Combining metabolomics and machine learning to discover biomarkers for early-stage breast cancer diagnosis

Nguyen Ky Anh; Anbok Lee; Nguyen Ky Phat; Nguyen Thi Hai Yen; Nguyen Quang Thu; Nguyen Tran Nam Tien; Ho-Sook Kim; Tae Hyun Kim; Dong Hyun Kim; Hee-Yeon Kim; Nguyen Phuoc Long

doi:10.1371/journal.pone.0311810

Abstract

There is an urgent need for better biomarkers for the detection of early-stage breast cancer. Utilizing untargeted metabolomics and lipidomics in conjunction with advanced data mining approaches for metabolism-centric biomarker discovery and validation may enhance the identification and validation of novel biomarkers for breast cancer screening. In this study, we employed a multimodal omics approach to identify and validate potential biomarkers capable of differentiating between patients with breast cancer and those with benign tumors. Our findings indicated that ether-linked phosphatidylcholine exhibited a significant difference between invasive ductal carcinoma and benign tumors, including cases with inconsistent mammography results. We observed alterations in numerous lipid species, including sphingomyelin, triacylglycerol, and free fatty acids, in the breast cancer group. Furthermore, we identified several dysregulated hydrophilic metabolites in breast cancer, such as glutamate, glycochenodeoxycholate, and dimethyluric acid. Through robust multivariate receiver operating characteristic analysis utilizing machine learning models, either linear support vector machines or random forest models, we successfully distinguished between cancerous and benign cases with promising outcomes. These results emphasize the potential of metabolic biomarkers to complement other criteria in breast cancer screening. Future studies are essential to further validate the metabolic biomarkers identified in our study and to develop assays for clinical applications.

Citation: Anh NK, Lee A, Phat NK, Yen NTH, Thu NQ, Tien NTN, et al. (2024) Combining metabolomics and machine learning to discover biomarkers for early-stage breast cancer diagnosis. PLoS ONE 19(10): e0311810. https://doi.org/10.1371/journal.pone.0311810

Editor: Anil Bhatia, University of California Riverside, UNITED STATES OF AMERICA

Received: May 3, 2024; Accepted: September 25, 2024; Published: October 21, 2024

Copyright: © 2024 Anh et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: All relevant data for this study are within the paper and its Supporting information files.

Funding: This work was supported by 2022 Inje University Busan Paik Hospital Research Grant. The funding organization did not participate in the study design, data acquisition, data analysis and interpretation, or the content of the manuscript.

Competing interests: The authors have declared that no competing interests exist.

Introduction

Breast cancer (BC) ranks among the most prevalent malignant neoplasms in women. The World Health Organization (WHO) reported alarming estimates of over 2.3 million BC diagnoses and 685,000 BC fatalities in 2020 [1]. Particularly concerning fact is that BC is the foremost cause of cancer-related deaths in women under the age of 45 years [2]. The elevated mortality rate associated with BC in numerous countries can be attributed to inadequacies in screening, early detection, and diagnosis [1]. The importance of screening and detecting BC at an initial stage cannot be overstated, as this is pivotal in enhancing treatment efficacy and decreasing mortality. Approaches ranging from imaging techniques to molecular biomarkers [3, 4], not only strive for precise diagnosis but also aim to classify BC subtypes, thereby guiding oncological decision-making [5]. For example, BC can be classified as preinvasive (ductal carcinoma in situ and lobular carcinoma in situ) and invasive (ductal carcinoma and lobular carcinoma) based on histological information, or molecular subtypes such as luminal A, luminal B and triple-negative BC based on immunohistochemistry information [6].

In the past several decades, mammography has emerged as a principal tool for BC screening and has contributed to a decline in mortality rates [7]. Furthermore, the incorporation of artificial intelligence and machine learning is garnering attention because of their potential to improve the accuracy of BC diagnosis based on mammography images [8]. According to the Breast Imaging Reporting and Data System (BI-RADS) lexicon, mammography results can be divided into 7 categories. Category 0 stands for incomplete information. Categories 1 to 3 are related to cancer negative, benign or probably benign. Category 4 represents cases with a likelihood of BC. Category 5 and 6 indicated highly suggestive and biopsy-proven of malignancy, respectively [9]. Despite proven clinical merits, this technique is hampered by a high false-positive rate, which can contribute to misdiagnosis [7]. Moreover, conventional classification methods do not adequately address the diverse clinical trajectories of individual cancer cases [5]. Consequently, molecular classifications employing cutting-edge, high-throughput technologies, such as multi-omics, are under investigation for their potential to enhance BC diagnosis [5].

Blood-based biomarkers have shown promise in the early detection and diagnosis of BC. Cancer antigen 15–3 (MUC-1 antigen) and carcinoembryonic antigen are two serum biomarkers that have been applied in clinical settings [10, 11]. However, the limited sensitivity and selectivity of these markers can result in misdiagnosis [10]. This underscores the need to identify novel biomarkers with high sensitivity and selectivity to improve BC diagnosis.

Metabolic rewiring is a hallmark of cancer and is closely associated with tumor initiation, progression, metastasis, and resistance to antineoplastic drugs [12]. Untargeted metabolomics and lipidomics utilizing liquid chromatography—tandem mass spectrometry (LC-MS/MS) have demonstrated immense potential in the discovery of novel biomarkers and the generation of hypotheses concerning metabolic alterations [13]. Accordingly, several studies have explored the metabolic and lipid profiles of BC patients using high-throughput LC-MS/MS [14–16]. For example, L-octanoylcarnitine, 5-oxoproline, hypoxanthine, and docosahexaenoic acid have been identified as potential plasma biomarkers for BC diagnosis [16]. Moreover, L-arginine and arachidonic acid could be used for both detecting BC and predicting the efficacy of trastuzumab [17]. In addition, lipids play a significant role in cell signaling processes, which are linked to membrane properties, metabolism, and the invasive and metastatic behavior of tumor cells [14, 18]. Therefore, metabolomics and lipidomics research is indispensable for the early detection, accurate diagnosis, prognosis, and treatment of BC [16, 19].

This study aimed to employ a multimodal omics approach in conjunction with machine learning models to identify and validate potential endogenous biomarkers that can differentiate the metabolic and lipid profiles of BC versus benign patients. Our findings highlighted several metabolites and lipids, particularly ether-linked phosphatidylcholine (PC(O-)), which exhibited significant alterations in BC compared with benign tumors. Additionally, we identified a metabolism-centric biosignature that exhibited good performance in cases where mammography yielded suboptimal results. The insights of this study will potentially serve as a foundation for the development of supplementary tools enhancing the effectiveness of mammography in the screening and early diagnosis of BC.

Materials and methods

Clinical samples and ethical approval

Patients recruited between January 1^st, 2019, and July 30^th, 2022, who had available clinical data along with plasma samples stored in the Inje University College of Medicine Biobank were included in this study. All participants provided written informed consent for the use of their clinical information and plasma samples for research purpose of the biobank. The study was conducted with the approval of the Institutional Review Board of Inje University College of Medicine Busan Paik Hospital (IRB No. 2022-08-052). The specimens were acquired from the biobank after the IRB approval on September 16^th, 2022. The study used information distributed from the biobank and identifiable information was not obtained.

Clinical characteristics of the patients and study design

The design and workflow of the study are illustrated in Fig 1. For the training set, aimed at the discovery of potential biomarkers, patients who had concordant preoperative mammography (Mammography BI-RADS Assessment Categories) findings and immunohistochemistry results obtained from biopsy procedures were included [20]. Specifically, the cancer group comprised 13 patients diagnosed with invasive ductal carcinoma who had category 6 mammography findings. The benign group included 18 patients with benign tumors and mammography findings categorized as 1 or 2. For the validation set, patients with discordant findings between the mammography category and immunohistochemistry results were included. This set consisted of five patients with invasive ductal carcinoma and seven cases with benign tumors. The available clinical characteristics provided by the biobank are presented in Table 1.

Download:

Fig 1. Schematic diagram representing the study design and computational workflow.

https://doi.org/10.1371/journal.pone.0311810.g001

Download:

Table 1. The clinical characteristics of the patients.

https://doi.org/10.1371/journal.pone.0311810.t001

Reagents and chemicals

LC-MS-grade solvents including water, acetonitrile, methanol (MeOH), and isopropanol were sourced from Merck KGaA (Darmstadt, Germany). Formic acid, ammonium formate, ammonium acetate, methyl tert-butyl ether (MTBE), and toluene were procured from Sigma-Aldrich (St. Louis, Missouri, USA). Internal standards (IS) for metabolomics, such as acetyl-L-carnitine-(N-methyl-d3), L-phenyl-d5-alanine, L-tryptophan-(indole-d5), leucine enkephalin, SM(d18:1/15:0)-d9, and cholic acid-2,2,3,4,4-d5, were also obtained from Sigma-Aldrich (St. Louis, Missouri, USA). For lipidomics analysis, the LIPIDOMIX^® Mass Spec Standard and Deuterated Ceramide Mass Spec Standard were acquired from Avanti Polar Lipids (Alabama, USA). The UPLC ACQUITY ethylene bridged hybrid (BEH) C18 column (100 mm × 2.1 mm, 1.7 μm) was used for metabolomics, while the UPLC ACQUITY BEH C18 column (50 mm × 2.1 mm; 1.7 μm), paired with a UPLC BEH C18 VanGuard pre-column (5 mm × 2.1 mm; 1.7 μm) (Waters, Milford, MA, USA), was employed for lipidomics analysis.

Sample preparation

For untargeted metabolomics, we employed the extraction protocol outlined in our previous study [21]. Samples were stored at -80°C prior to any experiment. Initially, 50 μL of the plasma samples were thawed on ice for 30 min and then briefly vortexed for 10 s. Next, 150 μL MeOH at −20°C, containing premixed IS was added to the plasma. The mixtures were vortexed for 30 s and centrifuged at 14,000 rcf and 4°C for 2 min. Thereafter, 150 μL of the supernatant was transferred to a new tube and evaporated completely under a stream of nitrogen gas at room temperature. For LC-MS/MS analysis, the dried extracts were reconstituted in 200 μL 50% MeOH and centrifuged for 2 min at 14,000 rcf and 4°C. Subsequently, 100 μL of the supernatant from each sample was allocated for analysis, while the remaining 50 μL was used to create pooled quality control (QC) samples.

For lipidomics, we utilized an MTBE-based biphasic extraction method, which was based on previously established protocols with minor modifications [21, 22]. Briefly, 5 μL of each lipid IS mix was spiked into every sample (50 μL) after thawing for approximately 30 min, followed by brief vortexing. The mixtures were then incubated on ice for 20 min, with intermittent vortexing. Subsequently, 300 μL of MeOH and 1,000 μL MTBE, both at −20°C were added to the samples. The mixtures were vortexed for 10 s and then shaken for 20 min at 1,200 rpm and 4°C. Following this, 250 μL water was added to the samples, and the samples were vortexed for 20 s. The samples were then centrifuged at 14,000 rcf and 4°C for 2 min, after which 500 μL of the upper phase was collected and transferred to a new tube. The lipid extracts were dried under a stream of nitrogen gas at room temperature and stored at −80°C until analysis. For reconstitution, the dried extracts were dissolved in 200 μL of a MeOH/toluene mixture (9:1, v/v). QC samples were generated by pooling 50 μL of each sample, and the rest of the sample was used for untargeted lipidomics analysis.

Data acquisition using LC-MS/MS

Data acquisition for both metabolomics and lipidomics analyses was performed using the Shimadzu Nexera LC (Kyoto, Japan) system coupled with the X500R Quadrupole Time-of-Flight mass spectrometer (SCIEX, MA, USA). The autosampler temperature was maintained at 4°C. For untargeted metabolomics, the gradient method, injection volumes, data acquisition, and MS/MS settings were selected in accordance with our previously established protocol [21]. For lipidomics, we utilized the rapid LC method for untargeted lipidomics as described by Cajka et al. [23]. The injection volume was set at 1 μL for the positive ion (ESI+) mode and 3 μL for the negative ion (ESI-) mode. Data Dependent Acquisition was employed for data acquisition, using the same MS/MS parameters as in our prior study [21]. Mass calibration was executed after every four injections for metabolomics and after eight injections for lipidomics via the X500R’s calibrant delivery system to ensure the quality of analysis.

Data processing and treatment

The raw mass spectrometry data files (.wiff) were processed using MS-DIAL version 4.9.0. The parameters for MS-DIAL were based on our previous study [21]. For metabolomics, the retention time was corrected using internal standards. Statistically significant hydrophilic metabolites were identified based on the established in-house library [21], the public MS-DIAL libraries, and IRCCS Istituto Giannina Gaslini-Mass Spectra Library [24]. The data were subsequently normalized using a locally weighted scatterplot smoothing (LOWESS) algorithm and the IS-normalization method. For lipidomics, lipids were annotated using MS-DIAL built-in library and Fiehn’s lab lipidomics library [25, 26]. The lipidomics data were also LOWESS-normalized.

Exploratory data analysis

The aligned data exported from MS-DIAL were analyzed using the MetaboAnalyst 5.0 platform and the MetaboAnalystR package version 3.2.0 to retain only the features with a missing rate of 50% or less [27, 28]. Missing values in these features were then imputed using the k-nearest neighbors algorithm. Features displaying a relative standard deviation exceeding 25% in the QC samples were excluded. Both metabolomic and lipidomic profiles were visualized employing principal component analysis (PCA), wherein the data were log-transformed and Pareto-scaled. To explore distinctions between the cancer and benign groups, a partial least squares—discriminant analysis (PLS-DA) was conducted. The performance of the PLS-DA model was evaluated via a five-fold cross-validation method, with Q² used to determine the optimal model. Unless specified otherwise, data visualization was performed using the ggplot2 package (version 3.4.1) in R version 4.2.2.

Statistical analysis

In this study, the training set was used to identify differential molecules (DMs) in BC and potential biomarkers differentiating between the cancer and benign groups. Linear models incorporating age and BMI adjustments (for DMs) and classical univariate receiver operating characteristic (ROC) curve analysis were conducted using MetaboAnalyst 5.0. The thresholds for statistical significance in the linear models were a p-value of 0.05 and a false discovery rate (FDR) of 0.25. Additionally, features with an area under the ROC curve (AUC) ≥ 0.7 and p-value < 0.05 were chosen as potential biomarker candidates. All potential biomarker candidates identified in the training set underwent validation using the validation set. Biomarkers that showed consistency between the training and validation sets were subjected to classical univariate ROC analysis. The AUC and p-value obtained from the models were used to assess the ability of the biomarkers to differentiate cancer from benign patients. Candidates demonstrating robust performance and consistent expression between the training and validation set were selected for univariate ROC analysis using the validation data. Subsequently, the best performers were chosen to create a single biosignature. The diagnostic potential of this biosignature was then assessed by a multivariate machine learning-based ROC model using the validation set. Linear support vector machine (SVM) and random forest algorithms, in conjunction with age and BMI as covariates, were implemented.

Results

Data exploration revealed subtle differences in plasma metabolic profiles between cancer and benign groups

PCA was performed on the metabolomics data to examine sample trends without considering the origins of the samples. In the PCA scores plots between patient’s and QC samples, the QC samples clustered indicating repeatability and satisfactory data acquisition process (S1A, S1B Fig in S1 File). The PCA scores plot without QC samples of the metabolomics data in ESI+ mode hinted at a slight distinction between the cancer and benign groups, whereas the PCA scores plot of ESI- mode did not display any apparent separation (Fig 2A and 2B). PCA was also applied to the lipidomics data in both the ESI+ and ESI- modes. The PCA scores plots with QC samples showed clustering of QC samples similar to metabolomics analysis (S1C, S1D Fig in S1 File). The PCA scores plot of clinical samples in ESI+ mode exhibited a general overlap between the cancer and benign groups (Fig 2C). In line with the results from ESI+ mode, no significant separation was evident between the cancer and benign groups in the PCA scores plot of ESI- mode (Fig 2D). Notably, the variance explained by PC1 and PC2 was considerably below 50%, indicating that the relationship among features was complex, and a linear model utilizing the complete profile might not effectively capture the biological variance between the two groups.

Download:

Fig 2. PCA scores plot visualizing metabolomic and lipidomic profiles in the breast cancer and benign tumor control groups.

A. Metabolomics positive ion mode. B. Metabolomics negative ion mode. C. Lipidomics positive ion mode. D. Lipidomics negative ion mode.

https://doi.org/10.1371/journal.pone.0311810.g002

Subsequently, PLS-DA was conducted using the metabolic profiles to discern differences between the cancer and benign groups. Analysis using the ESI+ mode metabolomics data showed that PLS-DA could not efficiently distinguish the two groups (accuracy = 0.757, R² = 0.944, Q² < 0, Fig 3A). In accordance with the ESI+ mode, PLS-DA of ESI- mode data failed to reliably differentiate the two groups (accuracy = 0.600, R² = 0.997, Q² < 0, Fig 3B). The models demonstrated unsatisfactory predictive performance, as indicated by the negative Q² values. When analyzing lipidomics data, PLS-DA scores plot in ESI+ mode revealed some separation in lipid profiles between the cancer and benign groups but with limited predictive accuracy (accuracy = 0.803, R² = 0.982, Q² = 0.246, Fig 3C), as did PLS-DA in ESI- mode (accuracy = 0.672, R² = 0.573, Q² = 0.111, Fig 3D). The results suggest that, similar to metabolomics analysis, PLS-DA in the lipidomics analysis also possessed limited predictive ability (Q² < 0.4).

Download:

Fig 3. PLS-DA scores plot visualizing metabolomic and lipidomic profiles in the breast cancer and benign tumor control groups.

A. Metabolomics positive ion mode. B. Metabolomics negative ion mode. C. Lipidomics positive ion mode. D. Lipidomics negative ion mode.

https://doi.org/10.1371/journal.pone.0311810.g003

In summary, the data exploration implied subtle differences in plasma metabolic profiles between the cancer and benign groups. However, both PCA and PLS-DA revealed that the distinctions were not pronounced, and the predictive performance was limited.

Univariate biomarker analysis identified potential biomarker candidates

A linear model, adjusted for age and BMI, was employed to identify molecules with significant differences between the cancer and benign groups. The analysis utilizing the linear model with metabolomics data revealed 86 significant features in ESI+ mode (9 upregulated and 77 downregulated in cancer) and 52 significant features in ESI- mode (30 upregulated and 22 downregulated in cancer), based on p < 0.05. Notably, only the ether-linked lysophosphatidylcholine (LPC) (O-22:2) remained significant after applying an adjusted p-value (FDR < 0.25). Using lipidomics data, the linear model analysis yielded 141 significant features in ESI+ mode (25 upregulated and 116 downregulated in cancer) and 76 significant features in ESI- mode (28 upregulated and 48 downregulated in cancer), with p < 0.05. Among these, 30 lipids in ESI+ mode and 11 in ESI- mode retained significance after adjusting the p-value (FDR < 0.25). The significant lipids were classified into subclasses including LPC(O-), PC(O-), sphingomyelin (SM), and free fatty acids (FA).

The results from the linear model, adjusted for age and BMI, were in harmony with the observations made in the exploratory data analysis, wherein only a handful of metabolites and lipids fulfilled the criteria for statistical significance. Consequently, univariate ROC analysis was employed to further assess the potential of plasma polar metabolites and lipids to distinguish between cancer and benign cases. In the metabolomics analysis, two annotated features in ESI+ mode and six in ESI- mode exhibited an AUC ≥ 0.7 for differentiating between the cancer and benign groups. Among these metabolites, deoxycholic acid glycine conjugate/glycoursodeoxycholic acid and LPC(O-22:2) demonstrated good performance (AUC > 0.8). In the lipidomics data analysis, a total of 53 lipids in ESI+ mode and 17 in ESI- mode achieved an AUC ≥ 0.7 (p < 0.05). Remarkably, PC(O-48:8) and PC(O-42:2) exhibited exceptional performance in differentiating between the cancer and benign groups (AUC > 0.9). Moreover, 14 lipids in ESI+ mode and 5 lipids in ESI- mode achieved an AUC > 0.8.

In summary, the univariate ROC analysis identified 73 potential biomarker candidates in BC (5 of which were detected in both ion modes), while the linear model detected 38 DMs in BC (4 of which were detected in both ion modes). The statistical attributes of these DMs and biomarker candidates are provided in S1 Table in S1 File.

External validation highlighted the potential of metabolism-centric biomarkers in aiding BC diagnosis

External validation was conducted to evaluate the potential diagnostic abilities of biomarkers identified in the training set, utilizing a distinct dataset. Initially, biomarker candidates with an AUC ≥ 0.7 and p < 0.05 in the univariate ROC analysis were chosen. In instances where candidates were detected in both ion modes, data from the ESI+ mode was selected. Subsequently, the expression direction of the biomarker candidates was compared between the training and validation datasets. A total of 61 candidates that demonstrated consistency in expression trends between both datasets were selected for further validation (S2 Fig in S1 File).

As part of the exploratory data analysis, PCA and PLS-DA were performed using the biomarker candidates to distinguish between cancer and benign cases within the validation set. Importantly, the PLS-DA scores plot displayed a distinct separation between the two groups along with strong predictive ability (accuracy = 0.933, R² = 0.996, Q² = 0.697, Fig 4A and 4B). This highlights the potential of the chosen biomarker candidates in differentiating between cancer and benign cases.

Download:

Fig 4. External validation of the metabolism-centric biosignature.

A. PCA scores plot. B. PLS-DA scores plot. C. Linear SVM ROC curve. D. Random Forest ROC curve. Abbreviations: CI, Confidence interval.

https://doi.org/10.1371/journal.pone.0311810.g004

Additionally, univariate ROC analysis was employed in the validation set to assess the classification performance of the biomarker candidates. Specifically, seven biomarker candidates demonstrated robust performance in differentiating between the cancer and benign groups and were thus chosen to create a composite signature (Table 2). Interestingly, though consistent as a biomarker, PC(O-) did not exhibit as of strong performance in differentiating between cancer and benign groups in the validation set as it did in the training set. This implies that individual biomarkers may not consistently deliver reliable results for precise BC diagnostics.

Download:

Table 2. The classic univariate ROC analysis on the validation set of significant biomarker candidates.

https://doi.org/10.1371/journal.pone.0311810.t002

Consequently, multivariate machine learning models, namely linear SVM and random forest, incorporating age, BMI, and the refined signature, were employed to classify BC cases in the validation set. Notably, the machine learning model exhibited exceptional performance in distinguishing between cancer and benign cases (Fig 4C and 4D). The multivariate ROC analysis employing linear SVM yielded an AUC of 0.996 (95% CI, 1.000–1.000), while the random forest model yielded an AUC of 0.985 (95% CI, 0.875–1.000).

These outcomes of the multivariate ROC analysis using machine learning models suggest that metabolism-centric biomarkers hold promise for enhancing the accuracy of BC screening and diagnosis.

Discussion

Early detection of BC is crucial for reducing patient mortality. However, mammography may not be sufficiently effective for the screening and accurate diagnosis of BC due to the inherent molecular heterogeneity of the disease [5]. Moreover, high-throughput technologies have provided multiple opportunities for biomarker discovery and validation, which can significantly enhance BC diagnosis and subtyping [5, 29]. Numerous studies have documented alterations in the metabolome and lipidome of BC patients, and these perturbations in small endogenous molecules are associated with the progression and metastatic potential of BC [30, 31]. Additionally, there is growing evidence supporting that multiple-marker biosignatures are inherently more robust and reliable in diverse clinical settings compared with single biomarkers [32]. Machine learning has also emerged as a pivotal tool in the research pipeline for biomarker discovery and validation [33]. Conventional statistical methods adequately characterize population interferences from a sample. On the other hand, machine learning can recognize potential predictive patterns [34]. Therefore, machine learning can empower exploratory omics-based biomarker studies for human diseases. In this study, we employed a multimodal omics data mining approach coupled with machine learning modeling to identify potential markers for early-stage BC detection.

Seven biomarkers for differentiating cancer and benign were confirmed by the validation process. These biomarkers exhibited outstanding performance in both the linear SVM and random forest models, with AUC values exceeding 0.9. This suggests significant promise of plasma metabolites in aiding the early-stage screening and diagnosis of BC. Among these, certain hydrophilic metabolites such as glutamate and glycochenodeoxycholate were altered in BC patients. For instance, glutamate levels were elevated in the cancer group compared with the benign group. Past research has indicated that accumulation of glutamate plays a pivotal role in energy provision, promotion of signaling pathways, and progression of tumors [35]. The elevated glutamate levels observed in our study may be partly attributed to the dysregulation of glutamine metabolism in cancer cells, particularly enhancement of glutaminolysis, which converts glutamine to glutamate [36]. Furthermore, we observed upregulation of glycochenodeoxycholate levels in BC patients relative to benign patients. This finding aligns with a previous study that reported increased bile acid concentrations in the serum of BC patients compared with healthy controls [37]. Collectively, our findings cohere with earlier reports on metabolic biomarkers in BC, supporting their potential role in BC screening and diagnosis.

The association between lipidomic alterations and BC invasiveness was evident in our study. Notably, PC(O-) exhibited significant differences between the invasive ductal carcinoma and benign tumor groups. The roles of PC(O-) in cancer have been well-documented [38]. It has been reported that ether lipids are involved in membrane trafficking and cell signalling, and are enriched in cancer cells [39]. For instance, some PC(O-) species have been linked to metabolic pathways that provide energy for cancer progression and activate oncogenic signaling pathways, promoting tumor growth [40]. Furthermore, interventional studies on human have suggested that increased circulated ether-linked phosphatidylcholine level could be a predictive biomarker for the progression of prostate cancer or colorectal cancer [41, 42]. Circulating lipid profiles also associated with treatment resistance in prostate cancer, further implicate the important role of lipids in cancer pathophysiology [43]. However, the univariate ROC analysis using the validation set showed subpar performance of PC(O-) compared with SM, which demonstrated better predictive ability. The type of mammography employed in the validation set could be related to this discrepancy. SM has been reported to participate in several intrinsic and extrinsic pathways that mediated cell proliferation and apoptosis via regulation of SM and ceramide balance [44, 45]. Of note, imbalance of SM and Ceramide could cause abnormal apoptotic activity that led to BC cell proliferation [46]. Additionally, alterations in SM metabolism have been correlated with tumor growth and drug resistance [47]. PC and SM lipid subclasses were altered in the plasma of BC patients compared with healthy controls [39]. It is important to note that the exact lipid species were not readily matched between the previous report and our study, possibly due to the difference of the study subjects, i.e., benign tumors rather than healthy controls. Moreover, we detected alterations in the plasma levels of other lipid subclasses, including triglycerides and free fatty acids. Changes in plasma TG levels have been detected in BC [48] and different stages of colorectal cancer [49].

This study has several limitations. First, the sample size was small, and various molecular subtypes were included within each group of interest. However, our focus was on early-stage BC profiles, which are critical for enhancing patient outcomes. The generalizability of our findings may be limited due to the small sample size and high heterogeneity of the molecular subtypes of BC patients. We tried to minimize the over-optimistic validation results by further filtering biomarker candidates (derived from the training set) using the concordance examination and the results of the univariate ROC analysis in the validation set. Then, only several of the most promising biomarker candidates, ranked based on the AUC of the ROC curve, were used for the machine learning model training and cross-validation in the validation set. However, the feature selection procedure was not conducted as part of the machine learning model training process, and the models were established and evaluated on the validation set; it is potentially subjected to selection bias [50, 51]. Second, the study was limited to one BC subtype, specifically invasive ductal carcinoma. This limits the generalizability of our findings to other BC subtypes. Third, the study only assessed relative changes in biomarker levels between cancer and benign patients, without providing quantitative measurements. The implementation of quantitative bioassays is necessary for a more comprehensive validation of the candidate biomarkers. Fourth, there was heterogeneity regarding metastatic status and neoadjuvant chemotherapy between the training and the validation cohort, which may affect the metabolome and lipidome of included patients. Subsequent studies are needed to validate our findings. Fifth, our study compared the early-stage BC patients to individuals with non-malignant tumors. The two groups had highly overlapping clinical manifestations, resulting in subtle differences in plasma metabolic profiles. Only one biomarker detected in the metabolomics analysis remained statistically significant after p-value adjustment. Therefore, pathway analysis for further biological interpretation was not conducted.

Conclusion

In conclusion, this study employed untargeted metabolomics and lipidomics analyses, coupled with robust feature selection and machine learning modeling, to identify and partially validate potential biomarkers aiding the screening and diagnosis of BC. While many studies on BC typically compared patients of all stages to healthy controls, our study differentiated patient samples to analyze differences between invasive BC and benign tumors. Furthermore, this study offers insights into alterations in hydrophilic and hydrophobic metabolites associated with BC, with a particular focus on the roles of PC(O-) and SM. Future research is imperative to thoroughly validate these biomarkers and to develop robust assay methods. Such biomarkers could serve as valuable tools to enhance screening and diagnosis of BC.

Supporting information

S1 File. Supplementary information of the study.

https://doi.org/10.1371/journal.pone.0311810.s001

(DOCX)

S1 Checklist. Human participants research checklist.

https://doi.org/10.1371/journal.pone.0311810.s005

(DOCX)

Acknowledgments

The biospecimens of benign and malignant breast tumors for this study were provided by Inje University, Paik Hospital Biobank. Graphical abstract and Fig 1 were created using biorender.com.

References

1. Arnold M, Morgan E, Rumgay H, Mafra A, Singh D, Laversanne M, et al. Current and future burden of breast cancer: Global statistics for 2020 and 2040. The Breast. 2022;66:15–23. pmid:36084384
- View Article
- PubMed/NCBI
- Google Scholar
2. Anastasiadi Z, Lianos GD, Ignatiadou E, Harissis HV, Mitsis M. Breast cancer in young women: an overview. Updates in Surgery. 2017;69(3):313–7. pmid:28260181
- View Article
- PubMed/NCBI
- Google Scholar
3. Jafari SH, Saadatpour Z, Salmaninejad A, Momeni F, Mokhtari M, Nahand JS, et al. Breast cancer diagnosis: Imaging techniques and biochemical markers. Journal of Cellular Physiology. 2018;233(7):5200–13. pmid:29219189
- View Article
- PubMed/NCBI
- Google Scholar
4. Moyya PD, Asaithambi M. Radiomics—Quantitative Biomarker Analysis for Breast Cancer Diagnosis and Prediction: A Review. Curr Med Imaging. 2022;18(1):3–17. Epub 2021/03/04. pmid:33655872.
- View Article
- PubMed/NCBI
- Google Scholar
5. Tsang JYS, Tse GM. Molecular Classification of Breast Cancer. Advances in Anatomic Pathology. 2020;27(1). pmid:31045583
- View Article
- PubMed/NCBI
- Google Scholar
6. Harbeck N, Penault-Llorca F, Cortes J, Gnant M, Houssami N, Poortmans P, et al. Breast cancer. Nature Reviews Disease Primers. 2019;5(1):66. pmid:31548545
- View Article
- PubMed/NCBI
- Google Scholar
7. Narayan AK, Lee CI, Lehman CD. Screening for Breast Cancer. Med Clin North Am. 2020;104(6):1007–21. Epub 2020/10/26. pmid:33099447.
- View Article
- PubMed/NCBI
- Google Scholar
8. Rodriguez-Ruiz A, Lång K, Gubern-Merida A, Broeders M, Gennaro G, Clauser P, et al. Stand-Alone Artificial Intelligence for Breast Cancer Detection in Mammography: Comparison With 101 Radiologists. JNCI: Journal of the National Cancer Institute. 2019;111(9):916–22. pmid:30834436
- View Article
- PubMed/NCBI
- Google Scholar
9. D’Orsi CJ, Sickles EA, Mendelson EB, Morris EA, et al. ACR BI-RADS^® Atlas, Breast Imaging Reporting and Data System. Reston, VA, American College of Radiology. 2013.
10. Lee JS, Park S, Park JM, Cho JH, Kim SI, Park B-W. Elevated levels of serum tumor markers CA 15–3 and CEA are prognostic factors for diagnosis of metastatic breast cancers. Breast Cancer Research and Treatment. 2013;141(3):477–84. pmid:24072270
- View Article
- PubMed/NCBI
- Google Scholar
11. Di Gioia D, Stieber P, Schmidt GP, Nagel D, Heinemann V, Baur-Melnyk A. Early detection of metastatic disease in asymptomatic breast cancer patients with whole-body imaging and defined tumour marker increase. British Journal of Cancer. 2015;112(5):809–18. pmid:25647014
- View Article
- PubMed/NCBI
- Google Scholar
12. Danzi F, Pacchiana R, Mafficini A, Scupoli MT, Scarpa A, Donadelli M, et al. To metabolomics and beyond: a technological portfolio to investigate cancer metabolism. Signal Transduction and Targeted Therapy. 2023;8(1):137. pmid:36949046
- View Article
- PubMed/NCBI
- Google Scholar
13. Alonso A, Marsal S, Julià A. Analytical Methods in Untargeted Metabolomics: State of the Art in 2015. Frontiers in Bioengineering and Biotechnology. 2015;3. pmid:25798438
- View Article
- PubMed/NCBI
- Google Scholar
14. Agarwala PK, Aneja R, Kapoor S. Lipidomic landscape in cancer: Actionable insights for membrane-based therapy and diagnoses. Medicinal Research Reviews. 2022;42(2):983–1018. pmid:34719798
- View Article
- PubMed/NCBI
- Google Scholar
15. Jasbi P, Wang D, Cheng SL, Fei Q, Cui JY, Liu L, et al. Breast cancer detection using targeted plasma metabolomics. Journal of Chromatography B. 2019;1105:26–37. pmid:30562627
- View Article
- PubMed/NCBI
- Google Scholar
16. Park J, Shin Y, Kim TH, Kim D-H, Lee A. Plasma metabolites as possible biomarkers for diagnosis of breast cancer. PLOS ONE. 2019;14(12):e0225129. pmid:31794572
- View Article
- PubMed/NCBI
- Google Scholar
17. Mao C, Wang M, Li L, Tang J-H. Circulating metabolites serve as diagnostic biomarkers for HER2-positive breast cancer and have predictive value for trastuzumab therapy outcomes. Journal of Clinical Laboratory Analysis. 2022;36(2):e24212. pmid:34994982
- View Article
- PubMed/NCBI
- Google Scholar
18. Beloribi-Djefaflia S, Vasseur S, Guillaumond F. Lipid metabolic reprogramming in cancer cells. Oncogenesis. 2016;5(1):e189–e. pmid:26807644
- View Article
- PubMed/NCBI
- Google Scholar
19. Wei Y, Jasbi P, Shi X, Turner C, Hrovat J, Liu L, et al. Early Breast Cancer Detection Using Untargeted and Targeted Metabolomics. Journal of Proteome Research. 2021;20(6):3124–33. pmid:34033488
- View Article
- PubMed/NCBI
- Google Scholar
20. He P, Cui L-G, Chen W, Yang R-L. Subcategorization of Ultrasonographic BI-RADS Category 4: Assessment of Diagnostic Accuracy in Diagnosing Breast Lesions and Influence of Clinical Factors on Positive Predictive Value. Ultrasound in Medicine & Biology. 2019;45(5):1253–8. pmid:30799123
- View Article
- PubMed/NCBI
- Google Scholar
21. Yen NTH, Anh NK, Jayanti RP, Phat NK, Vu DH, Ghim J-L, et al. Multimodal plasma metabolomics and lipidomics in elucidating metabolic perturbations in tuberculosis patients with concurrent type 2 diabetes. Biochimie. 2023;211:153–63. pmid:37062470
- View Article
- PubMed/NCBI
- Google Scholar
22. Barupal DK, Fan S, Wancewicz B, Cajka T, Sa M, Showalter MR, et al. Generation and quality control of lipidomics data for the alzheimer’s disease neuroimaging initiative cohort. Sci Data. 2018;5:180263. Epub 2018/11/21. pmid:30457571.
- View Article
- PubMed/NCBI
- Google Scholar
23. Cajka T, Hricko J, Rudl Kulhava L, Paucova M, Novakova M, Kuda O. Optimization of Mobile Phase Modifiers for Fast LC-MS-Based Untargeted Metabolomics and Lipidomics. International Journal of Molecular Sciences [Internet]. 2023; 24(3). pmid:36768308
- View Article
- PubMed/NCBI
- Google Scholar
24. Lavarello C, Barco S, Bartolucci M, Panfoli I, Magi E, Tripodi G, et al. Development of an Accurate Mass Retention Time Database for Untargeted Metabolomic Analysis and Its Application to Plasma and Urine Pediatric Samples. Molecules. 2021;26(14). Epub 2021/07/25. pmid:34299531.
- View Article
- PubMed/NCBI
- Google Scholar
25. Kind T, Liu KH, Lee DY, DeFelice B, Meissen JK, Fiehn O. LipidBlast in silico tandem mass spectrometry database for lipid identification. Nat Methods. 2013;10(8):755–8. Epub 2013/07/03. pmid:23817071.
- View Article
- PubMed/NCBI
- Google Scholar
26. Tsugawa H, Ikeda K, Takahashi M, Satoh A, Mori Y, Uchino H, et al. A lipidome atlas in MS-DIAL 4. Nat Biotechnol. 2020;38(10):1159–63. Epub 2020/06/17. pmid:32541957.
- View Article
- PubMed/NCBI
- Google Scholar
27. Pang Z, Chong J, Zhou G, de Lima Morais DA, Chang L, Barrette M, et al. MetaboAnalyst 5.0: narrowing the gap between raw spectra and functional insights. Nucleic Acids Res. 2021;49(W1):W388–W96. Epub 2021/05/22. pmid:34019663.
- View Article
- PubMed/NCBI
- Google Scholar
28. Pang Z, Chong J, Li S, Xia J. MetaboAnalystR 3.0: Toward an Optimized Workflow for Global Metabolomics. Metabolites. 2020;10(5). Epub 2020/05/13. pmid:32392884.
- View Article
- PubMed/NCBI
- Google Scholar
29. Claudino WM, Quattrone A, Biganzoli L, Pestrin M, Bertini I, Di Leo A. Metabolomics: Available Results, Current Research Projects in Breast Cancer, and Future Applications. Journal of Clinical Oncology. 2007;25(19):2840–6. pmid:17502626
- View Article
- PubMed/NCBI
- Google Scholar
30. Mrowiec K, Kurczyk A, Jelonek K, Debik J, Giskeødegård GF, Bathen TF, et al. Association of serum metabolome profile with the risk of breast cancer in participants of the HUNT2 study. Frontiers in Oncology. 2023;13. pmid:37007110
- View Article
- PubMed/NCBI
- Google Scholar
31. His M, Gunter MJ, Keski-Rahkonen P, Rinaldi S. Application of Metabolomics to Epidemiologic Studies of Breast Cancer: New Perspectives for Etiology and Prevention. Journal of Clinical Oncology. 2023;42(1):103–15. pmid:37944067
- View Article
- PubMed/NCBI
- Google Scholar
32. Pedersen JL, Barry SE, Bokil NJ, Ellis M, Yang Y, Guan G, et al. High sensitivity and specificity of a 5-analyte protein and microRNA biosignature for identification of active tuberculosis. Clinical & Translational Immunology. 2021;10(6):e1298. pmid:34188917
- View Article
- PubMed/NCBI
- Google Scholar
33. Xu C, Jackson SA. Machine learning and complex biological data. Genome Biology. 2019;20(1):76. pmid:30992073
- View Article
- PubMed/NCBI
- Google Scholar
34. Bzdok D, Altman N, Krzywinski M. Statistics versus machine learning. Nature Methods. 2018;15(4):233–4. pmid:30100822
- View Article
- PubMed/NCBI
- Google Scholar
35. Budczies J, Pfitzner BM, Györffy B, Winzer K-J, Radke C, Dietel M, et al. Glutamate enrichment as new diagnostic opportunity in breast cancer. International Journal of Cancer. 2015;136(7):1619–28. pmid:25155347
- View Article
- PubMed/NCBI
- Google Scholar
36. Altman BJ, Stine ZE, Dang CV. From Krebs to clinic: glutamine metabolism to cancer therapy. Nature Reviews Cancer. 2016;16(10):619–34. pmid:27492215
- View Article
- PubMed/NCBI
- Google Scholar
37. Li L, Zheng X, Zhou Q, Villanueva N, Nian W, Liu X, et al. Metabolomics-Based Discovery of Molecular Signatures for Triple Negative Breast Cancer in Asian Female Population. Scientific Reports. 2020;10(1):370. pmid:31941951
- View Article
- PubMed/NCBI
- Google Scholar
38. Messias MCF, Mecatti GC, Priolli DG, de Oliveira Carvalho P. Plasmalogen lipids: functional mechanism and their involvement in gastrointestinal cancer. Lipids in Health and Disease. 2018;17(1):41. pmid:29514688
- View Article
- PubMed/NCBI
- Google Scholar
39. Skotland T, Kavaliauskiene S, Sandvig K. The role of lipid species in membranes and cancer-related changes. Cancer and Metastasis Reviews. 2020;39(2):343–60. pmid:32314087
- View Article
- PubMed/NCBI
- Google Scholar
40. Benjamin DI, Cozzo A, Ji X, Roberts LS, Louie SM, Mulvihill MM, et al. Ether lipid generating enzyme AGPS alters the balance of structural and signaling lipids to fuel cancer pathogenicity. Proc Natl Acad Sci U S A. 2013;110(37):14912–7. Epub 2013/08/28. pmid:23980144.
- View Article
- PubMed/NCBI
- Google Scholar
41. Patel N, Vogel R, Chandra-Kuntal K, Glasgow W, Kelavkar U. A Novel Three Serum Phospholipid Panel Differentiates Normal Individuals from Those with Prostate Cancer. PLOS ONE. 2014;9(3):e88841. pmid:24603597
- View Article
- PubMed/NCBI
- Google Scholar
42. Zhao Z, Xiao Y, Elson P, Tan H, Plummer SJ, Berk M, et al. Plasma Lysophosphatidylcholine Levels: Potential Biomarkers for Colorectal Cancer. Journal of Clinical Oncology. 2007;25(19):2696–701. pmid:17602074
- View Article
- PubMed/NCBI
- Google Scholar
43. Lin H-M, Yang X, Centenera MM, Huynh K, Giles C, Dehairs J, et al. Circulating Lipid Profiles Associated With Resistance to Androgen Deprivation Therapy in Localized Prostate Cancer. JCO Precision Oncology. 2024;(8):e2400260. pmid:39074346
- View Article
- PubMed/NCBI
- Google Scholar
44. Hannun YA, Obeid LM. Principles of bioactive lipid signalling: lessons from sphingolipids. Nature Reviews Molecular Cell Biology. 2008;9(2):139–50. pmid:18216770
- View Article
- PubMed/NCBI
- Google Scholar
45. Zalba S, ten Hagen TLM. Cell membrane modulation as adjuvant in cancer therapy. Cancer Treatment Reviews. 2017;52:48–57. pmid:27889637
- View Article
- PubMed/NCBI
- Google Scholar
46. Li Y, Dong J, Ding T, Kuo M-S, Cao G, Jiang X-C, et al. Sphingomyelin Synthase 2 Activity and Liver Steatosis. Arteriosclerosis, Thrombosis, and Vascular Biology. 2013;33(7):1513–20. pmid:23640498
- View Article
- PubMed/NCBI
- Google Scholar
47. Kang YP, Yoon J-H, Long NP, Koo G-B, Noh H-J, Oh S-J, et al. Spheroid-Induced Epithelial-Mesenchymal Transition Provokes Global Alterations of Breast Cancer Lipidome: A Multi-Layered Omics Analysis. Frontiers in Oncology. 2019;9. pmid:30949448
- View Article
- PubMed/NCBI
- Google Scholar
48. Franky Dhaval S, Shilin Nandubhai S, Pankaj Manubhai S, Patel HR, Prabhudas Shankerbhai P. Significance of alterations in plasma lipid profile levels in breast cancer. Integr Cancer Ther. 2008;7(1):33–41. Epub 2008/02/23. pmid:18292593.
- View Article
- PubMed/NCBI
- Google Scholar
49. Liu T, Peng F, Yu J, Tan Z, Rao T, Chen Y, et al. LC-MS-based lipid profile in colorectal cancer patients: TAGs are the main disturbed lipid markers of colorectal cancer progression. Analytical and Bioanalytical Chemistry. 2019;411(20):5079–88. pmid:31201454
- View Article
- PubMed/NCBI
- Google Scholar
50. Vabalas A, Gowen E, Poliakoff E, Casson AJ. Machine learning algorithm validation with a limited sample size. PLOS ONE. 2019;14(11):e0224365. pmid:31697686
- View Article
- PubMed/NCBI
- Google Scholar
51. Kuhn Max, Johnson Kjell. Applied Predictive Modeling. Springer New York, NY. 2013.

[ref1] 1. Arnold M, Morgan E, Rumgay H, Mafra A, Singh D, Laversanne M, et al. Current and future burden of breast cancer: Global statistics for 2020 and 2040. The Breast. 2022;66:15–23. pmid:36084384
View Article
PubMed/NCBI
Google Scholar

[2] View Article

[3] PubMed/NCBI

[4] Google Scholar

[ref2] 2. Anastasiadi Z, Lianos GD, Ignatiadou E, Harissis HV, Mitsis M. Breast cancer in young women: an overview. Updates in Surgery. 2017;69(3):313–7. pmid:28260181
View Article
PubMed/NCBI
Google Scholar

[6] View Article

[7] PubMed/NCBI

[8] Google Scholar

[ref3] 3. Jafari SH, Saadatpour Z, Salmaninejad A, Momeni F, Mokhtari M, Nahand JS, et al. Breast cancer diagnosis: Imaging techniques and biochemical markers. Journal of Cellular Physiology. 2018;233(7):5200–13. pmid:29219189
View Article
PubMed/NCBI
Google Scholar

[10] View Article

[11] PubMed/NCBI

[12] Google Scholar

[ref4] 4. Moyya PD, Asaithambi M. Radiomics—Quantitative Biomarker Analysis for Breast Cancer Diagnosis and Prediction: A Review. Curr Med Imaging. 2022;18(1):3–17. Epub 2021/03/04. pmid:33655872.
View Article
PubMed/NCBI
Google Scholar

[14] View Article

[15] PubMed/NCBI

[16] Google Scholar

[ref5] 5. Tsang JYS, Tse GM. Molecular Classification of Breast Cancer. Advances in Anatomic Pathology. 2020;27(1). pmid:31045583
View Article
PubMed/NCBI
Google Scholar

[18] View Article

[19] PubMed/NCBI

[20] Google Scholar

[ref6] 6. Harbeck N, Penault-Llorca F, Cortes J, Gnant M, Houssami N, Poortmans P, et al. Breast cancer. Nature Reviews Disease Primers. 2019;5(1):66. pmid:31548545
View Article
PubMed/NCBI
Google Scholar

[22] View Article

[23] PubMed/NCBI

[24] Google Scholar

[ref7] 7. Narayan AK, Lee CI, Lehman CD. Screening for Breast Cancer. Med Clin North Am. 2020;104(6):1007–21. Epub 2020/10/26. pmid:33099447.
View Article
PubMed/NCBI
Google Scholar

[26] View Article

[27] PubMed/NCBI

[28] Google Scholar

[ref8] 8. Rodriguez-Ruiz A, Lång K, Gubern-Merida A, Broeders M, Gennaro G, Clauser P, et al. Stand-Alone Artificial Intelligence for Breast Cancer Detection in Mammography: Comparison With 101 Radiologists. JNCI: Journal of the National Cancer Institute. 2019;111(9):916–22. pmid:30834436
View Article
PubMed/NCBI
Google Scholar

[30] View Article

[31] PubMed/NCBI

[32] Google Scholar

[ref9] 9. D’Orsi CJ, Sickles EA, Mendelson EB, Morris EA, et al. ACR BI-RADS^® Atlas, Breast Imaging Reporting and Data System. Reston, VA, American College of Radiology. 2013.

[ref10] 10. Lee JS, Park S, Park JM, Cho JH, Kim SI, Park B-W. Elevated levels of serum tumor markers CA 15–3 and CEA are prognostic factors for diagnosis of metastatic breast cancers. Breast Cancer Research and Treatment. 2013;141(3):477–84. pmid:24072270
View Article
PubMed/NCBI
Google Scholar

[35] View Article

[36] PubMed/NCBI

[37] Google Scholar

[ref11] 11. Di Gioia D, Stieber P, Schmidt GP, Nagel D, Heinemann V, Baur-Melnyk A. Early detection of metastatic disease in asymptomatic breast cancer patients with whole-body imaging and defined tumour marker increase. British Journal of Cancer. 2015;112(5):809–18. pmid:25647014
View Article
PubMed/NCBI
Google Scholar

[39] View Article

[40] PubMed/NCBI

[41] Google Scholar

[ref12] 12. Danzi F, Pacchiana R, Mafficini A, Scupoli MT, Scarpa A, Donadelli M, et al. To metabolomics and beyond: a technological portfolio to investigate cancer metabolism. Signal Transduction and Targeted Therapy. 2023;8(1):137. pmid:36949046
View Article
PubMed/NCBI
Google Scholar

[43] View Article

[44] PubMed/NCBI

[45] Google Scholar

[ref13] 13. Alonso A, Marsal S, Julià A. Analytical Methods in Untargeted Metabolomics: State of the Art in 2015. Frontiers in Bioengineering and Biotechnology. 2015;3. pmid:25798438
View Article
PubMed/NCBI
Google Scholar

[47] View Article

[48] PubMed/NCBI

[49] Google Scholar

[ref14] 14. Agarwala PK, Aneja R, Kapoor S. Lipidomic landscape in cancer: Actionable insights for membrane-based therapy and diagnoses. Medicinal Research Reviews. 2022;42(2):983–1018. pmid:34719798
View Article
PubMed/NCBI
Google Scholar

[51] View Article

[52] PubMed/NCBI

[53] Google Scholar

[ref15] 15. Jasbi P, Wang D, Cheng SL, Fei Q, Cui JY, Liu L, et al. Breast cancer detection using targeted plasma metabolomics. Journal of Chromatography B. 2019;1105:26–37. pmid:30562627
View Article
PubMed/NCBI
Google Scholar

[55] View Article

[56] PubMed/NCBI

[57] Google Scholar

[ref16] 16. Park J, Shin Y, Kim TH, Kim D-H, Lee A. Plasma metabolites as possible biomarkers for diagnosis of breast cancer. PLOS ONE. 2019;14(12):e0225129. pmid:31794572
View Article
PubMed/NCBI
Google Scholar

[59] View Article

[60] PubMed/NCBI

[61] Google Scholar

[ref17] 17. Mao C, Wang M, Li L, Tang J-H. Circulating metabolites serve as diagnostic biomarkers for HER2-positive breast cancer and have predictive value for trastuzumab therapy outcomes. Journal of Clinical Laboratory Analysis. 2022;36(2):e24212. pmid:34994982
View Article
PubMed/NCBI
Google Scholar

[63] View Article

[64] PubMed/NCBI

[65] Google Scholar

[ref18] 18. Beloribi-Djefaflia S, Vasseur S, Guillaumond F. Lipid metabolic reprogramming in cancer cells. Oncogenesis. 2016;5(1):e189–e. pmid:26807644
View Article
PubMed/NCBI
Google Scholar

[67] View Article

[68] PubMed/NCBI

[69] Google Scholar

[ref19] 19. Wei Y, Jasbi P, Shi X, Turner C, Hrovat J, Liu L, et al. Early Breast Cancer Detection Using Untargeted and Targeted Metabolomics. Journal of Proteome Research. 2021;20(6):3124–33. pmid:34033488
View Article
PubMed/NCBI
Google Scholar

[71] View Article

[72] PubMed/NCBI

[73] Google Scholar

[ref20] 20. He P, Cui L-G, Chen W, Yang R-L. Subcategorization of Ultrasonographic BI-RADS Category 4: Assessment of Diagnostic Accuracy in Diagnosing Breast Lesions and Influence of Clinical Factors on Positive Predictive Value. Ultrasound in Medicine & Biology. 2019;45(5):1253–8. pmid:30799123
View Article
PubMed/NCBI
Google Scholar

[75] View Article

[76] PubMed/NCBI

[77] Google Scholar

[ref21] 21. Yen NTH, Anh NK, Jayanti RP, Phat NK, Vu DH, Ghim J-L, et al. Multimodal plasma metabolomics and lipidomics in elucidating metabolic perturbations in tuberculosis patients with concurrent type 2 diabetes. Biochimie. 2023;211:153–63. pmid:37062470
View Article
PubMed/NCBI
Google Scholar

[79] View Article

[80] PubMed/NCBI

[81] Google Scholar

[ref22] 22. Barupal DK, Fan S, Wancewicz B, Cajka T, Sa M, Showalter MR, et al. Generation and quality control of lipidomics data for the alzheimer’s disease neuroimaging initiative cohort. Sci Data. 2018;5:180263. Epub 2018/11/21. pmid:30457571.
View Article
PubMed/NCBI
Google Scholar

[83] View Article

[84] PubMed/NCBI

[85] Google Scholar

[ref23] 23. Cajka T, Hricko J, Rudl Kulhava L, Paucova M, Novakova M, Kuda O. Optimization of Mobile Phase Modifiers for Fast LC-MS-Based Untargeted Metabolomics and Lipidomics. International Journal of Molecular Sciences [Internet]. 2023; 24(3). pmid:36768308
View Article
PubMed/NCBI
Google Scholar

[87] View Article

[88] PubMed/NCBI

[89] Google Scholar

[ref24] 24. Lavarello C, Barco S, Bartolucci M, Panfoli I, Magi E, Tripodi G, et al. Development of an Accurate Mass Retention Time Database for Untargeted Metabolomic Analysis and Its Application to Plasma and Urine Pediatric Samples. Molecules. 2021;26(14). Epub 2021/07/25. pmid:34299531.
View Article
PubMed/NCBI
Google Scholar

[91] View Article

[92] PubMed/NCBI

[93] Google Scholar

[ref25] 25. Kind T, Liu KH, Lee DY, DeFelice B, Meissen JK, Fiehn O. LipidBlast in silico tandem mass spectrometry database for lipid identification. Nat Methods. 2013;10(8):755–8. Epub 2013/07/03. pmid:23817071.
View Article
PubMed/NCBI
Google Scholar

[95] View Article

[96] PubMed/NCBI

[97] Google Scholar

[ref26] 26. Tsugawa H, Ikeda K, Takahashi M, Satoh A, Mori Y, Uchino H, et al. A lipidome atlas in MS-DIAL 4. Nat Biotechnol. 2020;38(10):1159–63. Epub 2020/06/17. pmid:32541957.
View Article
PubMed/NCBI
Google Scholar

[99] View Article

[100] PubMed/NCBI

[101] Google Scholar

[ref27] 27. Pang Z, Chong J, Zhou G, de Lima Morais DA, Chang L, Barrette M, et al. MetaboAnalyst 5.0: narrowing the gap between raw spectra and functional insights. Nucleic Acids Res. 2021;49(W1):W388–W96. Epub 2021/05/22. pmid:34019663.
View Article
PubMed/NCBI
Google Scholar

[103] View Article

[104] PubMed/NCBI

[105] Google Scholar

[ref28] 28. Pang Z, Chong J, Li S, Xia J. MetaboAnalystR 3.0: Toward an Optimized Workflow for Global Metabolomics. Metabolites. 2020;10(5). Epub 2020/05/13. pmid:32392884.
View Article
PubMed/NCBI
Google Scholar

[107] View Article

[108] PubMed/NCBI

[109] Google Scholar

[ref29] 29. Claudino WM, Quattrone A, Biganzoli L, Pestrin M, Bertini I, Di Leo A. Metabolomics: Available Results, Current Research Projects in Breast Cancer, and Future Applications. Journal of Clinical Oncology. 2007;25(19):2840–6. pmid:17502626
View Article
PubMed/NCBI
Google Scholar

[111] View Article

[112] PubMed/NCBI

[113] Google Scholar

[ref30] 30. Mrowiec K, Kurczyk A, Jelonek K, Debik J, Giskeødegård GF, Bathen TF, et al. Association of serum metabolome profile with the risk of breast cancer in participants of the HUNT2 study. Frontiers in Oncology. 2023;13. pmid:37007110
View Article
PubMed/NCBI
Google Scholar

[115] View Article

[116] PubMed/NCBI

[117] Google Scholar

[ref31] 31. His M, Gunter MJ, Keski-Rahkonen P, Rinaldi S. Application of Metabolomics to Epidemiologic Studies of Breast Cancer: New Perspectives for Etiology and Prevention. Journal of Clinical Oncology. 2023;42(1):103–15. pmid:37944067
View Article
PubMed/NCBI
Google Scholar

[119] View Article

[120] PubMed/NCBI

[121] Google Scholar

[ref32] 32. Pedersen JL, Barry SE, Bokil NJ, Ellis M, Yang Y, Guan G, et al. High sensitivity and specificity of a 5-analyte protein and microRNA biosignature for identification of active tuberculosis. Clinical & Translational Immunology. 2021;10(6):e1298. pmid:34188917
View Article
PubMed/NCBI
Google Scholar

[123] View Article

[124] PubMed/NCBI

[125] Google Scholar

[ref33] 33. Xu C, Jackson SA. Machine learning and complex biological data. Genome Biology. 2019;20(1):76. pmid:30992073
View Article
PubMed/NCBI
Google Scholar

[127] View Article

[128] PubMed/NCBI

[129] Google Scholar

[ref34] 34. Bzdok D, Altman N, Krzywinski M. Statistics versus machine learning. Nature Methods. 2018;15(4):233–4. pmid:30100822
View Article
PubMed/NCBI
Google Scholar

[131] View Article

[132] PubMed/NCBI

[133] Google Scholar

[ref35] 35. Budczies J, Pfitzner BM, Györffy B, Winzer K-J, Radke C, Dietel M, et al. Glutamate enrichment as new diagnostic opportunity in breast cancer. International Journal of Cancer. 2015;136(7):1619–28. pmid:25155347
View Article
PubMed/NCBI
Google Scholar

[135] View Article

[136] PubMed/NCBI

[137] Google Scholar

[ref36] 36. Altman BJ, Stine ZE, Dang CV. From Krebs to clinic: glutamine metabolism to cancer therapy. Nature Reviews Cancer. 2016;16(10):619–34. pmid:27492215
View Article
PubMed/NCBI
Google Scholar

[139] View Article

[140] PubMed/NCBI

[141] Google Scholar

[ref37] 37. Li L, Zheng X, Zhou Q, Villanueva N, Nian W, Liu X, et al. Metabolomics-Based Discovery of Molecular Signatures for Triple Negative Breast Cancer in Asian Female Population. Scientific Reports. 2020;10(1):370. pmid:31941951
View Article
PubMed/NCBI
Google Scholar

[143] View Article

[144] PubMed/NCBI

[145] Google Scholar

[ref38] 38. Messias MCF, Mecatti GC, Priolli DG, de Oliveira Carvalho P. Plasmalogen lipids: functional mechanism and their involvement in gastrointestinal cancer. Lipids in Health and Disease. 2018;17(1):41. pmid:29514688
View Article
PubMed/NCBI
Google Scholar

[147] View Article

[148] PubMed/NCBI

[149] Google Scholar

[ref39] 39. Skotland T, Kavaliauskiene S, Sandvig K. The role of lipid species in membranes and cancer-related changes. Cancer and Metastasis Reviews. 2020;39(2):343–60. pmid:32314087
View Article
PubMed/NCBI
Google Scholar

[151] View Article

[152] PubMed/NCBI

[153] Google Scholar

[ref40] 40. Benjamin DI, Cozzo A, Ji X, Roberts LS, Louie SM, Mulvihill MM, et al. Ether lipid generating enzyme AGPS alters the balance of structural and signaling lipids to fuel cancer pathogenicity. Proc Natl Acad Sci U S A. 2013;110(37):14912–7. Epub 2013/08/28. pmid:23980144.
View Article
PubMed/NCBI
Google Scholar

[155] View Article

[156] PubMed/NCBI

[157] Google Scholar

[ref41] 41. Patel N, Vogel R, Chandra-Kuntal K, Glasgow W, Kelavkar U. A Novel Three Serum Phospholipid Panel Differentiates Normal Individuals from Those with Prostate Cancer. PLOS ONE. 2014;9(3):e88841. pmid:24603597
View Article
PubMed/NCBI
Google Scholar

[159] View Article

[160] PubMed/NCBI

[161] Google Scholar

[ref42] 42. Zhao Z, Xiao Y, Elson P, Tan H, Plummer SJ, Berk M, et al. Plasma Lysophosphatidylcholine Levels: Potential Biomarkers for Colorectal Cancer. Journal of Clinical Oncology. 2007;25(19):2696–701. pmid:17602074
View Article
PubMed/NCBI
Google Scholar

[163] View Article

[164] PubMed/NCBI

[165] Google Scholar

[ref43] 43. Lin H-M, Yang X, Centenera MM, Huynh K, Giles C, Dehairs J, et al. Circulating Lipid Profiles Associated With Resistance to Androgen Deprivation Therapy in Localized Prostate Cancer. JCO Precision Oncology. 2024;(8):e2400260. pmid:39074346
View Article
PubMed/NCBI
Google Scholar

[167] View Article

[168] PubMed/NCBI

[169] Google Scholar

[ref44] 44. Hannun YA, Obeid LM. Principles of bioactive lipid signalling: lessons from sphingolipids. Nature Reviews Molecular Cell Biology. 2008;9(2):139–50. pmid:18216770
View Article
PubMed/NCBI
Google Scholar

[171] View Article

[172] PubMed/NCBI

[173] Google Scholar

[ref45] 45. Zalba S, ten Hagen TLM. Cell membrane modulation as adjuvant in cancer therapy. Cancer Treatment Reviews. 2017;52:48–57. pmid:27889637
View Article
PubMed/NCBI
Google Scholar

[175] View Article

[176] PubMed/NCBI

[177] Google Scholar

[ref46] 46. Li Y, Dong J, Ding T, Kuo M-S, Cao G, Jiang X-C, et al. Sphingomyelin Synthase 2 Activity and Liver Steatosis. Arteriosclerosis, Thrombosis, and Vascular Biology. 2013;33(7):1513–20. pmid:23640498
View Article
PubMed/NCBI
Google Scholar

[179] View Article

[180] PubMed/NCBI

[181] Google Scholar

[ref47] 47. Kang YP, Yoon J-H, Long NP, Koo G-B, Noh H-J, Oh S-J, et al. Spheroid-Induced Epithelial-Mesenchymal Transition Provokes Global Alterations of Breast Cancer Lipidome: A Multi-Layered Omics Analysis. Frontiers in Oncology. 2019;9. pmid:30949448
View Article
PubMed/NCBI
Google Scholar

[183] View Article

[184] PubMed/NCBI

[185] Google Scholar

[ref48] 48. Franky Dhaval S, Shilin Nandubhai S, Pankaj Manubhai S, Patel HR, Prabhudas Shankerbhai P. Significance of alterations in plasma lipid profile levels in breast cancer. Integr Cancer Ther. 2008;7(1):33–41. Epub 2008/02/23. pmid:18292593.
View Article
PubMed/NCBI
Google Scholar

[187] View Article

[188] PubMed/NCBI

[189] Google Scholar

[ref49] 49. Liu T, Peng F, Yu J, Tan Z, Rao T, Chen Y, et al. LC-MS-based lipid profile in colorectal cancer patients: TAGs are the main disturbed lipid markers of colorectal cancer progression. Analytical and Bioanalytical Chemistry. 2019;411(20):5079–88. pmid:31201454
View Article
PubMed/NCBI
Google Scholar

[191] View Article

[192] PubMed/NCBI

[193] Google Scholar

[ref50] 50. Vabalas A, Gowen E, Poliakoff E, Casson AJ. Machine learning algorithm validation with a limited sample size. PLOS ONE. 2019;14(11):e0224365. pmid:31697686
View Article
PubMed/NCBI
Google Scholar

[195] View Article

[196] PubMed/NCBI

[197] Google Scholar

[ref51] 51. Kuhn Max, Johnson Kjell. Applied Predictive Modeling. Springer New York, NY. 2013.

Combining metabolomics and machine learning to discover biomarkers for early-stage breast cancer diagnosis

Combining metabolomics and machine learning to discover biomarkers for early-stage breast cancer diagnosis

Figures

Abstract

Introduction

Materials and methods

Clinical samples and ethical approval

Clinical characteristics of the patients and study design

Reagents and chemicals

Sample preparation

Data acquisition using LC-MS/MS

Data processing and treatment

Exploratory data analysis

Statistical analysis

Results

Data exploration revealed subtle differences in plasma metabolic profiles between cancer and benign groups

Univariate biomarker analysis identified potential biomarker candidates

External validation highlighted the potential of metabolism-centric biomarkers in aiding BC diagnosis

Discussion

Conclusion

Supporting information

S1 File. Supplementary information of the study.

S2 File.

S3 File.

S1 Data.

S1 Checklist. Human participants research checklist.

Acknowledgments

References