Abstract
The proteomic content of the endometrial fluid (EF) from patients with endometriosis has been investigated, but the lipidomic profile has not been analyzed yet in detail.
This study is a comparative untargeted lipidomic analysis of human EF obtained from 35 patients (12 endometriosis and 23 controls). Global differential lipidomic profile was analyzed in both groups by ultrahigh performance liquid chromatography coupled to mass spectrometry. A total of 123 out of the 457 metabolites identified in the EF were found to be significantly differentially expressed between ovarian endometriosis (OE) versus controls. Univariate statistical analysis showed reduced levels of saturated diacylglycerols and saturated triacylglycerols in endometriosis patients. A predictive model was generated using the 123 differentially expressed metabolites. Using this model, we were able to correctly classify 86% of the samples. This study identified the lipidomic profile in the EF associated with OE, suggesting that EF analysis could be considered as a minimally invasive approach for the diagnosis of endometriosis. In conclusion, the lipidomic profile of the EF is different between samples from patients with OE and controls.
Introduction
Endometriosis is one of the most common reproductive disorders affecting women and is characterized by the presence of endometrial tissue outside the uterine cavity. The nature of this disease is heterogeneous and includes different anatomical entities such as ovarian, peritoneal, and deep infiltrating endometriosis. All of them cause in a different range pelvic pain, dysmenorrhea, dyspareunia, painful defecation, and/or infertility [1]. While the estimated prevalence of endometriosis is 6%–10% in the general female population, an estimated 35%–50% of infertile women are affected by the disease [2].
The standard diagnosis is based on surgical inspection of the pelvic organs and histological identification. Surgical removal of ectopic endometrial lesions has been the therapeutic approach for more than one century although this is a chronic and recurrent condition. Both medical and surgical treatments are not without significant side effects and are often unsuitable for couples seeking fertility treatments [3], and do not address recurrence [4]. Due to the invasiveness of the diagnostic procedure, it can take up to 12 years before affected women obtain a diagnosis and receive appropriate treatment [5]. A reduction in the time to diagnosis would largely improve the field of endometriosis management at different levels [6], and ultimately improve the quality of life of affected women.
Although over 100 potential biomarkers of endometriosis have been reported [7], none of them have proven to be unequivocally clinically useful for diagnosis [7]. “Omics” technologies can detect and analyze hundreds of markers in the same experiment and some initial “-omics” approaches in endometriosis patients, including transcriptomic and proteomic analysis of blood [8] and endometrial fluid (EF) [9], have been used to evaluate different degrees of disease severity. Indeed, EF is an excellent body fluid in which to develop new minimally invasive diagnostics for any endometrial disease or even to predict implantation [10].
Lipids are increasingly known to be important bioactive signaling molecules involved in a diverse range of cellular processes. For example, ceramides and sphingoid bases modulate many apoptotic signaling events [11], whereas their phosphorylation or glycosylation can lead to the production of mitogenic factors like CER-1-phosphate (C1P) and glucosylceramide (GlcCer) [12]. Altered sphingolipid metabolism flux has been observed in the serum, peritoneal fluid, and endometrial tissue in women with endometriosis [12], suggesting that lipidomics may be a possibility in the diagnosis of endometriosis.
In this study, we aimed to investigate the differential lipidomic profile involved in the EF during the window of implantation from patients with endometriosis versus controls as the basis to understand and build up a minimally invasive diagnostic test for this pathological condition.
Material and Methods
Participants and study design
Twelve patients with ovarian endometriosis (OE) patients and 23 controls without endometriosis (NE) were recruited for the study. Mean age of the NE group was 29 years old compared to 35 years old in the endometriosis group. This difference was statistically significant (P < 0.05). Mean BMI of the NE group was 22.85 kg/m2 compared to 22.67 kg/m2 found in the OE group.
The inclusion criteria of the NE group were as follows: women, age 18–37, with normal physical examination, no family history of hereditary disease, negative screening for sexually transmitted diseases, normal karyotype, and a BMI between 18 and 28 kg/m2. Donors with infertility problems or with history of endometriosis were excluded from the control group. For the endometriosis group, women aged 18–39 were histologically confirmed in patients undergoing laparoscopy and/or positive presence of an endometrioma by vaginal ultrasound. The exclusion criteria were as follows: presence of hydrosalpinx, recurrent miscarriage, uterine abnormalities, endometrial polyp, adenomyosis, or myomas. For both groups, the use of oral contraceptives, GnRH analogs, or any other hormonal treatment 3 months before the procedure was an exclusion criteria. This study was approved by the Ethical Committee of IVI (1006-C-072-MR-iG). All participants who met the study criteria signed a consent form. A pelvic ultrasound was performed around days 10–12 of the menstrual cycle to confirm follicular growth. An additional scan was performed if necessary. Participants began self-monitoring the endogenous luteinizing hormone (LH) surge using urinary dipsticks. When the surge was detected, EF collection was scheduled 7 days later (LH+7).
A total of 35 EF samples were collected, frozen, and analyzed by ultrahigh performance liquid chromatography coupled to mass spectrometry (UPLC-MS) for global lipidomic profiling.
Endometrial fluid collection
Endometrial fluid was extracted while patients laid in lithotomy position. The cervix was cleansed after inserting the speculum and an empty flexible catheter (Wallace, Smith Medical International) was gently introduced 6 cm transcervically into the uterine cavity guided by abdominal ultrasound. Suction was gradually applied with a 10 mL syringe. To prevent contamination by cervical mucus during catheter removal, suction was dropped at the entrance of the internal cervical os and cervical mucus was aspirated prior to EF aspiration. Approximately 20–40 μL of EF was obtained per patient, centrifuged at 3000 g to avoid cell contamination and debris and was frozen in liquid nitrogen and stored at –80°C until analysis.
Metabolite extraction
There is no single platform or method to analyze the entire metabolome of a biological sample mainly due to the wide concentration range of metabolites coupled to their extensive chemical diversity. Therefore, metabolite extraction was accomplished by fractionating the EF samples into pools of lipid species with similar physicochemical properties using appropriate combinations of organic solvents.
Endometrial fluid samples were thawed on ice and mixed with 100 μL of water, 200 μL of methanol, and 150 μL of chloroform/methanol (2:1). The extraction solvents were spiked with metabolites not detected in unspiked human EF extracts. Samples were precipitated by centrifugation at 16,000 g for 20 min at room temperature. The following platforms were used according to the target analyte chemical class (Supplemental Table 1).
Platform 1: Fatty acyls, bile acids, and lysoglycerophospholipids; 200 μL of supernatants were collected, dried, and reconstituted in 60 μL of methanol before being transferred to vials for UPLC-MS (Waters Corp., Milford, USA) analysis.
Platform 2: Glycerolipids, cholesteryl esters (ChoE), sphingolipids, and glycerophospholipids (GLP) profiling; 200 μL of supernatants were collected, mixed with sodium chloride (50 mM). Samples were incubated for 30 min at –20 °C. After centrifugation at 16,000 g for 15 min at 4°C, 35 μL of the organic phase was collected and the solvent was removed. Dried extracts were then reconstituted in 60 μL of acetonitrile/isopropanol solution (50:50), and transferred to vials for UPLC-MS (Waters Corp.) analysis.
Platform 3: Amino acids (AA) profiling; 5 μL aliquots from the extracts prepared for platform 1 were transferred to microtubes and derivatized for AA analysis (Accq Tag Ultra Derivatization Kit; Waters Corp.).
Additionally, the study samples were injected together with six repeat injections of a pooled sample (quality control [QC]), which were used to assess the reproducibility of the analysis process. The QC sample is a pool of all samples included in the study (5 μL per sample) and was used as reference sample. The injection of the pooled sample was evenly distributed over the batches and extracted and analyzed at the same time as the individual samples.
For each of the three analytical platforms, sample preparation order was randomized from the first step of the metabolite extraction and re-randomized from the sample injection order to ensure the absence of nonsystematic biases. Only the QC sample was injected uniformly interspersed throughout the entire batch run.
Liquid chromatography-mass spectrometry analysis
A different UPLC-MS (Waters Corp.) method was used for each platform; chromatographic separation and mass spectrometric detection conditions employed are summarized in Supplemental Table 1. A reference sample was analyzed before the entire set of randomized sample injections in order to examine the retention time stability (generally <6 s variation, injection to injection), mass accuracy (platforms 1 and 2 [generally < 3 ppm for m/z 400–1000, and <1.2 mDa for m/z 50–400]), and sensitivity of the system throughout the course of the run. For each injection batch, the overall quality of the analysis procedure was monitored using five repeat extracts of the QC sample.
Data preprocessing
All data were processed using the TargetLynx application manager for MassLynx 4.1 software (Waters Corp.). A set of predefined retention time, mass-to-charge ratio pairs, Rt-m/z corresponding to metabolites included in the analysis were entered into the program. Associated extracted ion chromatograms (mass tolerance window = 0.05 Da) were then peak-detected and noise-reduced in both the LC and MS domains such that only true metabolite-related features were processed by the software. A list of chromatographic peak areas was then generated for each sample injection. The ion intensities for each peak detected were then normalized, within each sample, to the sum of the peak intensities in that sample. There was no significant correlation (F < Fcrit) between the sum of the peak intensities used for the intrabatch normalization and the groups being compared in this study.
The LC-MS features were identified prior to the analysis, either by comparison of their accurate mass spectra and chromatographic Rt with those of available reference standards or, where these were not available, by accurate mass MS/MS fragment ion analysis. Briefly, the identified ion features in the methanol extract (platform 1) included fatty acids (FA), acyl carnitine (AC), N-acylethanolamine, lysophospholipids, bile acids, and oxidized FAs. The chloroform/methanol lipid extract (platform 2) provided coverage over glycerolipids, ChoE, sphingolipids, and GLP. Lipids classification used in this study follows the comprehensive classification system proposed by Fahy et al. [13] under the leadership of the International Lipid Classification and Nomenclature Committee and expressed in the LIPID MAPS initiative (LIPID Metabolites And Pathways Strategy; http://www.lipidmaps.org).
Representative MS detection response curves were generated for identified metabolites using an internal standard for each chemical class included in the analysis. Assuming similar detector response levels for all metabolites belonging to a given chemical class allowed for a linear detection range to be defined for each variable. Maximum values were defined as those at which the detector response became nonlinear with respect to the concentration of the representative internal standard. Variables were not considered for further analysis where more than 30% of the data points were found outside their corresponding linear detection range.
Statistical analysis
The Shapiro-Wilk test showed that more than half of the metabolites did not match a normally distributed population. The lambda parameter of the Box-Cox transformation was close to 0.5, indicating that the optimal transformation that makes the data conform to a normal shape is the square root of the original data; therefore, the value of each metabolite was first square root transformed and multivariate and univariate analyses applied afterwards.
Univariate statistical analyses were performed calculating group percentage changes and unpaired Student t-test p-value (or Welch t-test where unequal variances were found) for comparing endometriosis (E) patients with nonendometriosis (NE) participants (Statistical software package R v.3.1.1 (R Development Core Team, 2011; http://cran.r-project.org with MASS package).
Multivariate data analysis was achieved using nonsupervised principal components analysis (PCA) [14] and/or supervised orthogonal partial least-squares to latent structures (OPLS) approaches [15,16] (SIMCA, version14.1.0; Umetrics, Umeå Sweden).
In order to improve the understanding about which groups of metabolites and pathways are involved in the differences found between lipidomic profiles in more detail, we performed a metabolite group enrichment analysis. This analysis consisted in a proportion test comparison with a Fisher exact test distinguishing between OE and NE. In this approach, we used an adjusted p-value for multiple comparison correction. Exploratory analysis using PCA of samples was performed to evaluate possible clinical and experimental variable effects as patient age or experimental batch to reduce the risk of bias in our study. Although the mean age of the OE patients was significantly higher than NE patients because endometriosis patients usually are older than NE patients, we cannot appreciate any related cluster to metabolite profile and age in our study (Supplemental Figure 1).
Predictive diagnostic model
Principal component and clustering exploratory analyses were performed using the Prcomp and hclust functions in R programming basic statistics (R 3.2 version) [17]. For outlier detection, the factoExtra R library was implemented with a 95% confidence interval for each class. A prediction model was trained to define lipidomic patterns for OE to classify patients. A bioinformatics pipeline was implemented to predict metabolomics using a random forest model for missing values considering each class separately (OE versus NE) as recommended by Gromski et al. [18], using the missForest R library from the Bioconductor software [19]. The support vector machine algorithm was implemented using the caret R library from the CRAN project [20] comparing significantly expressed metabolites in the OE samples with those in the NE group. Cross-validation process was performed with the strategy of Leave Group Out Cross Validation selecting 80% of the samples as training and leaving 20% out for testing. Results obtained with this cross-validation strategy avoid bias and overfitting. To avoid the unbalanced effect of sample size groups, we performed a balanced replacement model with the equal number of samples in each group and the same cross-validation parameters as our proposed model.
Results
Different lipidomic profiling in endometrial fluid from patients with endometriosis
All EFs were analyzed by LC-MS. A total of 457 metabolites were detected and subsequently underwent univariate and multivariate data analysis.
We began grouping the samples using a PCA scores plot, comparing OE with NE groups (Figure 1A). Thus, the PCA provided no separation of samples according to diagnosis (Figure 1A), patient age, or batch effect (data not shown). A supervised OPLS-Discriminant Analysis (DA) model was calculated to achieve maximum separation between the two groups of samples. Figure 1B shows the supervised OPLS-DA model scores plot and Figure 1C shows the loading plot displaying the metabolites responsible for the patterns seen among samples. Metabolites falling further away from the plot origin have a stronger impact on the model; furthermore, positively correlated variables are grouped together, while negatively correlated variables are positioned on opposite sides of the original plot in diagonally opposed quadrants. Metabolomic differences between the two groups were due mostly to several lysophosphatidylinositols, AC, and polyunsaturated triacylglycerols (TAG) with long fatty acyl chains that were found in higher amounts in OE samples. Additionally, several ceramides (CER), sphingomyelins (SM), and TAGs with shorter acyl chains were found in higher concentrations in the NE group.
PCA and supervised OPLS-DA score plots of endometriosis (E) patients and nonendometriosis (NE) controls. (A) First and second components in PCA explain 35.0% and 13.4% of the variability between samples, respectively. A slight separation of samples according to the disorder was found. The degree of fit and predictive abilities of the model, R2 and Q2 parameters, respectively, were: A = 2; R2X = 0.48, Q2X = 0.36. (B) A supervised OPLS-DA model was calculated to achieve maximum separation between the two groups of samples. Scores plot show a near perfect separation between groups. (C) The loadings plot displays the variables responsible for the patterns seen among samples in OPLS-DA model. Metabolites lying further away from the plot origin have a stronger impact on the model; furthermore, variables positively correlated are grouped together while variables negatively correlated are positioned on opposite sides of the plot origin in diagonally opposed quadrants. Metabolomic differences between the two groups were due mostly to several LPI, AC, and polyunsaturated TAG with long or very long fatty acyl chains which were higher in OE samples and to several CER, SM, and TAGs with shorter acyl chains which were higher in the NE group. Model diagnostics are (R2X = 0.46, R2Y = 1, Q2X = 0.52). AA, amino acids; TAG, triacylglycerols; DAG, diacylglycerols; PC, phosphatidylcholines; LPC, lysophosphatidylcholines; DAG, diacylglicerols; PE, phosphatidylethanolamines; LPE, lysophosphatidylethanolamines; PI, phosphatidylinositols; LPI, lysophosphatidylinositols; LPG, lysophosphatidylglycerols; BA, bile acids; ST, steroids; NEFA, nonesterified fatty acids; PUFA, polyunsaturated fatty acids; MUFA, monounsaturated fatty acids; SFA, saturated fatty acids; oxFA, oxidized fatty acids; NAE, N-acyl ethanolamines; AC, acyl carnitines; SM, sphingomyelins; CMH, monohexosylceramides; CER, ceramides.
We next performed a univariate analysis with all the metabolites found in both groups. After obtaining the raw intensity data, average group intensities, and fold-changes, we applied an unpaired Student t-test to each individual metabolite and of each metabolic class. Considering the metabolite classification (the metabolic classes calculated as the sum of the normalized areas of all the metabolites with the same chemical characteristics), univariate statistical analysis showed reduced levels of saturated diacylglycerols and saturated TAG in OE patients compared to NE participants (Table 1). Sphingolipid monohexosylceramides (CMH) and ceramides (CER) were also lower in the samples of women with endometriosis. Lastly, lower amounts of 1-ether, 2-acylglycerophosphocholines (MEMAPC) plasmanyles, and monoacylglycerophosphoethanolamines (MAPC) with fatty acyls esterified to the sn-2 position (2-MAPE) were also observed in this group (Table 1).
Differences between the lipids found in the endometrial fluids of endometriosis (E) versus nonendometriosis (NE) patients according to metabolic class and ratios.
| Classification | E Mean | OE SD | NE Mean | NE SD | log2(fold-change) | Student t-test (p-value) |
|---|---|---|---|---|---|---|
| Saturated DAG | 0.02 | 0.01 | 0.03 | 0.01 | –0.34 | 0.0081 |
| Saturated TAG | 0.2 | 0.07 | 0.32 | 0.12 | –0.67 | 0.00061 |
| CMH | 0.05 | 0.02 | 0.07 | 0.04 | –0.57 | 0.048 |
| CER | 0.13 | 0.05 | 0.18 | 0.06 | –0.45 | 0.018 |
| MEMAPC O plasmanyles | 0.57 | 0.23 | 0.75 | 0.26 | –0.39 | 0.046 |
| 2-MAPE | 0.59 | 0.12 | 0.73 | 0.15 | –0.31 | 0.016 |
| LPC/LPE | 1.38 | 0.25 | 1.16 | 0.25 | 0.26 | 0.024 |
| Classification | E Mean | OE SD | NE Mean | NE SD | log2(fold-change) | Student t-test (p-value) |
|---|---|---|---|---|---|---|
| Saturated DAG | 0.02 | 0.01 | 0.03 | 0.01 | –0.34 | 0.0081 |
| Saturated TAG | 0.2 | 0.07 | 0.32 | 0.12 | –0.67 | 0.00061 |
| CMH | 0.05 | 0.02 | 0.07 | 0.04 | –0.57 | 0.048 |
| CER | 0.13 | 0.05 | 0.18 | 0.06 | –0.45 | 0.018 |
| MEMAPC O plasmanyles | 0.57 | 0.23 | 0.75 | 0.26 | –0.39 | 0.046 |
| 2-MAPE | 0.59 | 0.12 | 0.73 | 0.15 | –0.31 | 0.016 |
| LPC/LPE | 1.38 | 0.25 | 1.16 | 0.25 | 0.26 | 0.024 |
DAG, diacylglycerols; TAG, triacylglycerols; CMH, monohexosylceramides; CER, ceramides; MEMAPC, 1-ether, 2-acylglycerophosphocholines; MAPE, monoacylglycerophosphoethanolamines; LPC, lysophosphatidylcholines; LPE, lysophosphatidylethanolamines.
Differences between the lipids found in the endometrial fluids of endometriosis (E) versus nonendometriosis (NE) patients according to metabolic class and ratios.
| Classification | E Mean | OE SD | NE Mean | NE SD | log2(fold-change) | Student t-test (p-value) |
|---|---|---|---|---|---|---|
| Saturated DAG | 0.02 | 0.01 | 0.03 | 0.01 | –0.34 | 0.0081 |
| Saturated TAG | 0.2 | 0.07 | 0.32 | 0.12 | –0.67 | 0.00061 |
| CMH | 0.05 | 0.02 | 0.07 | 0.04 | –0.57 | 0.048 |
| CER | 0.13 | 0.05 | 0.18 | 0.06 | –0.45 | 0.018 |
| MEMAPC O plasmanyles | 0.57 | 0.23 | 0.75 | 0.26 | –0.39 | 0.046 |
| 2-MAPE | 0.59 | 0.12 | 0.73 | 0.15 | –0.31 | 0.016 |
| LPC/LPE | 1.38 | 0.25 | 1.16 | 0.25 | 0.26 | 0.024 |
| Classification | E Mean | OE SD | NE Mean | NE SD | log2(fold-change) | Student t-test (p-value) |
|---|---|---|---|---|---|---|
| Saturated DAG | 0.02 | 0.01 | 0.03 | 0.01 | –0.34 | 0.0081 |
| Saturated TAG | 0.2 | 0.07 | 0.32 | 0.12 | –0.67 | 0.00061 |
| CMH | 0.05 | 0.02 | 0.07 | 0.04 | –0.57 | 0.048 |
| CER | 0.13 | 0.05 | 0.18 | 0.06 | –0.45 | 0.018 |
| MEMAPC O plasmanyles | 0.57 | 0.23 | 0.75 | 0.26 | –0.39 | 0.046 |
| 2-MAPE | 0.59 | 0.12 | 0.73 | 0.15 | –0.31 | 0.016 |
| LPC/LPE | 1.38 | 0.25 | 1.16 | 0.25 | 0.26 | 0.024 |
DAG, diacylglycerols; TAG, triacylglycerols; CMH, monohexosylceramides; CER, ceramides; MEMAPC, 1-ether, 2-acylglycerophosphocholines; MAPE, monoacylglycerophosphoethanolamines; LPC, lysophosphatidylcholines; LPE, lysophosphatidylethanolamines.
A total of 123 of 457 (27%) compounds showed significant differences (P < 0.05) between the OE and NE groups (Figure 2). Ninety-five were detected at a lower concentration in the OE group; mainly sphingolipids, GLP, and glycerolipids. The 28 metabolites found in higher amounts in OE patients were primarily mono- and polyunsaturated TAG with 52 carbon atoms. A volcano plot (Figure 2A) highlights the most significant differences among metabolites and the positive or negative fold-change seen in OE patients compared to the NE group. Significant differences with a p-value < 0.001 were phosphatidylcholines (PC 22:6/0:0) found in higher concentrations in the E samples and ceramides (CER d18:1/21:0, CER d18:1/23:0), phosphatidylcholines (PC O-42:6), 1-ether, 2-acylglycerophosphocholines, sphingomyelins (SM d18:1/25:0), and triacylglycerols (TAG 46:0, TAG 48:0, TAG 48:1, and TAG 50:4) found in lower amounts in the EF of women with endometriosis.
Volcano plot and heat map. (A) Volcano plot showing the most significant metabolites found by univariate analysis. The levels of 123 out of 457 metabolites were significantly different in the ovarian endometriosis group (OE) compared to the nonendometriosis group (NE). The volcano plot summarizes both fold-change and t-test criteria for all metabolites. It is a scatter-plot of the negative log10-transformed p-values from the t-test plotted against the log2 fold change. Negative values indicate downregulated metabolites in OE patients, while positive values reflect upregulated metabolites in OE patients. Metabolites with statistically significant differential levels according to the t-test lie above a horizontal threshold line. Metabolites with large fold-change values lie far from the vertical threshold line at log2 fold change = 0, indicating whether the metabolite is up or downregulated. (B) Heat map: green and red colors indicate higher drops and elevations of the metabolite levels, respectively. Gray lines correspond to significant fold-changes of individual metabolites and darker gray colors highlight statistical significance (P < 0.05, P < 0.01, or P < 0.001). The metabolites have been ordered in the heat map according to carbon number and unsaturation degree of acyl chains. AA, amino acids; AC, acyl carnitines; BA, bile acids; CER, ceramides; CMH, monohexosylceramides; ChoE, cholesteryl esters; DAG, diacylglycerols; LPC, lysophosphatidylcholines; LPE, lysophosphatidylethanolamines; LPG, lysophosphatidylglycerols; LPI, lysophosphatidylinositols; MUFA, monounsaturated fatty acids; NAE, N-acyl ethanolamines; PC, phosphatidylcholines; PE, phosphatidylethanolamines; PI, phosphatidylinositols; PUFA, polyunsaturated fatty acids; SFA, saturated fatty acids; SM, sphingomyelins; ST, steroids; TAG, triacylglycerols.
To summarize the univariate results, we used a heat map (Figure 2B) to display the fold-change of the 457 metabolic features included in the analysis together with the unpaired Student t-test comparison. The heat map allows the visualization of the different TAG levels according to the number of carbon atoms and the degree of unsaturation; TAG species with shorter acyl chains and unsaturation degree are lower in endometriosis samples, while the contrary was observed for species containing longer acyl chains esterified to glycerol and higher numbers of double bonds. Among AA, only cysteine was significantly increased (fold-change = 0.6, p = 0.01). ACs were higher in the EF of women with endometriosis. Significant increments were found for the species with saturated short and medium fatty acyl chains [AC(6:0), AC(8:0), and AC(10:0)]. Glycerophospholipids showed a trend to be reduced in OE samples, except for LPC and LPI classes. A significant drop in the concentration of MEMAPC plasmanyles, and of some phosphatidylethanolamines plasmalogen species, was noted in the fluid of women with endometriosis, along with lower levels of the sphyngolipids CER, SM, and CMH.
In summary, 123 of the 457 (27%) metabolites measured were found significantly differentially expressed in the EF of OE patients compared to NE; 95 of these were in lower concentrations in endometriosis patients.
Finally, to analyze the differences in the lipidomic profile between OE and controls, we used a metabolite group enrichment analysis perspective. This approach analyzes the proportion of metabolites from the same group independently on the concentration value. Glycerolipids and GLP were overrepresented (FDR < 0.05) (Table 2), meaning that these pathways are somehow altered in endometriosis.
Metabolite group enrichment analysis.
| Class | metRatio | bgRatio | FDR |
|---|---|---|---|
| Glycerolipids | 47/123 | 88/457 | 1.17E-09 |
| Glycerophospholipids | 39/123 | 216/457 | 0.0046 |
| Nonesterified fatty acids | 1/123 | 22/457 | 0.0574 |
| Sphingolipids | 17/123 | 48/457 | 2.24E-01 |
| Amino acid derivatives | 1/123 | 1/457 | 0.3355 |
| Fatty esters | 4/123 | 9/457 | 0.3355 |
| Bile acids | 2/123 | 11/457 | 0.8435 |
| Oxidized fatty acids | 4/123 | 18/457 | 1 |
| Class | metRatio | bgRatio | FDR |
|---|---|---|---|
| Glycerolipids | 47/123 | 88/457 | 1.17E-09 |
| Glycerophospholipids | 39/123 | 216/457 | 0.0046 |
| Nonesterified fatty acids | 1/123 | 22/457 | 0.0574 |
| Sphingolipids | 17/123 | 48/457 | 2.24E-01 |
| Amino acid derivatives | 1/123 | 1/457 | 0.3355 |
| Fatty esters | 4/123 | 9/457 | 0.3355 |
| Bile acids | 2/123 | 11/457 | 0.8435 |
| Oxidized fatty acids | 4/123 | 18/457 | 1 |
metRatio is the proportion of differential expressed metabolites belonging to the same group in the total of metabolites differentially expressed in this comparison; bgRatio is the proportion of total differential expressed metabolites from the same group in the total of metabolites detected in this study.
Metabolite group enrichment analysis.
| Class | metRatio | bgRatio | FDR |
|---|---|---|---|
| Glycerolipids | 47/123 | 88/457 | 1.17E-09 |
| Glycerophospholipids | 39/123 | 216/457 | 0.0046 |
| Nonesterified fatty acids | 1/123 | 22/457 | 0.0574 |
| Sphingolipids | 17/123 | 48/457 | 2.24E-01 |
| Amino acid derivatives | 1/123 | 1/457 | 0.3355 |
| Fatty esters | 4/123 | 9/457 | 0.3355 |
| Bile acids | 2/123 | 11/457 | 0.8435 |
| Oxidized fatty acids | 4/123 | 18/457 | 1 |
| Class | metRatio | bgRatio | FDR |
|---|---|---|---|
| Glycerolipids | 47/123 | 88/457 | 1.17E-09 |
| Glycerophospholipids | 39/123 | 216/457 | 0.0046 |
| Nonesterified fatty acids | 1/123 | 22/457 | 0.0574 |
| Sphingolipids | 17/123 | 48/457 | 2.24E-01 |
| Amino acid derivatives | 1/123 | 1/457 | 0.3355 |
| Fatty esters | 4/123 | 9/457 | 0.3355 |
| Bile acids | 2/123 | 11/457 | 0.8435 |
| Oxidized fatty acids | 4/123 | 18/457 | 1 |
metRatio is the proportion of differential expressed metabolites belonging to the same group in the total of metabolites differentially expressed in this comparison; bgRatio is the proportion of total differential expressed metabolites from the same group in the total of metabolites detected in this study.
Predictive diagnostic model
The 123 metabolic species with significantly different amounts between the two groups used to train a predictive model to distinguish between E and NE samples (Figure 3). Twelve E samples and 23 NE samples were included as a training set (Figure 3B). The support vector machine probabilistic model for the training set, using cross-validation, was implemented to distinguish endometriosis-positive samples. The resultant model predicts OE with 0.857 accuracy (95% C.I, 0.697, 0.95), 1.000 specificity, and 0.583 sensitivity. The confusion matrix is detailed in Figure 3A. With this model, we were able to classify 86% of the samples correctly. One hundred per cent of the NE samples and 58% of the OE samples were correctly diagnosed, while 42% of the OE samples were miss-classified (Figure 3A and 3B). Looking at the cluster tree (Figure 3B), we were able to detect a well-defined area with a molecular lipidomic profile of endometriosis, although some misclassification of OE appear in the area of NE patients. This fact could be due to the heterogeneity between endometriosis samples.
Predictive diagnostic model. (A) Training set confusion matrix 100-fold/leave 20% group out cross-validation. (B) Unsupervised clustering by Euclidean distance and WardD agglomerative minimum variance method. Miss-classification of endometriosis is indicated with an asterisk in the samples. OE, ovarian endometriosis; NE, nonendometriosis patients marked area defines the well classified branch of endometriosis patients. (C) Metabolite importance by group in endometriosis classification; X-axis represents the metabolite's importance in endometriosis prediction (0–1 range, average, and 95% confidence interval is shown).
The unbalanced effect of sample size (12:23) was incorporated into the calculations and it was determined that there was no proportion effect in the model nor in the prediction parameters (data not shown). The importance of each group of metabolites to the endometriosis diagnostic model is shown in Figure 3C. The fatty ester group represented by ACs and the uniquely expressed AA cysteine are the principal responsible features in endometriosis classification.
Discussion
Lipidomics is an emerging discipline in systems biology that aims to investigate and model lipids at the global level. In this study, we used LC-MS/MS-based lipidomics in a case–control study for the lipid profiling to evaluate possible changes in the EF of patients with OE at the time of embryonic implantation. To our knowledge, this is the first study to evaluate the lipidomic profile in the EF of E versus NE patients.
Although several studies have attempted to identify endometriosis using lipids as biomarkers of the disease, none have analyzed EF for these potential biomarkers. Vouk et al. [21] studied the serum of OE patients using a similar mass spec method and described eight differential lipid metabolites. Among them were the hydroxysphingomyelins SMOH C16:1 and SMOH C22:2, the sphingomyelin SMC16:1, and five ether-phospholipids (acyl-alkyl-phosphatidylcolines) as well as two saturated 2-acyl-1-alkyl-sn-glycero-3-phosphocholines (plasmanylcholines). The authors developed a model to discriminate endometriosis and nonendometriosis using several lipids with a sensitivity of 90.0% and a specificity of 84.3%. In our study, we found the sphingolipids CMH and ceramides in lower levels in the EF samples of women with endometriosis, confirming that these lipids are likely important markers of the disease.
In another study, Lee et al. [12] evaluated sphingolipid metabolism in peritoneal fluid, serum, and endometrial tissue. They found an altered sphingolipid metabolism flux in all tissues studied and a strong GlcCer correlation in patients with endometriosis, suggesting that GlcCer, a mitogenic factor, could be a prime candidate for lesion subsistence ectopically. Another study, which found deregulated levels of different lipids, such as sphingosine-1-phospate (S1P), sphingosine kinases SPHK1 and SPHK2, and increased phosphatases (SGPP1) in ectopic tissue compared to eutopic control tissue, also suggests that this lipid pathway might be important in the grafting and survival of endometriosis lesions [22].
It has been suggested that endometriotic cells have altered intracellular signals of apoptosis [23]. Sphingomyelin and ceramides are closely associated with the apoptosis pathways and the Fas-FasL system and ectopic and eutopic stromal endometrial cells from women with endometriosis have a damaged ceramide-downstream pathway to apoptosis [11]. We found some altered sphingolipids in EF from women with OE. Sphingolipids are bioactive compounds and their biosynthesis and metabolism modulate a range of cellular processes including proliferation, migration, and apoptosis; an imbalance in their metabolism has been previously linked to endometriosis pathophysiology [12].
We also discovered that cysteine was significantly increased in OE patients. This aa is important not only as a precursor for protein and antioxidant glutathione biosynthesis, but also in the maintenance of physiological redox conditions. There is evidence indicating that oxidative stress has an active role in the development of endometriosis [24]; thus, increased cysteine levels in EF could be an indication of alterations to redox balance.
We found increased levels of AC in the EF of women with OE. Increased levels of AC are associated with cell beta-oxidation dysfunction, and previous lipidomics studies have verified that different concentrations of AC due to the efficiency of mitochondrial membrane-bound enzymes are involved in the beta-oxidation process related to various levels of inflammation [21].
Finally, although this is a pilot study with a low number of samples, our predictive model classified 58% of OE cases and 100% of NE samples correctly, and another different cohort of patients should be employed to validate our model. Endometriosis classification has been considered extremely difficult [25]. For example, CA125, one of the most studied blood biomarkers of endometriosis, has been recently analyzed in a meta-analysis [26], obtaining a 93% specificity (95% CI 89%–95%) and 52% sensitivity (95% CI 38%–66%), while our model showed a 100% specificity and 58.3% sensitivity.
Although differential lipidomic profiles have been found, we are aware of the limitations of this predictive model but it might be improved with a larger sample size and deep phenotyping of the population investigated.
Supplementary data
Supplementary data are available at BIOLRE online.
Supplemental Figure 1. Principal component analysis (PCA) of the metabolomic profile for nonendometriosis (NE) and ovarian endometriosis (OE) samples related to age. First and second components in PCA analysis explained 36.0% and 12.4% of the variability between samples, respectively. Samples are colored depending on the range of age. Although the major part of endometriosis patients (OE) are older than donors (purple triangles), we can appreciate that there is no sample separation according to age.
Supplemental Table 1. UPLC-MS analysis methods.
Acknowledgments
We would like to thank Cristina Alonso, Miriam Perez and Patricia Sebastián-León for their kind help with the metabolomic analysis.
References
Author notes
Grant support: F.D.'s participation in this work was supported by the Spanish Ministry of Economy and Competitiveness, through the Miguel Servet Programme (CP13/00075) co-founded by FEDER and Basque Country ETORGAI grant: ER-2012/00020.
