A Scoring Model Using Multi-Metabolites Based on Untargeted Metabolomics for Assessing Dyslipidemia in Korean Individuals with Obesity

TL;DR

This study develops a metabolite-based scoring model to assess dyslipidemia in obese Korean individuals, showing promising diagnostic accuracy when combined with traditional lipid profiles.

Contribution

The novel contribution is the development and validation of a metabolite risk score (MRS) for dyslipidemia using untargeted metabolomics.

Findings

N-acetylisoputreanine-γ-lactam and eicosapentaenoic acid were identified as key metabolites for dyslipidemia.

The MRS model achieved 85.4% accuracy in the discovery set and 76.1% in the replication set.

Combining MRS with lipid profiles improved diagnostic accuracy to 91.8% in the discovery set and 86.0% in the replication set.

Abstract

Background/Objectives: Metabolite risk score (MRS), which considers the collective effects of metabolites closely reflecting a phenotype, is a new approach for disease assessment, moving away from focusing solely on individual biomarkers. This study aimed to investigate a metabolite panel for dyslipidemia and verify the diagnostic efficacy of MRS on dyslipidemia. Methods: Key metabolite identification and MRS establishment were conducted in the discovery set, and MRS validation was performed in the replication set, with 50 healthy individuals and 50 dyslipidemia patients in each set. The MRS was constructed using key metabolites, identified via UPLC-MS/MS analysis, employing a weighted approach based on linear regression analysis. Results: N-acetylisoputreanine-γ-lactam and eicosapentaenoic acid were identified as key metabolites for dyslipidemia and were utilized for establishing the MRS. In addition to the MRS model, a conventional dyslipidemia diagnostic model based on lipid profiles, as well as a combined model (MRS + lipid profiles), were also established. In the discovery set, the MRS model diagnosed dyslipidemia with 85.4% accuracy. When combined with lipid profiles, accuracy improved to 91.8%. In the replication set, the MRS demonstrated diagnostic power with 76.1% accuracy, while the combined model achieved 86.0% accuracy for dyslipidemia assessment. Conclusions: The MRS alone indicated sufficient assessment power in a real-world setting, despite a slight reduction in assessment ability when validated in the replication set. At this stage, therefore, the MRS serves as an auxiliary tool for disease diagnosis. This first attempt to apply MRS for dyslipidemia may offer a foundational concept for MRS in this disease.

Figures

Full text

1. Introduction

Dyslipidemia is a chronic disease that shows elevated levels of triglyceride (TG), low-density lipoprotein-cholesterol (LDL-C), and total cholesterol (TC) and reduced levels of high-density lipoprotein-cholesterol (HDL-C) in the blood. The incidence of dyslipidemia is closely linked to the risk of vascular diseases such as atherosclerotic events and stroke and is often observed with obesity and type 2 diabetes simultaneously as well [1,2,3]. According to the dyslipidemia fact sheet in Korea published in 2022, the mean prevalence of dyslipidemia from 2016 to 2020 was 48.2%, indicating a quite high proportion within the population [1]. Moreover, individuals with type 2 diabetes, hypertension, and obesity are accompanied by dyslipidemia at a ratio of 87.1%, 72.1%, and 55.4%, respectively [1]. Given that lipid disorders and subsequent events, including coronary artery disease and ischemic stroke, are regarded as leading causes of premature death even globally [3], it is crucial to monitor and manage dyslipidemia effectively.

Metabolites serve as biosignatures that closely mirror an individual’s phenotypic information since they are downstream products of metabolic processes [4]. Abnormalities in metabolism can lead to alterations in metabolites, making their accumulation or deficiency indicative hallmarks of diseases [4]. Hence, metabolomics has been presented as a promising tool for investigating diseases’ biomarkers and status. Researchers have attempted to elucidate what novel biomarkers are for dyslipidemia and the biological meaning of altered metabolites observed in patients with dyslipidemia via metabolomics study. Du et al. [5] revealed 69 plasma metabolites indicating disturbed regulation of inflammation and lipid metabolism in dyslipidemia patients; notably, indole-3-propionic acid particularly emerged as a promising biomarker. Ke et al. [6] demonstrated that 46 metabolites implicated in macrophage activation and oxidized (ox-) LDL formation were identified as potential biomarkers for dyslipidemia in the UK population. Wang et al. [7] showed a positive association between plasma retinol levels and dyslipidemia in the Chinese population.

Like previous studies, most research has been focused on investigating metabolic pattern changes and identifying the best novel biomarkers for a specific disease. However, diseases arise from the interactions between metabolites significantly related to the disease. In this context, metabolite risk score (MRS), which considers the comprehensive effects among metabolites, can be a novel concept for approaching disease progression. So far, a limited number of studies have investigated MRS. A study reported that future weight gain could be predicted by constructing an MRS using 8 metabolites [8]. Another recent study validated that an adipose tissue-derived MRS, incorporating hundreds of metabolites, aids in assessing the risk of type 2 diabetes before the elevation of fasting glucose levels [9]. Furthermore, mild cognitive impairment was found to be predictable based on MRS established by 61 metabolites [10]. The MRS studies indicated that this scoring system demonstrates improved assessment capability for diseases. Hence, the development of MRS is crucial to prevent future disease onset.

To date, there has been no research on MRS for dyslipidemia. Therefore, the present study aimed to establish and validate an MRS for the diagnosis of dyslipidemia in a Korean population by screening significant metabolites via ultra-performance liquid chromatography (UPLC)-tandem mass spectrometry (MS/MS). The decision to focus on MRS as a diagnostic tool was based on two key considerations. First, current standard diagnostic criteria for dyslipidemia have certain limitations. For instance, the Friedewald equation, which is commonly used in clinical settings due to its cost-effectiveness and convenience, becomes inaccurate when TG levels exceed 400 mg/dL. In addition, standard blood tests are unable to measure highly atherogenic subtypes such as small dense LDL. Second, since our cohort is cross-sectional in design, a diagnostic approach is more methodologically appropriate than a predictive one. Finally, this study contributes to the advancement of novel approaches to the use of MRS in dyslipidemia diagnosis.

2. Materials and Methods

2.1. Collection of Plasma Samples and Study Design

A total of 100 fasting plasma samples from healthy individuals and another 100 fasting plasma samples from dyslipidemia patients were acquired through the Korea Biobank Network, including the Biobank of Korea-Chungbuk National University Hospital (CBNUH) and the Biobank Ajou University Hospital under the institutional review board (IRB)-approved protocols (Inha University IRB, approved No.: 21004-1AR). The IRB waived the requirement to obtain written informed consent from study participants, as all samples and data were procured from the Biobanks with explicit authorization. The Biobanks randomly selected samples from their pools based on the following inclusion criteria: (1) Korean males and females, (2) aged between 40 and 60, and (3) Korean Standard Classification of Diseases (KCD) code Z00 for healthy individuals, and (4) KCD code E78.2 and E78.5 for dyslipidemia patients. The general information was also collected as well: age, sex, body weight, body mass index (BMI), systolic/diastolic blood pressures (BPs), and fasting levels of glucose, lipid profiles [TG, HDL-C, LDL-C, TC], aspartate aminotransferase (AST), alanine aminotransferase (ALT), and γ-glutamyltransferase (γ-GTP).

This study was designed as a cross-sectional, two-phase approach comprising a biomarker discovery stage (discovery) and an independent validation stage (replication). As illustrated in Figure 1, each of the two groups (healthy and dyslipidemia; n = 100 each) was further randomly divided into a discovery set and a replication set in a 1:1 ratio. This sample size was determined using G*Power ver. 3.1.9.7 (Franz Faul, University of Kiel, Kiel, Germany), which achieved a statistical power of 0.8 with an α error probability of 0.05 and an effect size exceeding 0.5.

Key metabolites were identified and used to establish the MRS in the discovery set. Subsequently, the diagnostic performance of the MRS for dyslipidemia was validated in the replication set. The replication set was re-allocated based on the MRS (re-grouped replication set) to further assess its accuracy in distinguishing actual disease conditions, and the results were compared with those of the original replication set (Figure 1).

2.2. Biochemical Analysis

Inflammatory and oxidative stress conditions were additionally evaluated using cytokines [interleukin (IL)-1β, IL-6, tumor necrosis factor (TNF)-α, interferon (IFN-γ)] and ox-LDL. Cytokines were analyzed by commercial kits, Human High Sensitivity T cell Magnetic Bead Panel (Merck Millipore, Billerica, MA, USA), and the resulting reactions were analyzed by a MAGPIX system (Luminex Corporation, Austin, TX, USA). Ox-LDL was assessed using Oxidized LDL kits (Mercodia, Uppsala, Sweden), and the resulting reactions were detected by SpectraMax 190 (Molecular Devices Corp., Sunnyvale, CA, USA).

2.3. Untargeted Metabolomics

The 100 μL plasma samples were combined with 300 μL of 70% cold acetonitrile (Thermo Fisher Scientific, Fair Lawn, NJ, USA). After vortexing, the mixtures were incubated (4 °C, 10 min). Subsequently, the mixtures were centrifuged at 13,000 rpm for 15 min at 4 °C. The supernatant was collected and transferred to a new vial, followed by drying using SpeedVac (Gyrozen, Gyeonggi-do, Republic of Korea). 10% methanol (J.T.Baker^®^; Avantor, Radnor, PA, USA) containing internal standards [ISTDs; L-leucine-^13^C_6_ (Sigma-Aldrich, St. Louis, MO, USA) and stearic-d_35_ acid (Sigma-Aldrich, St. Louis, MO, USA)] was utilized to re-dissolve the dried residues. Pooled plasma samples for quality control purposes were prepared using the same procedure.

5 μL of the prepared samples were injected into the LC system [Acquity UPLC-BEC-C18 column (2.1 × 10 mm, 1.7 μm); Waters, Milford, MA, USA] equipped with Ultimate 3000 RSLC System (Thermo Fisher Scientific, Bremen, Germany). Quality control samples were injected every 10th run. Mobile phases A and B were 0.1% formic acids (Merck Millipore, Billerica, MA, USA) dissolved in LC-MS grade water (Thermo Fisher Scientific, Fair Lawn, NJ, USA) and acetonitrile (Thermo Fisher Scientific, Fair Lawn, NJ, USA), respectively. Changes in the gradient of the mobile phases ranging from 0% to 100% were applied over 17 min at a flow rate of 0.4 mL/min. MS analysis was conducted using electrospray ionization (ESI) in both positive and negative modes with a Q-Exactive Orbitrap Plus MS (Thermo Fisher Scientific, Waltham, MA, USA). A full MS-ddms2 scan type [80~1000 mass-to-charge (m/z)] was employed. The following MS conditions were used: spray voltage (kV), 3.5; sheath gas flow rate, 50 arbitrary units; auxiliary gas, 13 arbitrary units; S-lens radio frequency level, 55; and capillary temperature, 370 °C.

2.4. Establishment of MRS and Its Validation

2.4.1. Major and Key Metabolites Related to Dyslipidemia in the Discovery Set

To select major metabolites, identified metabolites were screened based on variable importance in projection (VIP) value of ≥1.5 and false rate discover (FDR)-adjusted p-value (q-value) of <0.05 in the discovery set. These major metabolites were further screened to find out key metabolites associated with dyslipidemia through the following methods: first, a linear regression analysis using the stepwise approach was employed to verify dyslipidemia-related metabolites; and then, the verified metabolites were assessed using a receiver operating characteristic (ROC) curve analysis to evaluate their assessment ability for dyslipidemia. Finally, metabolites demonstrating significance in the ROC curve analysis were determined as key metabolites and utilized for establishing MRS.

2.4.2. MRS Establishment in the Discovery Set

The key metabolites were converted into MRS using a weighted approach, establishing the equation of the MRS as ∑β_i_M_i_, where β_i_ represents the value of standardized β-coefficient and M_i_ represents the simple score of each key metabolite. A linear regression analysis with the enter method was employed to obtain the standardized β-coefficient values. Subsequently, ROC curve analysis was performed to determine the cut-off values of the key metabolites, converting their relative intensities into a simple score; if the relative intensity was above the cut-off value, the simple score was 1; otherwise, it was 0.

2.4.3. MRS Validation in Both Discovery and Replication Set

The MRS model, individual key metabolite models, and a lipid profile model were constructed and compared with each other within the discovery set to evaluate the assessment ability of the MRS for dyslipidemia. Note that the lipid profile model consisted of TG, HDL-C, LDL-C, and TC, which are conventional markers for dyslipidemia. Then, the models were validated to determine whether the assessment power of these models remained consistent in the replication set.

2.5. Statistical Analysis

The details are provided in the supplementary information. Briefly, for putative metabolite identification and relative peak intensity data acquisition, Compound Discoverer ver. 3.3 SP 2 software (Thermo Fisher, Waltham, MA, USA) was used. SIMCA 17 software (Sartorius-Umetrics, Göttingen, Germany) was utilized for orthogonal partial least squares-discriminant analysis (OPLS-DA), loading plot generation, permutation tests, and VIP score analysis.

Statistical analysis was conducted using IBM SPSS Statistics 28.0 (IBM Corp., Armonk, NY, USA). This included the comparison of variables between groups, linear regression analysis, and ROC curve analysis. A significance threshold of two-tailed p < 0.05 was used for all analyses. For the comparison of metabolite relative intensities, meanwhile, q-values were additionally computed to adjust for FDR, using the R software ver. 4.4.2 (R Foundation for Statistical Computing, Vienna, Austria) with package fdrtool ver. 1.2.18 [https://CRAN.R-project.org/package=fdrtool (accessed on 1 November 2024)], with significance defined as q < 0.05 to mitigate type 1 error.

3. Results

3.1. General Clinical/Biochemical Markers Between the Groups in the Discovery Set

As shown in Table 1, individuals in the dyslipidemia group were significantly older than those in the healthy group. Additionally, both body weight and BMI were significantly higher in the dyslipidemia group than in the healthy group. Although the sex distribution was not equal between the groups, it did not show a statistically significant difference (Table 1). While sex differences may influence body weight due to physical and physiological differences between males and females, the presence of only three more males and four fewer females in the dyslipidemia group is unlikely to account for the observed 7.1 kg difference in average body weight. Therefore, the increase in body weight primarily reflects the metabolic characteristics of the dyslipidemia group rather than being driven solely by sex distribution. Consequently, the three variables that showed significant differences between the groups (i.e., age, body weight, and BMI) were used as confounding factors for adjustment. After adjusting for these confounding factors, systolic BP, glucose levels, and TNF-α retained their statistical significance, and diastolic BP became significantly elevated. Of note, individuals in the dyslipidemia group appear to exhibit early-stage hypertension, a well-known complication associated with dyslipidemia. In contrast, the statistical significance of TG, TC, AST, ALT, IL-1β, IL-6, and ox-LDL disappeared after adjustment.

3.2. UPLC-MS/MS Analysis

In the ESI positive and negative modes, a total of 452 and 258 metabolites, respectively, were detected and putatively identified. Among these, unidentified metabolites and drugs were excluded. Finally, 128 and 69 metabolites in ESI positive and negative modes remained, respectively, and utilized for the subsequent analysis (Table S1).

3.2.1. Metabolic Characteristics Between the Healthy and Dyslipidemia Groups in the Discovery Set

In both ESI positive and negative modes, the OPLS-DA results indicated that the groups significantly discriminated with enough R^2^Y and Q^2^Y values (both R^2^Y and Q^2^Y >0.5) and pCV-ANOVA values less than 0.05 (Figure 2A,B). The permutation test results also demonstrated group discrimination without overfitting in both modes (Figure S1). In ESI positive mode, the three most abundant subclasses of the detected metabolites were acylcarnitines (ACs), amino acids, and lysophosphatidylcholines (LPCs); thus, a loading plot was consequently generated using these three metabolites’ subclasses (Figure 2C). As shown in the plot, a distinct pattern of the metabolites was observed; particularly, most of short-chain ACs (L-acetylcarnitine, propionylcarnitine, 3-hydroxyisovalerylcarnitine, butyrylcarnitine, and hexanoylcarnitine) and amino acids were prevalent in the dyslipidemia group, whereas long-chain ACs with acyl chains over C18 (palmitoylcarnitine, linoleylcarnitine, oleoylcarnitine, stearoylcarnitine, arachidylcarnitine, hexacosanoylcarnitine) were plenty in the healthy group (Figure 2C). In the ESI negative mode, a loading plot was created based on the top 3 abundant metabolites’ subclasses, including C24 bile acids, unsaturated fatty acids (UFAs), and saturated fatty acids (SFAs) (Figure 2D). Similar to the ESI positive mode, a clearly different pattern of the metabolites was also observed in ESI negative mode; most of the UFAs and SFAs showed elevated levels in the dyslipidemia group (Figure 2D). Based on the results, we confirmed that altered metabolic patterns distinguish between the healthy and dyslipidemia groups.

3.2.2. Major Metabolite Screening and Key Metabolite Selection

According to the VIP score analysis, a total of 11 and 4 metabolites had a VIP score of over 1.5 in ESI positive and negative modes, respectively (Figure S2). Most of them had significant q-values (q < 0.05) except for 3 metabolites in ESI positive mode, resulting in 12 metabolites being selected as major metabolites (Figure 3).

In the ESI positive mode, levels of six metabolites (hexanal, butyl butyrate, ethyl acetate, triethylene glycol monobutyl ether, sphinganine 1-phosphate, and hexaethylene glycol) were decreased, while levels of two metabolites (L-alanyl-L-isoleucine and N-acetylisoputreanine-γ-lactam) were increased in the dyslipidemia group compared to the healthy group (Figure 3A–H). In the ESI negative mode, one metabolite (testosterone sulfate) showed a lower level, whereas three metabolites [1-phenyl-1,3-octadecanedione, docosapentaenoic acid (DPA), and eicosapentaenoic acid (EPA)] exhibited higher levels in the dyslipidemia group than in the healthy group (Figure 3I–L).

Using the 12 major metabolites, linear regression analysis with a stepwise method was performed to select notable dyslipidemia-related key metabolites. Among them, hexanal and butyl butyrate were ruled out because of multicollinearity problems during the analysis. As a result, ethyl acetate, hexaethylene glycol, N-acetylisoputreanine-γ-lactam, and EPA emerged as candidate metabolites. Consequently, ROC analysis revealed that two metabolites, N-acetylisoputreanine-γ-lactam and EPA, displayed significant diagnostics ability for dyslipidemia, thereby being selected as key metabolites for constructing MRS.

3.3. Establishment of MRS and Its Assessment Ability for Dyslipidemia

As mentioned above, N-acetylisoputreanine-γ-lactam and EPA were utilized for MRS establishment. To score the two metabolites, first, cut-off values (relative peak intensities) of them were obtained through Youden’s index method (N-acetylisoputreanine-γ-lactam: 17622907.3327901; and EPA: 45400161.1054486). If the relative peak intensities of the metabolites were above the cut-off values, they were simply scored as 1; otherwise, they were simply scored as 0. Then, the weight values (standardized β from linear regression analysis) of each metabolite were determined, as shown in Table S2, and were used to multiply the simple score. Consequently, the equation of the MRS, the final scoring, is as follows: ∑β_i_M_i_, where β_i_ and M_i_ represent the value of standardized β and simple score of each key metabolite, respectively.

Additional diagnostics models were established to compare the assessment ability of the MRS on dyslipidemia. These models consisted of each metabolite, the lipid profile (TG, HDL-C, LDL-C, and TC), and a combination of the MRS and lipid profile. In Figure 4A, in the discovery set, the MRS model significantly diagnosed dyslipidemia with an accuracy of 85.4%, demonstrating enhanced assessment ability compared to the lipid profile, the conventional diagnostic factor, which had an accuracy of 76.7%. In addition, the combination model of the MRS and lipid profile showed the highest assessment ability with 91.8% accuracy, suggesting the potential of the MRS as an adjuvant tool.

These models were locked down and applied to the replication set to verify whether their assessment ability remained intact. Unfortunately, the assessment ability of the MRS decreased by about 3% compared to that of the lipid profile, although statistical significance remained (Figure 4B). This might be attributed to the reduced assessment ability of N-acetylisoputreanine-γ-lactam (56.2%) (Figure 4B). In other words, this metabolite was not effectively applicable for use in the other independent population. Nevertheless, EPA exhibited sufficient assessment ability for dyslipidemia, and the combination model (MRS + lipid profile) was validated as having enhanced assessment ability compared to the conventional model (lipid profile) (Figure 4B). These results may emphasize the role of MRS as an assistant tool for dyslipidemia diagnostics.

3.4. Discrimination Performance of the MRS in the Re-Grouped Replication Set

We investigated how accurately patients could be classified when applying the MRS to actual practice. The replication set was re-grouped into healthy and dyslipidemia groups according to the MRS. Individuals with an MRS less than its median value ([MRS]median: 0.520) were re-assigned to the healthy group (n = 40), while those with an MRS equal or greater than the [MRS]median were included in the dyslipidemia group (n = 60). Of the individuals, 31 from the healthy group (77.5%) and 41 from the dyslipidemia group (68.3%) were correctly identified as healthy individuals and dyslipidemia patients, respectively (Table S3), suggesting that MRS can be utilized as an auxiliary tool for assessing disease status. In addition, the MRS was deemed effective in discriminating between the groups, as evidenced by significant pCV-ANOVA values and permutation test results (Figure 5), which can be used for a complementary metric when the Q2Y value is less than 0.5 [11]. Based on the results, we expect that the application of the MRS may contribute to managing the risk of dyslipidemia development but still need to improve.

4. Discussion

In the present study, N-acetylisoputreanine-γ-lactam and EPA were selected as a key metabolite for MRS on dyslipidemia; however, only EPA was able to be reproduced in the independent replication set. Even though N-acetylisoputreanine-γ-lactam alone did not significantly diagnose dyslipidemia in the replication set, the MRS and the combined use of the lipid profile still suggest that MRS has relevance for dyslipidemia in the Korean population. Therefore, this study demonstrated the potential of MRS as an assistive tool for dyslipidemia assessment.

N-acetylisoputreanine-γ-lactam, also known as N-(3-acetamidopropyl)pyrrolidin-2-one and acisoga, is a unique catabolite of N1-acetylspermidine [12]. This acetylated compound is one of the polyamines generated during polyamine metabolism, which originates from amino acids, arginine, and ornithine [13,14]. Spermidine/spemine-N1-acetyltransferase (SSAT), a rate-limiting enzyme essential for maintaining cellular polyamine homeostasis [14,15], plays a crucial role in polyamine metabolism by adding an acetyl group to the spermidine, thereby producing N1-acetylspermidine [14]. N1-acetylspermidine is a substrate for peroxisomal N1-acetylpolyamine oxidase (PAO), which generates H_2_O_2_ during N1-acetylspermidine degradation, resulting in oxidative stress [16,17]. Additionally, N1-acetylspermidine is also catabolized by copper-dependent amine oxidase (CuAO) and aldehyde dehydrogenase (ALDH) and converted into N-acetylisoputreanine-γ-lactam [16,17].

Whereas polyamines such as spermidine and N1-acetylspermidine have been well-studied regarding their roles in nucleic acid and protein formation, protection against oxidative stress, ion channel action, and cell survival [18,19], the exact biological functions of N-acetylisoputreanine-γ-lactam remain unclear. However, recent studies indicated its possibilities as a biomarker for heart failure, atrial fibrillation, Parkinson’s disease, and ovarian cancer, as elevated levels of N-acetylisoputreanine-γ-lactam have been observed in these conditions [20,21,22,23]. In the discovery set of the present study, N-acetylisoputreanine-γ-lactam also showed a significant increase in the dyslipidemia group, suggesting its potential as a biomarker for dyslipidemia. Unfortunately, in the replication set, the result was not validated even though the levels of this metabolite were higher in the dyslipidemia group than in the healthy group (log_2_ fold change of the dyslipidemia group relative to the healthy group in the replication set = 0.566, q = 0.252). There is a lack of research and evidence elucidating the associations between N-acetylisoputreanine-γ-lactam and dyslipidemia. As such, it remains uncertain whether this observation is merely by chance. One possible hypothesis is that N-acetylisoputreanine-γ-lactam is not directly or strongly related to dyslipidemia but may instead reflect potential chronic complications induced by dyslipidemia, such as the cardiovascular diseases and cancers mentioned earlier. This association needs to be further investigated through repeated studies on this polyamine metabolite in the context of dyslipidemia, given that many diseases have been shown to be linked to this metabolite.

EPA is a long-chain polyunsaturated fatty acid (PUFA) classified as ω-3 (C20:5, n-3). In humans, it is synthesized from α-linolenic acid (C18:3, n-3), an essential fatty acid that must be consumed from food [24]. However, the efficacy of EPA synthesis is very low; thus, it mostly needs to be obtained from the diet [25]. Many studies have demonstrated that EPA supplementation is beneficial to lowering blood lipids, especially in decreasing TG levels [26,27,28]. Moreover, studies have reported that the intake of ω-3 PUFAs, such as EPA, improves cardiovascular risk factors, including platelet aggregation, atherogenic lipoprotein particles, and C-reactive protein levels [29,30,31].

Regarding the EPA increment in the dyslipidemia group of the present study, although it is not possible to know for certain because of the inability to collect dietary information from the study individuals, it is thought to be due to an overall increase in fatty acid turnover rather than EPA supplementation. Indeed, not only EPA but also all other UFAs and SFAs, with the exception of capric acid (C10:0), were elevated in the dyslipidemia group. This pattern could be typical in dyslipidemia associated with obesity, as well as insulin resistance [32,33], as observed in our study population; the dyslipidemia group of the discovery set exhibited higher BMI and fasting glucose values, and even higher HbA1c levels as well [healthy group (n = 17): 5.427 ± 0.07, dyslipidemia group (n = 43): 6.542 ± 0.24, p^a^ < 0.001, p^b^ < 0.032]. In obesity, there is an increased flux of free fatty acids (FFAs) into the liver, where excess FFAs are used to produce TGs that are then released into the circulation [32]. Additionally, insulin resistance, accompanied by obesity, suppresses insulin from inhibiting hormone-sensitive lipase activity in adipose tissue, thereby promoting the breakdown of stored fat and releasing it as FFAs into the bloodstream [33]. The increased influx of FFAs into the liver via blood circulation perpetuates a vicious circle. In this study, only EPA was chosen as a key metabolite among the FFAs for dyslipidemia assessment. Given the focus of many studies on dietary EPA supplementation and its beneficial effects in various diseases, these results prompt consideration of the endogenous increase in EPA in disease conditions—whether it occurs naturally alongside other FFAs or through complementary mechanisms to combat such inflammatory or oxidative stress conditions.

In addition to the key metabolites, there is an overall difference in metabolite patterns between the healthy and dyslipidemia groups, particularly in ACs and amino acids. Among the ACs, short-chain ACs specifically were increased in the dyslipidemia group. The increase of these metabolites is related to a high risk of cardiovascular diseases, stroke, and type 2 diabetes [34,35,36], as dyslipidemia is also known to be associated with these conditions. The dyslipidemia group of the present study showed increased amino acid levels as well. Researchers have indicated that increased levels of branched-chain amino acids (BCAAs) are significantly related to the risk of TG level elevation and insulin resistance [37,38,39]. Moreover, BCAAs generate short-chain ACs (especially C3 and C5 ACs) as intermediate metabolites [40]. Hence, the increase in short-chain ACs and amino acids in the dyslipidemia group of this study, showing higher levels of glucose, insulin resistance (as described in the previous paragraph), and TG, is consistent with previous studies. Although the present study statistically selected only two metabolites, N-acetylisoputreanine-γ-lactam and EPA, the evidence regarding short-chain ACs and amino acids in dyslipidemia suggests that their potential applications in MRS panels are worth considering in the future.

The present study suggests the use of MRS as an assessment tool for diseases. However, it still has the limitation that MRS alone cannot yet fully screen diseases with very high diagnostic accuracy. In addition, although MRS enhanced diagnostic performance as an adjuvant tool for dyslipidemia, its practical application may not yet be timely in terms of cost-effectiveness, as metabolomics-based profiling is still required to detect the identified key metabolites. The development of point-of-care diagnostic devices or platforms capable of measuring these key metabolites in a simple and convenient manner will be crucial to enhance its practicality. Nevertheless, based on our findings, MRS holds promise as an auxiliary tool, particularly in combination with conventional markers, and this study provides a conceptual framework for its implementation in clinical settings [41]. For MRS to be applied as a professional stand-alone tool for disease diagnostics, future research should focus on expanding metabolite panels to include markers that are sensitive, specific, and reliable for dyslipidemia, as well as on establishing standardized MRS protocols. With continued advancements, MRS may ultimately overcome the limitations of current molecular diagnostic standards.