Lipidomics-based tissue heterogeneity in specimens of luminal breast cancer revealed by clustering analysis of mass spectrometry imaging: A preliminary study

Abstract

Cancer tissues reflect a greater number of pathological characteristics of cancer compared to cancer cells, so the evaluation of cancer tissues can be effective in determining cancer treatment strategies. Mass spectrometry imaging (MSI) can evaluate cancer tissues and even identify molecules while preserving spatial information. Cluster analysis of cancer tissues’ MSI data is currently used to evaluate the phenotype heterogeneity of the tissues. Interestingly, it has been reported that phenotype heterogeneity does not always coincide with genotype heterogeneity in HER2-positive breast cancer. We thus investigated the phenotype heterogeneity of luminal breast cancer, which is generally known to have few gene mutations. As a result, we identified phenotype heterogeneity based on lipidomics in luminal breast cancer tissues. Clusters were composed of phosphatidylcholine (PC), triglycerides (TG), phosphatidylethanolamine, sphingomyelin, and ceramide. It was found that mainly the proportion of PC and TG correlated with the proportion of cancer and stroma on HE images. Furthermore, the number of carbons in these lipid class varied from cluster to cluster. This was consistent with the fact that enzymes that synthesize long-chain fatty acids are increased through cancer metabolism. It was then thought that clusters containing PCs with high carbon counts might reflect high malignancy. These results indicate that lipidomics-based phenotype heterogeneity could potentially be used to classify cancer for which genetic analysis alone is insufficient for classification.

Introduction

In vivo, cancer tissues are not simply cancer cells alone; rather, they are a mixture of cancer cells and surrounding cells. Although cellular and genomic analyses have been used to classify cancer cells in order to determine treatment strategies [1–4], cancer tissues have more complex characters than the cellular unit, and the ability to characterize cancer tissues in greater detail will enable more accurate cancer classification. Mass spectrometry imaging (MSI) is one of the techniques that can evaluate cancer tissues; it can identify and visualize molecules on tissue sections while maintaining the molecules’ positional information. MSI is also suitable for measuring the molecular weight of lipids and fatty acids, and MSI has shown that the expression of certain lipids and fatty acids is enhanced in cancer tissues, unlike normal tissues [5,6].

Lipid molecules are very diverse due to their vast combinations of carbons double bonds, and the wide variety of expressions of these lipid molecules plays an important role in reflecting the complex pathology of cancer tissues [7,8]. It has been reported that multivariate grouping (by a cluster analysis) of these data may be able to describe the heterogeneity of cancer tissues [9].

In breast cancer, the subtype determined by immunohistochemistry of estrogen receptor (ER), progesterone receptor (PgR), and human epidermal growth factor receptor type 2 (HER2) is one of the prognostic factors as well as a factor in the clinical stage, histological grade, and the marker Ki-67 [10–12]. The triple-negative (ER-, PgR-, and HER2-negative) subtype of breast cancer was reported to have a poor prognosis compared to luminal (ER-positive and/or PgR-negative and HER2-negative) breast cancer [11]. Additional prognostic factors among these subtypes are known, such as the recurrence score of Oncotype DX for luminal breast cancer [13]. It has also been suggested that changes in the metabolism of a wide variety of lipids in breast cancer (including phosphatidylcholine, phosphatidylethanolamine, sphingomyelin, and triglycerides) are associated with the tumor grade and prognosis [6,7].

We conducted the present study to investigate whether a cluster analysis of MSI data can represent heterogeneity in luminal breast cancer tissues based on the type of lipid molecules and their expression levels (lipidomics).

Materials and methods

Clinical materials

We collected tissue samples of 12 cases of invasive breast cancer surgeries from among the surgeries performed at Hamamatsu University School of Medicine between February 2013 and May 2017. Tissues of the primary lesions were collected at the time of surgery. All of the patients were Japanese women who were aged 42–69 years (mean age, 54 yrs). The primary lesions were histologically diagnosed as breast cancer by experienced pathologists. The tumor-node-metastasis (TNM) classifications were T1bN0M0–T2N0M0 (Stages 1–2A). The breast cancer specimens were all positive for immunohistochemical staining of estrogen receptor and progesterone receptor and negative for immunohistochemical staining of HER2.

The collected human breast cancer specimens were immediately frozen by liquid N2. All specimens were kept in a cryostat chamber (CM1950; Leica, Wetzler, Germany) at −20°C for approx. 30 min just before cryo-sectioning. Thereafter, 10-μm-thick sections were cut from the unfixed tissue specimens and mounted on uncoated glass slides (Matsunami, Osaka, Japan) for desorption electrospray ionisation (DESI)-MSI measurement. The same specimens were stained with hematoxylin and eosin stain (H&E) after MSI measurement. Based on the findings of the H&E-stained images, three of the 12 tissues were diagnosed as inadequate for evaluation in this study, and the remaining nine patients’ tissues were included in the analysis.

Breast cancer subtypes (luminal A and luminal B)

In this study, we defined ’luminal A’ as ER+ and/or PgR+, Ki-67 low and HER2-, and ’luminal B’ as ER+ and/or PgR+, Ki-67 high and HER2-. The samples with low Ki-67 values (<14%) are classified as luminal A, and those with high Ki-67 values were classified as luminal B. The 14% cut-off point for the Ki-67 values is derived from a comparison with gene array data as a prognostic factor [11].

DESI-MSI

In this study, DESI-MSI was used as the MSI technique. DESI-MSI does not require tissue pretreatment and can be performed at atmospheric pressure. It also provides the soft ionization and visualization of molecules in their original state. DESI-MSI data were acquired in positive ion mode at m/z range 100–1100 on a quadrupole time-of-flight mass spectrometer (Xevo G2-XS, Waters, Milford, MA, USA). The mass spectrometer was first calibrated with 500 mM sodium formate dissolved in 2-propanol: water (90:10, v/v) prior to measurement. The spray solvent (98% MeOH, 2% water) was delivered at 3 μL/min using a solvent pump. In this study, all samples were measured at the same time (4 March 2018). In this case, the results are not biased by the experimental equipment or measurement conditions, but the storage time of the samples may affect the results.

Optimization of the DESI ionization source was performed to obtain the maximum signal intensity from the tissues. using the following parameters: capillary voltage: 4.0 kV, source temperature: 120°C, nebulizing gas (nitrogen gas) pressure: 0.4 MPa, incidence angle of sprayer: 80°, inlet to sprayer distance: ~5 mm, inlet to surface distance: ~0.5 mm, and surface to sprayer distance: ~2 mm. After the optimization of the DESI source, samples were placed on the 2D stage and the data were acquired using the following parameters: scan speed: 300 μm/sec, pixel size: 100 μm, mass resolution: 22000, and mass window: 0.02 Da. The limit of detection of DESI was 1 μg/ml for phosphatidylcholine (16:00/16:00).

Data import and preprocess

The DESI-MSI data were acquired with MassLynx ver. 4.1 (Waters) and processed by HDImaging ver. 1.4 (Waters). All data were normalized by the total ion current (TIC) and converted to imzml data using the HDImaging software. Then, the imzml data were converted to IMDX format using the IMDX converter ver. 1.12.0.10258 (Shimadzu, Kyoto, Japan). The data were processed using the centroid method. Mass correction was processed with m/z 309.2036 ([C₁₆H₃₀O₄+Na]⁺) as a lock mass. This ion is detected daily as the background ion by our equipment. In the present case, this ion was detected as m/z 309.20303. The peak width was set up as 50 ppm. The converted data were imported into IMAGEREVEAL™ MS ver. 1.3 (Shimadzu) for further analysis.

Dimensionality reduction

We performed dimensionality reduction before K-means clustering. The most common method of dimensionality reduction is a principal component analysis (PCA) [14], but a nonlinear dimensionality-reduction technique, i.e., uniform manifold approximation and projection (UMAP) has recently been used [15]. In the present study, we performed dimensionality reduction with UMAP with the following parameters: n_neighbors  =  5, n_components  =  3, random_state  =  42 [16].

Multivariate analysis

K-means (for a non-hierarchical cluster analysis) was selected as the multivariate analysis mode [17,18]. K-means was performed with the number of clusters from 2 to 15. We evaluated the clustering by creating silhouette plots (python 3.10, scikit-learn 1.2.1) to check the validity of the clustering analysis in the range of 2 to 15 clusters (S1 Fig). Among clusters 2 to 15, peak silhouette coefficients were found at 4 and 13. Since the aim of this study was to evaluate the heterogeneity of the cancer tissue and to verify whether it could be applied clinically as a preliminary study, the number of clusters was set at 13 by comparison with HE-stained images. As shown in S2 Fig, the selection of 13 clusters confirms that even detailed pathological findings can be reflected.

Partial least squares (PLS)

We used the above-described IMAGEREVEAL™ MS software to perform a linear regression analysis, i.e., the partial least squares (PLS) method (also known as partial least squares regression). The PLS method extracts principal components that are related to the dependent variable in high-dimensional data. We set ROIs on the clusters. The PLS parameters were set as 1 (Y-value) ROI and 0 (Y-value) for the area excluding the ROI. In the present study, we focused on the 30 molecules with high PLS coefficients; since MS/MS measurements were not performed, we evaluated the 30 molecules whose expression matched those known to be upregulated in breast cancer in the previous reports [7,19–24].

Ethics statement

All procedures were carried out following relevant national and international guidelines, including the Declaration of Helsinki in its present form. This study was approved by the Ethics Committee of the Hamamatsu University School of Medicine (project identification code: 14–114). All patients were provided with sufficient information, and a written informed consent for their materials to be used was obtained from all patients before their surgery. All patients were recruited to the study between February 2013 and May 2017.

Data confidentiality

Only the authors affiliated with 1st Department of Surgery, Hamamatsu University School of Medicine have access to patient-identifiable information. The patient identifiable code table is stored in an environment that is not connected to the Internet.

Results

The results of the cluster analysis of the MSI data from the nine luminal breast cancer cases are presented in Fig 1. We found out that the lipidomics-based clusters were not only consistent with H&E findings, but also exhibited intra-tumor heterogeneity. Despite similar pathological findings on HE-stained images, the findings were classified into different clusters (Fig 2). In Fig 2A, heterogeneity in cancerous tissue by clusters 1 and 2 was observed. Fig 2B also shows heterogeneity in cancerous tissue by clusters 6 and 9. Compared to Fig 2A, the cancerous tissue included stroma tissue in Fig 2B. Fig 2C shows heterogeneity in cancerous and stroma tissue by clusters 7 and 8. Fig 2D shows heterogeneity in cancerous tissue by clusters 4 and 10. Compared to Fig 2A, the glandular duct structure was preserved in Fig 2D. Thus, heterogeneity was represented by different clusters in cases of subtle differences in pathology.

[Figure omitted. See PDF.]

The cluster image and H&E images of nine breast cancer tissues are shown. The cluster image consists of 13 color clusters. The similarity of colors has nothing to do with the nature of the cluster. The lengths of the scale bars are as follows. A: 2500 μm, B: 2500 μm, C: 2500 μm, D: 1250 μm, E: 2500 μm, F: 500 μm, G: 2500 μm, H: 1250 μm, and I: 1250 μm.

[Figure omitted. See PDF.]

The cancerous areas are circled with a white dotted line. a: The cancerous area is classified by clusters 1 and 2. b: The cancerous area is classified by clusters 6 and 9. c: The cancerous area is classified by clusters 7 and 8. d: The cancerous area is classified by clusters 4 and 10.

We then looked more closely at the pathological findings that corresponded to these clusters. The pathological findings of each cluster are shown in Table 1. This essentially showed that the clusters corresponded to the proportion of cancers and stroma. For example, cancer findings were predominant in clusters 1, 2, and 11, etc., whereas stroma was predominant in clusters 3, 5 and 7, etc. As the next step, we examined lipid molecules expressed in lipidomics-based clusters. The lipid molecules representing each cluster with the highest contribution were identified by the PLS analysis. They included the two main lipid classes (triglycerides (TG) and phosphatidylcholine (PC)), and the three sub lipid classes (phosphatidylethanolamine (PE), sphingomyelin (SM), ceramide (Cer)) (Table 2). Interestingly, clusters diagnosed as C(cancer) in the HE results (Table 1) were composed mainly of PCs, whereas clusters diagnosed as S(stroma) were composed mainly of TG. In addition, clusters diagnosed as C = S were a mixture of PC and TG. Representative findings of the HE-stained images corresponding to each lipid composition are shown in Fig 3. HE findings consistent with cluster 1, which is mainly composed of PC, presenting a dense finding of cancer cells (Fig 3A). HE findings consistent with cluster 4, which is a mixture of PC and TG, show an equal mix of cancer cells and stroma. HE findings consistent with cluster 5, which is predominantly TG, show a sparse presence of cancer cells in the stroma. In addition to PC and TG, PE (cluster 3, 8, 9 and 10), SM (cluster 2) and Cer (cluster 1, 6) were also characteristic lipid classes of the clusters. These clusters were also basically consistent with cancer findings. Furthermore, these lipid classes differed in the length of their side chains and the number of double bonds depending on the cluster. For example, in cluster 1, all PCs tended to have more carbons than the other clusters, with 36 or more carbons. Double bonds also tended to be more than 3, indicating a higher number of double bonds. Conversely, PCs in cluster 13 tended to have a low carbon number of 13. For TG, clusters 3, 4, and 5 had more than 50 carbons, while clusters 6 and 8 had less than 50 carbons. For clusters 7 and 12, molecules reported to be upregulated in breast cancer did not show high PLS coefficients.

[Figure omitted. See PDF.]

a. Cluster 1, which is mainly composed of PC, is consistent with a dense finding of cancer cells in the HE stained images. b. Cluster 4, which is a mixture of PC and TG, is consistent with an equal mix of cancer cells and stroma finding in the HE stained image. c. Cluster 5, which is predominantly TG, show a sparse presence of cancer cells in the stroma in the HE stained image.

[Figure omitted. See PDF.]

[Figure omitted. See PDF.]

Discussion

Heterogeneity of breast cancer is classified into genotype and phenotype [25]. Genotype heterogeneity has been studied using next-generation sequencing [26]. Although gene mutations in breast cancer are less common than in other cancers [27], various genetic tests have been applied in breast cancer. For example, the predictor analysis of microarray 50 (PAM50) test classifies intrinsic subtypes based on 50 genes [1]. The Oncotype DX test measures 21 genes to predict the prognosis and treatment response in luminal breast cancers [2,3]. The measurement of the tumor mutational burden (TMB) can predict the response to immunotherapy based on the amount of genetic mutations [4,28]. Genotype heterogeneity is thus known to be associated with malignancy and prognosis. However, no matter how advanced genomic methods become, the pathological approach to assessing cancer phenotypes will not become unnecessary, but rather increasingly important: phenotypes can reflect the consequences of genomic abnormalities, and it is important to note that such cancer-associated molecular abnormalities may be targets for novel malignancy drugs. Indeed, it has been observed that genotype heterogeneity does not always correlate with phenotype heterogeneity in breast cancer [25]. For example, although a correlation between genotype heterogeneity and prognosis has been reported for HER2-positive breast cancer [29], a study examining phenotype heterogeneity at the protein level with MSI reported an inverse correlation with prognosis [30].

We thus decided to use MSI to examine the phenotype heterogeneity of luminal breast cancer. The luminal breast cancer was indicated to have relatively low genotype heterogeneity in a previous study [1], but the present cluster images clearly show tissue heterogeneity at the lipid molecular level in the luminal breast cancer tissues (Figs 1 and 2). Our finding that phenotype heterogeneity at the lipid molecular level may exist even in luminal breast cancer with few genetic mutations.

The heterogeneity obtained in this study was based on the five lipid classes, i.e., PC, TG, PE, SM, Cer (Table 2). Abnormalities of lipid metabolism in cancer are well known, and a correlation between prognosis and lipids in breast cancer has been noted [30]. PC that composes the cell membrane of cancer cells are different from those of normal cells, since the synthesis of cell membranes occurs actively during cell proliferation. TG serves as a store of fatty acids as ingredient of cell membranes and energy in cancer tissues. PE, while constituting the cell membrane, is also involved in cell motility and cancer cell invasion and metastasis, like PC [31]. SM is the most abundant lipid in the outer layer of the plasma membrane and has been shown to promote cancer development and progression by regulating cell proliferation and migration capacity [30]. SM is degraded to Cer by acid sphingomyelinase (ASM) [32]. In comparison with HE findings, the ratio of PC to TG correlated with the ratio of cancer to stroma. To the best of our knowledge, there are no reports of papers in which the ratio of PC to TG was consistent with the pathological findings of HE-stained images (Fig 3). Based on this, we indicated that the dense areas of cancer cells reflect the abundance of PCs as a result of cancer cell proliferation (Fig 3A). And in areas with sparse cancer cells, the tendency to proliferate is weaker, reflecting the abundance of TG, which stores fatty acids as a source for cell membrane proliferation and energy (Fig 3C). The fact that the clusters with intermediate PC and TG proportions were also consistent with findings of only intermediate cancer and stroma supports these hypotheses (Fig 3B). In addition, not only the ratio of PC to TG, but also the number of carbons and double bonds of the lipid molecules in these lipid classes varied widely. It is known that in cancer, in addition to increased lipid metabolism, elongases [33] that increase the number of carbons and desaturases [5] that increase the number of double bonds are activated. For example, cluster 1 has a high number of carbons and double bonds in the PC and can be considered a cluster that represents highly malignant findings. Cluster 13 can be considered a relatively low malignancy cluster due to the low number of carbons and double bonds in the PC. For TGs, for example, cluster 5 would be underestimated under current assessment criteria because the findings are consistent with stroma and not cancer. However, given the high number of carbons and double bonds in TGs, they could potentially contribute to the malignant potential of cancerous tissue. It is interesting to note that cancer tissues show such a wide diversity in terms of location, suggesting the importance of studying the tissue rather than just the cancer cells.

PE, SM and Cer were also represented in clusters consistent with cancer findings, but no specific findings were found. Therefore, they can be considered to provide information not found in the HE findings. For example, loss of SM has been shown to enhance anti-tumor immunity via Th1 and cytotoxic T cells, and an increased SM/Cer ratio is known to cause cancer immune evasion [31]. Many lipid molecules might be involved in cancer immunity, and the number of reports concerning lipids and cancer immunity have increased in recent years. Among the types of breast cancer, immunotherapy has been applied to the triple-negative type [34], but in a randomized clinical trial of patients with luminal breast cancer, the addition of immunotherapy did not improve progression-free survival or overall survival [35]. It was suggested that this is because luminal breast cancers are associated with lower rates of programmed cell death ligand 1 (PD-L1) positivity [36], lower levels of TILs [37], and a lower median TMB [4]. However, clinical trials of bevacizumab, a humanized anti-VEGF (vascular endothelial growth factor) monoclonal antibody, in combination with immunotherapy are underway in anticipation of bevacizumab’s ability to suppress immune regulation (NCT04732598) [38]. Our present investigation revealed that there is lipidomics-based phenotype tissue heterogeneity even though luminal breast cancer has few gene mutations. If there is a type of luminal breast cancer that responds well to immunotherapy, immune-associated lipids may be related to the treatment response.

Future perspectives

During this study period, the Oncotype DX genetic test for luminal breast cancer was developed, allowing patient stratification. However, patient stratification in breast cancer remains a challenge, such as stratifying early recurrence groups in triple-negative types. Lehmann classification [39], claudin-low [40] based on genetic analysis contribute to the stratification of triple-negative breast cancer patients. For example, Lehman et al. classified triple-negative breast cancer into 7 subtypes (basal-like 1, basal-like 2, immunomodulatory, mesenchymal, mesenchymal stem-like, luminal androgen receptor, and unstable) and Masuda et al. reported that the pCR rates for neoadjuvant chemotherapy were different between these subtypes [41]. Although they contribute to stratification, they have not yet led to change clinical strategy for breast cancer treatments. We would like to apply the lipidomics-based phenotype heterogeneity assessment method developed in this preliminary study to triple-negative breast cancer to achieve stratification of early recurrence in triple-negative breast cancer, which is considered to have the worst prognosis in the breast cancer field.

Conclusion

This preliminary study demonstrated that clustering analysis of MSI data can be used to elucidate the lipidomics-based tissue heterogeneity of luminal breast cancer, which is known to have few gene mutations.

Supporting information

S1 Fig. Silhouette plot method.

The vertical axis is the silhouette coefficient and the horizontal axis is the number of clusters.

https://doi.org/10.1371/journal.pone.0283155.s001

(PPTX)

S2 Fig. Representative examples of clusters showing pathological trends.

a. The white dotted line corresponds to the area of cluster 1. Cluster 1 corresponds to the cancer findings in the H&E images. The black dotted line corresponds to the area of cluster 7. Cluster 7 corresponds to the stromal findings in the H&E images. b. The white dotted line corresponds to the area of cluster 2. Cluster 2 corresponds to the cancer findings in the H&E images. The black dotted line corresponds to the area of cluster 3. Cluster 3 corresponds to the stromal findings in the H&E images.

https://doi.org/10.1371/journal.pone.0283155.s002

(PPTX)

Acknowledgments

We acknowledge the contributions to the bioinformatics analysis by Manabu Machida from the Department of Systems Molecular Anatomy, Institute for Medical Photonics Research, Preeminent Medical Photonics Education & Research Center, Shizuoka, Japan.

Citation: Aramaki S, Tsuge S, Islam A, Eto F, Sakamoto T, Oyama S, et al. (2023) Lipidomics-based tissue heterogeneity in specimens of luminal breast cancer revealed by clustering analysis of mass spectrometry imaging: A preliminary study. PLoS ONE 18(5): e0283155. https://doi.org/10.1371/journal.pone.0283155

About the Authors:

Shuhei Aramaki

Roles: Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Validation, Visualization, Writing – original draft

Affiliations: Department of Cellular and Molecular Anatomy, Hamamatsu University School of Medicine, Hamamatsu, Shizuoka, Japan, Department of Radiation Oncology, Hamamatsu University School of Medicine, Hamamatsu, Shizuoka, Japan, First Department of Pathology, Hamamatsu University School of Medicine, Hamamatsu, Shizuoka, Japan

Shogo Tsuge

Roles: Conceptualization, Methodology, Validation, Visualization

Affiliation: Department of Cellular and Molecular Anatomy, Hamamatsu University School of Medicine, Hamamatsu, Shizuoka, Japan

Ariful Islam

Roles: Conceptualization, Data curation, Methodology

Affiliation: Department of Cellular and Molecular Anatomy, Hamamatsu University School of Medicine, Hamamatsu, Shizuoka, Japan

ORICD: https://orcid.org/0000-0001-7497-3540

Fumihiro Eto

Roles: Conceptualization, Data curation

Affiliation: Department of Cellular and Molecular Anatomy, Hamamatsu University School of Medicine, Hamamatsu, Shizuoka, Japan

Takumi Sakamoto

Roles: Data curation, Methodology, Validation

Affiliation: Department of Cellular and Molecular Anatomy, Hamamatsu University School of Medicine, Hamamatsu, Shizuoka, Japan

ORICD: https://orcid.org/0000-0003-2271-9975

Soho Oyama

Roles: Investigation, Visualization

Affiliation: Department of Cellular and Molecular Anatomy, Hamamatsu University School of Medicine, Hamamatsu, Shizuoka, Japan

Wenxin Li

Roles: Conceptualization, Investigation, Methodology

Affiliation: Department of Radiation Oncology, Hamamatsu University School of Medicine, Hamamatsu, Shizuoka, Japan

Chi Zhang

Roles: Conceptualization, Investigation, Writing – review & editing

Affiliation: Department of Cellular and Molecular Anatomy, Hamamatsu University School of Medicine, Hamamatsu, Shizuoka, Japan

Shinichi Yamaguchi

Roles: Data curation, Methodology, Software

Affiliation: Analytical & Measuring Instruments Division, Shimadzu Corporation, Kyoto, Japan

Daiki Takatsuka

Roles: Conceptualization, Resources, Writing – review & editing

Affiliation: 1st Department of Surgery, Hamamatsu University School of Medicine, Hamamatsu, Shizuoka, Japan

Yuko Hosokawa

Roles: Conceptualization, Data curation

Affiliation: 1st Department of Surgery, Hamamatsu University School of Medicine, Hamamatsu, Shizuoka, Japan

A. S. M. Waliullah

Roles: Writing – review & editing

Affiliation: Department of Cellular and Molecular Anatomy, Hamamatsu University School of Medicine, Hamamatsu, Shizuoka, Japan

ORICD: https://orcid.org/0000-0001-6778-379X

Yutaka Takahashi

Roles: Data curation, Methodology, Writing – review & editing

Affiliation: Department of Cellular and Molecular Anatomy, Hamamatsu University School of Medicine, Hamamatsu, Shizuoka, Japan

Kenji Kikushima

Roles: Supervision

Affiliation: Department of Cellular and Molecular Anatomy, Hamamatsu University School of Medicine, Hamamatsu, Shizuoka, Japan

ORICD: https://orcid.org/0000-0001-8674-9554

Tomohito Sato

Roles: Supervision

Affiliations: Department of Cellular and Molecular Anatomy, Hamamatsu University School of Medicine, Hamamatsu, Shizuoka, Japan, 1st Department of Surgery, Hamamatsu University School of Medicine, Hamamatsu, Shizuoka, Japan, International Mass Imaging Center, Hamamatsu University School of Medicine, Hamamatsu, Shizuoka, Japan

Kei Koizumi

Roles: Conceptualization, Supervision

Affiliation: 1st Department of Surgery, Hamamatsu University School of Medicine, Hamamatsu, Shizuoka, Japan

Hiroyuki Ogura

Roles: Supervision

Affiliation: 1st Department of Surgery, Hamamatsu University School of Medicine, Hamamatsu, Shizuoka, Japan

Tomoaki Kahyo

Roles: Conceptualization, Project administration, Supervision, Writing – review & editing

Affiliations: Department of Cellular and Molecular Anatomy, Hamamatsu University School of Medicine, Hamamatsu, Shizuoka, Japan, International Mass Imaging Center, Hamamatsu University School of Medicine, Hamamatsu, Shizuoka, Japan

Satoshi Baba

Roles: Data curation, Visualization

Affiliation: Department of Diagnostic Pathology, Hamamatsu University School of Medicine, Hamamatsu, Japan

Norihiko Shiiya

Roles: Supervision

Affiliation: 1st Department of Surgery, Hamamatsu University School of Medicine, Hamamatsu, Shizuoka, Japan

Haruhiko Sugimura

Roles: Supervision

Affiliation: First Department of Pathology, Hamamatsu University School of Medicine, Hamamatsu, Shizuoka, Japan

Katsumasa Nakamura

Roles: Conceptualization, Supervision, Writing – review & editing

Affiliation: Department of Radiation Oncology, Hamamatsu University School of Medicine, Hamamatsu, Shizuoka, Japan

Mitsutoshi Setou

Roles: Supervision

E-mail: [email protected]

Affiliations: Department of Cellular and Molecular Anatomy, Hamamatsu University School of Medicine, Hamamatsu, Shizuoka, Japan, International Mass Imaging Center, Hamamatsu University School of Medicine, Hamamatsu, Shizuoka, Japan

ORICD: https://orcid.org/0000-0002-4672-187X

References

1. Perou CM, Sørlie T, Eisen MB, van de Rijn M, Jeffrey SS, Rees CA, et al. Molecular portraits of human breast tumours. Nature. 2000;406(6797):747–52. pmid:10963602.

2. Geyer CE Jr, Tang G, Mamounas EP, Rastogi P, Paik S, Shak S, et al. 21-Gene assay as predictor of chemotherapy benefit in HER2-negative breast cancer. NPJ Breast Cancer. 2018;4:37. Epub 2018/11/21. pmid:30456299; PubMed Central PMCID: PMC6235896

3. McVeigh TP, Kerin MJ. Clinical use of the Oncotype DX genomic test to guide treatment decisions for patients with invasive breast cancer. Breast Cancer (Dove Med Press). 2017;9:393–400. Epub 2017/06/16. pmid:28615971; PubMed Central PMCID: PMC5459968.

4. Luen S, Virassamy B, Savas P, Salgado R, Loi S. The genomic landscape of breast cancer and its interaction with host immunity. Breast. 2016;29:241–50. pmid:27481651.

5. Ide Y, Waki M, Hayasaka T, Nishio T, Morita Y, Tanaka H, et al. Human breast cancer tissues contain abundant phosphatidylcholine(36∶1) with high stearoyl-CoA desaturase-1 expression. PLoS One. 2013;8(4):e61204. Epub 2013/04/25. pmid:23613812; PubMed Central PMCID: PMC3629004.

6. Hosokawa Y, Masaki N, Takei S, Horikawa M, Matsushita S, Sugiyama E, et al. Recurrent triple-negative breast cancer (TNBC) tissues contain a higher amount of phosphatidylcholine (32:1) than non-recurrent TNBC tissues. PLoS One. 2017;12(8):e0183724. Epub 2017/08/24. pmid:28832678; PubMed Central PMCID: PMC5568295.

7. Eiriksson FF, Nøhr MK, Costa M, Bödvarsdottir SK, Ögmundsdottir HM, Thorsteinsdottir M. Lipidomic study of cell lines reveals differences between breast cancer subtypes. PLoS One. 2020;15(4):e0231289. Epub 20200414. pmid:32287294; PubMed Central PMCID: PMC7156077.

8. Santoro AL, Drummond RD, Silva IT, Ferreira SS, Juliano L, Vendramini PH, et al. In Situ DESI-MSI Lipidomic Profiles of Breast Cancer Molecular Subtypes and Precursor Lesions. Cancer Res. 2020;80(6):1246–57. Epub 20200107. pmid:31911556.

9. Jones EA, van Remoortere A, van Zeijl RJ, Hogendoorn PC, Bovée JV, Deelder AM, et al. Multiple statistical analysis techniques corroborate intratumor heterogeneity in imaging mass spectrometry datasets of myxofibrosarcoma. PLoS One. 2011;6(9):e24913. Epub 2011/10/08. pmid:21980364; PubMed Central PMCID: PMC3183001.

10. Elston CW, Ellis IO. Pathological prognostic factors in breast cancer. I. The value of histological grade in breast cancer: experience from a large study with long-term follow-up. Histopathology. 1991;19(5):403–10. pmid:1757079.

11. Howlader N, Cronin KA, Kurian AW, Andridge R. Differences in Breast Cancer Survival by Molecular Subtypes in the United States. Cancer Epidemiol Biomarkers Prev. 2018;27(6):619–26. Epub 20180328. pmid:29593010.

12. Petrelli F, Viale G, Cabiddu M, Barni S. Prognostic value of different cut-off levels of Ki-67 in breast cancer: a systematic review and meta-analysis of 64,196 patients. Breast Cancer Res Treat. 2015;153(3):477–91. Epub 20150904. pmid:26341751.

13. Dowsett M, Cuzick J, Wale C, Forbes J, Mallon EA, Salter J, et al. Prediction of risk of distant recurrence using the 21-gene recurrence score in node-negative and node-positive postmenopausal patients with breast cancer treated with anastrozole or tamoxifen: a TransATAC study. J Clin Oncol. 2010;28(11):1829–34. Epub 20100308. pmid:20212256.

14. McCombie G, Staab D, Stoeckli M, Knochenmuss R. Spatial and spectral correlations in MALDI mass spectrometry images by clustering and multivariate analysis. Anal Chem. 2005;77(19):6118–24. Epub 2005/10/01. pmid:16194068.

15. Smets T, Verbeeck N, Claesen M, Asperger A, Griffioen G, Tousseyn T, et al. Evaluation of Distance Metrics and Spatial Autocorrelation in Uniform Manifold Approximation and Projection Applied to Mass Spectrometry Imaging Data. Anal Chem. 2019;91(9):5706–14. Epub 20190425. pmid:30986042.

16. Leland McInnes JH, Nathaniel Saul, and Lukas Großberger . UMAP: Uniform Manifold Approximation and Projection. Journal of Open Source Software. 2018;3(29).

17. Bednarz H, Roloff N, Niehaus K. Mass Spectrometry Imaging of the Spatial and Temporal Localization of Alkaloids in Nightshades. J Agric Food Chem. 2019;67(49):13470–7. Epub 20190723. pmid:31334645.

18. Sarkari S, Kaddi CD, Bennett RV, Fernandez FM, Wang MD. Comparison of clustering pipelines for the analysis of mass spectrometry imaging data. Annu Int Conf IEEE Eng Med Biol Soc. 2014;2014:4771–4. pmid:25571059.

19. Cífková E, Holčapek M, Lísa M, Vrána D, Gatěk J, Melichar B. Determination of lipidomic differences between human breast cancer and surrounding normal tissues using HILIC-HPLC/ESI-MS and multivariate data analysis. Anal Bioanal Chem. 2015;407(3):991–1002. Epub 20141029. pmid:25352274.

20. Silva CL, Perestrelo R, Sousa-Ferreira I, Capelinha F, Câmara JS, Petković M. Lipid biosignature of breast cancer tissues by matrix-assisted laser desorption/ionization time-of-flight mass spectrometry. Breast Cancer Res Treat. 2020;182(1):9–19. Epub 20200515. pmid:32415496.

21. 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(9):3236–45. Epub 20110317. pmid:21415164.

22. Dória ML, Cotrim CZ, Simões C, Macedo B, Domingues P, Domingues MR, et al. Lipidomic analysis of phospholipids from human mammary epithelial and breast cancer cell lines. J Cell Physiol. 2013;228(2):457–68. pmid:22767159.

23. Huang YC, Chung HH, Dutkiewicz EP, Chen CL, Hsieh HY, Chen BR, et al. Predicting Breast Cancer by Paper Spray Ion Mobility Spectrometry Mass Spectrometry and Machine Learning. Anal Chem. 2020;92(2):1653–7. Epub 2019/12/07. pmid:31809016.

24. Moro K, Kawaguchi T, Tsuchida J, Gabriel E, Qi Q, Yan L, et al. Ceramide species are elevated in human breast cancer and are associated with less aggressiveness. Oncotarget. 2018;9(28):19874–90. Epub 20180413. pmid:29731990; PubMed Central PMCID: PMC5929433.

25. Beca F, Polyak K. Intratumor Heterogeneity in Breast Cancer. Adv Exp Med Biol. 2016;882:169–89. pmid:26987535.

26. Yates LR, Gerstung M, Knappskog S, Desmedt C, Gundem G, Van Loo P, et al. Subclonal diversification of primary breast cancer revealed by multiregion sequencing. Nat Med. 2015;21(7):751–9. Epub 20150622. pmid:26099045; PubMed Central PMCID: PMC4500826.

27. Morris LG, Riaz N, Desrichard A, Şenbabaoğlu Y, Hakimi AA, Makarov V, et al. Pan-cancer analysis of intratumor heterogeneity as a prognostic determinant of survival. Oncotarget. 2016;7(9):10051–63. pmid:26840267; PubMed Central PMCID: PMC4891103.

28. Banerji S, Cibulskis K, Rangel-Escareno C, Brown KK, Carter SL, Frederick AM, et al. Sequence analysis of mutations and translocations across breast cancer subtypes. Nature. 2012;486(7403):405–9. Epub 20120620. pmid:22722202; PubMed Central PMCID: PMC4148686.

29. Shen T, Nitta H, Wei L, Parwani AV, Li Z. HER2 intratumoral heterogeneity is independently associated with distal metastasis and overall survival in HER2-positive breast carcinomas. Breast Cancer Res Treat. 2020;181(3):519–27. Epub 20200425. pmid:32335801.

30. Gawin M, Kurczyk A, Niemiec J, Stanek-Widera A, Grela-Wojewoda A, Adamczyk A, et al. Intra-Tumor Heterogeneity Revealed by Mass Spectrometry Imaging Is Associated with the Prognosis of Breast Cancer. Cancers (Basel). 2021;13(17). Epub 20210827. pmid:34503159; PubMed Central PMCID: PMC8431441.

31. Li H, Huang F, Fan L, Jiang Y, Wang X, Li J, et al. Phosphatidylethanolamine-binding protein 4 is associated with breast cancer metastasis through Src-mediated Akt tyrosine phosphorylation. Oncogene. 2014;33(37):4589–98. Epub 20131125. pmid:24276246.

32. Perry DK. Serine palmitoyltransferase: role in apoptotic de novo ceramide synthesis and other stress responses. Biochim Biophys Acta. 2002;1585(2–3):146–52. pmid:12531548.

33. Wang Y, Cai X, Zhang S, Cui M, Liu F, Sun B, et al. HBXIP up-regulates ACSL1 through activating transcriptional factor Sp1 in breast cancer. Biochem Biophys Res Commun. 2017;484(3):565–71. Epub 20170126. pmid:28132807.

34. Taniguchi M, Okazaki T. The role of sphingomyelin and sphingomyelin synthases in cell death, proliferation and migration-from cell and animal models to human disorders. Biochim Biophys Acta. 2014;1841(5):692–703. Epub 20131217. pmid:24355909.

35. Zheng K, Chen Z, Feng H, Chen Y, Zhang C, Yu J, et al. Sphingomyelin synthase 2 promotes an aggressive breast cancer phenotype by disrupting the homoeostasis of ceramide and sphingomyelin. Cell Death Dis. 2019;10(3):157. Epub 20190215. pmid:30770781; PubMed Central PMCID: PMC6377618.

36. Jung IY, Kim YY, Yu HS, Lee M, Kim S, Lee J. CRISPR/Cas9-Mediated Knockout of DGK Improves Antitumor Activities of Human T Cells. Cancer Res. 2018;78(16):4692–703. Epub 20180702. pmid:29967261.

37. Denkert C, von Minckwitz G, Darb-Esfahani S, Lederer B, Heppner BI, Weber KE, et al. Tumour-infiltrating lymphocytes and prognosis in different subtypes of breast cancer: a pooled analysis of 3771 patients treated with neoadjuvant therapy. Lancet Oncol. 2018;19(1):40–50. Epub 20171207. pmid:29233559.

38. Hegde PS, Wallin JJ, Mancao C. Predictive markers of anti-VEGF and emerging role of angiogenesis inhibitors as immunotherapeutics. Semin Cancer Biol. 2018;52(Pt 2):117–24. Epub 20171208. pmid:29229461.

39. Masuda H, Baggerly KA, Wang Y, Zhang Y, Gonzalez-Angulo AM, Meric-Bernstam F, et al. Differential response to neoadjuvant chemotherapy among 7 triple-negative breast cancer molecular subtypes. Clin Cancer Res. 2013;19(19):5533–40. Epub 20130815. pmid:23948975; PubMed Central PMCID: PMC3813597.

40. Prat A, Parker JS, Karginova O, Fan C, Livasy C, Herschkowitz JI, et al. Phenotypic and molecular characterization of the claudin-low intrinsic subtype of breast cancer. Breast Cancer Res. 2010;12(5):R68. Epub 20100902. pmid:20813035; PubMed Central PMCID: PMC3096954.

41. Lehmann BD, Bauer JA, Chen X, Sanders ME, Chakravarthy AB, Shyr Y, et al. Identification of human triple-negative breast cancer subtypes and preclinical models for selection of targeted therapies. J Clin Invest. 2011;121(7):2750–67. pmid:21633166; PubMed Central PMCID: PMC3127435.