- Research article
- Open Access
- Open Peer Review
Metabolomic profiles in breast cancer:a pilot case-control study in the breast cancer family registry
- Marcelle M. Dougan†1, 2Email author,
- Yuqing Li†1, 3,
- Lisa W. Chu1,
- Robert W. Haile1, 4,
- Alice S. Whittemore1,
- Summer S. Han1,
- Steven C. Moore5,
- Joshua N. Sampson5,
- Irene L. Andrulis6,
- Esther M. John1, 3 and
- Ann W. Hsing1, 7
© The Author(s). 2018
- Received: 9 March 2017
- Accepted: 25 April 2018
- Published: 5 May 2018
Metabolomics is emerging as an important tool for detecting differences between diseased and non-diseased individuals. However, prospective studies are limited.
We examined the detectability, reliability, and distribution of metabolites measured in pre-diagnostic plasma samples in a pilot study of women enrolled in the Northern California site of the Breast Cancer Family Registry. The study included 45 cases diagnosed with breast cancer at least one year after the blood draw, and 45 controls. Controls were matched on age (within 5 years), family status, BRCA status, and menopausal status. Duplicate samples were included for reliability assessment. We used a liquid chromatography/gas chromatography mass spectrometer platform to measure metabolites. We calculated intraclass correlations (ICCs) among duplicate samples, and coefficients of variation (CVs) across metabolites.
Of the 661 named metabolites detected, 338 (51%) were found in all samples, and 490 (74%) in more than 80% of samples. The median ICC between duplicates was 0.96 (25th – 75th percentile: 0.82–0.99). We observed a greater than 20% case-control difference in 24 metabolites (p < 0.05), although these associations were not significant after adjusting for multiple comparisons.
These data show that assays are reproducible for many metabolites, there is a minimal laboratory variation for the same sample, and a large between-person variation. Despite small sample size, differences between cases and controls in some metabolites suggest that a well-powered large-scale study is likely to detect biological meaningful differences to provide a better understanding of breast cancer etiology.
Metabolomics is the systematic survey of the small molecules (< 1 k Dalton in size) that are the products of metabolism in biological systems [1, 2]. A metabolic phenotype represents the collection of metabolites within the body which reflects influences from both genetic and lifestyle/environmental factors. Because metabolites include the intermediate- and end-products of the cellular processes, metabolomics provides a functional readout of the physiological state of health and disease. Changes in energy metabolism within cells are one of the hallmarks of carcinogenesis. Under aerobic conditions, normal cells metabolize energy by first converting glucose into pyruvate and then to carbon dioxide, and under anaerobic conditions, cells metabolize by glycolysis. However, the converse is true for cancer cells, where under aerobic conditions, energy metabolism occurs largely by glycolysis, i.e., “aerobic glycolysis” . Thus, a characterization of metabolic processes may provide new insights into carcinogenesis. In recent years, metabolomics has emerged as an important tool for the identification of biomarkers in a growing number of applications, including early disease detection, monitoring of disease progression, and investigation of metabolic pathways. The application of metabolomics has yielded novel signatures predicting the occurrence and progression of complex diseases, including cancers of the breast , prostate, colon, and kidney [5–8].
Most metabolomics studies of breast cancer to date have been conducted in tumor tissues or cell lines and with the goals of distinguishing cancer from normal tissue and cancers with metastasis from those without, as well as identifying therapeutic targets [9–11]. Data from these studies have suggested that metabolomic profiles may differ by pathological and molecular subtype of breast cancer. In large scale epidemiologic studies, blood and urine are more readily available than tissue. Because blood and urine serve as transporters of nutrients and wastes to and from cells for excretion, and maintain homeostasis of essential molecules and fluid levels, they are sensitive indicators of health and perturbations from diseases. Several studies have measured urinary metabolic profiles and found promising candidate markers for early detection and monitoring of breast cancer progression [12–14]. However, studies using pre-diagnostic blood are limited. To our knowledge, there has been only one previously published study on metabolomics and breast cancer risk using pre-diagnostic blood , warranting additional studies to replicate the findings in other populations. We conducted a pilot study to generate preliminary data to assess whether circulating metabolomic profiles could be detected in pre-diagnostic plasma samples of women enrolled in the Breast Cancer Family Registry (BCFR) cohort, and to evaluate the reproducibility of metabolomic assays.
Pre-diagnostic plasma samples were obtained from the BCFR, an international prospective cohort of breast cancer families established in 1995  . For this pilot study, samples were selected from the Northern California site (NC-BCFR), which enrolled women with newly diagnosed breast cancer (probands) identified through the population-based cancer registry of the San Francisco Bay area and family members . At baseline, participants completed a risk factor questionnaire and provided a blood sample. During follow-up, newly diagnosed breast cancer cases were identified among family members who were unaffected at baseline. This pilot study included 45 women who were diagnosed with breast cancer at least one year after the blood draw (cases) and 45 women who did not develop breast cancer (controls). Of the 45 cases, 72% of the cases were confirmed via cancer registry linkage or pathology reports; the remainder were self-reported. Controls were matched to cases on family status (a sister was selected if available; if more than one sister was available, we selected the sister closest in age), age at blood draw (±5 years), menopausal status at diagnosis, and number of affected first degree relatives (1, 2, or ≥3). The age range for cases was 26–80 years (average 52.4 years), and that for controls was 36–73 years (average 53 years).
Plasma samples obtained from cases and matched controls were aliquoted into 200 μl ethylenediaminetetraacetic acid (EDTA) plasma vials. Case-control sets were assayed in the same batch and adjacent to each other in sequence. Samples were identified by specimen ID only, and laboratory technicians were masked to the case-control status of samples. The samples were assayed on the Discovery HD4 platform, a mass spectrometry-based metabolomics profiling platform, at Metabolon (Durham, NC, USA). This method combines automated sample extraction processing, an ultrahigh performance liquid chromatography/electrospray ionization tandem mass spectrometry (UHPLC/MS) with additional gas chromatography mass spectrometry (GC/MS) platform. Peaks were quantified using area-under-the-curve and metabolite levels were generated. The metabolite data were normalized to a median of 1.00 to correct for variation resulting from instrument tuning differences. Metabolites not detected in individual samples were imputed with the minimum value for that metabolite. The data were then log-transformed to reduce non-normality. Duplicates from 10 controls were included to assess assay reproducibility. Data from the duplicates were used to assess intra-class correlation (ICC) and were averaged for the case-control analysis.
To assess the reliability of our assay results, we calculated coefficients of variation (CVs) and ICCs across duplicate samples. Coefficient of variation is a measure of dispersion, that describes the amount of variability relative to the mean. For samples measured using the same method, a low (~ 10%) variability within subjects and high variability across subjects is desirable. Intra-class correlations describe the degree to which duplicate samples agree: a value between 0.75 and 1 indicates excellent agreement.
We used the variance component from a one-way analysis of variance (ANOVA) model to estimate the ICC for replicate samples, and estimated confidence intervals for the ICCs using the Smith method  in R (ICC-Package) . We calculated CVs across named metabolites, and used principal component analysis  to identify the important components (groups of metabolites) in each sample (including the duplicates). We used paired t-tests to examine differences in the normalized metabolite levels between cases and controls. We used non-linear modeling to examine whether normalized metabolite levels were associated with age at blood draw. We evaluated quadratic and cubic models and used the Akaike Information Criterion (AIC) to evaluate the best model fit. We also evaluated using ANOVA with robust variance to examine differences in metabolite levels by the number of affected first-degree relatives (1, 2, or ≥3), and among cases with available information, by estrogen receptor (ER) status (positive/negative) and progesterone receptor (PR) status (positive/negative). Due to the limited sample size of this pilot study, further analysis of subgroups by age was not statistically meaningful.
Characteristics of Study Subjects
Cases (n = 45)
Controls (n = 45)
Age at blood draw, years
Age at diagnosis, years
Post-menopausal at blood draw, %
Number of affected first-degree relatives, %
BRCA1 Mutation, %
BRCA2 Mutation, %
Estrogen Receptor Status, %
Progesterone Receptor Status, %
Human Epidermal Growth Factor Receptor 2 Status, %
Summary Stage, %
Regional, nodes only
Distribution of observed coefficient of variation for named and most common1 metabolites among 45 cases and 45 controls
Most common1 metabolites
Lower 95% CI
Upper 95% CI
Differences by key breast cancer risk factors among metabolites that showed a greater than 20% case-control difference
Age at blood drawa (Adjusted R2)
Number of affected first-degree relatives mean levels (standard deviation)
n = 90
1 (n = 27)
2 (n = 25)
3 (n = 38)
1-linoleoyl-GPA (18:2) c
1-palmitoleoyl-2-linoleoyl-GPC (16:1/18:2) 3
1-palmitoyl-GPG (16:0) c
Polyunsaturated fatty acid (n3 and n6)
Alanine and aspartate
Glycolysis, Gluconeogenesis, and Pyruvate
Secondary bile acid
Medium chain fatty acid
Fatty acid (acyl carnitine)
Glycine, serine, and threonine
Differences by key breast cancer variables among metabolites that showed a greater than 20% case-control difference
ER Status mean levels (standard deviation)
PR Status mean levels (standard deviation)
ER + (n = 23)
ER - (n = 10)
PR + (n = 24)
PR - (n = 9)
1-(1-enyl-palmitoyl)-2-oleoyl-GPC (P-16:0/18:1) b
1-linoleoyl-GPA (18:2) b
1-palmitoleoyl-2-linoleoyl-GPC (16:1/18:2) b
1-palmitoyl-GPG (16:0) 2
Polyunsaturated fatty acid (n3 and n6)
Alanine and aspartate
Glycolysis, Gluconeogenesis, and Pyruvate
Secondary bile acid
Medium chain fatty acid
Fatty acid (acyl carnitine)
Glycine, serine, and threonine
We examined whether metabolites differed by the number of affected first-degree relatives. Overall, there was no clear monotonically increasing or decreasing pattern by the number of affected relatives: mean levels were similar for women with one or three affected relatives, and lower or higher for those with two affected relatives. For example, for 2-ethylphenylsulfate, the mean levels were 0.68, 0.97, and 0.64, for women with one, two, or three affected relatives, respectively (p < 0.001).
In some metabolites, we found differences by ER status. For example, cases with ER+ breast cancer had higher mean laurylcarnitine level than those with ER- breast cancer (1.16 vs. 0.83, p = 0.04). Conversely, the mean indoleacetyleglutamine level was lower for ER+ breast cancer cases than ER- cases (0.46 vs. 0.54, p = 0.04).
Finally, we examined whether metabolite levels differed by PR status. The mean asparagine level for PR+ cases was 1.12, compared to 1.18 for PR- cases (p = 0.02). For adrenate (22:4n6), the mean level for PR+ cases was 1.07 compared to 0.69 for PR- cases, and for N-(2-furoyl)glycine, the mean level for PR+ cases was 2.87 compared to 7.54 for PR- cases. However, none of the associations remained significant after adjusting for multiple comparisons.
These data, despite small numbers, suggest that a large number of metabolites have detectable levels, with good reproducibility, as suggested by high ICCs and reasonable CVs. We also showed that for most metabolites, the within-person variance is small, while the between-person variance is much larger. We found that some metabolites have a greater than 20% case-control difference. Finally, we showed that some metabolites (including N-(2-furoyl)glycine in the xenobiotics pathway) differed by key breast cancer risk factors such as the number of affected family members, although these associations were not significant after adjusting for multiple comparisons, likely due to the small sample size. Taken together, these results suggest that a large-scale study (~ 1000) would be well-powered to detect meaningful biological and statistically significant differences between cases and controls to provide a better understanding of breast cancer etiology across a wide spectrum of risks, and among high-risk women in particular.
Metabolomics profiles are becoming increasingly utilized in epidemiological studies to predict the risk of chronic diseases, including breast cancer; however, data from prospective studies are limited. In the first prospective study of metabolomics and breast cancer risk, Kuhn et al.  found that phosphatidylcholines were associated with breast cancer risk. That study included 362 sporadic breast cancer cases, and measured only 120 metabolites. To date, there are no metabolomics data on women at increased risk of breast cancer due to their family history of breast cancer. Clearly additional data from prospective studies are needed to further examine the role of metabolomics in breast carcinogenesis.
The assay performance on our samples, measured by ICCs and CVs, is consistent with earlier studies that have examined the utility of metabolomics in epidemiological research among participants of the Shanghai Physical Activity Study . In that study, the variability in a large subset of metabolites was assessed and the intraclass correlation was high (median 0.8). Similar assay performance was also observed in a nested case-control study of metabolomics and colorectal cancer risk  that included 254 cases and 254 matched controls from the Prostate, Lung, Colorectal and Ovarian Cancer study. In that study, which used a metabolomics platform similar to the one used in our pilot study, the median intraclass correlation was 0.86 (25th–75th percentile: 0.64–0.92).
Consistent with our observation that age at blood collection was associated with metabolite levels, Saito et al.  also reported that certain metabolites were associated with age at blood draw in a Japanese population. Because the populations in the two studies are quite different, a direct comparison is not possible. Similarly, Tang et al.  reported that metabolites in tumor tissues were associated with ER status, and also with BRCA1-associated tumors. However, studies utilizing human plasma are limited.
One notable limitation of our study is that we did not match on the duration of storage time between cases and controls. Post-hoc analysis revealed that among case-control pairs, 35 pairs (78%) had a difference of less than 3 years of storage duration, while 1 pair had a difference of more than 10 years. Further analyses showed that while there were no appreciable differences in the analyses by age, there were differences in 5 metabolites when evaluating levels by the number of affected relatives. Future studies should match on calendar year of blood draw (hence storage duration) within case-control matched sets.
Our study is among the first to examine the association between metabolomics and breast cancer risk using pre-diagnostic plasma samples. Despite the limited sample size, we were able to find a larger than 20% case-control difference in several metabolites, although we cannot rule out the possibility that the presence of asymptomatic preclinical breast cancer may have affected metabolite levels in cases. Such bias is possible but should be minimal as we excluded cases diagnosed with breast cancer within 12 months after the blood draw in order to limit the potential for preclinical disease to influence metabolite levels. Finally, our study is also among the first to examine reliability across more than 600 metabolites.
In conclusion, findings from this study suggest that metabolomics can be used reliably in large-scale epidemiologic studies of breast cancer to detect meaningful differences in risk.
We would like to gratefully acknowledge all the families that participate in the Northern California site of the BCFR.
AWH, LWC, ASW, and EMJ designed and executed the study. YSL and MMD performed the data analysis and drafted the manuscript. RWH, SSH, SCM, JNS, and IA substantially contributed to the analysis and interpretation of data. All authors (MMD, YSL, LWC, RWH, ASW, SSH, SCM, JSN, ILA, EMJ, and AWH) reviewed and provided critical feedback for intellectual content prior to submission. All authors read and approved the final manuscript.
The study was funded by an innovation grant from the Stanford Cancer Institute.
Availability of data and materials
The datasets used and/or analysed during the current study available from the corresponding author on reasonable request.
Ethics approval and consent to participate
All participants provided written informed consent. The study was approved by the Institutional Review Board of the Cancer Prevention Institute of California.
The authors declare no conflict of interest.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
- Johnson CH, Manna SK, Krausz KW, Bonzo JA, Divelbiss RD, Hollingshead MG, Gonzalez FJ. Global metabolomics reveals urinary biomarkers of breast cancer in a mcf-7 xenograft mouse model. Metabolites. 2013;3(3):658–72.View ArticlePubMedPubMed CentralGoogle Scholar
- Lin NU, Vanderplas A, Hughes ME, Theriault RL, Edge SB, Wong Y-N, Blayney DW, Niland JC, Winer EP, Weeks JC. Clinicopathologic features, patterns of recurrence, and survival among women with triple-negative breast cancer in the national comprehensive Cancer network. Cancer. 2012;118(22):5463–72.View ArticlePubMedPubMed CentralGoogle Scholar
- Hanahan D, Weinberg Robert A. Hallmarks of Cancer: the next generation. Cell. 2011;144(5):646–74.View ArticlePubMedGoogle Scholar
- Kuhn T, Floegel A, Sookthai D, Johnson T, Rolle-Kampczyk U, Otto W, von Bergen M, Boeing H, Kaaks R. Higher plasma levels of lysophosphatidylcholine 18:0 are related to a lower risk of common cancers in a prospective metabolomics study. BMC Med. 2016;14:13.View ArticlePubMedPubMed CentralGoogle Scholar
- Bertini I, Cacciatore S, Jensen BV, Schou JV, Johansen JS, Kruhøffer M, Luchinat C, Nielsen DL, Turano P. Metabolomic NMR fingerprinting to identify and predict survival of patients with metastatic colorectal cancer. Cancer Res. 2012;72(1):356–64.View ArticlePubMedGoogle Scholar
- Mondul AM, Moore SC, Weinstein SJ, Karoly ED, Sampson JN, Albanes D. Metabolomic analysis of prostate cancer risk in a prospective cohort: the alpha-tocolpherol, beta-carotene cancer prevention (ATBC) study. Int J Cancer. 2015;137:2124–32.View ArticlePubMedPubMed CentralGoogle Scholar
- Mondul AM, Moore SC, Weinstein SJ, Männistö S, Sampson JN, Albanes D. 1-stearoylglycerol is associated with risk of prostate cancer: results from serum metabolomic profiling. Metabolomics : Official journal of the Metabolomic Society. 2014;10(5):1036–41.View ArticleGoogle Scholar
- Sreekumar A, Poisson LM, Rajendiran TM, Khan AP, Cao Q, Yu J, Laxman B, Mehra R, Lonigro RJ, Li Y, et al. Metabolomic profiles delineate potential role for sarcosine in prostate cancer progression. Nature. 2009;457:910–14.Google Scholar
- Budczies J, Brockmoller SF, Muller BM, Barupal DK, Richter-Ehrenstein C, Kleine-Tebbe A, Griffin JL, Oresic M, Dietel M, Denkert C, et al. Comparative metabolomics of estrogen receptor positive and estrogen receptor negative breast cancer: alterations in glutamine and beta-alanine metabolism. J Proteome. 2013;94:279–88.View ArticleGoogle Scholar
- Henneges C, Bullinger D, Fux R, Friese N, Seeger H, Neubauer H, Laufer S, Gleiter CH, Schwab M, Zell A, et al. Prediction of breast cancer by profiling of urinary RNA metabolites using support vector machine-based feature selection. BMC Cancer. 2009;9:104.View ArticlePubMedPubMed CentralGoogle Scholar
- Kanaan YM, Sampey BP, Beyene D, Esnakula AK, Naab TJ, Ricks-Santi LJ, Dasi S, Day A, Blackman KW, Frederick W, et al. Metabolic profile of triple-negative breast cancer in African-American women reveals potential biomarkers of aggressive disease. Cancer Genomics Proteomics. 2014;11(6):279–94.PubMedGoogle Scholar
- Asiago VM, Alvarado LZ, Shanaiah N, Gowda GAN, Owusu-Sarfo K, Ballas RA, Raftery D. Early detection of recurrent breast cancer using metabolite profiling. Cancer Res. 2010;70(21):8309–18.View ArticlePubMedPubMed CentralGoogle Scholar
- Kim Y, Koo I, Jung BH, Chung BC, Lee D. Multivariate classification of urine metabolome profiles for breast cancer diagnosis. BMC Bioinformatics. 2010;11(Suppl 2):S4.View ArticlePubMedPubMed CentralGoogle Scholar
- Slupsky CM, Steed H, Wells TH, Dabbs K, Schepansky A, Capstick V, Faught W, Sawyer MB. Urine metabolite analysis offers potential early diagnosis of ovarian and breast cancers. Clinical cancer research : an official journal of the American Association for Cancer Research. 2010;16(23):5835–41.View ArticleGoogle Scholar
- John EM, Hopper JL, Beck JC, Knight JA, Neuhausen SL, Senie RT, Ziogas A, Andrulis IL, Anton-Culver H, Boyd N, et al. The breast Cancer family registry: an infrastructure for cooperative multinational, interdisciplinary and translational studies of the genetic epidemiology of breast cancer. Breast Cancer Res. 2004;6(4):R375–89.View ArticlePubMedPubMed CentralGoogle Scholar
- Terry MB, Phillips KA, Daly MB, John EM, Andrulis IL, Buys SS, Goldgar DE, Knight JA, Whittemore AS, Chung WK, et al. Cohort profile: the breast Cancer prospective family study cohort (ProF-SC). Int J Epidemiol. 2016;45(3):683–92.Google Scholar
- John EM, Miron A, Gong G, Phipps AI, Felberg I, Li FP, West DW, AS W. Prevalence of pathogenic BRCA1 mutation carriers in five US racial/ethnic groups. JAMA. 2007;298(24):2869–76.View ArticlePubMedGoogle Scholar
- Smith CAB. On the estimation of intraclass correlation. Ann Hum Genet. 1956;21:363–73.View ArticleGoogle Scholar
- Wolack M: ICC: facilitating estimation of the intraclass correlation coefficient. 2015 Available at: https://cranr-projectorg/web/packages/ICC/ICCpdf Accessed 8 Sep 2016 .Google Scholar
- Kim J-O MC. In: Uslaner EM, editor. FACTOR ANALYSIS statistical methods and practical issues. Iowa City, IA: Sara Miller McCune; 1978.Google Scholar
- Sampson JN, Boca SM, Shu XO, Stolzenberg-Solomon RZ, Matthews CE, Hsing AW, Tan YT, Ji BT, Chow WH, Cai Q, et al. Metabolomics in epidemiology: sources of variability in metabolite measurements and implications. Cancer Epidemiol Biomark Prev. 2013;22(4):631–40.View ArticleGoogle Scholar
- Cross AJ, Moore SC, Boca S, Huang WY, Xiong X, Stolzenberg-Solomon R, Sinha R, Sampson JN. A prospective study of serum metabolites and colorectal cancer risk. Cancer. 2014;120(19):3049–57.View ArticlePubMedPubMed CentralGoogle Scholar
- Saito K, Maekawa K, Kinchen JM, Tanaka R, Kumagai Y, Saito Y. Gender- and age-associated differences in serum metabolite profiles among Japanese populations. Biol Pharm Bull. 2016;39(7):1179–86.View ArticlePubMedGoogle Scholar
- Tang X, Lin CC, Spasojevic I, Iversen ES, Chi JT, Marks JR. A joint analysis of metabolomics and genetics of breast cancer. Breast Cancer Res. 2014;16(4):415.View ArticlePubMedPubMed CentralGoogle Scholar