Clinically meaningful molecular subtypes for classification of breast cancers have been established, however, initiation and progression of these subtypes remain poorly understood. The recent development of desorption electrospray ionization-mass spectrometry imaging (DESI-MSI) facilitates the convergence of analytical chemistry and traditional pathology, allowing chemical profiling with minimal tissue pretreatment in frozen samples. Here, we characterized the chemical composition of molecular subtypes of breast cancer with DESI-MSI. Regions of interest were identified, including invasive breast cancer (IBC), ductal carcinoma in situ (DCIS), and adjacent benign tissue (ABT), and metabolomic profiles at 200 μm elaborated using Biomap software and the Lasso method. Top ions identified in IBC regions included polyunsaturated fatty acids, deprotonated glycerophospholipids, and sphingolipids. Highly saturated lipids, as well as antioxidant molecules [taurine (m/z 124.0068), uric acid (m/z 167.0210), ascorbic acid (m/z 175.0241), and glutathione (m/z 306.0765)], were able to distinguish IBC from ABT. Moreover, luminal B and triple-negative subtypes showed more complex lipid profiles compared with luminal A and HER2 subtypes. DCIS and IBC were distinguished on the basis of cell signaling and apoptosis-related ions [fatty acids (341.2100 and 382.3736 m/z) and glycerophospholipids (PE (P-16:0/22:6, m/z 746.5099, and PS (38:3), m/z 812.5440)]. In summary, DESI-MSI identified distinct lipid composition between DCIS and IBC and across molecular subtypes of breast cancer, with potential implications for breast cancer pathogenesis.

Significance:

These findings present the first in situ metabolomic findings of the four molecular subtypes of breast cancer, DCIS, and normal tissue, and add to the understanding of their pathogenesis.

Breast cancer is the most common cancer affecting women and represents a complex group of diseases that exhibit great variability at clinical presentation and biologic aggressiveness. Numerous clinical studies of breast cancer coupled with advances in genomic profiling provided data that support the existence of clinically useful molecular subtypes (1). In particular, assessments of hormonal receptor [estrogen receptor (ER) and progesterone receptor (PR)] and HER2 status have contributed to translation of this molecular subtype into clinics (2). Despite all genomic characterization, our understanding of initiation and progression of breast cancer remains incomplete. Cell processes driving invasiveness from ductal carcinoma in situ (DCIS) stage and differentiation into each molecular subtype of breast cancer are complex, and analyses beyond genomics are needed to better understand them.

Metabolic reprogramming in cancer cells, referred to as the Warburg effect, was described nearly a century ago and the interest on this effect has been recently renewed as it is considered one of the hallmarks of cancer cells (3, 4). Lipids play a key role in cell membranes' structure and trafficking and are substrates for energy production (5), and de novo lipogenesis has roles in many other cellular processes, such as oxidative stress–induced cell death, regulation of chemotherapeutic agent uptake (6), and the generation of signaling molecules (5). Genomic changes in breast cancer, such as deletion of chromosome 8p, lead to de novo lipogenesis and that loss has been recognized as a hallmark of aggressive cancers (7–10).

The convergence of two powerful disciplines, pathology and analytical chemistry, has facilitated studies of discrete molecular changes that occur in cancer cells. In situ mass spectrometry has helped to elucidate the complex connections between metabolism, genomics, and cellular functions (11). Desorption electrospray ionization (DESI) represents a recent and robust advance among ambient ionization techniques that allows scientists to study cancer cell metabolism coupled to mass spectrometry imaging (MSI) with minimal sample pretreatment (12–15). Global levels of metabolites and their spatial distributions with a 200 μm resolution can be determined from frozen tissue samples and comparisons across normal, precancerous, or invasive regions of interest (ROI) stained with hematoxylin and eosin (H&E) bring new insights to cancer pathogenesis. To date, desorption electrospray ionization-mass spectrometry imaging (DESI-MSI) has been used to study the metabolite profiles of various types of cancer tissues, including breast (16, 17), brain (18), gastric (14), prostate (11), and has been considered a promising tool diagnosis and tumor margin evaluation (19).

The aim of this study was to integrate lipidomic and metabolomic data to distinct molecular subtypes of breast cancer using in situ DESI-MSI. Metabolomic profiles of frozen slides containing paired tumor and normal samples were created in two-dimensional (2D) maps. Small molecules 2D maps were overlaid with digital H&E images of the same samples, harboring invasive breast cancer (IBC), in situ lesions, and normal tissue. These data were examined considering the ROI to better understand the lipidomic differences across subtypes of breast cancer.

Ethics statements

This study was approved by the Institutional Research Board of the AC Camargo Cancer Center (São Paulo, Brazil; No. 1830/13) and adheres to the tenets of the Declaration of Helsinki. Written informed consent for collection of tissues for research was obtained from all patients.

Patient samples

Frozen tumor samples were collected from 68 patients. These samples included 122 tissues (67 breast tumor tissues and 55 paired normal tissues) representing different molecular subtypes of breast cancer: luminal A (n = 13), luminal B (n = 17), triple-negative (n = 12), HER2+ (n = 10), pure DCIS (n = 5, of which, 2 are ER+/HER2, 2 are ER+/HER2+, and 1 is ER/HER2+), and DCIS adjacent to an invasive tumor (n = 10, of which, 6 are ER+/HER2, 3 are ER/HER2+, and 1 is ER/HER2+; Supplementary Table S1). The samples were resected with sterile surgical blades and were snap-frozen in liquid nitrogen within 30 minutes of collection. The samples were stored at −140°C until they were sectioned. At least three sequential sections (each 10 μm thick) were prepared from each sample with a Leica Multicut 2045 Cryotome. The cryotome chamber was chilled to −29°C and the specimen holder was chilled to −20°C. The samples were kept on dry ice before and after the sections were cut and they were subsequently stored at −80°C. For DESI- MSI, samples were transported on dry ice. Prior to DESI-MSI, the glass slides were dried in a desiccator for approximately 15 minutes. After DESI-MSI, the slides were stained with H&E and scanned with the Aperio XT Scanscope (Leica Biosystems). Digital H&E images were overlaid with their corresponding 2D metabolic images.

Mass spectrometry imaging

A 2D DESI Source (Prosolia, Inc.) coupled to a QExactive HF hybrid quadrupole orbitrap mass spectrometer (Thermo Fisher Scientific) was operated at 70,000 resolving power for tissue imaging. Data were acquired in the negative-ion mode (−5 kV) over a mass-to-charge (m/z) ratio range of m/z 100–1,200 for 42 samples (batch 1 from sample A1 to A42) and over a range of m/z 100–1,000 for 26 samples (batch 2 from sample A43 to A68). Analyses were performed in a matrix design, with 200 μm spatial resolution. A histologically compatible solvent, 100% methanol (HPLC grade; Merck), was used for analyses at a flow rate of 1.5 μL/min. Nitrogen was used as the sheath gas for electrospray nebulization at a pressure of 175 psi. DESI-MSI data were collected from entire tissue sections. These data were used to generate a m/z spectra every 200 μm, and these spectra were used to build 2D metabolic maps. XCalibur 2.2 Software (Thermo Fisher Scientific) was used to acquire data and Firefly was used to convert processed mass spectral data into image files (.img). These files were read by the biomedical image analysis software, Biomap (freeware, https://ms-imaging.org/wp/). All imaging experiments were performed under identical experimental conditions, including identical geometrical parameters. Data were preprocessed by background subtraction and normalization based on total ion current. DESI-MSI intersection and intrasample reproducibility were assessed by analyzing different tissue sections of mouse brain tissue on different days under the same experimental conditions. Very similar mass spectra profiles were observed confirming that DESI-MSI is reproducible between serial sections of tissue and from scan to scan within one experiment. Full spectra profile reproducibility was also investigated, and %RSD values of approximately 1% were obtained, showing that similar mass spectra profiles are obtained from similar samples.

Histopathologic and molecular classifications

After DESI-MSI, the same tissue sections were stained with H&E for histopathologic evaluation by an expert pathologist. Careful histologic evaluations were performed by a breast pathologist (V.P. de Andrade) who used digital images to assign the following ROI: adjacent benign tissue or normal lobules (ABT), DCIS, and IBC. Molecular classification of each sample was recorded from the original diagnostic IHC and ISH assays. These classifications included the following antibody reagents as appropriate: a rabbit mAb–recognizing ER (SP1, Roche), an anti-PR rabbit monoclonal primary antibody (1E2, Roche), a rabbit monoclonal primary antivirus antibody for HER2 [IHC, Anti-HER-2/neu (4B5), Roche] or a dual color-dual hapten ISH antibody (as needed), and a rabbit primary anti-Ki-67 antibody (30-9, Roche). The luminal A subtype was defined as ER+ and/or PR+, HER2, and Ki-67 < 20%. The luminal B subtype was defined as ER+ and/or PR+, as well as PR < 20%, HER2+, or Ki67 >20%. The HER2 subtype was defined as ER/PR and HER2+. The triple-negative subtype was defined as ER, PR, and HER2. IHC was performed on automated protocols recommended for the Roche/Ventana Benchmark XT Autostainer (Roche/Ventana).

Imaging evaluation

ROI were overlaid on 2D molecular images with Biomap software to correlate m/z spectra with ABT, DCIS, and IBC tissues (Fig. 1). A corresponding spectrum was generated for each selected ROI and was exported as a .txt file for statistical analysis.

Figure 1.

Representative selections of ROI based on H&E staining and overlaid on the DESI-MSI images. A, A DCIS sample (sample A5) histologic evaluation and attribution of ROI in H&E staining; the black square selected in the figure is magnified and represents DCIS. B, Imported H&E image in Biomap software with the ROI defined. C, Overlaid H&E and DESI-MSI images. D, ROI defined in the H&E image overlaid on the DESI-MSI image. E, A luminal B sample (sample A2) histologic evaluation and attribution of ROI in H&E staining. The squares selected in the figure are magnified and represent IBC (yellow), ABT (benign; blue), and DCIS (black). F, The 2D molecular images obtained for the anions of m/z 187.0065 and 303.2331 for sample A2.

Figure 1.

Representative selections of ROI based on H&E staining and overlaid on the DESI-MSI images. A, A DCIS sample (sample A5) histologic evaluation and attribution of ROI in H&E staining; the black square selected in the figure is magnified and represents DCIS. B, Imported H&E image in Biomap software with the ROI defined. C, Overlaid H&E and DESI-MSI images. D, ROI defined in the H&E image overlaid on the DESI-MSI image. E, A luminal B sample (sample A2) histologic evaluation and attribution of ROI in H&E staining. The squares selected in the figure are magnified and represent IBC (yellow), ABT (benign; blue), and DCIS (black). F, The 2D molecular images obtained for the anions of m/z 187.0065 and 303.2331 for sample A2.

Close modal

Statistical analysis

The large amount of molecular data obtained for each sample hindered a direct interpretation of the data. Mass spectra .txt files obtained from ROIs were imported into the R environment for statistical analysis using the Lasso method (least absolute shrinkage and selection operator; refs. 20, 21). Forty-two samples analyzed by DESI-MSI (from sample A1 to A42) were analyzed by Lasso, the other 26 samples (from sample A43 to A68) were not included in the Lasso analyses because their mass spectra were obtained over a different m/z range. Lasso is a method of regression analysis with a regularization term, which leads to variable selection. Briefly, Lasso minimizes the usual sum of squared errors (as other regression methods for supervised learning); however, while performing this, it restrains the sum of the absolute values of the regression coefficients to be less than a fixed value. This constraint shrinks the coefficient estimates toward zero and forces some of them to be exactly equal to zero. Models generated from the Lasso are therefore simpler, sparse, and generally more accurate and much easier to interpret. Lasso regularization also avoids overfitting to data and selects variables that are most important to the model. Particularly, when applied to a classification problem, Lasso will select a subset of the variables that most discriminate the sample groups. In this study, Lasso was initially applied to m/z values of both invasive and benign tissues that were obtained from ROI. The resulting models contained sets of variables that were able to discriminate IBC from ABTs. Next, Lasso was applied to ratios of m/z values for invasive and ABTs to generate models that could discriminate the following groups: (i) IBC × DCIS and (ii) subtypes of breast tumors (e.g., luminal A, luminal B, triple-negative, and HER2+). Briefly, a mathematical weight for each statistically informative m/z value was calculated according to Lasso depending on the importance of the mass-spectral feature in characterizing a certain group. Features that were not statistically informative received a weight of zero and were disregarded. An anion whose peak height (abundance) was important to characterize a certain group received a positive weight value. In contrast, important ions with low abundances, or those that were absent, received a negative weight. Because the features selected according to Lasso can occur at a valley or a shoulder of an actual mass spectra peak, m/z ion peaks were selected by characterizing the nearest mass spectra peak to the statistically selected feature. The generated models were cross-validated according to the leave one out method, with each sample in the set being ignored during data modeling at least once.

Database searches

Each ion was selected on the basis of visual inspection of 2D molecular images and the Lasso method. Each ion was subsequently compared with available literature and identified on the basis of searches of lipid maps, metabolomics, human metabolome database data available at www.lipidmaps.org, www.metabolomicsworkbench.org, and www.hmdb.ca respectively, based on high mass accuracy and isotopic distribution considering a mass error of 5 ppm. When the isomerism of the double bonds in the fatty-acid chains of complex lipids, and isobaric species, resulted in more than one proposed structure, the information about the category to which they belonged was considered.

Metabolites identification

Selected ions were submitted, after ultra-high-performance liquid chromatography (UHPLC) separation, to higher energy collision-induced dissociation and compared the corresponding fragmentation profile with that of the standard from the human metabolome database (HMDB) to confirm the identity of this species. To each tissue sample, 400 μL of methanol (HPLC grade; Merck), were added and samples were vortexed for 3 minutes, followed by centrifugation at 13,000 rpm for 5 minutes. Samples were analyzed on a Q Exactive Orbitrap System (Thermo Fisher Scientific) connected to an Ultimate 3000 RS UHPLC chromatographic system (Dionex). Mobile phases were water + 0.1% formic acid (A) and acetonitrile (B; HPLC grade; Merck), and the column used was an Acquity HSS C18 2.1 × 100 mm, 1.8 μm (Waters). Separation was performed on a 6-minute gradient from 2% to 100% B, followed by cleaning and conditioning steps. Column (40°C) and samples (10°C) were kept at a constant temperature. The Orbitrap was run on MS Scans on both polarities, using the following parameters: capillary voltage +3.5/−2.5 kV; capillary temperature 275°C; sheath gas 55; auxiliary gas 15; spare gas 3; probe heater temperature 450° C; and S-Lens RF level 50. MS scans were acquired from m/z 50 to 750, with 70.000 resolution, automatic gain control target of 3e6 and maximum IT time of 200 ms. Whenever possible, collision energy was tuned for optimized fragmentation.

Comparison between areas of normal tissue versus tumor tissue

An inspection of the 2D molecular images that were collected from the 122 samples (67 tumors and 55 ABTs) resulted in the observation of more than 100 ions that delimited ROI. Figure 2 and Supplementary Fig. S1 (SI Appendix) show representative mass spectra that were extracted from ABT and IBC regions of two samples for each analyzed subtype. Figure 1E and F provides an example of the ROI that were identified with H&E staining and that corresponded with 2D maps of lipid and metabolite distribution for a sample of luminal B subtype (sample A2), and Supplementary Fig. S2 (SI Appendix) shows examples for each subtypes analyzed.

Figure 2.

A, Representative DESI-MSI data (m/z 100–1200) in the negative-ion mode obtained from the region delimited by the black square on H&E staining and overlaid on the DESI-MSI image of ABT (benign) from a HER2 subtype sample (sample A43); the 2D molecular image presented was obtained for the anion of m/z 187.0065. B, IBC regions from a HER2 subtype sample (sample A43); the 2D molecular image presented was obtained for the anion of m/z 303.2331.

Figure 2.

A, Representative DESI-MSI data (m/z 100–1200) in the negative-ion mode obtained from the region delimited by the black square on H&E staining and overlaid on the DESI-MSI image of ABT (benign) from a HER2 subtype sample (sample A43); the 2D molecular image presented was obtained for the anion of m/z 187.0065. B, IBC regions from a HER2 subtype sample (sample A43); the 2D molecular image presented was obtained for the anion of m/z 303.2331.

Close modal

The 2D ion images obtained from the DESI-MSI analyses performed exhibited high heterogeneity in their molecular distribution (Fig. 2; Supplementary Fig. S1, SI Appendix). The anions identified mainly included metabolites, xenobiotics, and deprotonated lipids (Supplementary Tables S2 and S3, SI Appendix). Most of the metabolites, such as glutamine (m/z 145.0612) and glutamate (m/z 146.0453), were present at higher relative abundances in the IBC regions than in the ABT regions, (Fig. 3; Supplementary Fig. S3, SI Appendix). Similarly, anions from antioxidant molecules, such as taurine (m/z 124.0068), uric acid (m/z 167.0210), ascorbic acid (m/z 175.0241), and glutathione (m/z 306.0765), were present at higher relative abundances in the IBC tissues (Fig. 3; Supplementary Fig. S3, SI Appendix). Meanwhile, in the ABT regions, the most abundant anions included caprylic acid (m/z 143.1072), palmitoleic acid (m/z 253.2173), oleic acid (m/z 281.2487), and some organosulfur compounds such as benzyl sulfate (m/z 187.0065) dodecylbenzenesulfonic acid (m/z 325.1841), and sulfated steroids (m/z 367.1585 and 395.1897). The anion of m/z 187.0065 was the one that presented the best capacity to delimit the ABT (Fig. 4; Supplementary Fig. S4, SI Appendix). Some anions from xenobiotics, including picric acid (m/z 281.2487) and pantoprazole (m/z 382.067; Fig. 5), were also observed at higher relative abundances in ABT. In the m/z 400–950 range, the spectra for the IBC regions exhibited greater complexity and higher abundances than the ABT regions (Fig. 2; Supplementary Fig. S1, SI Appendix). HER2 tumors presented a different profile, with much lower relative abundances for the IBC and ABT regions when compared with other subtypes (Fig. 2; Supplementary Fig. S1, SI Appendix). Polyunsaturated fatty acids (FA) were among the top abundant anions in the IBC tissues. These polyunsaturated FAs included deprotonated eicosanoids (e.g., arachidonic acid, m/z 303.2331), glycerophospholipids [GP; e.g., glycerophosphoinositol (PI; PI(38:4)), m/z 885.5497 and glycerophosphoserine (PS; PS(36:1)), m/z 788.5438], and sphingolipids [SP; e.g., Cer(d42:2), m/z 682.5906].

Figure 3.

The individual distribution profiles of 15 molecules mapped throughout a breast tissue containing luminal A IBC and DCIS (sample A67). A, The histologic evaluation and assignment of ROI was performed. The outlines selected in the figure are magnified and represent IBC (yellow) and DCIS (green). B, The 2D molecular images obtained for the anions of m/z 124.0068, 145.0612, 146.053, 167.0210, 175.0241, 187.0065, 248.9286, 279.2330, 303.2331, 306.0765, 600.5140, 682.5906, 788.5438, 794.5443, and 885.5497.

Figure 3.

The individual distribution profiles of 15 molecules mapped throughout a breast tissue containing luminal A IBC and DCIS (sample A67). A, The histologic evaluation and assignment of ROI was performed. The outlines selected in the figure are magnified and represent IBC (yellow) and DCIS (green). B, The 2D molecular images obtained for the anions of m/z 124.0068, 145.0612, 146.053, 167.0210, 175.0241, 187.0065, 248.9286, 279.2330, 303.2331, 306.0765, 600.5140, 682.5906, 788.5438, 794.5443, and 885.5497.

Close modal
Figure 4.

The 2D molecular image obtained for the ion of m/z 187.0065 in breast tissue containing DCIS, IBC, and ABT (benign) regions. A, The histologic evaluation and assignment of ROI performed for a DCIS sample with luminal B as invasive component (sample A2). B, The 2D molecular image obtained for the anion of m/z 187.0065 for the sample A2. C, The histologic evaluation and assignment of ROI performed for a DCIS sample (sample A48). D, The 2D molecular image obtained for the anion of m/z 187.0065 for the sample A48. The squares selected in the figure are magnified and represent DCIS (black), IBC (yellow), and ABT (blue). This anion presents high relative abundance in ABTs than in IBC in all subtypes analyzed.

Figure 4.

The 2D molecular image obtained for the ion of m/z 187.0065 in breast tissue containing DCIS, IBC, and ABT (benign) regions. A, The histologic evaluation and assignment of ROI performed for a DCIS sample with luminal B as invasive component (sample A2). B, The 2D molecular image obtained for the anion of m/z 187.0065 for the sample A2. C, The histologic evaluation and assignment of ROI performed for a DCIS sample (sample A48). D, The 2D molecular image obtained for the anion of m/z 187.0065 for the sample A48. The squares selected in the figure are magnified and represent DCIS (black), IBC (yellow), and ABT (blue). This anion presents high relative abundance in ABTs than in IBC in all subtypes analyzed.

Close modal
Figure 5.

The distribution profile of the molecule detected by the anion of m/z 382.067, mapped throughout a breast tissue containing luminal B IBC with ABT (sample A2; A); luminal A IBC (sample A66; B); HER2 IBC (sample A31; C); DCIS with ABT (sample A48; D); triple-negative IBC (sample A7; E). For each sample, its H&E staining is presented; the anion was assigned as pantoprazole and has higher relative abundance in ABTs.

Figure 5.

The distribution profile of the molecule detected by the anion of m/z 382.067, mapped throughout a breast tissue containing luminal B IBC with ABT (sample A2; A); luminal A IBC (sample A66; B); HER2 IBC (sample A31; C); DCIS with ABT (sample A48; D); triple-negative IBC (sample A7; E). For each sample, its H&E staining is presented; the anion was assigned as pantoprazole and has higher relative abundance in ABTs.

Close modal

It was further observed that unsaturated GPs, including PS, glycerophosphoethanolamine (PE), phosphatidylcholine (PC), PI, and glycerophosphoglycerol (PG), presented different degrees of unsaturation in different ROI. The amount of saturated FAs was relatively higher in the IBC regions than in the other ROI (Fig. 6; Supplementary Fig. S5, SI Appendix). Supplementary Table S4 (SI Appendix) lists some of the molecules that were less unsaturated and in which subtypes were observed.

Figure 6.

The 2D molecular images for nine ions of different m/z identified for a triple-negative sample (sample A7), which represent decreasing degrees of unsaturation for different categories of lipids. A, The histologic evaluation and assignment of ROI was performed. B, The 2D molecular images obtained for the anions of m/z 327.2330, 329.2490, 331.2640, 758.4990, 760.5124, 771.5179, 773.5328, 861.5500, and 863.5651. PS, phosphatidylserine; PG, phosphatidylglycerol; PI, phosphatidylinositol.

Figure 6.

The 2D molecular images for nine ions of different m/z identified for a triple-negative sample (sample A7), which represent decreasing degrees of unsaturation for different categories of lipids. A, The histologic evaluation and assignment of ROI was performed. B, The 2D molecular images obtained for the anions of m/z 327.2330, 329.2490, 331.2640, 758.4990, 760.5124, 771.5179, 773.5328, 861.5500, and 863.5651. PS, phosphatidylserine; PG, phosphatidylglycerol; PI, phosphatidylinositol.

Close modal

The Lasso method identified 18 ions with different m/z values in the m/z 100–800 m/z range as being important to the morphologic characteristics of the ROI identified and for the differentiation of IBC from ABT regions (Fig. 3; Supplementary Fig. S3; Supplementary Table S5 A, SI Appendix). The metabolites selected included glutamic acid (m/z 146.0453) and ascorbic acid (m/z 175.0241). The xenobiotics selected included 2-(4′-chlorophenyl)-3,3-dichloropropenoate (m/z 248.9286). The former two metabolites were more concentrated in IBC regions, whereas the latter was present in greater abundance in ABT regions. In the range of m/z 300–800, Lasso selected anions from unsaturated FAs including eicosanoid such as arachidonic acid (m/z 303.2331), sphingolipids such as Cer(d36:1) (m/z 600.5140) and glycerophospholipids such as PS(36:1) (m/z 788.5438). Figure 3 and Supplementary Fig. S3 present comparisons of ion images obtained with results obtained by Lasso.

Comparison between IBC and DCIS tissues

The results obtained from the DESI-MSI analyses of DCIS samples (68% ER+, 37% HER2+, and 6% triple-negative) showed lower abundances and complexity compared with the luminal B and triple-negative IBC subtypes and higher abundances and complexity compared with the IBC HER2 subtype. Lasso subsequently selected four ions of different m/z that were present at higher abundances in DCIS. These ions included FAs (m/z 341.2100 and 382.3736) and GPs (PE(P-16:0/22:6, m/z 746.5099 and PS(38:3), m/z 812.5440) (Supplementary Table S5B, SI Appendix). Figure 7 and Supplementary Fig. S6, SI Appendix, show the 2D molecular images for PE(P-16:0/22:6) m/z 746.5099 and PS(38:3) m/z 812.5440 in five DCIS samples.

Figure 7.

The 2D molecular images for a DCIS sample (A40) for two anions that were selected as important for classifying DCIS samples. A, The histologic evaluation and assignment of ROI was performed. The black square selected in the figure is magnified and represents DCIS. B, The 2D molecular images obtained for anions of m/z 746.5099 and 812.5440.

Figure 7.

The 2D molecular images for a DCIS sample (A40) for two anions that were selected as important for classifying DCIS samples. A, The histologic evaluation and assignment of ROI was performed. The black square selected in the figure is magnified and represents DCIS. B, The 2D molecular images obtained for anions of m/z 746.5099 and 812.5440.

Close modal

Molecular subtypes

The spectra of each molecular subtype exhibited distinct metabolomic profiles, with differences evident within the m/z 400–950 range (Fig. 2; Supplementary Fig. S1, SI Appendix). For example, the HER2 subtypes showed a much lower abundance of anions compared with the other subtypes (Fig. 2; Supplementary Fig. S1, SI Appendix). In contrast, the triple-negative and luminal B subtypes presented the highest relative abundance of anions, and this was mostly attributed to the presence of glycerolipids, GPs, and SPs (Supplementary Fig. S1, SI Appendix).

Lasso also selected 13 anions for the differentiation of molecular subtypes (Supplementary Table S6, SI Appendix). Five anions were selected as important for assigning a luminal A subtype. These included two anions with positive weights [linoleic acid, m/z 279.2330, and Cer(d40:1), m/z 656.5760] and three anions with negative weights (Unknown-3, m/z 107.9900 and GPs, m/z 765.5190 and m/z 772.5020). Four lipids with positive weights were selected as important for assigning the HER2 subtype. The anions were those of m/z 359.2720 (C24H39O2, assigned as steroid (ST) or FA); the PE, C43H77NO7P (m/z 751.5530); the PG, PG(36:3) (m/z 771.5280); and the PI, PI(36:4) (m/z 857.5380). For classification of the luminal B subtype, two GPs [PG(32:3), m/z 751.4860 and PI(36:2), m/z 861.5330] were selected, whereas classification of the triple-negative samples included ascorbic acid (m/z 175.0241) with a positive weight and the GP (m/z 751.4860) with a negative weight.

Metabolite identification

Supplementary Figure S7A (SI Appendix) shows the extracted ion chromatogram (m/z 384) for sample A2. Supplementary Fig. S7B shows the exact mass spectrum (electrospray in the positive-ion mode), from which the molecular formula was calculated as C16H16F2N3O4S, compatible with the protonated molecule [M + H]+ of pantoprazole. To confirm this hypothesis, this cation was fragmented to generate its MS/MS spectrum (Supplementary Fig. S7C). Comparison of this spectrum with the ones at the HMDB database confirmed the identity of the proposed structure of pantoprazole. Supplementary Fig. S8A shows the extracted ion chromatogram (m/z 187) for sample A2, whereas Supplementary Fig. S8B shows the MS spectrum of m/z 187 (electrospray in the negative-ion mode). As Supplementary Fig. S8B clearly shows, this ion coelutes with several interfering species, some of them being isobaric, which in that case did not allow the acquisition of a proper fragmentation spectrum. Nevertheless, data processing resulted in the molecular formula of C7H7O4S, corresponding to the deprotonated molecule of benzyl sulfate. Closer inspection in the isotopic pattern of this ion is also compatible with the presence of a sulfur atom in this species, whose identity should be confirmed by further experiments.

DESI-MSI is a fast and robust tissue-based imaging method that does not require fixation, staining, or any complex preparation protocol; however, it allows detection of numerous metabolites and their relative abundances. DESI-MSI has been able to distinguish benign from malignant metabolomic profiles (11) and was recently proven to be a reproducible technique for rapid breast cancer diagnosis (22). It has been applied to many types of tumors. In Supplementary Table S7 (SI Appendix), DESI-MSI data from thyroid (23), brain (18), gastric (14), pancreas (24), and prostate tumors are compared with our data, including, in bold and underlined, the lipids that presented different degrees of saturation in each type of tumor. Herein we demonstrate the capacity of DESI-MSI to differentiate molecular subtypes of breast cancer, and to provide a better understanding of the metabolism inherent to particular ROI in breast tissue. Two-dimensional molecular images (with a spatial resolution of 200 μm) were correlated with histology to provide valuable information regarding the metabolism of specific regions of IBC, DCIS, and distinct molecular subtypes of breast cancer. Differences in molecular profiles between DCIS and IBC were observed before morphologic changes become apparent in malignant cells. Because very few frozen DCIS samples are available from biobanks due to challenges in macroscopic identification, our understanding of their biology and behavior has been therefore incomplete (25). In 2008, Castro and colleagues showed that gene subsets are differentially expressed between pure DCIS and the in situ component of lesions that coexist with invasive ductal carcinoma (26). We examined 15 DCIS, with 5 pure DCIS and 10 DCIS with invasive component, the larger cohort using DESI-MSI to date.

Lipidomic profiling was able to distinguish DCIS from IBC. When Lasso was applied, anions from four molecules were selected for identification of DCIS, two FAs and two GPs. Among these FAs anions, the most important one selected by Lasso was from a docosanoid molecule whose signaling activity contributes to an inflammatory mechanism (27). Docosanoids are produced from the oxidation of polyunsaturated fatty acids by more than one pathway including the oxygenation via cyclooxygenases (COX; ref. 28), lipoxygenases (29), and cytochrome P450 monooxygenases. COX enzymes control a wide spectrum of processes (30) and many studies, from 1977 to the present, show the involvement of cyclooxygenase, including COX-2 and COX-1, in breast cancer (31–33). The two GPs are intercorrelated in mammalian cells and include a PE and a PS (34). This PS is primarily localized to the internal leaflet of the plasma membrane. In apoptotic cells, this PS translocates to the external leaflet, and this translocation is considered a crucial step in the recognition of apoptotic cells by macrophages (35). Taken together, these results suggest that mechanisms mediating signaling and apoptosis are more active in DCIS compared with IBC.

Reprogrammed activities for supporting cell survival under stressful conditions or allowing cells to grow and proliferate at abnormal levels included altered bioenergetics, enhanced biosynthesis, and redox balance (36). Changes in levels of some metabolites involved in these processes were revealed by DESI-MSI. The higher relative abundance of glutamine in IBC regions of luminal subtypes could be one example of enhanced biosynthesis and altered bioenergetics. Glutaminolysis is an important anaplerotic flux in cancer, which generates TCA cycle intermediates (36). The relative abundance of glutamine in the two more aggressive subtypes examined (e.g., HER2 and triple-negative) was lower than in the luminal subtypes, suggesting that more aggressive tumors are characterized by higher glutamine consumption, and this trend is consistent with a previous report that described low glutamine availability in aggressive tumor tissues (36). This data also may suggest that (i) these tumors could have other way to enhanced biosynthesis and obtain energy as described previously by Choi and colleagues; and (ii) protein expression of glycolysis markers such as Glut-1, CAIX, and MCT-4 was highest in HER2 and lower in luminal A and B (37). HER2 subtype is characterized by enhanced glycolytic metabolism and HER2‐mediated expression of glycolysis‐related genes (38), whereas less aggressive luminal subtypes appear to rely on a balance between de novo fatty acid synthesis and oxidation as sources for both biomass and energy requirements (39).

The HER2 tumors presented a singular lipid profile, which is consistent with data from previous transcriptomic and genomic studies, whereas HER2+ breast cancers have a molecular signature that distinguish these cancers from other types of breast cancer (2). This trend is also consistent with the observation that amplification of HER2 does not have a dramatic effect on lipid metabolism in breast cancer (40).

A higher relative abundance of antioxidant molecules was detected in the IBC when compared with ABT. These results are consistent with the model in which the elevated activity of tumor cells neutralizes the large amount of reactive oxygen species (ROS) that is generated in tumors (41). Cancer cells present higher level of ROS-scavenging enzymes than normal cells, such as glutathione peroxidase, preventing ROS-mediated activation of death-inducing pathway (36). Another possible strategy for the neutralization of ROS in cancer cells, which is also consistent with our data, is an increase of GP saturation degree. A higher concentration of saturated and monounsaturated GPs in membranes has shown the potential to confer extra protection from oxidative and chemotherapeutic damage induced by reducing lipid peroxidation (6). A higher concentration of lipids with higher degrees of saturation in the membrane can also facilitate the formation of relatively ordered regions (e.g., lipid rafts), which can recruit other specific lipids and proteins, and regulate many cellular processes, including immune signaling and host–pathogen interactions (42). A higher degree of saturation is also indicative of de novo lipogenesis. Indeed, higher levels of lipids in cancer and their precursors lesions have been found to unexpectedly undergo exacerbated endogenous fatty acid biosynthesis (43). The terminal catalytic step in FA synthesis is done by fatty acid synthase (FASN); its gene overexpression and increased activity are among the most frequent alterations in cancer cells (43).

Higher levels of lipids in cancer tissues have previously been described in DESI-MSI studies conducted with various types of cancer, including prostate (11), gastric (14), and breast cancers (16, 17). The higher level of lipids in cancer is also consistent with a usual cataplerotic reaction involving the Krebs cycle in tumor cells whereby the cycle is altered from producing citrate to contributing to lipogenesis (43). The various classes of lipids that have been identified have been associated with specific functions, including proinflammatory responses (eicosanoids; ref. 27), biological membrane composition (phospholipids), apoptotic mechanisms (PSs; ref. 44), and secondary messengers (PIs; ref. 45).

Some of the most abundant ions detected in benign breast tissues derive from organosulfur compounds, including benzyl sulfate (m/z 187.0063) and steroids sulfated (m/z 395.1897 and 465.3040; Fig. 4). And indeed, sulfation is an important process in the metabolism and inactivation of estrogen, and thus modulates the concentrations of active estrogens (46). The sulfation of estrogens is catalyzed by several members of a superfamily of cytosolic sulfotransferase (SULT) enzymes (47). Estrogen SULT activity has been demonstrated in a variety of steroid target tissues in human, including breast, and may well be important in affecting the biologic activity of estrogens within those tissues (48). Xenobiotics in ABT was also observed, including pantoprazole (m/z 382.0670). Higher levels of xenobiotic molecules are consistent with the contemporary view that adipose tissue is widely contaminated with various lipophilic xenobiotics, such as organochlorines, and these are agents that can act on hormones. Deposits of lipophilic xenobiotics in fat tissue have been observed in breast tissue, and they may contribute to the development of breast tumors (49).

Overall, DESI-MSI is an easy-to-use technique, comparison with pathologic slides is feasible and straightforward, but the analysis of the huge amount of data it generates can be rather complex. On the basis of our expertise with genomic analyses, we extrapolated metabolomic and lipidomic data analyses by using Lasso. However, improvements in the computational methods currently available for data acquisition and processing are needed to successfully translate this new technology into clinical practice.

In conclusion, we have demonstrated that data from both DESI-MSI and morphologic inspection can be combined, and that this approach can be used to obtain metabolomic profiles able to characterize normal parenchyma, DCIS, and various molecular subtypes of breast cancer. These findings also reveal a potential role for lipids in the development of selected molecular subtypes. For example, lipids that mediate signaling involved in inflammation and apoptotic mechanisms were found to have a key role in differentiating DCIS from IBC, whereas lipids with an overall higher degree of saturation and antioxidant molecules appeared to have essential roles in IBC. The results obtained for xenobiotic molecules were also promising, indicating that these molecules should be targets for further studies in breast cancer.

No potential conflicts of interest were disclosed.

Conception and design: A.L. Santoro, M.N. Eberlin, V.P. Andrade

Development of methodology: A.L. Santoro, I.T. Silva

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.L. Santoro, P.H. Vendramini, M.B.C. Lemos, V.P. Andrade

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A.L. Santoro, R.D. Drummond, I.T. Silva, M.B.C. Lemos, V.P. Andrade

Writing, review, and/or revision of the manuscript: A.L. Santoro, R.D. Drummond, L. Juliano, M.B.C. Lemos, M.N. Eberlin, V.P. Andrade

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A.L. Santoro, S.S. Ferreira, V.P. Andrade

Study supervision: L. Juliano, M.N. Eberlin, V.P. Andrade

The authors would like to thank the Biobank from the AC Camargo Cancer Center for providing breast cancer samples and sample preparation and Apex Science for technical assistance for MS validation.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1.
Heng
YJ
,
Lester
SC
,
Tse
GM
,
Factor
RE
,
Allison
KH
,
Collins
LC
, et al
The molecular basis of breast cancer pathological phenotypes
.
J Pathol
2017
;
241
:
375
91
.
2.
Perou
CM
,
Sørlie
T
,
Eisen
MB
,
Rijn
MV
,
Jeffrey
SS
,
Rees
CA
, et al
Molecular portraits of human breast tumours
.
Nature
2000
;
406
:
747
52
.
3.
Hanahan
D
,
Weinberg
RA
. 
The hallmarks of cancer: the next generation
.
Cell
2011
;
144
:
646
74
.
4.
Heiden
MGV
,
Cantley
LC
,
Thompson
CB
. 
Understanding the warburg effect: the metabolic requirements of cell proliferation
.
Science
2009
;
324
:
1029
33
.
5.
Röhrig
F
,
Schulze
A
. 
The multifaceted roles of fatty acid synthesis in cancer
.
Nat Rev Cancer
2016
;
16
:
732
49
.
6.
Rysman
E
,
Brusselmans
K
,
Scheys
K
,
Timmermans
L
,
Derua
R
,
Munck
S
, et al
De novo lipogenesis protects cancer cells from free radicals and chemotherapeutics by promoting membrane lipid saturation
.
Cancer Res
2010
;
70
:
8117
27
.
7.
Wishart
DS
. 
Is cancer a genetic disease or a metabolic disease?
EBioMedicine
2015
;
2
:
478
9
.
8.
Boroughs
LK
,
DeBerardinis
RJ
. 
Metabolic pathways promoting cancer cell survival and growth
.
Nat Cell Biol
2015
;
17
:
351
9
.
9.
DeBenardinis
RJ
,
Lum
JJ
,
Hatzivassiliou
G
,
Thompson
CB
. 
The biology of cancer: metabolic reprogramming fuels cell growth and proliferation
.
Cell Metab
2008
;
7
:
11
20
.
10.
Cai
Y
,
Crowther
J
,
Pastor
T
,
Abbasi
LA
,
Baietti
MF
,
De Troyer
M
, et al
Loss of chromosome 8p governs tumor progression and drug response by altering lipid metabolism
.
Cancer Cell
2016
;
29
:
751
66
.
11.
Banerjee
S
,
Zare
RN
,
Tibshirani
RJ
,
Kunder
CA
,
Nolley
R
,
Fan
R
, et al
Diagnosis of prostate cancer by desorption electrospray ionization mass spectrometric imaging of small metabolites and lipids
.
Proc Natl Acad Sci U S A
2017
;
114
:
3334
9
.
12.
Eberlin
LS
,
Ferreira
CL
,
Dill
AL
,
Ifa
DR
,
Cooks
RG
. 
Desorption electrospray ionization mass spectrometry for lipid characterization and biological tissue imaging
.
Biochim Biophys Acta
2011
;
1811
:
946
60
.
13.
Eberlin
LS
,
Norton
I
,
Orringer
D
,
Dunn
IF
,
Liu
X
,
Ide
JL
, et al
Ambient mass spectrometry for the intraoperative molecular diagnosis of human brain tumors
.
Proc Natl Acad Sci U S A
2013
;
110
:
1611
6
.
14.
Eberlin
LS
,
Tibshirani
RJ
,
Zhang
J
,
Longacre
TA
,
Berry
GJ
,
Bingham
DB
, et al
Molecular assessment of surgical-resection margins of gastric cancer by mass-spectrometric imaging
.
Proc Natl Acad Sci U S A
2014
;
111
:
2436
41
.
15.
Eberlin
LS
,
Gabay
M
,
Fan
AC
,
Gouw
AM
,
Tibshirani
RJ
,
Felsher
DW
, et al
Alteration of the lipid profile in lymphomas induced by MYC overexpression
.
Proc Natl Acad Sci U S A
2014
;
111
:
10450
55
.
16.
Calligaris
D
,
Caragacianu
D
,
Liu
X
,
Norton
I
,
Thompson
CJ
,
Richardson
AL
, et al
Application of desorption electrospray ionization mass spectrometry imaging in breast cancer margin analysis
.
Proc Natl Acad Sci U S A
2014
;
111
:
15184
9
.
17.
Guenther
S
,
Muirhead
LJ
,
Speller
AV
,
Golf
O
,
Strittmatter
N
,
Ramakrishnan
R
, et al
Spatially resolved metabolic phenotyping of breast cancer by desorption electrospray ionization mass spectrometry
.
Cancer Res
2015
;
75
:
1828
37
.
18.
Eberlin
LS
,
Norton
I
,
Dill
AL
,
Golby
AJ
,
Ligon
KL
,
Santagata
S
, et al
Classifying human brain tumors by lipid imaging with mass spectrometry
.
Cancer Res
2012
;
72
:
645
54
.
19.
Zhang
J
,
Rector
J
,
Lin
JQ
,
Young
JH
,
Sans
M
,
Katta
N
, et al
Nondestructive tissue analysis for ex vivo and in vivo cancer diagnosis using a handheld mass spectrometry system
.
Sci Transl Med
2018
;
9
:
1
11
.
20.
James
G
,
Witten
D
,
Hastie
T
,
Tibshirani
R
.
An introduction to statistical learning: with applications in R
.
New York
:
Springer-Verlag
; 
2014
. p.
203
64
.
21.
Tibshirani
R
. 
Regression shrinkage and selection via the lasso
.
J Royal Stat Soc Series B
1996
;
58
:
267
88
.
22.
Porcari
AM
,
Zhang
J
,
Garza
KY
,
Rodrigues-Peres
RM
,
Lin
JQ
,
Young
JH
, et al
A multi-center study using desorption electrospray ionization mass spectrometry imaging for breast cancer diagnosis
.
Anal Chem
2018
;
90
:
11324
32
.
23.
Zhang
J
,
Feider
CL
,
Nagi
C
,
Yu
W
,
Carter
SA
,
Suliburk
J
, et al
Detection of metastatic breast and thyroid cancer in lymph nodes by desorption electrospray ionization mass spectrometry imaging
.
J Am Soc Mass Spectrom
2017
;
28
:
1166
74
.
24.
Eberlin
LS
,
Margulis
K
,
Planell-Mendez
I
,
Zare
RN
,
Tibshirani
R
,
Longacre
TA
, et al
Pancreatic cancer surgical resection margins: molecular assessment by mass spectrometry imaging
.
PLoS Med
2016
;
13
:
e1002108
.
25.
Carraro
DM
,
Elias
EV
,
Andrade
VP
. 
Ductal carcinoma in situ of the breast: morphological and molecular features implicated in progression
.
Biosci Rep
2014
;
34
:
19
28
.
26.
Castro
NP
,
Osório
CABT
,
Torres
C
,
Bastos
EP
,
Neto
MM
,
Soares
FA
, et al
Evidence that molecular changes in cells occur before morphological alterations during the progression of breast ductal carcinoma
.
Breast Cancer Res
2008
;
10
:
R87
.
27.
Dennis
EA
,
Norris
PC
. 
Eicosanoid storm in infection and inflammation
.
Nat Rev Immunol
2015
;
15
:
511
23
.
28.
Smith
WL
,
DeWitt
DL
,
Garavito
RM
. 
Cyclooxygenases: structural, cellular, and molecular biology
.
Annu Rev Biochem
2000
;
69
:
145
82
.
29.
Kuhn
H
,
O'Donnell
VB
. 
Inflammation and immune regulation by 12/15- lipoxygenases
.
Prog Lipid Res
2006
;
45
:
334
56
.
30.
Lawrence
T
,
Willoughby
DA
,
Gilroy
DW
. 
Anti-inflammatory lipid mediators insights into the resolution of inflammation
.
Nat Rev Immunol
2002
;
2
:
787
95
.
31.
Brueggemeier
RW
,
Quinn
AL
,
Parrett
ML
,
Joarder
FS
,
Harris
RE
,
Robertson
FM
. 
Correlation of aromatase and cyclooxygenase gene expression in human breast cancer specimens
.
Cancer Lett
1999
;
140
:
27
.
32.
Bennett
A
,
Charlier
EM
,
McDonald
AM
,
Simpson
JS
,
Stamford
IF
,
Zebro
T
. 
Prostaglandins and breast cancer
.
Lancet
1977
;
2
:
624
.
33.
Hwang
D
,
Scollard
D
,
Byrne
J
,
Levine
E
. 
Expression of cyclooxygenase-1 and cyclooxygenase-2 in human breast cancer
.
J Natl Cancer Inst
1998
;
90
:
455
.
34.
Bleijerveld
OB
,
Houweling
M
,
Thomas
MJ
,
Cui
Z
. 
Metabolipidomics: profiling metabolism of glycerophospholipid species by stable isotopic precursors and tandem mass spectrometry
.
Anal Biochem
2006
;
352
:
1
14
.
35.
Mariño
G
,
Kroemer
G
. 
Mechanisms of apoptotic phosphatidylserine exposure
.
Cell Res
2013
;
23
:
1247
8
.
36.
DeBenardinis
RL
,
Chandel
NS
. 
Fundamentals of cancer metabolism
.
Sci Adv
2016
;
2
:
e1600200
.
37.
Choi
J
,
Kim
DH
,
Jung
WH
,
Koo
JS
. 
Metabolic interaction between cancer cells and stromal cells according to breast cancer molecular subtype
.
Breast Cancer Res
2013
;
15
:
R78
.
38.
Ferrari
A
,
Vincent-Salomon
A
,
Pivot
X
,
Sertier
AS
,
Thomas
E
, et al
A whole-genome sequence and transcriptome perspective on HER2- positive breast cancers
.
Nat Commun
2016
;
7
:
12222
.
39.
Monaco
ME
. 
Fatty acid metabolism in breast cancer subtypes
.
Oncotarget
2017
;
8
:
29487
500
.
40.
Hilvo
M
,
Denkert
C
,
Lehtinen
L
,
Müller
B
,
Brockmöller
S
,
Seppänen- Laakso
T
, et al
Novel theranostic opportunities offered by characterization of altered membrane lipid metabolism in breast cancer progression
.
Cancer Res
2011
;
71
:
3236
45
.
41.
Panieri
E
,
Santoro
MM
. 
ROS homeostasis and metabolism: a dangerous liason in cancer cells
.
Cell Death Dis
2016
;
7
:
e2253
.
42.
Sezgin
E
,
Levental
I
,
Mayor
S
,
Eggeling
C
. 
The mystery of membrane organization: composition, regulation and roles of lipid rafts
.
Nat Rev Mol Cell Biol
2017
;
18
;
361
74
.
43.
Menendez
JA
,
Lupu
R
. 
Fatty acid synthase and the lipogenic phenotype in cancer pathogenesis
.
Nat Rev Cancer
2007
;
7
:
763
77
.
44.
Fadok
VA
,
Voelker
DR
,
Campbell
PA
,
Cohen
JJ
,
Bratton
DL
,
Henson
PM
. 
Exposure of phosphatidylserine on the surface of apoptotic lymphocytes triggers specific recognition and removal by macrophages
.
J Immunol
1992
;
148
:
2207
16
.
45.
Vanhaesebroeck
B
,
Leevers S
J
,
Ahmadi
K
,
Timms
J
,
Katso
R
,
Driscoll
PC
, et al
Synthesis and function of 3-phosphorylated inositol lipids
.
Ann Rev Biochem
2001
;
70
:
535
602
.
46.
Falany
JL
,
Falany
CN
. 
Expression of cytosolic sulfotransferases in normal mammary epithelial cells and breast cancer cell lines
.
Cancer Res
1996
;
56
:
1551
5
.
47.
Weinshilboum
RM
,
Otterness
DM
,
Aksoy
IA
,
Wood
TC
,
Her
C
,
Raftogianis
RB
. 
Sulfation and sulfotransferases 1: sulfotransferase molecular biology: cDNAs and genes
.
FASEB J
1997
;
11
:
3
14
.
48.
Zhu
BT
,
Conney
AH
. 
Functional role of estrogen metabolism in target cells: review and perspectives
.
Carcinogenesis
1998
;
19
:
1
27
.
49.
Rudel
RA
,
Ackerman
JM
,
Attfield
KR
,
Brody
JG
. 
New exposure biomarkers as tools for breast cancer epidemiology, biomonitoring, and prevention: a systematic approach based on animal evidence
.
Environ Health Perspect
2014
;
122
:
881
95
.