and specificity values were computed with bootstrap resampling and the averaging methods described by Fawcett (21). All ROC analysis were performed using the R "pROC" package (22). To assess the robustness of each protein panel, the “leave-one-out” cross-validation procedure was performed by applying to each sample in the dataset the logistic regression model adjusted to the remaining samples on the dataset, hence deriving a new ROC curve and afterward performing the usual ROC analysis. In a similar way, the discrimination power of the diagnostic protein panel was further validated by applying to each sample of an independent set of samples (cohort 2) the logistic regression model adjusted to the initial set (cohort 1), hence deriving a new ROC curve and afterward performing the usual ROC analysis.
ELISA
The concentrations of MMP9 and KPYM isoform M1/M2 were quantified in the soluble fraction of uterine aspirates with commercially available ELISA kits (R&D Systems and USCN Life Science and Technology Company, respectively) according to the manufacturer's protocol. For MMP9, 105 uterine aspirates were analyzed using 1:10, 1:100, or 1:1,000 dilutions. For KPYM, only 39 uterine aspirate samples could be analyzed using 1:2, 1:4, or 1:10 dilutions due to a lack of sample material. All samples were assayed in duplicates and the average values were reported as ng/mL. The linear correlation between the results from LC-PRM and ELISA assays was calculated using the Pearson correlation coefficient.
Results
Subject characteristics
A total of 116 patients entering the EC diagnostic process due to common symptoms in EC, mainly AUB (86%), and/or a thickened endometrium based on transvaginal ultrasonography (79%), were included in the study. The mean age of the patients was 61 years (range, 30–93 years) with 79% postmenopausal women and 21% premenopausal women. Among the 69 EC cases, the two most common histologic types, endometrioid, and serous histologies, were represented in 49 (71%) and 20 (29%) cases, respectively. A total of 47 women without EC included as controls suffered from benign uterine pathologies [polyps (36%), myomas (15%), and endometrial hyperplasia (19%)], or were women with a normal endometrium (30%). Clinical and demographical data from women included in this study are summarized in Table 1.
Clinical characteristics of women enrolled in the study
. | EEC (n = 49) . | SEC (n = 20) . | Non-EC control (n = 47) . |
---|---|---|---|
Age (years) | |||
Median | 67 | 73 | 53 |
Minimum | 37 | 51 | 30 |
Maximum | 87 | 93 | 80 |
Collection center | |||
VHIR | 41 | 12 | 37 |
Lleida | 5 | 8 | — |
Freiburg | 3 | — | 10 |
Uterine condition | |||
Premenopausal | 7 | 1 | 16 |
Postmenopausal | 42 | 19 | 31 |
Histologic gradea | |||
Grade 1 | 5 | — | |
Grade 2 | 33 | — | |
Grade 3 | 10 | 20 | |
FIGO stage | |||
IA | 25 | 5 | |
IB | 13 | — | |
II | 9 | 3 | |
IIIA | — | 2 | |
IIIB | — | 1 | |
IIIC1 | — | — | |
IIIC2 | 1 | 6 | |
IVA | 1 | 2 | |
IVB | — | 1 | |
Myometrial invasion | |||
<50% | 30 | 12 | |
>50% | 19 | 8 | |
Lymphovascular invasion | |||
Yes | 9 | 11 | |
No | 40 | 9 |
. | EEC (n = 49) . | SEC (n = 20) . | Non-EC control (n = 47) . |
---|---|---|---|
Age (years) | |||
Median | 67 | 73 | 53 |
Minimum | 37 | 51 | 30 |
Maximum | 87 | 93 | 80 |
Collection center | |||
VHIR | 41 | 12 | 37 |
Lleida | 5 | 8 | — |
Freiburg | 3 | — | 10 |
Uterine condition | |||
Premenopausal | 7 | 1 | 16 |
Postmenopausal | 42 | 19 | 31 |
Histologic gradea | |||
Grade 1 | 5 | — | |
Grade 2 | 33 | — | |
Grade 3 | 10 | 20 | |
FIGO stage | |||
IA | 25 | 5 | |
IB | 13 | — | |
II | 9 | 3 | |
IIIA | — | 2 | |
IIIB | — | 1 | |
IIIC1 | — | — | |
IIIC2 | 1 | 6 | |
IVA | 1 | 2 | |
IVB | — | 1 | |
Myometrial invasion | |||
<50% | 30 | 12 | |
>50% | 19 | 8 | |
Lymphovascular invasion | |||
Yes | 9 | 11 | |
No | 40 | 9 |
aOne case with undetermined grade.
Final diagnosis was performed on the basis of the highest level of diagnostic evaluation. For all EC cases, this was the histopathologic analysis of the surgical specimen after hysterectomy. For non-EC controls, final diagnosis was achieved with the histologic examination of uterine aspirates (12%), or hysteroscopy-guided biopsies (88%).
LC-PRM analysis
The general workflow followed in this study is shown in Fig. 1. The fluids of uterine aspirates of 116 patients were processed in duplicates and analyzed by PRM acquisition on a hybrid quadrupole orbitrap mass spectrometer. In each sample, the relative abundance of 52 candidate biomarkers prioritized in a previous study (15) was measured by the analysis of 98 peptides (∼2 peptides per protein; Supplementary Table S2). A spectral similarity score was calculated to confirm the identity of the peptides and to detect interferences in the MS signal. Five peptides not detected in more than 50% of the samples were removed, leading to a total of 51 proteins robustly measured with 93 peptides. This data set included 21,576 measured pairs of endogenous/synthetic peptides (L/H) and 92.7% were validated by spectral matching. The rejected values were due to different causes including MS signal below the limits of detection that were replaced by an estimation of the background (6.8% cases), and high signals showing interferences that had to be manually reviewed (0.5%). The reproducibility of the analytical workflow was evaluated by duplication of the sample preparation and the coefficient of variation (CV%) was below 15% for 98% of the detected peptides, with an average of 3.6% (Supplementary Fig. S1A). Moreover, peptides derived from the same protein showed a high correlation, and consequently, they could be used interchangeably to evaluate the protein levels (Supplementary Fig. S1B and Supplementary Table S3). Only three proteins (KPYM, OSTP, and ROA2) displayed correlation coefficients below 0.85 because they were monitored by isoform-specific peptides. These peptides were considered as derived from different proteins.
Overview of the study design. Schematic representation of the steps followed for the development of the EC diagnostic panel (top), and the predictive panel (bottom). FC, fold change.
Overview of the study design. Schematic representation of the steps followed for the development of the EC diagnostic panel (top), and the predictive panel (bottom). FC, fold change.
Diagnostic biomarkers
The relative concentration levels of the 51 proteins were compared in 116 uterine aspirates belonging with 69 patients with EC and 47 control patients in order to identify the proteins that optimally allows for the detection of EC. A total of 28 proteins showed significantly higher levels in the fluid fraction of uterine aspirates from EC patients (adjusted P value <0.05, and fold change >2) and presented high accuracy to individually discriminate between EC and control cases (AUC values higher than 0.75; Table 2). These results further confirmed previous results reported by our group in which the levels of these proteins were measured in a simplified set of only postmenopausal women including 20 EEC cases and 18 age-matched controls (Table 2; ref. 15). Although postmenopausal women represent the vast majority of patients with EC, up to 14% of patients with EC will be in the premenopause (23) and should be considered when evaluating biomarkers for EC diagnosis. Consequently, 21% patients included in the present study were premenopausal patients. Moreover, less frequent but more aggressive SEC cases were also included. Importantly, the validated biomarkers showed potential to detect EC independently of the endometrial status and the histologic type of the EC cases. The five best individual biomarkers, measured by the AUC values, were LDHA with 0.91 (95% CI, 0.856–0.957), KPYM isoform M1-M2 with 0.90 (95% CI, 0.841–0.953), MMP9 with 0.89 (95% CI, 0.827–0.950), NAMPT with 0.88 (95% CI, 0.824–0.942), and SPIT1 with 0.88 (95% CI, 0.814–0.948).
Statistical results of the 28 proteins showing significant differences between patients with EC (n = 69) and non-EC control women (n = 47)
Cohort 1 (n = 116) . | Cohort 2 (n = 38) (Martinez-Garcia et al 2016) . | ||||||||
---|---|---|---|---|---|---|---|---|---|
Protein ID . | FC: All EC cases/All Control . | Adjusted P value . | AUC . | FC: early stage IA/All Control . | Adjusted P value . | AUC . | FC: EEC/Control . | Adjusted P value . | AUC . |
LDHA | 5.49 | 1.E−11 | 0.91 | 5.34 | 2.E−07 | 0.89 | 6.21 | 1.E−04 | 0.91 |
KPYM: isoform M1–M2 | 5.67 | 1.E−11 | 0.90 | 5.48 | 2.E−07 | 0.90 | 5.37 | 1.E−04 | 0.91 |
aKPYM: isoform M1–M3 | 3.39 | 9.E−05 | 0.72 | 2.82 | 2.E−02 | 0.66 | 3.11 | 1.E−02 | 0.75 |
MMP9 | 11.40 | 2.E−11 | 0.89 | 9.55 | 2.E−07 | 0.89 | 5.68 | 1.E−04 | 0.91 |
NAMPT | 3.84 | 4.E−11 | 0.88 | 3.61 | 2.E−06 | 0.85 | 4.24 | 3.E−04 | 0.88 |
SPIT1 | 3.92 | 5.E−11 | 0.88 | 3.49 | 6.E−07 | 0.87 | 3.35 | 1.E−04 | 0.93 |
CADH1 | 3.33 | 5.E−11 | 0.88 | 3.44 | 5.E−07 | 0.87 | 3.75 | 9.E−05 | 0.94 |
ENOA | 3.66 | 1.E−10 | 0.87 | 3.55 | 4.E−07 | 0.87 | 3.78 | 1.E−04 | 0.92 |
PERM | 8.39 | 4.E−10 | 0.86 | 9.02 | 2.E−07 | 0.90 | 14.14 | 6.E−05 | 0.97 |
CAPG | 3.74 | 8.E−10 | 0.85 | 3.20 | 8.E−07 | 0.86 | 3.56 | 5.E−04 | 0.85 |
CH10 | 3.08 | 1.E−09 | 0.85 | 2.98 | 8.E−06 | 0.82 | 2.29 | 6.E−03 | 0.77 |
CTNB1 | 3.89 | 2.E−09 | 0.84 | 3.85 | 2.E−06 | 0.84 | 4.19 | 3.E−04 | 0.88 |
K2C8 | 3.02 | 2.E−09 | 0.84 | 2.98 | 2.E−06 | 0.84 | 3.61 | 3.E−04 | 0.88 |
CLIC1 | 2.91 | 4.E−09 | 0.84 | 2.86 | 2.E−06 | 0.84 | 2.79 | 5.E−04 | 0.86 |
PDIA1 | 2.68 | 4.E−09 | 0.83 | 2.59 | 4.E−06 | 0.83 | 3.30 | 3.E−04 | 0.89 |
PRDX1 | 2.85 | 5.E−09 | 0.83 | 2.70 | 2.E−06 | 0.85 | 4.21 | 2.E−04 | 0.90 |
CD44 | 2.71 | 6.E−09 | 0.83 | 2.51 | 2.E−06 | 0.84 | 2.58 | 4.E−04 | 0.86 |
MIF | 2.85 | 9.E−09 | 0.83 | 2.77 | 1.E−05 | 0.81 | 4.21 | 3.E−04 | 0.88 |
FABP5 | 3.19 | 1.E−08 | 0.82 | 2.73 | 2.E−06 | 0.84 | 3.89 | 6.E−04 | 0.85 |
XPO2 | 4.68 | 2.E−08 | 0.81 | 3.92 | 8.E−05 | 0.77 | 4.03 | 1.E−03 | 0.83 |
TPIS | 2.35 | 7.E−08 | 0.80 | 2.37 | 2.E−05 | 0.80 | 2.93 | 3.E−04 | 0.87 |
CASP3 | 3.49 | 1.E−07 | 0.80 | 3.52 | 5.E−06 | 0.83 | 4.91 | 2.E−04 | 0.91 |
GSTP1 | 2.88 | 5.E−07 | 0.79 | 2.74 | 3.E−05 | 0.80 | 2.91 | 2.E−03 | 0.82 |
ANXA1 | 4.06 | 7.E−07 | 0.78 | 5.03 | 8.E−06 | 0.82 | 4.80 | 3.E−03 | 0.80 |
NGAL | 3.63 | 3.E−06 | 0.77 | 4.27 | 3.E−05 | 0.80 | 4.41 | 4.E−03 | 0.79 |
ANXA2 | 3.22 | 3.E−06 | 0.76 | 3.72 | 6.E−05 | 0.79 | 4.83 | 4.E−04 | 0.87 |
GTR1 | 3.17 | 6.E−06 | 0.76 | 2.89 | 3.E−03 | 0.71 | 1.47 | 3.E−02 | 0.71 |
OSTP: isoform A, B, D | 2.28 | 7.E−06 | 0.76 | 1.71 | 1.E−02 | 0.68 | 9.02 | 4.E−04 | 0.87 |
MUC1 | 2.38 | 7.E−06 | 0.76 | 1.89 | 2.E−04 | 0.77 | 3.60 | 1.E−03 | 0.84 |
Cohort 1 (n = 116) . | Cohort 2 (n = 38) (Martinez-Garcia et al 2016) . | ||||||||
---|---|---|---|---|---|---|---|---|---|
Protein ID . | FC: All EC cases/All Control . | Adjusted P value . | AUC . | FC: early stage IA/All Control . | Adjusted P value . | AUC . | FC: EEC/Control . | Adjusted P value . | AUC . |
LDHA | 5.49 | 1.E−11 | 0.91 | 5.34 | 2.E−07 | 0.89 | 6.21 | 1.E−04 | 0.91 |
KPYM: isoform M1–M2 | 5.67 | 1.E−11 | 0.90 | 5.48 | 2.E−07 | 0.90 | 5.37 | 1.E−04 | 0.91 |
aKPYM: isoform M1–M3 | 3.39 | 9.E−05 | 0.72 | 2.82 | 2.E−02 | 0.66 | 3.11 | 1.E−02 | 0.75 |
MMP9 | 11.40 | 2.E−11 | 0.89 | 9.55 | 2.E−07 | 0.89 | 5.68 | 1.E−04 | 0.91 |
NAMPT | 3.84 | 4.E−11 | 0.88 | 3.61 | 2.E−06 | 0.85 | 4.24 | 3.E−04 | 0.88 |
SPIT1 | 3.92 | 5.E−11 | 0.88 | 3.49 | 6.E−07 | 0.87 | 3.35 | 1.E−04 | 0.93 |
CADH1 | 3.33 | 5.E−11 | 0.88 | 3.44 | 5.E−07 | 0.87 | 3.75 | 9.E−05 | 0.94 |
ENOA | 3.66 | 1.E−10 | 0.87 | 3.55 | 4.E−07 | 0.87 | 3.78 | 1.E−04 | 0.92 |
PERM | 8.39 | 4.E−10 | 0.86 | 9.02 | 2.E−07 | 0.90 | 14.14 | 6.E−05 | 0.97 |
CAPG | 3.74 | 8.E−10 | 0.85 | 3.20 | 8.E−07 | 0.86 | 3.56 | 5.E−04 | 0.85 |
CH10 | 3.08 | 1.E−09 | 0.85 | 2.98 | 8.E−06 | 0.82 | 2.29 | 6.E−03 | 0.77 |
CTNB1 | 3.89 | 2.E−09 | 0.84 | 3.85 | 2.E−06 | 0.84 | 4.19 | 3.E−04 | 0.88 |
K2C8 | 3.02 | 2.E−09 | 0.84 | 2.98 | 2.E−06 | 0.84 | 3.61 | 3.E−04 | 0.88 |
CLIC1 | 2.91 | 4.E−09 | 0.84 | 2.86 | 2.E−06 | 0.84 | 2.79 | 5.E−04 | 0.86 |
PDIA1 | 2.68 | 4.E−09 | 0.83 | 2.59 | 4.E−06 | 0.83 | 3.30 | 3.E−04 | 0.89 |
PRDX1 | 2.85 | 5.E−09 | 0.83 | 2.70 | 2.E−06 | 0.85 | 4.21 | 2.E−04 | 0.90 |
CD44 | 2.71 | 6.E−09 | 0.83 | 2.51 | 2.E−06 | 0.84 | 2.58 | 4.E−04 | 0.86 |
MIF | 2.85 | 9.E−09 | 0.83 | 2.77 | 1.E−05 | 0.81 | 4.21 | 3.E−04 | 0.88 |
FABP5 | 3.19 | 1.E−08 | 0.82 | 2.73 | 2.E−06 | 0.84 | 3.89 | 6.E−04 | 0.85 |
XPO2 | 4.68 | 2.E−08 | 0.81 | 3.92 | 8.E−05 | 0.77 | 4.03 | 1.E−03 | 0.83 |
TPIS | 2.35 | 7.E−08 | 0.80 | 2.37 | 2.E−05 | 0.80 | 2.93 | 3.E−04 | 0.87 |
CASP3 | 3.49 | 1.E−07 | 0.80 | 3.52 | 5.E−06 | 0.83 | 4.91 | 2.E−04 | 0.91 |
GSTP1 | 2.88 | 5.E−07 | 0.79 | 2.74 | 3.E−05 | 0.80 | 2.91 | 2.E−03 | 0.82 |
ANXA1 | 4.06 | 7.E−07 | 0.78 | 5.03 | 8.E−06 | 0.82 | 4.80 | 3.E−03 | 0.80 |
NGAL | 3.63 | 3.E−06 | 0.77 | 4.27 | 3.E−05 | 0.80 | 4.41 | 4.E−03 | 0.79 |
ANXA2 | 3.22 | 3.E−06 | 0.76 | 3.72 | 6.E−05 | 0.79 | 4.83 | 4.E−04 | 0.87 |
GTR1 | 3.17 | 6.E−06 | 0.76 | 2.89 | 3.E−03 | 0.71 | 1.47 | 3.E−02 | 0.71 |
OSTP: isoform A, B, D | 2.28 | 7.E−06 | 0.76 | 1.71 | 1.E−02 | 0.68 | 9.02 | 4.E−04 | 0.87 |
MUC1 | 2.38 | 7.E−06 | 0.76 | 1.89 | 2.E−04 | 0.77 | 3.60 | 1.E−03 | 0.84 |
NOTE: Proteins with potential as diagnostic markers are shown. All of them have an adjusted P value <0.05, fold change >2 and AUC > 0.75. The performance of these proteins as early-stage EC biomarkers was assessed by comparing EC cases in the early-stage IA with non-EC controls. The statistic results of a previous study performed with these proteins in a limited cohort of 38 age-matched patients are shown.
Abbreviation: FC, fold change.
aPerformance of KPYM is isoform specific.
Early-stage EC diagnostic biomarkers
Early detection of EC is directly associated with a high rate of survival for the patients. Thus, the suitability of the 51 proteins as markers for early detection of EC was evaluated. A total of 30 EC cases diagnosed at IA stage according to the International Federation of Gynecology and Obstetrics (FIGO) staging, that is, tumors confined to the uterine corpus with less than 50% myometrial invasion (24, 25), were compared with the 47 non-EC controls. All proteins previously identified for EC diagnosis were found differentially abundant with adjusted P value <0.05 and fold change >2, except for OSTP and MUC1 (Table 2). Remarkably, the five proteins showing the best performance in discriminating all EC cases from non-EC controls were also able to accurately discriminate the very initial stage of the disease with AUC values over 0.84 (Table 2). In addition, PERM increased its individual potential to detect early-stage EC as follows: AUCEC was 0.86 (95% CI, 0.786–0.929) and AUCearly-EC was 0.90 (95% CI, 0.841–0.969).
Endometrial hyperplasia
Endometrial hyperplasia is a thickening of the endometrium caused by the excess of estrogen stimuli. Although it is a benign disease, it is considered a precursor lesion of EC, and should be distinctively diagnosed (26). As described previously, our biomarkers allowed for the accurate discrimination between EC and non-EC controls, including hyperplasia cases (Fig. 2A and Table 2). Interestingly, 16 proteins were found differentially abundant with adjusted P values <0.05 and fold changes higher than 2 (Supplementary Table S4) between hyperplasias (n = 9) and the other non-EC controls (n = 38). Four of them, NAMPT, ENOA, CATD, and GSTP1, showed AUC higher than 0.85. These biomarkers, if validated, open the avenue to individually diagnose hyperplasias from EC and other benign conditions.
Diagnostic performance of biomarkers in discriminating patients with EC from non-EC controls in the fluid fraction of uterine aspirates. A, Principal component analysis (PCA) plot constructed with the data of the differential proteins of the 116 patients included in the study. The plot clearly shows that most of the variance plotted in the x-axis allows for the differentiation of patients with EC compared with endometrial hyperplasias and other non-EC controls. B, Scattering plots depicting the distribution of the light/heavy (L/H) ratios obtained by LC-PRM across the 69 patients with EC and 47 controls of the two proteins that compose the diagnostic panel. C, ROC curves of EC versus non-EC controls women for the individual proteins of the panel and the 2-protein panel. D, Summary table of the results of the logistic regression model adjusted to the data of cohort 1 (n = 116) underlying the ROC analysis. The robustness of the panel was assessed by performing a "leave-one-out" cross-validation (CV) for each sample in the cohort 1, and by applying the model to an independent cohort 2 (n = 38) from a previous study. Sensitivity (SN) and specificity (SP) values obtained after "leave-one-out" cross-validation in cohort 1 and validation in cohort 2 are shown. E, Correlation between LC-PRM and ELISA assay results evaluated in 105 patients for MMP9 (68 patients with EC, in black; and 37 control patients, in red), and 39 patients for KPYM (32 patients with EC, in black; and seven control patients, in red).
Diagnostic performance of biomarkers in discriminating patients with EC from non-EC controls in the fluid fraction of uterine aspirates. A, Principal component analysis (PCA) plot constructed with the data of the differential proteins of the 116 patients included in the study. The plot clearly shows that most of the variance plotted in the x-axis allows for the differentiation of patients with EC compared with endometrial hyperplasias and other non-EC controls. B, Scattering plots depicting the distribution of the light/heavy (L/H) ratios obtained by LC-PRM across the 69 patients with EC and 47 controls of the two proteins that compose the diagnostic panel. C, ROC curves of EC versus non-EC controls women for the individual proteins of the panel and the 2-protein panel. D, Summary table of the results of the logistic regression model adjusted to the data of cohort 1 (n = 116) underlying the ROC analysis. The robustness of the panel was assessed by performing a "leave-one-out" cross-validation (CV) for each sample in the cohort 1, and by applying the model to an independent cohort 2 (n = 38) from a previous study. Sensitivity (SN) and specificity (SP) values obtained after "leave-one-out" cross-validation in cohort 1 and validation in cohort 2 are shown. E, Correlation between LC-PRM and ELISA assay results evaluated in 105 patients for MMP9 (68 patients with EC, in black; and 37 control patients, in red), and 39 patients for KPYM (32 patients with EC, in black; and seven control patients, in red).
Diagnostic biomarker panel
By using a homemade R script, individual EC biomarkers were combined into all possible panels from two to five proteins. The combination of MMP9 and KPYM significantly improved the diagnostic power of the individual biomarkers, achieving an AUC value of 0.96 (95% CI, 0.94–0.99), with 94.2% sensitivity and 87.2% specificity (Fig. 2B–D). The addition of more proteins only enhanced marginally the performance of the 2-protein panel and, therefore, were not considered. The robustness of the panel was assessed using two methods: a “leave-one-out” cross-validation procedure performed over the set of 116 samples (cohort 1), and a validation of the model over an independent set of 38 samples (cohort 2) from a previous study (Supplementary Table S5; ref. 15). The results of the logistic regression model underlying the ROC analysis in both independent datasets are shown in Fig. 2D. The MMP9–KPYM panel achieved sensitivity (specificity) of 89.86% (85.1%) and 100% (83.33%) in the first and second cross-validation, respectively. The diagnostic biomarker panel can differentiate EC from non-EC control patients with high accuracy in the fluid of the uterine aspirates. Importantly, in the cohort of patients included in this study, the histopathologic examination of uterine aspirates did not provide a proper diagnosis to 17% of women, either due to insufficient material or incorrect diagnosis. Importantly, the misdiagnosed patients were patients with EC, which were diagnosed with no malignancy. For this 17% of women, the biomarker panel combining MMP9 and KPYM achieved a correct diagnosis. Consequently, our biomarker panel was able to perfectly complement the current diagnostic procedure to provide final diagnosis to all women using this minimally invasive sampling.
Transferability to ELISA
In order to evaluate the feasibility of developing an assay that could be more easily deployed in a clinical environment, we assessed the transferability of the MS-based results to ELISA assays. The levels of the two proteins of the diagnostic panel, MMP9 and KPYM, were quantified by commercially available ELISA kits and the correlation with the results obtained by LC-PRM was evaluated. The quality of the ELISA performance was assessed by the median overall CV (%), which was 1.8% for MMP9 and 6.0% for KPYM. Absolute levels in the ELISA (ng/mL) and relative levels (light/heavy ratios) in the PRM acquisition showed a linear correlation for both biomarkers, with an R2 value of 0.93 for MMP9 and 0.61 for KPYM (Fig. 2E).
Predictive biomarkers
EEC is the most common histology in EC and has a good prognosis when compared with non-endometrioid EC cases (NEEC; ref. 23). NEEC represents about 20% of all EC cases but accounts for more than 50% of recurrences and deaths from EC. Among NEEC, the serous EC (SEC) is the most common subtype. We investigated the abundance of the 51 proteins in the cohort of 49 EEC and 20 SEC cases. The levels of nine proteins were significantly increased in uterine aspirate samples from EEC patients (adjusted P value < 0.05; ref. Fig. 3A). Among those, six proteins had a fold change higher than 2 and presented the highest individual AUC values: PIGR with 0.85 (95% CI, 0.734–0.958), CAYP1 with 0.83 (95% CI, 0.725–0.942), CTNB1 with 0.78 (95% CI, 0.670–0.895), SG2A1 with 0.77 (95% CI, 0.661–0.880), VIME with 0.76 (95% CI, 0.645–0.881), and WFDC2 with 0.74 (95% CI, 0.624–0.855).
Predictive performance of biomarkers in classifying EC cases in the most prevalent histologic subtypes in the fluid fraction of uterine aspirates. A, List of proteins showing statistical differences between patients diagnosed with EEC (n = 49) and SEC tumors (n = 20) with adjusted P value < 0.05. Two proteins, LEG1 and CAPG, showed an adjusted P value > 0.05, but P value < 0.05. B, ROC curves of EEC versus SEC cases for the individual proteins of the panel and the 3-protein panel. C, Summary table of the results of the logistic regression model adjusted to the data from the 69 patients underlying the ROC analysis. The robustness of the panel was assessed by performing a "leave-one-out" cross-validation for each sample in this cohort of patients.
Predictive performance of biomarkers in classifying EC cases in the most prevalent histologic subtypes in the fluid fraction of uterine aspirates. A, List of proteins showing statistical differences between patients diagnosed with EEC (n = 49) and SEC tumors (n = 20) with adjusted P value < 0.05. Two proteins, LEG1 and CAPG, showed an adjusted P value > 0.05, but P value < 0.05. B, ROC curves of EEC versus SEC cases for the individual proteins of the panel and the 3-protein panel. C, Summary table of the results of the logistic regression model adjusted to the data from the 69 patients underlying the ROC analysis. The robustness of the panel was assessed by performing a "leave-one-out" cross-validation for each sample in this cohort of patients.
Predictive biomarker panel
Following the same procedure as described before, all possible combinations of two and three proteins were evaluated among the diagnostic and predictive biomarkers to identify panels of proteins that will improve the outcome of individual biomarkers. A combination of three proteins, consisting of CTNB1, XPO2, and CAPG was the best-performing panel to discriminate between EEC and SEC in the fluid of uterine aspirate samples with an AUC of 0.99 (95% CI, 0.90–1; Fig. 3B). This panel achieved 95% (95% CI, 85%–100%) sensitivity and 95.9% (95% CI, 89.8%–100%) specificity (Fig. 3C). After completion of the “leave-one-out” cross-validation the values were 95% sensitivity and 89.8%, specificity.
Discussion
In this study, we defined two protein biomarker signatures to detect EC and to distinguish between EEC and SEC histologies using a liquid biopsy obtained from the female genital tract (i.e., the fluid of uterine aspirates). The current diagnostic procedure is based on the histologic examination of the limited cellular content in this sample and it is associated to important drawbacks: an average of 22% of undiagnosed patients (8), and up to 50% of incorrect histotype and/or grade assignment of EC cases (10). Our approach help to overcome these limitations by the identification of biomarkers in the fluid fraction of uterine aspirates.
The diagnostic panel is composed of two proteins. Metalloproteinase 9 (MMP9) is involved in the release of tumor promoting agents and in the degradation of extracellular matrix, which may promote cell migration and invasion, favoring tumor metastasis (27, 28). Increased levels of MMP9 have been detected by immunohistochemistry in endometrial tissue from patients with EC; whereas lower levels have been reported in patients with EC compared with healthy women in serum samples (29, 30). Pyruvate kinase (KPYM) is a key enzyme in glycolysis and is also involved in gene transcription (31). An increased aerobic glycolytic rate is a metabolic hallmark of malignant cells (32). In EC, higher levels of this protein have been reported in endometrial tissue from patients with EC compared with controls (33, 34).
When combined, MMP9 and KPYM form a powerful panel that can detect EC with 94.2% sensitivity and 87.2% specificity in the fluid fraction of minimally invasive uterine aspirates. This molecular panel, combined with the current diagnosis based on the histologic examination of the cells in these samples, permitted to achieve diagnosis in 100% of patients in our dataset. Consequently, implementation of this panel is expected to impact on the clinical scenario by precluding the use of subsequent invasive sampling methods, (i.e., dilatation and curettage or hysteroscopy).
A clinical strength of this investigation is that women enrolled in the study covered the broad variability of women entering the EC diagnostic process. Regarding patients with EC, the two most common histologies, EEC and SEC cases, were included. Patients without EC covered all women with suspicion of EC mainly due to AUB or thickening of the endometrium (16, 35). This included women suffering from benign pathologic conditions (mainly polyps, myomas, and endometrial hyperplasia), and women with normal endometrium. Moreover, this study included both premenopausal women with functional endometrium and postmenopausal women mostly presenting an atrophic endometrium. Despite the molecular differences that are known to exist between EEC and SEC tumors (36), and between endometrial tissues of women with different hormonal status (37), our diagnostic biomarkers allowed for the accurate differentiation between EC patients and non-EC women independently of these factors.
A limitation of this investigation is that we here reported several proteins that allowed for the discrimination of endometrial hyperplasias from both EC and control women. However, hyperplasias include a broad range of lesions with very different progression risk to carcinoma, from 1% to 3% for women with non-atypical simple endometrial hyperplasia, to almost 30% for patients with complex atypical hyperplasia (26). Because of the limited number of cases in this study, division into risk groups was not feasible. Therefore, further exploration of the levels of these biomarkers in the different subgroups is needed.
Regarding the role of the studied proteins to improve the risk group assignment of patients with EC, current major risk stratification systems, such as the European Society for Medical Oncology (ESMO) classification, focus on the histology, grade, myometrial invasion, and lymphovascular invasion of the tumors. As most information is not available preoperatively, histologic subtype and grade become key factors for risk group assignment and for the determination of the extent of the surgical staging procedure (10). Our data showed that the combination of three proteins (CTNB1, XPO2, and CPG) allowed for the accurate discrimination between EEC and SEC histologic EC types with a sensitivity of 95.0% and specificity of 95.9%.
These three proteins have been previously related to EC. β-Catenin (CTNB1) has an important role in epithelial cell–cell adhesion and in the transcription of essential genes responsible for cellular proliferation and differentiation in the Wnt-signaling pathway (38). In concordance with our observations, CTNB1 has been described in EEC tissue specimens, but not in SEC tumors (39, 40). Macrophage-capping protein (CAPG) modulates cell motility by remodeling actin filaments. It is involved in cell migration and invasiveness in several type of cancers (41). Unlike CTNB1, higher levels of CAPG have been reported in more aggressive SEC tumors compared with EEC cases at tissue level, in agreement with our results in uterine aspirates. Finally, exportin-2 (XPO2), also known as cellular apoptosis susceptibility protein, has a role in the mitotic spindle checkpoint. Depletion of XPO2 leads to cell-cycle arrest and, consequently, it has been associated with tumor proliferation (42); although it has also been related to tumor invasion and metastasis (43). Higher levels of XPO2 have been observed in many cancer types, including EC, and have been positively associated with a higher cancer grade and worse outcome of the patients (44). Although no significant differences in XPO2 levels were observed between EEC and SEC cases in our study, its inclusion in the panel formed by CTNB1 and CAPG significantly improved its performance.
This study was, however, limited to identify protein signatures that accurately distinguish among tumor grades, as the number of patients available for this comparison was too low to achieve statistically robust results. This should be tackled in a near future to aid in the risk group assignment of patients with EC. Moreover, current histopathologic risk stratification systems are limited to predict the risk of recurrence. Additional molecular information is needed to improve these stratification systems and guide a more precise surgery. Although it is not yet applied in the clinical practice, the novel molecular classification of EC developed by The Cancer Genome Atlas (TCGA) has demonstrated higher prognostic accuracy (45). The behavior of CAPG, XPO2, and CTNB1 expression in relation to the molecular classification was assessed using the available TCGA data (46). CAPG and XPO2 presented a significant mRNA upregulation in the serous-like group in contrast to the microsatellite instability and copy-number low subgroups that were characterized by a high number of CTNB1 missense mutations. The POLE group presented mixed features, with a high number of alterations in XPO2 and CTNB1 (Supplementary Fig. S2). Although these findings are at genomic level, they open up a route to assess the potential of this proteomic signature to discriminate molecular subgroups and, hence, further ameliorate the stratification of ECs.
An important strength of this study is manifested in the technical viewpoint. We here used the last generation of targeted MS method, the LC-PRM acquisition performed on a high-resolution accurate mass spectrometer. Unlike the vast majority of EC biomarker studies that use antibody-based approaches to test one or few proteins, this multiplexing-targeted MS-based approach presents two clear advantages. First, it increases the chances of finding a clinically useful biomarker; and second, it is better suited for the development of protein panels because combinations among all the studied proteins can be evaluated. Indeed, the best individual proteins do not necessarily perform well as a combination, as they can provide redundant information, as proved in both panels described in this study. Finally, the use of PRM acquisition facilitates the application of targeted MS to large cohorts of patients in comparison to the standard SRM acquisition (47, 48). Apart from the easier method development, the decrease of interferences and the use of spectral matching significantly facilitated the data processing. Moreover, we achieved a precise quantification of the 93 peptides (average CV of 3.6%).
As a first step toward clinical implementation, the diagnostic panel was evaluated by ELISA assay. Immunoassays continue to be the preferred method for clinical validation and further application in the clinical environment (49). The commercially available ELISA kit against MMP9 yielded results perfectly matching the LC-PRM data, but the KPYM ELISA did not perform up to the same standard. The development of a highly specific and reproducible multiplexed immunoassay for the protein panels would be an interesting approach although it presents important challenges, such as the design and validation of high-quality antibodies for each protein (associated to high costs and development time), assessment of cross-reactivity of reagents, and the need of high-level assay automation for the clinical application. Despite this, multiplexed immunoassays have tremendous potential for in vitro diagnostics, and several diagnostic protein arrays have been already approved by the FDA (50). Once the assay is optimized, the performance of the biomarker panels need to be validated in a prospective, multicentric study including a higher number of patients.
The workflow presented here based on targeted MS approaches, and LC-PRM in particular, until advanced stages of the biomarker pipeline, followed by immunoassay development for the best-performing panel of proteins, allows for a more efficient identification of a clinically useful panel of biomarkers, making a significant improvement over the current state of biomarker identification.
Remarkably, this study covers an important clinical need in EC. From one side, the diagnostic molecular signature identified in the fluid of the uterine aspirates would perfectly complement the current diagnostic procedure based on the cellular fraction of these samples, reducing the number of invasive biopsies needed. On the other hand, the uterine aspirate-based proteomic approach described here paves the way for the identification of proteomic signatures that classify EC tumors into more clinically relevant risk groups to help on surgical treatment prediction. Although further studies are needed in the future, a first step has been achieved with a signature to accurately classify the most common histologies. Altogether, the development of uterine aspirate-based biomarker signatures is expected to improve the management of patients with EC and save great health care costs.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Authors' Contributions
Conception and design: E. Martinez-Garcia, S. Cabrera, J. Reventos, B. Domon, E. Colas, A. Gil-Moreno
Development of methodology: E. Martinez-Garcia, A. Lesur, S. Cabrera, J. Reventos
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): E. Martinez-Garcia, A. Lesur, S. Cabrera, X. Matias-Guiu, M. Hirschfeld, J. Asberger
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): E. Martinez-Garcia, L. Devis, S. Cabrera, X. Matias-Guiu, M. de los Á. Casares de Cal, A. Gómez-Tato, J. Reventos, E. Colas, A. Gil-Moreno
Writing, review, and/or revision of the manuscript: E. Martinez-Garcia, A. Lesur, S. Cabrera, X. Matias-Guiu, J. van Oostrum, M. de los Á. Casares de Cal, A. Gómez-Tato, J. Reventos, B. Domon, E. Colas, A. Gil-Moreno
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): X. Matias-Guiu, M. Hirschfeld, J. Reventos
Study supervision: A. Lesur, J. Reventos, E. Colas, A. Gil-Moreno
Other (provided grants to finance the study): J. Reventos
Acknowledgments
The authors would like to acknowledge the work of all who have participated in the recruitment of clinical samples. We also thank the patients for their willingness to participate in the study. Samples from Hospital Arnau de Vilanova were obtained with the support of xarxa catalana de Banc de Tumors (PT 13/0010/0014).
Grant Support
This work was supported by the Spanish Ministry of Health (RD12/0036/0035), the Spanish Ministry of Economy and Competitivity (PI14/02043), the “Fondo Europeo de Desarrollo Regional” FEDER (RTC-2014-3110-1), the AECC (Grupos Estables de Investigacion 2011-AECC-GCB 110333 REVE), the Fundació La Marató TV3 (2/C/2013), the CIRIT Generalitat de Catalunya (2014 SGR 1330), the Fundación DEXEUS Salud de la Mujer (FSDF-2013-03). The Spanish Ministry of Economy and Competitiveness (IJCI-2015-25000), and the PERIS grant (Generalitat de Catalunya) granted Dr. Colás. The present work has been also funded by the "Fonds National de la Recherche du Luxembourg" (FNR) via the PEARL-CPIL program to B. Domon and an AFR grant to A. Lesur (PDR 2013-2, Project Reference 6835664).
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.