Abstract
In order to undertake an epidemiologic study relating levels of parent estrogens (estrone and estradiol) and estrogen metabolites (EMs) to other breast cancer risk factors, we have optimized methods for EM quantification with ultra high performance liquid chromatography-high resolution mass spectrometry (UHPLC-HRMS). A twostep approach was adopted; the first step comprised method development and evaluation of the method performance. The second step consisted of applying this method to quantify estrogens in postmenopausal women and determine if the observed patterns are consistent with the existing literature and prior knowledge of estrogen metabolism. First, 1-methylimidazole-2-sulfonyl chloride (MIS) was used to derivatize endogenous estrogens and estrogen metabolites in urine from study participants. Since C18 reversed phase columns have not been able to separate all the structurally related EMs, we used a C18-pentafluorophenyl (PFP) column. The parent estrogens and EMs were baseline resolved with distinct retention times on this C18-PFP column using a 30 min gradient. This method was used to quantify the parent estrogens and 13 EMs in urine samples collected in an initial pilot study involving males as well as pre- and peri-menopausal females to assess a range of EM levels in urine samples and enable comparison to the previous literature for assay evaluation. Detection limits ranged from 1 − 20 pg/mL depending on the EM. We evaluated matrix effects and interference as well as the intra- and inter-batch reproducibility including hydrolysis, extraction, derivatization and LC-MS analysis using charcoal-stripped human urine as a matrix. Methods were then applied to the measurement of estrogens in urine samples from 169 postmenopausal women enrolled in an epidemiological study to examine relationships between breast cancer risk, the intestinal microbiome, and urinary EMs. The results from our cohort are comparable to previous reports on urinary EMs in postmenopausal women and enabled thorough evaluation of the method.
Keywords: Derivatization, Estrogens, Liquid chromatography-high resolution mass, spectrometry, Metabolites
1. Introduction
Estrogens play an important role in the development of many female cancers such as breast, endometrial and ovarian cancer [1,2]. Epidemiologic studies have shown that elevated endogenous estrogen levels, specifically parent estrogens, estradiol and estrone, are associated with increased postmenopausal breast cancer risk [2–4]. However, the overall relationships of estrogen metabolism to breast cancer risk are still not fully understood, in part because of the challenges in detecting estrogen metabolites (EM) in human biofluids at the low concentrations observed in postmenopausal women [5]. Parent estrogens are metabolized by three competitive pathways, 2-, 4- and 16-pathways via irreversible hydroxylation at positions C2, C4 and C16 to catechol estrogens (2-hydroxyestrone, 4-hydroxyestrone, 2-hydroxyestradiol, and 4-hydroxyestradiol) and 16α-hydroxyestrone and subsequently to methoxylated metabolites [6]. EMs in the 2-pathway include 2-hydroxyestrone, 2-hydroxyestradiol, 2-methoxyestrone, 2-methoxyestradiol and 3-methoxyestrone. The 4-pathway consists of 4-methoxyestrone, 4-methoxyestradiol and 4-hydroxyestrone. EMs in the 16-pathway include 16α-Hydroxyestrone, 16-Ketoestradiol, 16- Epiestriol, Estriol and 17-Epiestriol. Among the EMs in these pathways, 2-hydroxymetabolites have little estrogenic activity and have been classified as anti-carcinogenic [7], whereas 16α-hydroxyestrone and 4-hydroxymetabolites are relatively more estrogenic and have genotoxic potential [6]. 2-Hydroxyestrone (2-OHE1) and 16α-hydroxyestrone (16α-OHE1) are produced by the 2- or 16- pathway, respectively, so the net effect could potentially depend on which metabolic pathway dominates within a woman [8]. The elevation of 16α-OHE1 has been associated with greater risk of hormonally related cancers, including breast cancer; therefore, the ratio of 2-OHE1 to 16α-OHE1 may be useful in measuring a women’s exposure to more potent metabolites [9–12]. Several factors are known to be associated with changes in the levels of 2- to 16-pathway ratio as measured in urine, including exercise, body weight, diet, hormone therapy and alcohol consumption [9,13,14]. Comparisons of other urinary estrogen metabolism pathway ratios also show association with breast cancer risk. For example, Moore et al. evaluated various metabolic pathway ratios in relation to breast cancer risk, specifically showing that higher ratios of 2-pathway metabolites to total estrogens/EMs and 2-pathway metabolites to parent estrogens were associated with 40% lower breast cancer risk [15]. Previous findings also suggest that low values for the 2- to 16-pathway ratio are associated with significant increases in breast cancer risk, while high values for this ratio are associated with lower risk [16]. Based on the accumulated evidence, all of these EMs are measured as a panel (see Table 1).
Table 1.
Targets of Quantification. EMs were derivatized with 1-methylimidazole-2-sulfonyl chloride (MIS) prior to analysis with UHPLC-HRMS. Three metabolites were derivatized on two separate functional groups, as noted below. N/A indicates standard not available.
Derivatized Estrogen Metabolite | Abbreviation | Metabolite [M+H]+ | Isotope Label | Standard [M+H]+ |
---|---|---|---|---|
Estrone-MIS | E1-MIS | 415.1686 | 13C3 | 418.1787 |
17β-Estradiol-MIS | E2-MIS | 417.1843 | 13C6 | 423.2044 |
16α-Hydroxyestrone-MIS | 16α-OHE1-MIS | 431.1635 | 13C3 | 434.1736 |
16-Ketoestradiol-MIS | 16-ketoE2-MIS | 431.1635 | D5 | 436.1949 |
Estriol-MIS | E3-MIS | 433.1792 | 13C3 | 436.1892 |
16-Epiestriol-MIS | 16-epiE3-MIS | 433.1792 | D2 | 435.1917 |
17-Epiestriol-MIS | 17-epiE3-MIS | 433.1792 | N/A | N/A |
2-Hydroxyestrone-MIS2 | 2-OHE1-MIS2 | 575.1629 | 13C6 | 581.1830 |
2-Hydroxyestradiol-MIS2 | 2-OHE2-MIS2 | 577.1785 | 13C6 | 583.1986 |
4-Hydroxyestrone-MIS2 | 4-OHE1-MIS2 | 575.1629 | 13C6 | 581.1830 |
2-Methoxyestrone-MIS | 2-MeOE1-MIS | 445.1792 | 13C6 | 451.1993 |
2-Methoxyestradiol-MIS | 2-MeOE2-MIS | 447.1948 | 13C6 | 453.2149 |
3-Methoxyestrone-MIS | 3-MeOE1-MIS | 445.1792 | 13C6 | 451.1993 |
4-Methoxyestrone-MIS | 4-MeOE1-MIS | 445.1792 | 13C6 | 451.1993 |
4-Methoxyestradiol-MIS | 4-MeOE2-MIS | 447.1948 | 13C6 | 453.2149 |
Urine samples are commonly used in epidemiology studies to measure biomarkers [17]. Human urine contains conjugated parent estrogens and their metabolites, including sulfate or glucuronide conjugates, and the levels of endogenous EMs are low in postmenopausal women, i.e. < 100 pg/mL [18]. Urine has several advantages as compared to blood, as it is non-invasive, inexpensive and relatively easy to collect for large-scale epidemiologic studies [19,20] and can provide a straightforward way to measure these biomarkers. Estrogens are metabolized in the liver, and some conjugates re-enter circulation from the intestines; both mechanisms lead to transport to the kidney for excretion [21]. A recent study showed moderate levels of correlation between serum and urinary estrogens/estrogen metabolites in premenopausal women; however, it was observed that urinary levels were less correlated with serum when evaluating EM pathways or relative total concentrations of estrogens and EMs [21]. However, evaluation of these differences in the serum of postmenopausal women and males remains challenging due to the low levels of circulating estrogens in these populations [19], so we chose to focus on measurements in urine.
Gas chromatography (GC) or liquid chromatography (LC) with mass spectrometry (MS) detection supports sensitive analytical methods that can measure these analytes in urine samples. However, with high-resolution LC and MS approaches, including the application of ultra-high performance liquid chromatography (UHPLC) to reduce analysis time, LC-MS has become the favored approach [22]. Although LC-MS is a sensitive and selective technique, estrogens and EMs are often below the limit of detection, particularly in urine collected from postmenopausal women [23]. Therefore, derivatization reagents like dansyl chloride (DNSC) have been used to improve sensitivity [24,25]. DNSC is a commonly used reagent that reacts selectively with aromatic hydroxyl groups in estrogens and their metabolites; this labeling strategy supports ionization and provides a fragment ion reporter that can be detected in tandem mass spectrometry (MS/MS) experiments. However, studies have shown that other derivatizing reagents, such as 1-methylimidazole-2-sulfonyl chloride (MIS), can be used to improve sensitivity for detection of estrogens [26]. Derivatization methods combined with LC-MS using a high resolution (HR) Orbitrap mass spectrometer have also been shown to improve sensitivity of estrogen measurements [26]. Previous LC-MS- and LC-MS/MS-based studies utilize derivatization to improve detection sensitivity of estrogens in biological matrices; however, these existing methods can still be optimized to improve sensitivity. Previously published liquid chromatography-selected reaction monitoring mass spectrometry (LC-SRM) assays use a total analysis time of 100 min to quantify EMs in human serum samples [27] or 88 min to quantify EMs in urine samples. Another published LC-MS assay to quantify estrogens in human serum samples utilized a 20 min separation, but could not chromatographically resolve all EMs [26]. Furthermore, SRM strategies select a wide m/z window (Q1 resolution typically set to 0.7) and rely solely on the detection of a fragment ion from the tagging reagent. To overcome these limitations, we tested three different mixed mode reversed phase columns to improve separation of the EMs and optimized gradient length to develop a robust and reproducible method using MIS for derivatization of 15 endogenous estrogens and EMs in urine for LC-HRMS quantitation that has the advantage of retaining the specific m/z of the derivatized parent compound. Here, we report using an ACE C18-pentafluorophenyl (PFP) column with HRMS detection to separate and quantify all 15 MIS derivatized estrogens and estrogen metabolites in human urine. Assay performance is examined in a large population of volunteer samples to evaluate its advantages, limitations, and overall performance. Finally, a quality control strategy using bioinformatics and mapping against known biology is described and applied to the dataset from the volunteer cohort.
2. Description of experiments
2.1. Reagents and materials
Estrogen and estrogen metabolite standards, estrone (E1), estradiol (E2), 16-epiestrol (16-epiE3), and estriol (E3) were purchased from MilliporeSigma (St. Louis, MO). 17-Epiestriol (17-epiE3), 16-ketoestradiol (16-ketoE2), 16α-Hydroxyestrone (16α-OHE1), 2-Hydroxyestrone (2-OHE1), 4-Hydroxyestrone (4-OHE1), 3-methoxyestrone (3-OHE1), 4-methoxyestrone (4-MeOE1), 2-methoxyestrone (2-MeOE1), 2-Methoxyestradiol (2-MeOE2); and 4-Methoxyestradiol (4-MeOE2) were purchased from Steraloids (Newport, RI). 2-Hydroxyestradiol (2-OHE2) was purchased from Cambridge Isotope Laboratories (Andover, MA). Deuterium-labeled internal standards, including D2-16epi-E3 and D5-16-ketoE2, were purchased from C/D/N Isotopes (Pointe-Claire, QC, Canada). The 13C-labeled internal standards, 13C3-E1, 13C6-E2, 13C3-16α-OHE1, 13C3-E3, 13C6-2-OHE1, 13C6-2-OHE2, 13C6-4-OHE1, 13C6-2-MeOE1, 13C6-2-MeOE2, 13C6-3-MeOE1, 13C6-4-MeOE1, and 13C6-4-MeOE2, were purchased from Cambridge Isotope Laboratories (Andover, MA). The derivatizing reagent, 1-methylimidazole-2-sulfonyl chloride (MIS), was purchased from MilliporeSigma (St. Louis, MO). HPLC grade dichloromethane was purchased from VWR International (Radnor, PA). HPLC grade water and acetonitrile were purchased from Honeywell - Burdick & Jackson (Muskegon, MI). Sodium bicarbonate, 1 M sodium acetate buffer solution, pH 4.6, resorufin, resorufin β-D-glucuronide and β-glucuronidase/sulfatase from Helix pomatia were purchased from MilliporeSigma (St. Louis, MO). Formic acid, L-ascorbic acid, and anhydrous acetone were purchased from Fisher Scientific (Hampton, NH). Charcoal-stripped human urine (Mass Spec Gold Urine, MSG5000, Lot #C09002) was purchased from Golden West Biologicals (Temecula, CA).
2.2. Stock solutions and standards
Stock solutions of EM standards and their respective stable isotope-labeled standards were prepared in 100% methanol at concentrations of 1 mg/mL and 20 µg/mL. Stock solutions were stored at −20 °C. To determine chromatographic elution order, 50 ng of EM standards were derivatized by adding 100 µL 1-methylimidazole-2-sulfonyl chloride at 3 mg/mL in anhydrous acetone and incubating for 15 min at 60 °C. Derivatized standards were dried and resuspended in 50 µL of loading solvent (aqueous 35% acetonitrile containing 0.1% formic acid).
Calibration standards were prepared by adding 10 µL of solution containing 500 ng/mL of Deuterium and 13C stable isotope-labeled standards (SIS) to add 5 ng of each SIS to each urine sample and varying amounts of a working EM standard solution at 0.10, 0.24, 1, 4, 20, 100, 500, 1000, 4000 and 6400 pg/mL to charcoal-stripped human urine. Calibration standards were analyzed in triplicate. Quality control (QC) samples were prepared in charcoal-stripped human urine by adding 4.0 pg/mL and 6.4 ng/mL synthetic EMs, and 5 ng of SIS EMs were added. QC samples were processed and analyzed with each batch of 12 samples to evaluate consistency of sample preparation and instrument performance. Human urine samples were randomized to avoid batch effects between samples collected at different sites or based on time of collection. One sample from each batch was re-analyzed as a duplicate in a final batch to further evaluate the consistency of the results.
2.3. Deglucuronidation efficiency and recovery of estrogen metabolites following hydrolysis
Because glucuronidated EM standards were not commercially available, we used resorufin β-D-glucuronide as an external standard to evaluate hydrolysis efficiency. Glucuronidated resorufin (140 ng) and D6-resorufin (70 ng) were spiked into 0.5 mL of charcoal-stripped human urine. The following enzyme amounts were tested: 0, 400, 1000, 1600 and 3200 units. Different incubation times at 60 °C were also evaluated, including 0, 15, 60, 240 and 1080 min. UHPLC-HRMS was used to evaluate levels of protonated glucuronidated resorufin after treatment with different amounts of enzyme as well as different incubation times to show complete loss of signal for the conjugated molecule. To determine the impact of deglucuronidation on EM levels in human urine, the levels of free EMs were compared to total EMs (without and with enzymatic deglucuronidation) using 6 replicates from each urine sample (n = 2; collected from post-menopausal female volunteers). Peak heights were used to calculate the amount of endogenous EM by comparison to the spiked internal standards. The average ratio was calculated by taking the light-to-heavy ratio of each EM in the untreated sample to quantify free EMs divided by the light-to-heavy ratio of each EM in the treated sample used to quantify total EMs across all the replicates.
2.4. Preparation of samples to evaluate matrix effects and percent recovery
Matrix effects on chromatographic separation and retention time were assessed. A fixed amount (5 ng) of estrogen and EM standards as well as the stable isotope-labeled internal standard mixture were spiked into 100% methanol (neat solvent) and derivatized with 1-methylimidazole-2-sulfonyl chloride in anhydrous acetone, as described above. Similarly, samples were prepared by spiking the same amount of estrogens and EMs and internal standard mixture in charcoal-stripped human urine. Retention time reproducibility was evaluated for the parent estrogens and EMs spiked into charcoal-stripped human urine and background free solvent. The goodness-of-fit was calculated based on the correlation coefficient (R2). In addition, underivatized human urine was compared to derivatized estrogens and EMs in solvent to determine any potential for interference in these measurements from other compounds. Briefly, duplicate 0.5 mL samples of human urine were processed using hydrolysis and extraction followed by drying and resuspension in 50 µL of loading solvent (aqueous 35% acetonitrile containing 0.1% formic acid). A mixture of estrogens and EM standards (25 ng) was derivatized, as described above. Aliquots (25 µL) of underivatized urine and derivatized standards were injected for LC-HRMS analysis. Data analysis was performed using Skyline (Version 20.1.0.76) [28] to compare between the underivatized urine and the derivatized EM standards. A mass tolerance was set to 5 ppm in Skyline. We also analyzed derivatized EMs from urine samples to determine if other molecules create interference after derivatization. We compared EMs spiked at 250 pg/mL into postmenopausal female urine and in “neat” solvent. Data analysis was performed using Skyline to extract peak heights with a mass tolerance of 5 ppm.
Matrix effects influencing ionization were also evaluated for the individual EMs. We compared the MS response (peak height) of individual EMs spiked at a given concentration (250, 1000, and 6400 pg/mL) into charcoal-stripped human urine (CSHU) to the MS response of the same individual EMs in the “neat” solvent. Using the ratio of signal in CSHU to signal in solvent [29], we determined the matrix effect. Values of 100% would indicate that the response was the same in neat solvent and matrix, ˃100% would indicate potential interference from the matrix, and < 100% would indicate loss of compound, interference in SIS, or ion suppression.
Recovery of estrogens and EMs was evaluated by spiking in 250, 1000 and 6400 pg/mL of synthetic EM standards and a fixed amount (5 ng) of stable isotope-labeled standards in 0.5 mL of post-menopausal female urine. Duplicate samples were prepared using hydrolysis, extraction, and derivatization, and then analyzed using LC-MS. A blank injection was run between each sample. The percent recovery of EMs was calculated by subtracting the measured concentration of unspiked urine from the measured concentration of urine spiked with standard then dividing by the expected concentration of the spiked amount. The results are reported as the mean of urine samples collected from two volunteers.
2.5. Evaluation of carryover
LC-MS carryover was evaluated by injecting EMs extracted from the urine of a pre-menopausal female on birth control (BC), selected based on her high EM levels; six replicates were analyzed for each of two different types of solvent blank injections. The first solvent blank was analyzed using the EMs LC-MS method directly after analysis of the urine sample. Following the analysis of the first solvent blank, which shows carryover without washing, a wash using the “saw-tooth” gradient was performed, and a second solvent blank was analyzed to evaluate carryover after washing. The carryover is expressed as the ratio of the EM peak height in the solvent blank against the EM peak height from the original urine sample.
2.6. Preparation of samples to assess accuracy and reproducibility
The intra-batch and inter-batch accuracy and reproducibility of estrogen metabolite quantification was determined by spiking EM standards at 250, 1000 and 6400 pg/mL with the addition of 5 ng of Deuterium- and 13C-labeled standards in 0.5 mL of charcoal-stripped human urine. Each sample was prepared independently, which included hydrolysis, extraction, derivatization, and LC-MS analysis repeated in triplicate on each of three consecutive days. The accuracy was assessed by comparing the mean determined concentration to the expected concentration. The percent accuracy was assessed using the Relative Mean Error (RME), ((mean measured value - theoretical value)/theoretical value) × 100); accuracy was considered acceptable for RME values ≤20%. The coefficient of variation (CV) was also used to assess the precision of the assay. CV values were accepted if they were ≤20% [30].
To assess reproducibility, estrogen metabolites from 13 postmenopausal women urine were processed in duplicates. A single urine sample from each batch was reprocessed and reanalyzed. The linear regression of the calibration curve was characterized by the slope, intercept and R-squared value of the sample processing duplicates.
2.7. Quality control to assess ion signal stability of stable isotope-labeled standards
The stability of stable isotope-labeled standards (SIS) in human urine was evaluated. A fixed amount of SIS EM mixture (5 ng) was spiked into each sample. Each sample was prepared independently, including hydrolysis, extraction, and derivatization, then analyzed using LC-MS and repeated across all 16 batches. The signal stability in the mass spectrometer was monitored for each sample and across all batches.
2.8. Urine samples
Urine samples were collected from women undergoing breast cancer screening at the Moffitt Cancer Center as well as from volunteers who responded to study recruitment flyers at the University of Florida (UF). Most of the women were White/non-Hispanic (82%) and the remaining women were Hispanic (8%), African American (7%) or other races including Asian, Native American, and more than 2 races (3%). All biological samples were collected under Institutional Review Board (IRB) approved protocols (UF: IRB 201500572 and IRB 201600709; Moffitt Cancer Center SRC 18419; Advarra IRB Pro00014574). All participants provided written informed consent prior to sample and survey data collection. In the initial method development part of the project, urine samples were collected in the morning from two males, three premenopausal females (one woman using oral contraceptives, one woman with no current contraceptive use, and one pregnant) and one perimenopausal female to test the assay in a pilot study. For the main epidemiologic study, a urine collection cup was used to capture ~50 mL of urine from individual volunteers; then 2.0 mL was aliquoted and stored at −80 °C for later quantitation of estrogens and EMs. The spot urine samples were collected from 169 postmenopausal women between 2016 and 2018 at the University of Florida (n = 63) and Moffitt Cancer Center (n = 106). The eligibility criteria included: (1) no previous diagnosis of breast cancer; (2) postmenopausal status; (3) no current or previous hormone therapy; (4) no hepatic or metabolic disorders; and (5) BMI ≤ 35. Samples were aliquoted immediately after collection and stored at −80 °C prior to analysis.
2.9. Urine derivatization
To each 0.5 mL aliquot of human urine, 5 ng of Deuterium- and 13C-labeled internal standards were added. Modified from previous methods [24,27], derivatized estrogens and EMs were hydrolyzed overnight at 37 °C using β-glucuronidase/sulfatase from Helix pomatia (1000 U) in 0.15 M sodium acetate buffer (pH 4.6) containing 2.5 mg L-ascorbic acid to deglucuronidate and desulfate the endogenous estrogens and EMs in the urine. Estrogens and EMs were extracted by adding the hydrolyzed sample to a solid support liquid extraction cartridge (Chem Elut, Agilent Technologies, Part number 12198003, 3 mL, unbuffered) and eluted with 3 mL of dichloromethane three times for a total of 9 mL eluent. Samples containing extracted estrogens and EMs were dried under a stream of nitrogen and resuspended in 75 µL of aqueous 100 mM sodium bicarbonate (pH 9.0) prior to the addition of 75 µL of 3 mg/mL of 1-methylimidazole-2-sulfonyl chloride in anhydrous acetone. The mixture was capped and incubated at 60 °C for 15 min. Derivatized samples were dried under a stream of nitrogen and re-suspended in 50 µL of loading solvent (35% acetonitrile containing 0.1% formic acid) for LC-MS analysis.
2.10. Liquid chromatography-mass spectrometry
Liquid Chromatography-High Resolution Mass Spectrometry (LCHRMS) was performed using a UHPLC (Vanquish, Thermo Scientific) interfaced with a Q Exactive Focus hybrid quadrupole-Orbitrap mass spectrometer (Thermo Scientific, San Jose, CA). A total of 25 µL of each standard or sample was injected onto the column. Chromatographic separation was performed on an ACE C18-PFP column, (2.1 mm ID × 100 mm length with 2 µm particle size) maintained at 40 °C using the following solvent system: Solvent A was composed of HPLC grade water containing 0.1% formic acid (FA); Solvent B was acetonitrile containing 0.1% FA. A linear gradient was programmed from 35 to 45% B over 30 min with a flow rate of 0.250 mL/min, then ramped from 45% B to 90% B from 30 to 35 min at a flow rate of 0.500 mL/min to wash the column followed by re-equilibration over 5 min at a flow rate of 0.500 mL/min, for a total of 45 min for each experiment. A “saw-tooth” wash was used between samples to remove the maximum amount of impurities from the column. The “saw-tooth” gradient started with 35% solvent B and held for 3 min, then ramped to 90% solvent B over 5 min, held at 90% solvent B for 3 min, then reverted back to 35% solvent B over 5 min, hold at 35% solvent B for 3 min, then ramped to 90% solvent B over 5 min, held at 90% solvent B for 3 min, then back to 35% solvent B over 3 min to equilibrate for 10 min at a flow rate of 250 µL/min for a total run time of 40 min.
The Q Exactive Focus mass spectrometer was equipped with a HESI probe and operated using these parameters for ionization in positive mode: sheath gas 50, auxiliary gas 10, sweep gas 1, spray voltage 3.5 kV, 70,000 resolution, 100 ms maximum injection time, and scan range from m/z 350 to m/z 620.
2.11. Creatinine normalization
To account for variations in renal function, each EM concentration (pg/mL) was normalized using creatinine [31] and expressed as pg/mg of creatinine. The creatinine assay was performed in the Clinical Laboratory at Moffitt Cancer Center on a Roche Cobas c501 analyzer (Roche Diagnostics, Indianapolis, IN).
2.12. Data analysis for quantitation of estrogen metabolites
LC-MS peak heights were determined using Skyline (Version 20.1.0.76) [28] and the peak height ratio of the estrogen to its internal standard was used to calculate injected EM amounts for sample comparison. Based on the similarities of chemical structure and retention time, D2-16-epiE3 was used as the internal standard for quantifying 17-epiE3, because a matched stable isotope-labeled standard was not available when we began the study. Estrogens and EMs concentrations were determined by factoring in the peak height ratio to the amount of stable isotope standard that was spiked into each sample. Peak height was chosen because it performed more reproducibly than peak area.
2.13. Bioinformatics
Cluster 3.0 (http://bonsai.hgc.jp/~mdehoon/software/cluster/software.htm) was used to log transform, center, and normalize EM levels prior to unsupervised hierarchical clustering of EMs and samples via average linkage. Data were visualized in a heat map with Java Treeview (jtreeview.sourceforge.net) and exported as postscript files. Postscript files were converted to PDF using GSview 5.0, imported into Inkscape, and transferred into PowerPoint for figure preparation. Violin plots were created in GraphPad Prism 8 and manually overlaid with analyte missingness and LLOQ values The metabolite data were log2 transformed, and principal components analysis (PCA) was performed using the non-linear iterative partial least-squares (NIPALS) approach [32] implemented in R (nipals package, v. 0.7). One metabolite (4-MeOE2) was removed due to high levels of missingness (not detected in 161 of 169 samples). Fourteen samples were removed for having a high number of missing metabolite measurements (6 or more of the 14 remaining metabolites) and/or based on visualization as an outlier in the PCA plot. Then, the remaining log2 transformed metabolite data were examined for co-expression of metabolites using Pearson’s correlation. Metabolite correlations were grouped by pathway and visualized as a heatmap using ComplexHeatmap [33] in R/Bioconductor. Strong positive correlations were assessed using a cutoff value of 0.7 from the Pearson correlation.
3. Results and discussion
3.1. Deglucuronidation efficiency and recovery of estrogen metabolites after hydrolysis
Endogenous estrogens and their metabolites are present in urine primarily as glucuronide and sulfate conjugates; therefore, an initial enzymatic hydrolysis step with β-glucuronidase/sulfatase from Helix pomatia was used to enable measurement of total EM levels. Using resorufin-glucuronide as an external standard, our results show that optimal hydrolysis time was achieved as early as 15 min into incubation with no significant degradation up to 18 h using 1000 units of enzyme (Supplemental Fig. 1). For convenience when processing large batches of samples, we hydrolyzed samples overnight for 18 h. To determine the potential impact of incomplete deglucuronidation, we quantified total EMs (with deglucuronidation) and free EMs (without deglucuronidation) in a subset of urine samples (Supplemental Table 1). Comparison of the ion signals of free EMs to total EMs ranged from 0 to 11% of the signal was recovered without enzymatic hydrolysis, indicating deglucuronidation and desulfation is needed to quantify the total levels of the EMs effectively.
3.2. Extracted ion chromatograms for estrogen metabolites show chromatographic separation for isobaric peaks
Separation of a mixture containing 15 derivatized estrogens and EMs was performed on C18-PFP (MAC-MOD, Chadds Ford, PA), Phenyl-X (Thermo Fisher Scientific, Waltham, MA) and biphenyl C18 reversed phase columns (Thermo Fisher Scientific, Waltham, MA). In general, the estrogens and EMs showed a similar elution pattern with small changes in retention times for some EMs, because of the differences between these stationary phases. The C18-PFP column showed baseline separation for all 15 estrogens and EMs analyzed using a 30 min gradient (Fig. 1A); the other columns resolved most compounds, but could not achieve baseline separation for 16-ketoE2 and 16α-OHE1 (Supplemental Fig. 2A and B). However, extension of the gradient and/or changes to the mobile phases may improve separation on the phenyl-X and biphenyl C18 columns. Due to the large sample number for this study and future work, we decided not to further optimize separations on those columns and focus on the C18-PFP column. Full scan electrospray MS in positive ion mode was used to identify and quantify MIS-derivatized estrogens and EMs as well as their corresponding stable isotope-labeled standards using peak heights recorded for ion signals at their exact m/z (Table 1). The protonated molecules or [M+H]+ for the derivatized estrogens and EMs with a single MIS tag (Fig. 1B and C) and two MIS tags were measured (Fig. 1D), depending on the number of aromatic hydroxyl groups that could be modified. The example of MIS derivatized 13C6-2-Hydroxyestradiol (Fig. 1C and D) shows that both the single and double derivatization can be detected with mass errors less than 5 ppm; however, MIS is used in excess as in these experiments, so derivatization of both hydroxyl groups is consistently dominant (~98% of ion signal), as seen with higher peak intensity in Fig. 1D relative to Fig. 1C. Hence, the double tagged metabolite is used for quantitation for 2-OHE1, 2-OHE2 and 4-OHE1, where 2 hydroxyl groups are expected to be labeled. The labeling pattern of the endogenous EMs and SIS was highly consistent in replicates and across samples, so quantification can be performed using only the ion signals of the doubly derivatized molecules.
Fig. 1.
Extracted Ion Chromatograms and Example Mass Spectra of Derivatized EMs. Example extracted ion chromatograms show separation and detection of a mixture of 15 MIS derivatized estrogens and estrogen metabolites using an ACE C18-PFP column (A). Each trace shows ion signal at a given mass-to-charge ratio (m/z), which can include multiple metabolites. For metabolites with the same m/z value, LC separation is achieved with full baseline resolution. Data are shown for: (i) Estriol, 16-Epiestriol and 17-Epiestriol; (ii) 16-Ketoestradiol and 16α-Hydroxyestrone; (iii) 2-Hydroxyestradiol; (iv) 4-Hydroxyestrone and 2-Hydroxyestrone; (v) 4-Methoxyestradiol and 2-Methoxyestradiol; (vi) 17β-Estradiol; (vii) 4-methoxyestrone, 3-methoxyestrone and 2-Methoxyestrone; and (viii) Estrone. Example mass spectra are shown for MIS-derivatized 13C3-Estrone (B), as well as singly (C) and doubly derivatized (D) 13C6-2-Hydroxyestradiol. The normalization level (NL) lists the intensity of the base peak for each spectrum.
Because these initial tests were performed in neat solvent, the estrogens and EMs were analyze in charcoal-stripped human urine (CSHU) to rule out changes in retention time (RT) due to the matrix. The retention times of the extracted EMs from charcoal-stripped human urine showed linear correlation to the RTs for EMs in solvent (Supplemental Fig. 3). Furthermore, the retention times for the 15 estrogens and EMs were also consistent throughout the analysis of the urine samples collected from this large study population (data not shown). Therefore, the chromatographic retention times along with a high resolution instrument capable of accurate mass measurement should allow for reliable detection and quantification of estrogens and EMs.
We also analyzed underivatized urine to determine whether there are any abundant small molecules in urine that could contribute interference. The underivatized human urine showed no discernable background interference from metabolites at the specified retention time and m/z of the derivatized estrogen metabolites (data not shown). We also analyzed derivatized urine to evaluate interference of other endogenous molecules after derivatization. The derivatized human urine did not show any notable interference from other metabolites at the specified retention time and m/z of most of the derivatized estrogen metabolites; the exceptions were 2-OHE1, 4-OHE1 and 2-OHE2 (Supplemental Table 2). Split peaks were observed for 2-OHE2 and D5-16-ketoE2 (Supplemental Fig. 4). However, using a mass tolerance of 5 ppm isolates 2-OHE2 from the co-eluting analyte, allowing 2-OHE2 to be effectively quantified. The split peak observed with D5-16-ketoE2 corresponded to the M + 1 isotope of D2-16-epiE3 based on similar retention time and a mass error of ≤2 ppm. Peak height was used for quantitation to avoid this interference; therefore, the observed split peak did not have a high impact 16-ketoE2 quantitation. Replacement of these SIS with other standards or increasing the length of the gradient could also eliminate this issue.
To ensure quantitative accuracy of the LC-MS method, sample carryover was evaluated to confirm there was no extraneous signal contributing to the EMs peak height. We tested carryover by injecting solvent blanks between each urine sample from a pre-menopausal female on BC, whose EMs are at the upper limit of quantitation. Chromatographic profiles of the two highest EMs (Supplemental Fig. 5A and B) show high intensity signals in the sample and no detected signal in the blanks. Bar graphs (Supplemental Fig. 5C) of peak heights for each endogenous EM detected in the urine of the pre-menopausal female on BC compared to solvent blank injections showed no carryover of endogenous EMs and less than 1% of carryover was observed for SIS EMs.
3.3. Matrix effects of estrogen metabolites
Urine is a complex matrix, which may contain substances that can interfere with assay measurements. To determine the extent of matrix interference when detecting EMs, we compared the peak height of EMs spiked at 250, 1000 and 6400 pg/mL into charcoal stripped human urine (CSHU) and in neat solvent (S). Matrix effects were observed based on the signal in urine as compared with LC Solvent A, which was expressed as a percentage (Table 2). For most analytes, significant matrix effects (> 15% of signal) were not observed. 17-epiE3 showed lower signal than expected in urine samples, indicating signal suppression, but this EM was also the only one without a matched SIS for comparison. Interference was also very low, as few values for EM signal in CSHU exceeded 100% of the values observed for EMs in solvent.
Table 2.
Matrix Effects on Detection of Estrogen Metabolites. We compared the peak height of EMs spiked at given concentrations (250, 1000 or 6400 pg/mL) into charcoal stripped human urine (CSHU) and in neat LC solvent A (S). Matrix effects were observed based on the recovery of signal (%) in urine as compared with LC Solvent A. For most analytes, significant matrix effects (> 20%) were not observed. Estrogen metabolite 17-epiE3 showed lower signal in urine samples, indicating significant signal suppression. Interference was also very low, as few values for EM signal in CSHU exceeded 100% of the values observed for EMs in solvent.
Metabolite | Average Recovery in CSHU vs. Neat Solvent (%) |
||
---|---|---|---|
250 pg/mL | 1,000 pg/mL | 6,400 pg/mL | |
E3 | 86 | 93 | 92 |
E1 | 87 | 94 | 92 |
2-OHE1 | 89 | 94 | 93 |
16-ketoE2 | 98 | 73 | 102 |
4-OHE1 | 100 | 113 | 122 |
16α-OHE1 | 94 | 98 | 92 |
2-MeOE1 | 102 | 109 | 98 |
4-MeOE2 | 86 | 93 | 91 |
E2 | 89 | 94 | 93 |
16-epiE3 | 93 | 91 | 95 |
17-epiE3* | 47 | 51 | 61 |
4-MeOE1 | 87 | 94 | 94 |
2-OHE2 | 108 | 110 | 97 |
3-MeOE1 | 88 | 93 | 91 |
2-MeOE2 | 89 | 93 | 93 |
No standard was available for 17-epiE3, so the closest analogue, D2-16-epiE3, was used.
3.4. Estrogens and EMs calibration curve and limit of quantitation
The calibration curve in charcoal-stripped human urine for most estrogens and estrogen metabolites was linear with an R2 ≥ 0.99 from 4 pg/mL to 6,400 pg/mL (Supplemental Fig. 6), so 4 pg/mL was chosen as the LLOQ (Supplemental Table 3). The calibration curves of 2 EMs showed linear signal down to 1 pg/mL: 3-MeOE1 and 4-MeOE1. Interference in CSHU was noted for 2-OHE2, which limited the LLOQ. Examination of the data from the pilot study indicated that the interference in the CSHU matrix was not observed in the collected urine samples. Therefore, 2 additional calibration curves were performed (Supplemental Fig. 7); the first used a pooled urine sample, while the second was a reverse calibration curve in CSHU. Both of these experiments led to 20 pg/mL values for the LLOQ of 2-OHE2. However, 2-OHE2 was detected and quantified in 102 samples out of 169 (60%) with values ranging from 41 to 1567 pg/mL, so this LLOQ value was not limiting in this experiment.
3.5. Estrogens and EMs recovery and reproducibility
We measured the percent recovery of known amounts of estrogens and EMs added to charcoal-stripped human urine at 250, 1000, and 6400 pg/mL, which includes values in the expected range of levels present in human urine as well as higher concentrations to serve as a positive control that produced strong ion signals [24]. Each sample was hydrolyzed, extracted and derivatized with MIS before UHPLC-HRMS analysis. Supplemental Table 4 shows the percent recovery of the known added amounts of estrogens and EMs which ranged from 65 to 139% for 250 pg/ml, 63–108% for 1000 pg/mL, and 67–132% for 6400 pg/mL. Among the metabolites, 4-OHE1 had a lower recovery at all concentrations tested and 2-OHE2 had a notably higher recovery at 250 pg/mL. The inter-batch accuracy values for a representative parent estrogen and selected EMs from the 2-, 4-, and 16-pathway spiked in charcoal-stripped human urine at 250, 1000 and 6,400 pg/mL are shown in Supplemental Fig. 8. The intra-batch reproducibility using process replicates of charcoal-stripped human urine spiked with 250, 1,000 and 6400 pg/mL of estrogen and EMs and analyzed over three days is shown in Table 3. The inter-batch precision of estrogens and EMs was determined using analysis of three independent batches, shown in Table 3. The intra- and inter-batch CVs were ≤ 20% for all analytes, except 17-epiE3 (which lacked the corresponding SIS). We also evaluated the ion signal stability of the estrogens and EMs stable isotope-labeled standards spiked in human urine as shown in Supplemental Fig. 9. Ion signals with mass error greater than 5 ppm and outliers, defined as having SIS ion signals greater than two standard deviations away from the mean, were removed from the data set. There were a small number of outliers detected for most of the estrogens and EMs, which were often observed in the same samples. In those cases, poor sample quality was likely the cause, so the measurements were discarded. EM loss could have occurred in those samples due to the problems with collection, sample processing, or storage conditions prior to receiving the samples.
Table 3. Inter- and Intra-batch Reproducibility of Full Process Replicates for LC-MS Quantification of Estrogen Metabolites.
On each of 3 days, three replicate samples of charcoal stripped human urine (CSHU) were spiked with 250, 1000 and 6400 pg/mL estrogen metabolites. Variability in LC-MS measurements are reported as CV (%). CV values should not exceed 20%; if it does then the source of the variation should be identified [30].
Metabolite | Inter-Batch CV (%) | Day 1 Intra-Batch CV (%) | Day 2 Intra-Batch CV (%) | Day 3 Intra-Batch CV (%) | ||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
250 pg/mL | 1000 pg/mL | 6400 pg/mL | 250 pg/mL | 1000 pg/mL | 6400 pg/mL | 250 pg/mL | 1000 pg/mL | 6400 pg/mL | 250 pg/mL | 1000 pg/mL | 6400 pg/mL | |
E3 | 13 | 10.1 | 12.5 | 1.4 | 1.2 | 3.5 | 5.8 | 7.2 | 3 | 4.2 | 2.2 | 7.6 |
E1 | 9.5 | 7.5 | 9.7 | 0.4 | 3.1 | 3.9 | 4.2 | 4.6 | 4.1 | 7.6 | 3 | 7.3 |
2-OHE1 | 7.1 | 5.5 | 6.9 | 3.1 | 4.6 | 3.3 | 3.6 | 7.7 | 4.9 | 4.9 | 0.9 | 7.8 |
16-ketoE2 | 5.3 | 5 | 7.3 | 0.7 | 1 | 4.8 | 6.6 | 6.5 | 3.4 | 5.7 | 6.2 | 6.2 |
4-OHE1 | 4.4 | 4.5 | 7.1 | 3.1 | 3 | 3.9 | 3.5 | 6.4 | 5 | 5.6 | 3.2 | 8 |
16α-OHE1 | 11 | 8.1 | 10.6 | 1.3 | 1.7 | 4.3 | 4.9 | 5.3 | 4.7 | 5.3 | 2.3 | 7 |
2-MeOE1 | 15.9 | 5.7 | 6.8 | 16.1 | 8.5 | 5.2 | 4.6 | 6.3 | 3.1 | 5.3 | 1.6 | 7.1 |
4-MeOE2 | 5.6 | 3.5 | 6.8 | 1.2 | 2.3 | 4.7 | 4.3 | 4.7 | 2.2 | 9.3 | 2.1 | 7.7 |
E2 | 13.6 | 10.5 | 10.9 | 1.7 | 1.7 | 4.5 | 6.4 | 6.8 | 3.6 | 4.6 | 2.5 | 8.4 |
16-epiE3 | 5.9 | 16.1 | 15.4 | 2.3 | 6 | 3.9 | 4.6 | 11.5 | 7.8 | 4.4 | 4.7 | 7.9 |
17-epiE3 | 16.1 | 20.1 | 15.6 | 4.8 | 6.4 | 6.3 | 6.8 | 12 | 8.5 | 14.9 | 20.1 | 4.5 |
4-MeOE1 | 12.7 | 11.5 | 11.4 | 0.7 | 1.9 | 4.5 | 4.3 | 4.4 | 2.8 | 5.1 | 2.7 | 5.8 |
2-OHE2 | 11.9 | 10.7 | 9.9 | 2.4 | 2.1 | 3.7 | 6.2 | 7 | 3 | 7.7 | 6.8 | 8.8 |
3-MeOE1 | 10.5 | 8.6 | 9.3 | 0.9 | 6.1 | 3.8 | 5 | 6 | 2.9 | 4 | 5.1 | 7.3 |
2-MeOE2 | 9.8 | 7.4 | 8.4 | 1 | 2.8 | 3.7 | 5.7 | 5.2 | 3.3 | 5.7 | 1.2 | 5.6 |
No standard was available for 17-epiE3, used D2-16-epiE3.
Urine samples (n = 13) were repeated as biological replicates to compare the reproducibility of EM quantification. In general, a strong linear correlation was observed for EMs between the first processing replicate and the second processing replicate as shown in Supplemental Fig. 10. Data missingness was consistent between replicates, and values are only plotted for EMs that were consistently detected.
3.6. Assay evaluation for EM quantitation
Urinary endogenous estrogens and EMs were measured in urine samples from two healthy males, three premenopausal females and one perimenopausal female to evaluate assay performance over a range of analyte concentrations. In Supplemental Fig. 11, we show the mean and standard deviation of 3 process replicates. A pregnant female showed high levels of all 15 estrogens measured, including E2, which is synthesized by the placenta in very large amounts during pregnancy [34], and E3, which is the main pregnancy hormone [35]. In general, our results show that males excrete lower amounts of estrogens and EMs than premenopausal women, consistent both with expectation and the previous literature [8]. As expected, there were significantly higher levels of urinary estrogen in a premenopausal female using an oral contraceptive (OC) compared with a premenopausal female not using an OC. The estrogens and EMs excreted in the highest amounts by the premenopausal female using an OC was estrone, 2-hydroxyestrone and estriol. The elevated amount of estrogens and EMs excreted in a female using OCs relates to the synthetic estrogen in the contraceptive pill and its metabolism [36]. Estrone and estriol were the most abundant estrogens excreted by the premenopausal female not using OCs. These results are consistent with a previous study of urinary estrogen levels in premenopausal women [37] and showed the initial proof-of-principle needed to move the assay forward to analyze the volunteer cohort.
3.7. Urinary EM quantitation in postmenopausal women
Using this optimized and characterized UHPLC-HRMS approach, we were able to quantify 15 estrogens and EMs in urine collected from 169 healthy postmenopausal women (Supplemental Table 5). These samples were a part of a larger sample set of 179 with losses due to poor performance in the assay, insufficient sample volume, or removal due to not meeting the volunteer eligibility criteria in the protocol. Urinary EM concentrations (calculated as pg/mg of creatinine) for 169 healthy postmenopausal women are shown in box plots in Fig. 2A. The plot shows that the inter-individual differences for each estrogen and EM were large, where the highest EM concentration exceeded by 20 times or more the lowest concentration for all the estrogens and EMs. The differences observed between the highest and lowest concentration for 3-MeOE1, 4-MeOE1 and 17-epiE3 were more than 200-fold and approximately 100-fold differences were observed for E1, 2-OHE1, 2-OHE2, 2-MeOE1 and 16-ketoE2. The differences were less than 100-fold for E2, 2-MeOE2, 4-OHE1, 16α-OHE1, E3 and 16-epiE3. To further explore differences between the study participants, we also examined EM expression levels normalized to the total and the relative contributions of different EM modification pathways. The two parent estrogens and 13 estrogen metabolites concentrations were expressed as a percentage of total EMs in each of the metabolic pathways (Fig. 2B). Parent estrogens, 2-pathway, 4-pathway and 16-pathway contain 20%, 20%, 12% and 48% of total EMs, respectively. Data are comparable to EM levels found among postmenopausal women in a previous study [16], with the exception of 16-ketoestradiol. Variation in the amount of 16-ketoestradiol measured may be attributed to the difference in internal standard used to quantify this EM. In the current study, we used D5-16-ketoE2 to quantify 16-ketoE2, whereas in the previous study D3-E3 was used as a surrogate SIS to quantify 16-ketoE2. As such, we expect these results to be more accurate based on the matched SIS.
Fig. 2.
Urinary EMs Measured in 169 Postmenopausal Women. Single 0.5 mL aliquots of urine sample were processed and analyzed with UHPLC-HRMS, as described in the methods, to quantify urinary EMs measured in postmenopausal women. Data are shown as mean and standard deviation of 169 excreted by postmenopausal women (A). The average endogenous estrogens and estrogen metabolites excreted in urine of postmenopausal women were normalized with creatinine and grouped by pathway; data are presented as a percentage of total estrogen in each of the metabolic pathways by taking the sum of estrogen metabolites in each respective pathway divided by the total amount in the sample and each individual EM divided by the total (B). The * indicates that 17-epiE3 was quantitated using D2-16epi-E3, because it is the closest structural analog and also had a similar retention time. Scatter plot of total levels of excreted EMs in 169 postmenopausal women (C). Heat map of urinary endogenous estrogens and estrogen metabolites excreted in 169 postmenopausal women (D). Highest EM levels are shown in red, lowest in blue, and gray indicates not detected. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
Total estrogen levels in the 169 postmenopausal women ranged from 1838 pg/mg of creatinine to 67,476 pg/mg of creatinine (Fig. 2C). Substantial variation was also observed for different individual estrogens and EMs among the postmenopausal women. As shown in a heat map visualizing data normalized across each EM to show the samples with the highest and lowest levels of each metabolite (Fig. 2D). As expected, the parent EMs clustered together in the dendrogram, as did the 16-pathway metabolites. Some 2-pathway metabolites grouped together, while others clustered with the 4-pathway metabolites. In the dendrogram of volunteers’ samples, two main groups were noted corresponding to consistently low levels of EMs and consistently high levels of EMS, while smaller groups were observed for elevated EMs levels in each pathway. The data were also visualized as a heat map normalized by sample to show the EMs highest and lowest levels in the urine of each woman (Supplemental Fig. 12); expression levels of E1, E3 and 16-ketoE2 were generally higher compared to other estrogens and EMs in these women, which was consistent with a previous study [8]. In addition, other comparisons can also be made with these data, e.g. the lower abundance of 2-hydroxylation pathway EMs when compared to 16-hydroxylation pathway EMs in most volunteer). The lowest frequency of detection for an EM, 4-MeOE2, was 5% of the study population; levels of that EM may be below the LLOQ for the assay in most of these postmenopausal women. The low amount of 4-MeOE2 excreted in postmenopausal women urine was also consistent with published studies [8], though the level of missingness was not reported previously.
3.8. Quality control using bioinformatics approaches and comparison with known biology
To further assess the results in the dataset, several metrics were calculated. First, the numbers of missing values for each sample and each analyte were calculated (see Fig. 3A and B as well as Supplemental Table 5). In addition, the average value, minimum value, and maximum value for each EM detected in urine were also calculated (ibid.). Missingness in the samples appeared to be primarily linked to the total level of estrogens and EMs, though some exceptions were noted (e.g. Sample 150). At the analyte level, missingness was primarily correlated with average observed EM level in the dataset, except for the 4-pathway metabolites, 4-OHE1 and 4-MeOE2, which are observed at relatively high levels when present and yet missing in a large number of samples. These data could indicate differences in biology between post-menopausal women and merit further investigation.
Fig. 3.
Investigation of Missing Values for EM Measurements in the Dataset. To determine whether missingness was driven by low signal or relevant to biology, values for the total EM level in pg/mg creatinine were plotted for each number of missing values in the sample (A). Higher missingness is associated with lower total EM levels, but other factors are also involved (e.g. Sample 150 has high levels of detected metabolites, but 9 missing values). Next, the number of missing values for each EM is plotted against the average level of that EM, when detected (B). Some metabolites detected at high levels in a subset of samples still have high levels of missingness in the dataset (e.g. 4-MeOE2 and 4-OHE1). To provide further detail, a violin plot of the EM measurements was prepared to compare detected levels, missing values (provided at the top of the graph), and LLOQ values (green bar) for each metabolite (C). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
Violin plots (Fig. 3C) were then used to examine the distribution of measured values for each metabolite; the numbers of missing values and the LLOQ values have been overlaid on the plot. The data reflect different types of missingness. Values for high abundance EMs (e.g. E3 and E1) were likely to be missing at random. Values for lower abundance analytes, where the violin plot nears the LLOQ value (e.g. 2OHE2), may be too low for detection or missing below the LLOQ. Using the assumption that these EM levels are expected to be a continuous variable, high missingness of higher abundance analytes, like 4-OHE1 and 4-MeOE2, is consistent with the hypothesis that some postmenopausal women may not excrete these EMs in their urine. If the EMs were present and not detected because they were below the LLOQ, then the levels were much lower than those recorded in the dataset and these EM measurements should be treated as a discontinuous variable.
In order to better understand relationships between the EMs, the first step was additional quality control to develop filters for the data. Principal component analysis with the NIPALS algorithm using the log transformed data was used to examine the dataset (Fig. 4A). Overall, the data clustered together, but a few outliers were identified: Samples 11, 17, 92, 138, 144, and 150. Several samples had already been flagged for removal from further analysis due to high missingness: 11, 70, 76, 98, 111, 119, 144, 149, 150, 164, and 167. Therefore, 14 samples in total, which are labeled on the PCA plot were removed (Fig. 4A). Because only 8 values could be measured for 4-MeOE2, that analyte was also excluded from further analysis. Next, co-expression analysis was performed and visualized in a heat map (Fig. 4B). As expected, the parent estrogens are highly related and metabolites within the pathways show higher levels of co-expression than EMs in different pathways. In particular, the group of 16-pathway metabolites had the highest values from Pearson correlation. One relationship between metabolites was not related to the estrogen metabolism pathways; 4-MeOE1 in the 4-pathway and 3-MeOE1 in the 2-pathway had a correlation value of 0.75, indicating some potential relationship between these metabolites in the dataset. Correlation values between these two metabolites and other methoxylated EMs had much lower values between 0.30 and 0.44, so this relationship is not solely defined by the activity levels of catechol-O-methyltransferases [6]. Using the pathway diagram redrawn from [38], correlation values were overlaid for each of the nearest neighbors to illustrate that the dataset is consistent with the expected biological inter-relationships between EMs (Fig. 4C). These additional strategies for quality control enable deeper insight into the dataset and confirm that the experimental data are consistent with known biology.
Fig. 4.
Correlation of Estrogen Metabolite Levels across the Volunteer Cohort. Samples were compared using principal component analysis (A); outliers and samples with high missingness are labeled in the plot. After removal of these samples and data for 4-MeOE2 (also due to high number of missing values), the correlation between metabolites was quantified and plotted as a heat map where the metabolites are grouped by pathway (B). Parent estrogens and 16-pathway metabolites show the highest levels of correlation. Values for metabolite correlations are added to the estrogen metabolism diagram, redrawn from Falk et al. [38], to show the relationships between nearest neighbors (C). Two values are shown for the correlations between E1 (top) and E2 (bottom) and the first metabolites in the 2-, 4-, and 16-pathways; NC indicates not calculated due to the high number of missing values for 4-MeOE2.
4. Conclusions
A LC-HRMS method was optimized using an ACE-PFP column to separate all 15 estrogens and EMs using a reduced total analysis time of 45 min, which allows for analysis of large numbers of urine samples at the scale required in epidemiologic studies. This method also provides HRMS quantification using the peak at the m/z of the MIS-derivatized estrogen rather than fragments generated from the derivatizing reagent. The derivatization approach and LC-HRMS method was validated using the chromatographic retention times and exact mass of the estrogens and EMs by assessing carryover, matrix effects, accuracy, reproducibility, and internal standard stability. An LLOQ between 1 and 20 pg/mL was achieved for the EMs, which are below the previously reported values of 40 pg/mL for EMs in human urine.[24] The assay was initially tested and refined using male and pre- and peri-menopausal female urine samples, where the results were found comparable to previously reported data. The method was then applied to the quantification of 2 parent estrogens and 13 EMs in 169 urine samples from postmenopausal women and produced quantitative results for these analytes comparable to those in previous studies. These analytical advances were paired with a bioinformatics strategy for data analysis that was informed by known biology. Given the established increase in risk of breast cancer associated with high levels of estrogens and lower EM ratios, monitoring of circulating estrogen using high throughput approaches such as the one presented here could have a place in targeted breast cancer surveillance and studies of novel risk reduction interventions.
Supplementary Material
Acknowledgements
This research was supported by pilot funding from the Florida Academic Cancer Center Alliance (FACCA) and the UF Health Cancer Center Bridge Funding. This work has been supported in part by the Proteomics and Metabolomics Core as well as the Tissue Core at the H. Lee Moffitt Cancer Center & Research Institute; an NCI designated Comprehensive Cancer Center (P30-CA076292). The authors would like to thank Martin Abrams at Moffitt Cancer Center Clinical Laboratory for the creatinine measurements and Volker Mai for participation in scientific discussions.
5. Financial support
This research was supported by pilot funding from the Florida Academic Cancer Center Alliance (FACCA to KME and LY), UF Health Cancer Center Bridge Funding (to LY) and Moffitt’s Cancer Center Support Grant (NCI P30-CA076292).
Footnotes
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Appendix A. Supplementary material
Supplementary data to this article can be found online at https://doi.org/10.1016/j.jchromb.2020.122288.
References
- [1].Henderson BE, Ross R, Bernstein L, Estrogens as a cause of human cancer: the Richard and Hinda Rosenthal Foundation award lecture, Cancer Res 48 (1988) 246–253. [PubMed] [Google Scholar]
- [2].Yager JD, Davidson NE, Estrogen carcinogenesis in breast cancer, N. Engl. J. Med 354 (2006) 270–282. [DOI] [PubMed] [Google Scholar]
- [3].Yager JD, Endogenous estrogens as carcinogens through metabolic activation, J. Natl. Cancer Inst. Monogr (2000) 67–73. [DOI] [PubMed]
- [4].Yue W, Yager JD, Wang JP, Jupe ER, Santen RJ, Estrogen receptor-dependent and independent mechanisms of breast cancer carcinogenesis, Steroids 78 (2013) 161–170. [DOI] [PubMed] [Google Scholar]
- [5].Ziegler RG, Fuhrman BJ, Moore SC, Matthews CE, Epidemiologic studies of estrogen metabolism and breast cancer, Steroids 99 (2015) 67–75. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [6].Samavat H, Kurzer MS, Estrogen metabolism and breast cancer, Cancer Lett 356 (2015) 231–243. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [7].Bradlow HL, Telang NT, Sepkovic DW, Osborne MP, 2-hydroxyestrone: the ‘good’ estrogen, J. Endocrinol 150 (Suppl) (1996) S259–265. [PubMed] [Google Scholar]
- [8].Xu X, Veenstra TD, Fox SD, Roman JM, Issaq HJ, Falk R, Saavedra JE, Keefer LK, Ziegler RG, Measuring fifteen endogenous estrogens simultaneously in human urine by high-performance liquid chromatography-mass spectrometry, Anal. Chem 77 (2005) 6646–6654. [DOI] [PubMed] [Google Scholar]
- [9].Dallal C, Taioli E, Urinary 2/16 estrogen metabolite ratio levels in healthy women: a review of the literature, Mutat. Res 705 (2010) 154–162. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [10].Sepkovic DW, Bradlow HL, Estrogen hydroxylation–the good and the bad, Ann. N. Y. Acad. Sci 1155 (2009) 57–67. [DOI] [PubMed] [Google Scholar]
- [11].Kabat GC, Chang CJ, Sparano JA, Sepkovie DW, Hu XP, Khalil A, Rosenblatt R, Bradlow HL, Urinary estrogen metabolites and breast cancer: a case-control study, Cancer Epidemiol., Biomark. Prevent.: Publ. Am. Assoc. Cancer Res., Am. Soc. Prevent. Oncol 6 (1997) 505–509. [PubMed] [Google Scholar]
- [12].Meilahn EN, De Stavola B, Allen DS, Fentiman I, Bradlow HL, Sepkovic DW, Kuller LH, Do urinary oestrogen metabolites predict breast cancer? Guernsey III cohort follow-up, Brit. J. Cancer 78 (1998) 1250–1255. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [13].Sowers MR, Crawford S, McConnell DS, Randolph JF Jr., Gold EB, Wilkin MK, Lasley B, Selected diet and lifestyle factors are associated with estrogen metabolites in a multiracial/ethnic population of women, J. Nutrit 136 (2006) 1588–1595. [DOI] [PubMed] [Google Scholar]
- [14].Modugno F, Kip KE, Cochrane B, Kuller L, Klug TL, Rohan TE, Chlebowski RT, Lasser N, Stefanick ML, Obesity, hormone therapy, estrogen metabolism and risk of postmenopausal breast cancer, Int. J. Cancer 118 (2006) 1292–1301. [DOI] [PubMed] [Google Scholar]
- [15].Moore SC, Matthews CE, Ou-Shu X, Yu K, Gail MH, Xu X, Ji BT, Chow WH, Cai Q, Li H, Yang G, Ruggieri D, Boyd-Morin J, Rothman N, Hoover RN, Gao YT, Zheng W, Ziegler RG, Endogenous estrogens, estrogen metabolites, and breast cancer risk in postmenopausal Chinese women, J. Natl. Cancer Inst 108 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
- [16].Sampson JN, Falk RT, Schairer C, Moore SC, Fuhrman BJ, Dallal CM, Bauer DC, Dorgan JF, Shu XO, Zheng W, Brinton LA, Gail MH, Ziegler RG, Xu X, Hoover RN, Gierach GL, Association of estrogen metabolism with breast cancer risk in different cohorts of postmenopausal women, Cancer Res 77 (2017) 918–925. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [17].Smolders R, Koch HM, Moos RK, Cocker J, Jones K, Warren N, Levy L, Bevan R, Hays SM, Aylward LL, Inter- and intra-individual variation in urinary biomarker concentrations over a 6-day sampling period. Part 1: metals, Toxicol. Lett 231 (2014) 249–260. [DOI] [PubMed] [Google Scholar]
- [18].Faqehi AMM, Cobice DF, Naredo G, Mak TCS, Upreti R, Gibb FW, Beckett GJ, Walker BR, Homer NZM, Andrew R, Derivatization of estrogens enhances specificity and sensitivity of analysis of human plasma and serum by liquid chromatography tandem mass spectrometry, Talanta 151 (2016) 148–156. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [19].Maskarinec G, Beckford F, Morimoto Y, Franke AA, Stanczyk FZ, Association of estrogen measurements in serum and urine of premenopausal women, Biomarkers Med 9 (2015) 417–424. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [20].Lasley BL, Mobed K, Gold EB, The use of urinary hormonal assessments in human studies, Ann. N. Y. Acad. Sci 709 (1994) 299–311. [DOI] [PubMed] [Google Scholar]
- [21].Coburn SB, Stanczyk FZ, Falk RT, McGlynn KA, Brinton LA, Sampson J, Bradwin G, Xu X, Trabert B, Comparability of serum, plasma, and urinary estrogen and estrogen metabolite measurements by sex and menopausal status, Cancer Causes Control: CCC 30 (2019) 75–86. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [22].Denver N, Khan S, Homer NZM, MacLean MR, Andrew R, Current strategies for quantification of estrogens in clinical research, J. Steroid Biochem. Mol. Biol 192 (2019) 105373. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [23].Wang Q, Bottalico L, Mesaros C, Blair IA, Analysis of estrogens and androgens in postmenopausal serum and plasma by liquid chromatography-mass spectrometry, Steroids 99 (2015) 76–83. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [24].Xu X, Keefer LK, Ziegler RG, Veenstra TD, A liquid chromatography-mass spectrometry method for the quantitative analysis of urinary endogenous estrogen metabolites, Nat. Protoc 2 (2007) 1350–1355. [DOI] [PubMed] [Google Scholar]
- [25].Nelson RE, Grebe SK, DJ O.K., Singh RJ, Liquid chromatography-tandem mass spectrometry assay for simultaneous measurement of estradiol and estrone in human plasma, Clin. Chem 50 (2004) 373–384. [DOI] [PubMed] [Google Scholar]
- [26].Li X, Franke AA, Improved profiling of estrogen metabolites by orbitrap LC/MS, Steroids 99 (2015) 84–90. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [27].Xu X, Roman JM, Issaq HJ, Keefer LK, Veenstra TD, Ziegler RG, Quantitative measurement of endogenous estrogens and estrogen metabolites in human serum by liquid chromatography-tandem mass spectrometry, Anal. Chem 79 (2007) 7813–7821. [DOI] [PubMed] [Google Scholar]
- [28].MacLean B, Tomazela DM, Shulman N, Chambers M, Finney GL, Frewen B, Kern R, Tabb DL, Liebler DC, MacCoss MJ, Skyline: an open source document editor for creating and analyzing targeted proteomics experiments, Bioinformatics (Oxford, England) 26 (2010) 966–968. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [29].Matuszewski BK, Constanzer ML, Chavez-Eng CM, Strategies for the assessment of matrix effect in quantitative bioanalytical methods based on HPLC-MS/MS, Anal. Chem 75 (2003) 3019–3030. [DOI] [PubMed] [Google Scholar]
- [30].Reed GF, Lynn F, Meade BD, Use of coefficient of variation in assessing variability of quantitative assays, Clin. Diagn. Lab. Immunol 9 (2002) 1235–1239. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [31].Miller RC, Brindle E, Holman DJ, Shofer J, Klein NA, Soules MR, O’Connor KA, Comparison of specific gravity and creatinine for normalizing urinary reproductive hormone concentrations, Clin. Chem 50 (2004) 924–932. [DOI] [PubMed] [Google Scholar]
- [32].Wold H, Estimation of principal components and related models by iterative least squares, in: Krishnaiah PR (Ed.), Multivariate analysis, Academic Press, NY, 1966, pp. 391–420. [Google Scholar]
- [33].Gu Z, Eils R, Schlesner M, Complex heatmaps reveal patterns and correlations in multidimensional genomic data, Bioinformatics (Oxford, England) 32 (2016) 2847–2849. [DOI] [PubMed] [Google Scholar]
- [34].Berkane N, Liere P, Oudinet JP, Hertig A, Lefevre G, Pluchino N, Schumacher M, Chabbert-Buffet N, From pregnancy to preeclampsia: a key role for estrogens, Endocr. Rev 38 (2017) 123–144. [DOI] [PubMed] [Google Scholar]
- [35].Kovacs K, Vasarhelyi B, Gyarmati B, Karvaly G, Estrogen metabolism during pregnancy, Orv. Hetil 160 (2019) 1007–1014. [DOI] [PubMed] [Google Scholar]
- [36].Maduka IC, Ezeonu FE, Neboh EE, Shu EN, Ikekpeazu EJ, Urinary estrogen levels in women on contraceptives in enugu, South-East Nigeria, J. Family Med. Primary Care 1 (2012) 39–42. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [37].Eliassen AH, Spiegelman D, Xu X, Keefer LK, Veenstra TD, Barbieri RL, Willett WC, Hankinson SE, Ziegler RG, Urinary estrogens and estrogen metabolites and subsequent risk of breast cancer among premenopausal women, Cancer Res 72 (2012) 696–706. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [38].Falk RT, Brinton LA, Dorgan JF, Fuhrman BJ, Veenstra TD, Xu X, Gierach GL, Relationship of serum estrogens and estrogen metabolites to postmenopausal breast cancer risk: a nested case-control study, Breast cancer Res.: BCR 15 (2013) R34. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.