Analysis of Serum Metabolic Profiles in Women with Endometrial Cancer and Controls in a Population-Based Case-Control Study

Abstract

Context:

Endometrial cancer is associated with metabolic disturbances related to its underlying risk factors, including obesity and diabetes. Identifying metabolite biomarkers associated with endometrial cancer may have value for early detection, risk assessment, and understanding etiology.

Objective:

The objective of the study was to evaluate the reliable measurement of metabolites in epidemiological studies with nonstandardized blood collection; confirm previously reported correlations of metabolites with body size; and assess differences in metabolite levels between cases and controls.

Design:

This was the Polish Endometrial Cancer Study (2001–2003).

Setting:

This study was a population-based case-control study.

Patients:

Patients included 250 cases and 250 controls.

Intervention:

The intervention included the measurement of serum metabolite levels of 15 amino acids, 45 acylcarnitines, and nine fatty acids.

Main Outcome Measure:

The main outcome measure was endometrial cancer.

Results:

Body mass index was correlated with levels of valine (r = 0.26, P = 3.4 × 10⁻⁵), octenoylcarnitine (r = 0.24, P = 1.5 × 10⁻⁴), palmitic acid (r = 0.26, P = 4.4 × 10⁻⁵), oleic acid (r = 0.28, P = 9.9 × 10⁻⁶), and stearic acid (r = 0.26, P = 2.9 × 10⁻⁵) among controls. Only stearic acid was inversely associated with endometrial cancer case status (quartile 4 vs. quartile 1: odds ratio 0.37, 95% confidence interval 0.20–0.69, P for trend = 1.2 × 10⁻⁴). Levels of the C5-acylcarnitines, octenoylcarnitine, decatrienoylcarnitine, and linoleic acid were significantly lower in cases than controls (odds ratios ranged from 0.21 to 0.38).

Conclusions:

These data demonstrate that previously reported variations in metabolomic profiles with body mass index can be replicated in population-based studies with nonfasting blood collection protocols. We also provide preliminary evidence that large differences in metabolite levels exist between cases and controls, independent of body habitus. Our findings warrant assessment of metabolic profiles, including the candidate markers identified herein, in prospectively collected blood samples to define biomarkers and etiological factors related to endometrial cancer.

Endometrial cancer is the most common gynecologic malignancy worldwide (1). Several lines of research suggest that metabolic dysfunction is involved in endometrial carcinogenesis. Metabolic dysfunction is a common consequence of obesity, a strong risk factor for endometrial cancer (2). Obesity causes disruptions in multiple metabolic pathways, including elevated levels of sex steroid hormones, insulin, and inflammatory mediators and lower levels of adiponectin (3). However, similar to endometrial cancer, metabolic abnormalities are found only in a subset of obese women and many nonobese women show comparable abnormalities (4). Therefore, defining biomarkers of the underlying pathophysiological disturbances associated with endometrial cancer risk may have value for early detection or risk prediction. In support of this view, higher levels of insulin (5–7) and sex steroid hormones (8–10) and lower levels of adiponectin (11, 12) are associated with elevated endometrial cancer risk, independent of body mass index, in case-control and cohort studies.

Recently profiling of metabolic intermediates, including amino acids, acylcarnitines, and fatty acids, has demonstrated a linkage between a subset of analytes and several chronic conditions and diseases. In a study of 73 obese and 67 lean subjects, elevated levels of branched chain amino acids were associated with obesity and were related to insulin resistance in a subset of nonobese individuals (13, 14). More recently, studies in the Framingham and Malmo longitudinal cohorts demonstrated that elevated levels of branched chain and aromatic amino acids were predictive of development of type 2 diabetes (15), and these same analytes are rapidly decreased in response to bariatric surgery, the current best therapy for improvement of glucose homeostasis in obese, type 2 diabetic subjects (16). These studies demonstrate the potential utility of metabolic profiling to define biomarkers related to underlying endometrial cancer risk factors, such as obesity and diabetes. However, there is a paucity of data directly exploring relationships of metabolic profiles to endometrial cancer in population-based epidemiological studies.

We sought to evaluate whether the panel of metabolites previously examined in relation to metabolic dysfunctional states could be reliably measured in epidemiologic study with nonstandardized blood collection to confirm previously reported correlations of metabolites with body size in population-based controls and to assess differences in metabolite levels between endometrial cancer cases and controls. Specifically, we analyzed levels of 15 amino acids, 45 acylcarnitines, and nine fatty acids, as measured in prior studies (13, 14, 17), among 250 incident cases and 250 controls included in a population-based case-control study. Our goal in this proof-of-principle study was to explore metabolic signatures in endometrial cancer cases and controls that could be validated in prospective settings.

Materials and Methods

Study population

Our study subjects were derived from the Polish Endometrial Cancer Study, a large, population-based, endometrial case-control study (18) conducted among women residing in two Polish cities, Warsaw and Lodz, from 2001 to 2003. Eligible cases were diagnosed with incident pathologically confirmed endometrial cancer. Study personnel identified cases through a rapid identification system at participating hospitals that cover 85% of all cases diagnosed in Lodz and Warsaw and through the local cancer registries to ensure complete case ascertainment. Eligible controls were identified and enrolled concurrently through the Polish Electronic System, a database of all Polish residents. Controls were limited to women with an intact uterus who had not been diagnosed with endometrial cancer.

In total, 551 of 695 eligible cases (79.3%) and 1925 of 2843 eligible controls (67.7%) were interviewed about demographic factors, reproductive history, exogenous hormone use, lifetime body size and physical activity, and medical history. Study nurses measured the waist circumference of 1562 of the controls who overlapped with controls from the breast cancer portion of the study (19). Medical records of case participants were obtained and abstracted for diagnostic parameters.

Trained study nurses collected venous blood from 85% of participating cases and 93% of participating controls. The median time from case diagnosis to blood draw was 55 d. For this project, we selected postmenopausal women with adequate sample volumes that were processed within 8 h of blood collection. From a pool of 426 postmenopausal cases, we selected 250 endometrial cancer cases with the shortest time between blood draw and processing (range 0.8–4.8 h, median 3.2 h). A total of 211 eligible postmenopausal controls (84.4%) were selected matched to cases on site, fasting time before blood draw (overnight, <4 h, and 4+ hours), and processing time of blood sample (±1 h). An additional 32 postmenopausal controls were matched to cases based on fasting time before blood draw (overnight, >4 h, and 4+ hours), and processing time of blood sample (±1 h); the remaining seven cases were matched only on processing time (±1 h).

Institutional review board approval was obtained from all participating institutions, and all respondents gave signed informed consent.

Quality control samples

Replicate serum samples from two pools [a high body mass index (BMI) pool and a low BMI pool] created from 24 ineligible cases with the highest or lowest BMI (high BMI range 37.4–54.4 kg/m² or low BMI range 16.1–20.1 kg/m²) were tested for quality control purposes in a masked fashion.

Assays

Using stable isotope dilution of internal standards (Supplemental Figure 1, published on The Endocrine Society's Journals Online web site at http://jcem.endojournals.org), we quantified 15 amino acids and 45 acylcarnitines in serum samples using tandem mass spectrometry and nine total fatty acids using gas chromatography/mass spectrometry, as previous described (13, 14). Each batch contained 14 case-control pairs and three quality control samples (including at least one high and one low pool) in random order within the batch, masked to case-control status and quality control samples.

Data filtering

Some assays for specific acylcarnitines included in our panel are most applicable to identifying inherited metabolic defects and thus are expected to be undetectable in the general population. The limit of quantification was estimated to achieve a coefficient of variation (CV) of ±15% for amounts that are of the order of 5% of the internal standard spike for the assay format for plasma; this corresponds to 0.25 μm/liter for acetyl carnitine and 0.05 μm/liter for other carnitines. We excluded 20 acylcarnitines that had more than 90% of the values at a level below 0.05 μm (Supplemental Table 1). Interbatch reproducibility (i.e. overall CV; see Statistical methods for further details) were also evaluated; three acylcarnitines had CV greater than 15%. Lastly, three analytes with intraclass correlation coefficients (ICC) less than 80% (see Statistical methods for further details) were also excluded.

Statistical methods

For all analyses the metabolite measurements were log transformed.

The ICC and overall CV were estimated as follows. First, the variability among subjects (σ_a²) was estimated from the 250 controls. Based on the pooled quality control samples, the variability between batches (σ_b²), and variability associated within serum samples measured in the same batch (σ_c²) were estimated using a nested, within-person ANOVA model. Then the variance estimates were used to compute the ICC as [σ_a²/(σ_a² + σ_b² + σ_c²)] and the between-batch CV as $\sqrt{\surd }$σ_b² + σ_c²).

To investigate the relationships of metabolites with body size, partial correlation coefficients of metabolites with BMI [weight (kilograms)/ height (meters)²] and waist circumference at reference date were estimated using the procedure PCORRMAT. Correlations were adjusted for age and matching factors (study site, postprandial time, and processing time relative to the blood draw). To estimate the associations of individual metabolites with endometrial cancer, multivariable unconditional logistic regression models were used to estimate odds ratios (OR) and 95% confidence intervals (CI). Minimally adjusted statistical models included age and matching factors; fully adjusted models also included known or suspected endometrial cancer risk factors (BMI, age at menarche, number of full term births, oral contraceptive use, age at menopause, use of exogenous hormones, smoking status, history of diabetes, and history of hypertension). Multivariable models of statistically significant analytes were stratified by BMI (<30.0 kg/m², ≥30 kg/m²). To evaluate whether main-effect associations were driven by cancer severity, we examined the relationship of grade of endometrioid adenocarcinomas with metabolites using a two-way ANOVA adjusted for age and matching factors. The statistical analyses were performed with STATA (version 11.1. Stata Corp., College Station, TX).

A Bonferroni corrected alpha level less than 2.9 × 10⁻⁴ [i.e. 0.05/(43 analytes times four comparisons)] was used to evaluate statistical significance.

Results

Description of study population

Case subjects who were included in the analysis were diagnosed predominantly with endometrioid tumors (84%). Many cases did not have surgery (43%); among those that did, most had their study blood specimen drawn more than a week after surgery (38%), fewer had blood collected before surgery (9%), and 10% had their blood drawn within a week after surgery. The cases were slightly younger, more recently postmenopausal, and slightly more obese than those not included (data not shown). Controls who were included in the analysis were more likely recruited from Warsaw than controls not included in the analysis. However, cases and controls were distributed similarly with respect to the key matching factors, processing and fasting time (Table 1).

Table 1.

Distribution of key factors by case-control status among study participants selected for metabolomic profile assays, Polish Endometrial Cancer Case-Control Study

Factor	Cases (n = 250)		Controls (n = 250)
	n	%	n	%
Age (yr)
<45	8	3.2	0	0.0
45–49	18	7.2	2	0.8
50–54	32	12.8	25	10.0
55–59	48	19.2	54	21.6
60–64	52	20.8	61	24.4
65–69	52	20.8	64	25.6
70+	40	16.0	44	17.6
Site
Warsaw	145	58.0	157	62.8
Lodz	105	42.0	93	37.2
Time since menopause (yr)
<1	36	14.4	2	0.8
1 to <2	18	7.2	9	3.6
2 to <5	24	9.6	15	6.0
5 to <10	16	6.4	55	22.0
10+	136	54.4	169	67.6
Body mass index (kg/m2)
<25	57	22.8	51	20.4
25 to <30	96	38.4	111	44.4
30+	94	37.6	86	34.4
Processing time (h)
<1	2	0.8	2	0.8
1 to <2	31	12.4	37	14.8
2 to <3	69	27.6	69	27.6
3 to <4	84	33.6	77	30.8
4–4.8	64	25.6	65	26.0
Fasting time (h)
<4	167	66.8	169	67.6
4 to <12	34	13.6	29	11.6
12+	49	19.6	52	20.8
Tumor histology
Endometroid adenocarcinoma	167	66.8
Endometroid with squamous differentiation	37	14.8
Endometroid with other features	5.0	2.0
Othera	41	16.4
Grade of endometrioid tumorsb
Well differentiated	92	56.4
Moderately differentiated	72	44.2
Poorly differentiated	23	14.1
Missing	63

^aOther tumors include papillary serous adenocarcinoma (n = 8), clear cell adenocarcinoma (n = 3), mucinous adenocarcinoma (n = 3), squamous carcinoma (n = 2), mixed types (n = 9), malignant mixed mullerian tumor (carcinoma; n = 1), other primary carcinoma (n = 7), and other malignant tumor (n = 8).

^bBy convention, histological grading applies only to endometrioid tumors.

Quality control results

Of the 43 analytes remaining after data filtering, the CV ranged from 0.28 to 14.50% and the ICC values ranged from 87.96 to 99.94%, indicating that variation among subjects was large compared with intra and interbatch analytical variability (Supplemental Table 1).

Correlation of metabolite levels and body size variables among controls

BMI data were available for 248 controls and waist circumference for 149. The correlation coefficient between BMI and waist circumference of the 148 controls with both measurements was 0.79. Using a stringent correlation of the alpha value to account for multiple comparisons, BMI was weakly correlated with six analytes and waist circumference with none of the analytes (Fig. 1). Four of the six analytes with statistically significant positive correlations with BMI were fatty acids: palmitic (rho = 0.20, P = 4.4 × 10⁻⁵), oleic (rho = 0.28, P = 9.9 × 10⁻⁶), stearic (rho = 0.26, P = 2.9 × 10⁻⁵), and dihomo-γ-linolenic acids (rho = 0.26, P = 4.4 × 10⁻⁵). BMI was not correlated with these analytes after adjustment for other fatty acids, yielding the following attenuated associations: palmitic acid (rho = −0.074, P = 0.25), stearic acid (rho = 0.046, P = 0.48), oleic acid (rho = 0.13, P = 0.053), and dihomo-γ-linolenic acid (rho = 0.13, P = 0.048). BMI was also positively correlated with levels of valine (rho = 0.26, P = 3.4 × 10⁻⁵) and octenoylcarnitine (rho = 0.24, P = 1.5 × 10⁻⁴).

Fig. 1.

Multivariable-adjusted Spearman correlation coefficients (adjusted for age, study site, postprandial time, and processing time relative to blood draw) of the relationship between body size and selected metabolites under study among 250 controls (Polish Endometrial Cancer Study). The gradient of color represents the correlation coefficients in which green cells represent inverse correlations and red cells represent positive correlations and the brightness of the hue represents the strength of the correlation ranging from −0.15 to 0.3 as displayed in the key.

Comparison of metabolites with endometrial cancer case-control status

Five metabolites were associated significantly with endometrial cancer case status (Table 2). Women with the highest quartile (Q) levels of the C5-acylcarnitines, octenoylcarnitine, decatrienoylcarnitine, linoleic acid, and stearic acid were more likely to be endometrial cancer cases than women in the lowest Q of these analytes. Multivariable adjustment including known and suspected endometrial cancer risk factors did not alter the OR compared with the minimally adjusted models. However, in a multivariable model that simultaneously included these six statistically significant analytes, the associations for decatrienoylcarnitine (P for trend = 0.40) and stearic acid (P for trend = 0.34) were no longer statistically significant. Associations for the C5-acylcarnitines [Q 4 vs. Q1: 0.47, 95% CI 0.24–0.93, P for trend = 0.028], octenoylcarnitine (Q4 vs. Q1: 0.29, 95% CI 0.10–0.85, P for trend = 0.028), and linoleic acid (Q4 vs. Q1: 0.33, 95% CI 0.13–0.80, P for trend = 0.014) were also attenuated but remained statistically significant. For the C5-acylcarnitines, we also examined a multivariable model with additional control for leucine/isoleucine and valine and the OR for the C5-acylcarnitines remained virtually unchanged and the P value for linear trend was only slightly attenuated (data not shown). For stearic acid, we further examined a multivariable model that included a ratio of stearic acid and its monounsaturated equivalent, oleic acid, and the results (data not shown) were similar to those for stearic acid alone. In addition, we included stearic acid and oleic acid in the same model as individual variables, and the OR for stearic acid (data not shown) were similar to those in Table 2.

Table 2.

Minimally and fully adjusted OR and 95% CI for the association between individual metabolites and endometrial cancer risk [Polish Endometrial Cancer Case-Control Study (250 cases and 250 controls)]

Analyte	Cases		Controls		Minimally adjusted modela				Fully adjusted modelb
	n	%	n	%	OR	95% CI		P for trendc	OR	95% CI		P for trendc
Isovalerylcarnitine/2-methylbutyrylcarnitine (the C5 acylcarnitines)
Q1	92	36.8	63	25.2	1.00				1.00
Q2	64	25.6	62	24.8	0.73	0.45	1.19		0.61	0.35	1.07
Q3	60	24.0	63	25.2	0.61	0.37	1.01		0.54	0.30	0.95
Q4	34	13.6	62	24.8	0.38	0.22	0.67	1.8 × 10−5	0.35	0.18	0.66	1.26 × 10−5
Octenoylcarnitine (C8:1)
Q1	110	44	63	25.2	1.00				1.00
Q2	69	27.6	62	24.8	0.70	0.44	1.12		0.65	0.38	1.12
Q3	47	18.8	63	25.2	0.44	0.27	0.72		0.38	0.22	0.68
Q4	24	9.6	62	24.8	0.24	0.14	0.43	3.2 × 10−6	0.19	0.10	0.37	2.75 × 10−6
Decatrienoylcarnitine (C10:3)
Q1	110	44	63	25.2	1.00				1.00
Q2	67	26.8	62	24.8	0.63	0.39	1.01		0.55	0.32	0.93
Q3	44	17.6	63	25.2	0.42	0.26	0.70		0.35	0.20	0.62
Q4	29	11.6	62	24.8	0.29	0.17	0.50	5.3 × 10−6	0.24	0.13	0.46	1.07 × 10−5
Linoleic acid (C18:2, ω-6)
Q1	107	42.8	63	25.2	1.00				1.00
Q2	61	24.4	62	24.8	0.56	0.35	0.91		0.51	0.29	0.88
Q3	59	23.6	63	25.2	0.54	0.33	0.88		0.50	0.29	0.88
Q4	23	9.2	62	24.8	0.21	0.12	0.39	1.6 × 10−7	0.19	0.10	0.37	8.33 × 10−7
Stearic acid (C18:0)
Q1	102	40.8	63	25.2	1.00				1.00
Q2	64	25.6	62	24.8	0.67	0.41	1.08		0.80	0.46	1.37
Q3	49	19.6	63	25.2	0.47	0.29	0.78		0.44	0.25	0.79
Q4	35	14	62	24.8	0.33	0.19	0.57	3.5 × 10−6	0.37	0.20	0.69	1.21 × 10−4

^aMinimally adjusted OR included the following variables in the model: age, site, BMI, hours between last meal and blood collection, hours between blood collection and blood processing, and assay batch.

^bFully adjusted models included the variables in the minimally adjusted models as well as age at menarche, number of full-term births, oral contraceptive use, age at menopause, use of exogenous hormones, smoking status, history of diabetes, and history of hypertension.

^cLog-transformed metabolites were treated as a continuous variable in statistical models.

The multivariable models of the statistically significant metabolites, the C5-acylcarnitines, octenoylcarnitine, decatrienoylcarnitine, linoleic acid, and stearic acid, were stratified using a BMI of 30 kg/m² as the cut point. In both the nonobese and obese subgroups, controls had higher levels than cases with increasing Q of these analytes (individual results not shown); the P values for interaction with BMI were not statistically significant (P ≥ 0.05). In sensitivity analyses, exclusion of 26 cases who had their blood drawn within a week after surgery did not affect metabolite associations (data not shown). The results for all analytes are presented in Supplemental Table 2.

Relation of metabolites with tumor characteristics

Endometrial cancers, particularly those of endometroid histology, present predominantly with localized stage disease; thus, advanced stage is not an important confounder of the relationships of metabolite levels with case-control status. Tumor grade was not significantly related to levels of measured analytes (Supplemental Table 3).

Discussion

This study provides proof-of-principle evidence that metabolic profiling may be useful for identifying endometrial cancer biomarkers among postmenopausal women. First, we demonstrate that metabolite profiling methods that include stable isotope dilution of internal standards to quantify the analytes of interest as used here (13, 14) yield satisfactory CV and ICC for most analytes. Second, using blood collected under nonexperimental conditions from population-based controls, we replicated previously reported findings based on clinical studies with stringent blood collection protocols (13, 14, 16, 17) showing that BMI is associated with altered levels of specific metabolites. Finally, we provide preliminary evidence suggesting that metabolomic profiles identifies endometrial cancer biomarkers, which are independent of obesity, other risk factors, and blood collection parameters.

Among controls, higher BMI was significantly correlated with increasing levels of valine, octenoylcarnitine, palmitic acid, oleic acid, stearic acid, and dihomo-γ-linolenic acid. Correlations of these analytes with waist circumference were in the same direction as found for BMI and of similar magnitude but were not considered statistically significant based on fewer subjects with available anthropometric data and stringent statistical correction for multiple comparisons. Previously, elevated BMI was linked to higher levels of valine, octenoylcarnitine (C18:1), palmitic acid, and stearic acid; valine has also been linked to insulin resistance (13).

A prior study of 27 obese individuals with elevated baseline levels of palmitic and stearic acids found that weight loss was associated with decreased levels of these saturated fatty acids (20). This study also found that dihomo-γ-linolenic acid (C20:3, ω-6), an omega-6 fatty acid found predominantly in vegetable oils, was elevated among obese compared with lean individuals (P < 0.0001) (13). Palmitic acid, which is less prone to desaturation than stearic acid (21), has been linked to insulin resistance and hypercholesterolemia (22, 23). Stearic acid might be converted to the monounsaturated omega-9 fatty acid, oleic acid (C18:1, ω-9), which is both endogenously produced and widely distributed in nature. High levels of oleic acid have been linked to obesity (13) and breast cancer risk (24, 25) and have been shown to decrease with weight loss (20). Correlations between high levels of branched chain amino acids and larger body size reported in the literature (13, 16) were also observed in our study, although the statistical significance for these analytes with BMI and waist circumference in our study fell below the stringent Bonferroni-corrected alpha level. The consistency of our results with prior clinical studies (13, 16) provides evidence that metabolite patterns are robust to a range of fasting states and processing times for the blood draw, which are difficult to control in large, population-based epidemiological studies.

In this analysis, levels of the C5-acylcarnitines, octenoylcarnitine, decatrienoylcarnitine, linoleic acid, and stearic acid were lower in cases than controls. The most robust associations were limited to the C5-acylcarnitines, octenoylcarnitine, and linoleic acid. The C5-acylcarnitines category is comprised of isovalerylcarnitine and 2-methylbutyrylcarnitine, which are five-carbon intermediate products of catabolism of the branched chain amino acids, leucine and isoleucine, respectively, that cannot be distinguished in this assay. Elevated levels of the C5-acylcarnitines are associated with insulin resistance and obesity in humans (13, 17) and in rodents fed experimental diets with supplemented branched chain amino acids (13). There is a brief but compelling body of literature, which support our findings with endometrial cancer, linking isovalerylcarnitine to proapoptotic activity via activation of calpain and the caspase system in human neutrophils and erythrocytes (26–28); however, more functional research is needed. Octenoylcarnitine is likely generated during β-oxidation of fatty acids but has not been previously identified to be associated with cancer or related risk factors. Linoleic acid (C18:2, ω-6) is an essential fatty acid with an unclear role in carcinogenesis. Although linoleic acid promotes mammary tumors in rodent models (29) and a case-control study found that high dietary intake of linoleic acid was associated with a higher risk of endometrial cancer (30), our results are consistent with a meta-analysis of cohort studies found that circulating levels of linoleic acid were associated with a 12% decrease in breast cancer risk (24), a site that has risk factors similar to that of endometrial cancer.

It is notable that stearic acid was associated with higher BMI, even though levels were lower among cases than controls in our study, highlighting that not all obesity related metabolic disturbances imply heightened endometrial cancer risk. In fact, analytes linked to high BMI were not associated with increased endometrial cancer risks in this study; conversely, the metabolites showing a reduced risk for endometrial cancer, in particular isovalerylcarnitine and linoleic acid, were not significantly associated with BMI. Reasons for this are not clear, but measurement of BMI alone is insufficient to accurately predict individual endometrial cancer risk. More precise models that include some of these metabolites might be required to capture the relevant metabolic changes related to endometrial cancer risk in obese and nonobese individuals.

Given that our samples were collected at the time of diagnosis, rather than prediagnostically, we cannot determine whether the profiles that we describe are markers of disease-related metabolic changes as opposed to markers of future risk and/or etiologically related changes. However, we believe that these concerns are somewhat lessened in endometrial cancer compared with some other tumor types because most endometrial cancers are organ confined at diagnosis and typically present with abnormal vaginal bleeding, rather than systemic effects related to advanced disease. Data suggest that established risk factors for endometrial cancers of the usual histological type (endometrioid) and their precursors (endometrial hyperplasia) are similar (31). Confirmation that metabolic profiles are related to risk in prediagnostic samples could lead to the development of strategies for identifying at risk women who could be targeted for preventive strategies.

Our study benefited from several strengths, including a population-based design with expected relationships for established endometrial cancer risk factors (18, 32). In addition, our assays performed reliably, without evidence of analytical drift between batches (33, 34), and we measured metabolic profiles in blood, which is more stable than in urine with dietary fluctuations (35). We controlled for collection parameters and fasting status as was possible given the study design (36). Some of our case samples were taken after diagnosis or treatment; however, only a small number had treatment within a week of their blood collection and exclusion of these samples did not materially affect our results or conclusions. Lifestyle changes in the short postdiagnostic interval preceding blood collection would likely have favored changes toward healthier lifestyle choices, thereby narrowing differences between cases and controls. To reduce chance findings secondary to multiple testing, we applied stringent statistical corrections that might have obscured true associations with body size or endometrial cancer. Thus, our results present the basis for prospective analysis of candidate markers but do not preclude future efforts to identify new analytes of interest through profiling.

Our results suggest that obesity, a strong risk factor for endometrial cancer, affects circulating levels of multiple metabolites, but assessment of BMI alone might not capture all relevant metabolic changes related to endometrial cancer risk. Thus, analysis of metabolic profiles may have value for early detection, risk assessment, and etiological studies of endometrial cancer. Specifically, our profiling approach provides the basis for identifying and studying a parsimonious panel of markers to understand mechanisms and risk related to endometrial cancer in hypothesis-driven, prospective analyses. Cohort studies will also permit further evaluation of factors that impact metabolic measurements, including fasting status, sample processing, and disease effects. Integrating metabolic studies with measurements of hormones and inflammatory mediators, genetic susceptibility, and analysis of tumor tissues might contribute to developing translational approaches for and a more complete biological understanding of endometrial cancer.