Characterizing lipoprotein profiles in coronary atherosclerosis development through quantitative lipidomics and proteomics approaches

Abstract

Lipoproteins, synthesized in the liver and secreted into the bloodstream, play a central role in lipid transport. Although their quality affects cardiovascular outcomes, conventional markers do not fully explain the disease risk. We characterized the lipoprotein composition during coronary atherosclerosis progression in myocardial infarction-prone rabbits, which consistently develop severe aortic lesions but show variable coronary stenosis at 20 months. This variation allowed for the exploration of coronary-specific factors that are often masked by other models. Using quantitative lipidomics and proteomics, we profiled 237 lipids and four apolipoproteins in lipoproteins every four months. Coronary severity was positively correlated with long-chain saturated ceramide levels in very-low- and low-density lipoprotein particles, specifically at eight months, reflecting an early disease stage. This correlation was independent of apolipoproteins and classical risk factors, such as cholesterol and triacylglycerol. Network analysis revealed that lipid class-specific clusters were associated with disease severity. This unbiased multi-omics approach provides novel insights into the contribution of lipoprotein composition to coronary atherosclerosis.

Introduction

Lipids such as cholesteryl esters (CEs) and triacylglycerols (TGs) are transported in the blood to various tissues as lipoproteins, which are covered with phospholipids and apolipoproteins¹. An imbalance in lipoprotein levels is commonly assumed to lead to the onset of atherosclerotic cardiovascular disease (ASCVD). Low-density lipoprotein cholesterol (LDL-C) and high-density lipoprotein cholesterol (HDL-C) levels are commonly used diagnostic markers of abnormal lipid metabolism. Although major cardiovascular events are reduced by decreasing the LDL-C concentration with statins, they are not completely suppressed^2,3. These cohort studies suggest that factors other than cholesterol concentration in lipoproteins can more precisely predict the severity of ASCVD. Recently, the quality of lipoproteins (e.g., small dense LDL and cholesterol efflux capacity of HDL) has attracted attention as a factor involved in cardiovascular events^4,5.

Lipid composition in lipoproteins is related to their quality because lipoprotein particles consist of lipid species and apolipoproteins. Lipidomics, which is a comprehensive analysis of biological lipids using mass spectrometry (MS), has focused on biomarker discovery and the elucidation of lipid metabolism in serious diseases such as atherosclerosis⁶, diabetes⁷, and Alzheimer’s disease⁸. Although the lipid composition in the plasma and lipoprotein fractions has been the focus^9–12, the alteration of lipid composition in lipoprotein particles associated with ASCVD is poorly understood. We have previously developed a novel lipoprotein profiling method to obtain quantitative values of lipid species and apolipoproteins in lipoprotein particles using lipidomics and proteomics techniques^13,14. Single apolipoprotein B-100 (apoB-100) is present in individual very low-density lipoprotein (VLDL) and LDL particles^15,16, and thus, the number of VLDL and LDL particles can be calculated from the apoB-100 concentration in each lipoprotein fraction determined using a targeted proteomics approach. By normalizing the lipidomics data of the lipoprotein fractions to particle numbers, the average number of lipid species in the VLDL and LDL particles can be estimated. In addition, our methodology targets apolipoprotein C-II (apoC-II), apolipoprotein C-III (apoC-III), and apolipoprotein E (apoE), which act as activators or inhibitors of lipoprotein lipase and LDL receptor ligands. The application of our lipoprotein profiling methodology can characterize the lipid composition of lipoprotein particles associated with ASCVD, while monitoring apolipoproteins involved in the regulation of lipoprotein metabolism.

In 1980, Watanabe et al. at Kobe University (Hyogo, Japan) established a line of rabbits with hypercholesterolemia called Watanabe heritable hyperlipidemic (WHHL) rabbits¹⁷. These rabbits exhibited defective LDL receptor function owing to a 12-nucleotide in-frame deletion in exon 4 of the LDL receptor gene, which resulted in the loss of four amino acids from the cysteine-rich ligand-binding domain of the protein¹⁸. Myocardial infarction-prone WHHL (WHHLMI) rabbits are a substrain developed by selectively breeding WHHL rabbits^19,20. These animals spontaneously develop coronary atherosclerosis and myocardial infarction between 11 and 35 months of age, with the incidence increasing from 23% to 97% after selective breeding¹⁹. Lipoprotein metabolism in rabbits is similar to that in humans, but differs from that in mice and rats^21–23. Rodents predominantly produce apolipoprotein B-48 (apoB-48)-containing VLDL because of the expression of the apoB mRNA editing enzyme, APOBEC1, in the liver, which edits the apoB transcript to generate apoB-48. In contrast, similar to humans, rabbits do not express APOBEC1 in the liver; therefore, only apoB-100 is synthesized in the hepatic tissues and incorporated into VLDL. Moreover, the activity of CE transfer proteins in plasma is high in WHHL rabbits, whereas mice and rats lack CE transfer protein activity²⁰. Plasma HDL levels were low in WHHL rabbits but high in mice and rats. In addition to lipoprotein metabolism, atherosclerosis (e.g., expression of VLDL receptors in lesions) and cardiac function (e.g., myosin type) also differ between human/rabbits and mice/rats²⁴. Based on this advantage, we characterized various metabolic changes in response to pitavastatin and D-47, a novel lipid-lowering drug that reduces blood lipid concentrations by a mechanism different from that of statins^25,26. Shiomi et al. reported that coronary atherosclerosis develops with age, and coronary lesions grow rapidly at approximately eight months^20,27,28. Using metabolomics and lipidomics techniques, we identified candidate serum markers for the rapid progression of coronary lesions at an early stage²⁸. Because the serum contains a wide variety of lipoproteins and extracellular vesicles, it is difficult to explain the cause of these alterations. Focusing on individual lipoproteins may provide additional insight.

In this study, we aimed to investigate the association between lipoprotein composition, including lipid species and apolipoproteins, and the severity of coronary atherosclerosis. To meet this demand, MS-based targeted quantitative lipidomics and proteomics techniques were applied to lipoprotein fractions and erythrocytes from WHHLMI rabbits. Twelve lipid classes, including fatty acids (FAs), lysophosphatidylcholines (LPCs), lysophosphatidylethanolamines (LPEs), phosphatidylcholines (PCs), phosphatidylethanolamines (PEs), phosphatidylserines (PSs), phosphatidylinositols (PIs), sphingomyelins (SMs), ceramides (Cers), CEs, monoacylglycerols (MGs), and diacylglycerols (DGs), were quantified at the molecular species level, allowing the determination of individual FA compositions, using normal-phase supercritical fluid chromatography (SFC) coupled to triple-quadrupole MS (QqQMS). In addition, apoB-100, apoC-II, apoC-III, and apoE were absolutely quantified using stable isotope-labeled peptides with nanoflow liquid chromatography (nano-LC) coupled to QqQMS. Using these advanced multi-quantitative omics techniques, we characterized the unique lipidomes of lipoprotein particles in the early stages of coronary atherosclerosis progression.

Results

Assessment of blood markers and pathological indicators

To characterize the compounds involved in the severity of coronary atherosclerosis, the lipid species and apolipoproteins of erythrocytes and/or lipoprotein fractions were analyzed in WHHLMI rabbits. The strategy used in this study is shown in Fig. 1a. All WHHLMI rabbits exhibited a severe aortic lesion phenotype, with lesion areas ranging 84–98%, as determined by calculating the ratio of the lesion area to the total luminal surface area (Supplementary Table 1). To evaluate coronary atherosclerosis, we focused on the left circumflex artery, one of the major coronary arteries and a frequent site of lesion development in WHHLMI rabbits²⁹. The degree of atherosclerosis was assessed using cross-sectional narrowing (CSN) across approximately 20 sections from each animal (Supplementary Table 1). In contrast to the uniformly severe aortic lesions, the frequency and severity of coronary lesions varied among WHHLMI rabbits (Supplementary Table 1). Based on a coronary severity score developed in our previous study, the ten rabbits were classified into a severe group (n = 6) and a mild group (n = 4) (Fig. 1b and Supplementary Table 1)²⁸. No significant differences in body weight were observed between the two groups (Fig. 1c). Despite the presence of severe aortic lesions in both groups, the coronary severity scores clearly differentiated the animals based on the incidence of coronary stenosis (Fig. 1b). Two rabbits classified as mild showed frequent coronary lesions (affecting approximately 70% and 90% of the examined sections, respectively); however, none of the rabbits in the mild group exhibited sections with >75% CSN (Supplementary Table 1).

Fig. 1. Multi-quantitative omics approach for characterizing coronary atherosclerosis development.

a Strategy for characterizing lipid composition in lipoprotein particles using quantitative lipidomics and proteomics techniques. b Severity of coronary and aortic lesions in WHHLMI rabbits at 20 months. Rabbits were divided into severe and mild groups based on the coronary severity score. Circles and triangles represent males (n = 4) and females (n = 6), respectively. c Body weight and hematocrit values of WHHLMI rabbits. Red and blue indicate severe and mild groups, respectively, while circles and triangles represent males and females, respectively. Error bars represent standard deviations of biological replicates. Statistical significance was assessed using Student’s or Welch’s t-test following an F-test for variance.

Erythrocytes and plasma were fractionated by centrifugation of blood collected from the marginal ear vein. In erythrocytes, hematocrit, an indicator of polycythemia, did not increase in the severe group (Fig. 1c). After further fractionation of lipoproteins by ultracentrifugation, total cholesterol, TG, and protein were measured using biochemical analyses (Supplementary Table 2). Total cholesterol and TG concentrations in lipoproteins, which are diagnostic factors for abnormal lipid metabolism in humans, did not change significantly, except for total cholesterol at 12 months (Supplementary Fig. 1).

Relationship between apolipoproteins and coronary atherosclerosis

To assess the differences in lipoprotein metabolism, apolipoproteins in the VLDL and LDL fractions were quantified using nano-LC/MS-based targeted proteomics (Fig. 2a, b and Supplementary Fig. 2a). Peptide sequences of rabbit apolipoproteins were searched using the National Center for Biotechnology Information (NCBI) protein database, and multiple reaction monitoring (MRM) transitions for trypsin-digested peptides were determined using Skyline software (Supplementary Table 3)³⁰. Based on the screening analyses of peptide fragments derived from apoB-100, apoC-II, apoC-III, and apoE, the targeted peptides used for quantification were selected¹⁴ (Fig. 2a), and their ¹³C- and ¹⁵N-labeled peptide internal standards were spiked into the peptide extracts (Fig. 2b). The selection criteria for the targeted peptides were as follows: i) the peptide sequences did not overlap with other sequences (not less than ten amino acids); ii) they had three or four different product ions; iii) they represented a good peak shape and high sensitivity; and iv) they did not co-elute with other sequences. The time-course of apolipoprotein levels is shown in Fig. 2c and Supplementary Table 4. Except for the apoB-100 concentration in the LDL fraction at 20 months, the apolipoprotein concentrations (apoB-100, apoC-II, apoC-III, and apoE) were not significantly different between the severe and mild groups. Although apolipoproteins are known to mediate lipid transport, they were not associated with the severity of coronary atherosclerosis in the WHHLMI rabbits analyzed in this study.

Fig. 2. Targeted proteomics of apolipoproteins using nano-LC/MS.

a Total ion chromatogram of LDL fractions using nano-LC/MS. b Quantification of apolipoproteins using stable-isotope-labeled synthetic standards. c Time-course changes in apolipoproteins within VLDL and LDL fractions. Red and blue indicate severe and mild groups, respectively, while circles and triangles represent males and females, respectively. Error bars indicate standard deviations of biological replicates. Statistical significance was assessed using Student’s or Welch’s t-test following an F-test for variance (^*p < 0.05).

Correlation between apoB-100 and lipid classes

Since apolipoproteins did not change significantly, other components, such as lipid species, seemed to be involved in lipoprotein quality. We further examined the differences in lipoprotein quality between the mild and severe groups. Reanalysis of serum lipidomics data from a previous study²⁸ revealed a continuous age-related trajectory in the principal component analysis (PCA), although distinct age-based clusters were not evident (Fig. 3a). In contrast, sex-based differences were not distinguishable. The lipid species in each sample were annotated using in-house lipid-screening methods developed in our previous study³¹. Our in-house lipid screening methods contain approximately 2500 lipid species, including 22 lipid classes with a variety of 23 FA side chains. Based on the screening of erythrocyte, serum, and lipoprotein fractions, 196, 232, and 237 lipid species were detected in each reference sample containing equal amounts of lipid extracts, respectively. Lipid classes were well separated within 20 min using a diethylamine (DEA) column in normal phase mode (Fig. 3b). Accordingly, the constituent lipid species were quantified by normalizing their ionization efficiencies using a synthetic internal standard mixture containing dodecanoyl- or heptadecanoyl-based lipids (Supplementary Fig. 2b and Supplementary Tables 5–8). The concentrations of each lipid class were then calculated by summing the levels of the corresponding lipid species. Since a single apoB-100, a non-exchangeable apolipoprotein, is contained in individual VLDL and LDL particles, it reflects the number of lipoprotein particles^15,16. The correlation between apoB-100 and each lipid class in the lipoprotein fractions was evaluated each month using Pearson’s correlation coefficient to characterize lipoprotein quality, regardless of the severity of coronary atherosclerosis (Fig. 3c, d). Although the concentrations of apoB-100 and most lipid classes were significantly correlated each month in the VLDL fractions, the number of lipid classes that correlated with apoB-100 was reduced in the LDL fractions.

Fig. 3. Targeted lipidomics of lipoprotein lipids using SFC/MS.

a Serum lipid profiles visualized to assess age, sex, and disease severity. PCA was performed on the autoscaled data (mean-centered and variance-scaled) to assess the variance explained by each principal component. Lipidomics data were reanalyzed from a previous study²⁸. b Separation of lipid classes in LDL fractions using normal-phase SFC/MS. The number of lipids detected is shown. c Correlation between apoB-100 and cholesterol levels in VLDL and LDL fractions at eight months. Red and blue represent severe and mild groups, respectively, while circles and triangles represent males and females, respectively. Correlation analysis was conducted using Pearson’s method. d Correlation between apoB-100 and lipid classes across at 8, 12, 16, and 20 months in VLDL and LDL fractions. Total cholesterol and TG levels were measured using biochemical assay kits.

Time-course alteration of lipid classes

We characterized the lipid classes in erythrocytes and lipoprotein particles involved in coronary atherosclerosis. The number of VLDL and LDL particles was estimated from the apoB-100 concentration analyzed by nano-LC/MS. The lipidomics data of the erythrocytes and lipoprotein fractions were normalized to the number of red blood cells and lipoproteins, respectively (Fig. 4a). A comparison of the lipid classes between the two groups is shown in Supplementary Fig. 3a. A decrease in serum ether PC and an increase in serum Cer levels were observed in the severe group at eight months (Fig. 4b, c). Focusing on lipoprotein particles, Cer levels in the VLDL and LDL particles increased significantly in the severe group at eight months (Fig. 4b, c). The change in serum Cer at eight months was the result of increased Cer concentrations in VLDL and LDL particles. Although the quantitative accuracy varied between the two assays, across different lipid classes, and between the VLDL and LDL fractions, we explored the compositional balance of lipoproteins by calculating the ratio of total phospholipids (measured using MS) to total cholesterol (quantified using a biochemical assay kit). This ratio in VLDL particles decreased only at 20 months (Supplementary Fig. 3b). This result indicates a relative enrichment of non-phospholipid components, such as core lipids (e.g., CEs), in VLDL particles at a later stage. In contrast, the SM/PC ratio remained constant throughout the observation period (Supplementary Fig. 3b). In erythrocytes, Cer did not change significantly from month to month, but PI decreased significantly at 16 months (Fig. 4b, c).

Fig. 4. Time-course of lipid classes in erythrocytes and lipoprotein particles.

a Number of erythrocytes and lipoprotein particles in blood. ApoB-100 molecules in 20 μL of VLDL and LDL fractions were calculated using Avogadro’s constant. Red and blue represent severe and mild groups, respectively, while circles and triangles represent males and females, respectively. Error bars indicate standard deviations of biological replicates. Statistical significance was assessed using Student’s or Welch’s t-test following an F-test for variance. These values were used for normalization of the lipidomics data. b Comparison of lipid classes between severe and mild groups using a volcano plot. Beige areas highlight significantly different lipid classes (p < 0.05). c Changes in representative lipid classes between severe and mild groups across ages. Red and blue indicate severe and mild groups, respectively, while circles and triangles represent males and females, respectively. Error bars represent standard deviations of biological replicates. Statistical significance was determined using Student’s or Welch’s t-test followed by F-test for variance (^*p < 0.05 and ^**p < 0.01). A complete time-course analysis of lipid classes is shown in Supplementary Fig. 3.

Unraveling lipid dynamics through blood compartment profiling in coronary atherosclerosis

We further examined the constituent lipid species in the erythrocytes and lipoprotein particles on a monthly basis to gain specific insights (Supplementary Fig. 4). Similar to the behavior of the lipid classes, serum lipid profiling at eight months characterized 18 lipid species that differed significantly between the severe and mild groups (p < 0.05) (Fig. 5a, b). Notably, nine Cer species, representing more than half of the detected Cer species, were significantly increased in the severe group (Fig. 5b). In lipoproteins, eight and fourteen lipid species in the VLDL and LDL particles, respectively, also exhibited significant changes (Fig. 5a, b). Among these, five Cer species in the LDL particle were increased in the severe group compared to those in the mild group, similar to the result of serum lipid profiling (Fig. 5b, c). In addition, two Cer species were increased in the VLDL particle (Fig. 5b, c). The number of Cer species that significantly changed ranked first or second among all lipid classes, with Cers in the serum and LDL particles showing the highest ratio of changes relative to the total number of lipids detected (Fig. 5b). In particular, saturated Cer species such as Cer 18:1;O2/18:0 and Cer 18:1;O2/22:0 were characterized in both serum and lipoprotein particles during the early stages of atherosclerosis (Fig. 5c).

Fig. 5. Increased Cers in lipoprotein particles with coronary lesion progression at eight months.

a Lipid profiling of erythrocytes, serum, and lipoprotein particles at eight months using a volcano plot. Black circles represent all lipids that showed significant changes (p < 0.05), while red circles indicate significantly altered Cer species. A complete profile of lipid species is shown in Supplementary Fig. 4. b Number of significantly altered lipid species in erythrocytes, serum, and lipoprotein particles. Black areas represent Cer species and the numbers in brackets indicate the number of altered lipids. c Comparison of Cer species in VLDL and LDL particles between severe (n = 6) and mild (n = 4) groups. Black and white bars represent the severe and mild groups, respectively. Error bars indicate standard deviations of biological replicates. Statistical significance was assessed using Student’s or Welch’s t-test after F-test for variance (^* p < 0.05, ^** p < 0.01). d Correlation between coronary severity scores and individual lipid species in LDL particles at eight months. Pearson’s correlation analysis was performed, focusing only on lipids detected in all samples. Lipids with significant correlations (|R | > 0.6) are displayed. Positively and negatively correlated lipids are shown in green and blue, respectively. The network of lipid species correlated with the coronary severity score (|R | > 0.6) was constructed based on pairwise correlation coefficients (|R | > 0.6). Nodes represented individual lipid species, with node size corresponding to the average lipid abundance. The color of the edges was assigned based on the direction of the correlation, with positive correlations in gray and negative correlations in beige.

To explore lipid dynamics from another perspective, we performed Pearson correlation analysis between lipid levels and the absolute coronary severity score (Fig. 5d). This revealed that while phospholipids and CEs showed significant negative correlations with the severity score, Cer 18:1;O2/18:0 was uniquely and significantly positively correlated. In addition, the network analysis highlighted the proximity of lipid classes, suggesting that lipids within the same class tend to behave similarly in the progression of coronary atherosclerosis (Fig. 5d). The clustering of lipids further highlighted the distinct and potentially class-specific roles in atherosclerosis, providing new insights into lipid dynamics in coronary atherosclerosis.

While the significance of Cers in serum and lipoproteins was most apparent at eight months, other lipid classes and time points also showed notable changes. For example, CE species decreased significantly in VLDL and LDL particles at eight months (Fig. 5a, b), while PC species in erythrocytes showed the highest number of alterations among all classes (Fig. 5b). At 16 months, erythrocyte lipids appeared to be affected by the progression of coronary atherosclerosis, with PE species emerging as prominent markers, as suggested by seven significantly altered PE and ether PE species (Supplementary Fig. 4). Early changes in Cer species were observed in VLDL and erythrocytes, but these changes were less consistent across time points (Supplementary Fig. 4). In serum and LDL particles, some Cer species showed significant changes throughout the study period, although the number of altered lipids was significantly reduced compared to that at eight months (Supplementary Fig. 4). These findings highlight the unique lipid dynamics across compartments and time points, and further emphasize the critical change in Cers at eight months in atherosclerosis progression.

Lipid remodeling during lipoprotein metabolism

The lipid composition of VLDL particles is altered during lipoprotein metabolism mediated by lipoprotein and hepatic lipases³². To assess these changes, we evaluated the intra-class composition of DGs, PCs, PEs, and PIs, all of which are either products or substrates of these lipases, at eight months (Supplementary Fig. 5). Differences in the peak areas between the two MRM transitions derived from the FA moieties within the same lipid species reflect variations in the fragmentation efficiency. Therefore, we calculated the normalization factors by comparing the fragmentation efficiencies across a sufficiently diverse set of detected lipids within each class. These correction factors were applied to normalize the MRM-based quantification of diacyl lipids, including DGs, PCs, PEs, and PIs.

PCs and PEs are synthesized from DGs using choline and ethanolamine phosphotransferases, respectively. However, the major FA compositions of DGs, PCs, and PEs differed. The predominant FAs in both DGs and PCs were FA 16:0, FA 18:1, and FA 18:2, each comprising approximately 20–30% of the total content (Supplementary Fig. 5a, b). FA 18:0-containing DGs were relatively scarce (approximately 4%), whereas FA 18:0 was relatively abundant in PCs (approximately 20%) (Supplementary Fig. 5a, b). In contrast, the major FAs in PEs were FA 18:0 and FA 18:2, each accounting for >20% (Supplementary Fig. 5a, b). Although the metabolic pathways diverge in phosphatidic acid toward the synthesis of PEs and PIs, their major FA compositions are similar.

The FA compositions of the DGs in the human plasma reference material (NIST SRM 1950) were primarily derived from FA 16:0, FA 18:1, and FA 18:2, consistent with those of WHHLMI rabbits (Supplementary Fig. 5c)^33,34. Similarly, FA 18:0 constituted only 2% of the total DGs in humans (Supplementary Fig. 5c). The predominant FA components of PEs, FA 18:0 and FA 18:2, were similar in humans and WHHLMI rabbits (Supplementary Fig. 5c). In both species, FA 20:4 was abundant in PEs, comprising >10% of the total (Supplementary Fig. 5c). These findings indicate that humans and WHHLMI rabbits share notable similarities not only in lipoprotein metabolism and atherosclerosis phenotypes but also in their plasma lipid profiles.

Although no significant changes in the overall DG and PC compositions were observed during lipoprotein metabolism, the amounts of certain DGs (such as those containing FA 16:0, FA 18:2, and FA 18:3) and PCs (such as those containing FA 16:0, FA 20:3, and FA 20:4) exhibited changes (Supplementary Fig. 5d). In addition, the DG, PC, PE, and PI compositions were largely comparable between the mild and severe groups, except for the FA 16:1-containing LDL-DG and FA 18:1-containing LDL-PEs (Supplementary Fig. 5a, b). Thus, quantitative intra-class compositional ratios were characterized throughout the progression of coronary atherosclerosis. Our approach may be useful for understanding the packaging of lipids into lipoproteins before VLDL is secreted into the bloodstream.

Discussion

The WHHLMI rabbit is an animal model of coronary atherosclerosis¹⁹. The severity of atherosclerotic lesions in the coronary arteries varied between rabbits, although atherosclerotic lesions in the aorta developed severely in each rabbit (Supplementary Table 1). Because of the uniformly severe aortic lesions observed in WHHLMI rabbits, this model is not well-suited for assessing the relationship between lipid profiles and aortic lesion progression. Therefore, we focused our analysis on lipoprotein profiles in coronary atherosclerosis, where differences in disease severity were more readily detectable. This is particularly important because the extensive burden of aortic lesions may mask serum or plasma biomarkers that are specific to coronary lesions. WHHLMI rabbits develop a broad spectrum of coronary atherosclerotic lesions, including plaques with a large lipid core covered by a thin fibrous cap, fatty streaks, early and advanced fibroatheromas, fibrous lesions, and advanced lesions with calcium accumulation and vasa vasorum³⁵. Matrix metalloproteinase-positive macrophages participate in the initiation, progression, and destabilization of coronary plaques in this model. These findings reveal that the pathophysiological mechanisms governing coronary lesions differ from those involved in the development of aortic lesions, further justifying the emphasis on coronary pathology in metabolic profiling. Since conventional factors (e.g., glucose, insulin, oxidized LDL, superoxide dismutase activity, and NADPH oxidase) did not change significantly between the severe and mild groups²⁸, it is suggested that other factors are involved in the severity of coronary atherosclerosis. The concentration of lipid species in Japanese white rabbits is extremely low compared to that in WHHLMI rabbits¹³. In fact, many lipid species were below the limit of detection in the Japanese white rabbits. Therefore, we investigated the correlation between blood lipid species and the progression of coronary atherosclerosis in WHHLMI rabbits.

The concentrations of total phospholipids, cholesterol, TGs, and proteins in plasma lipoproteins have been characterized in rabbits, including WHHL rabbits; however, the detailed lipoprotein composition remains unknown^36,37. We have previously developed a methodology for lipoprotein profiling using targeted lipidomics and proteomics approaches (Fig. 1a)¹⁴. The targeted lipidomics system has been described in a previous study³¹. Briefly, each lipid class was separated using SFC with the DEA-bonded silica column, because its stationary phase recognizes the head groups rather than the FA side chains of lipid species. Lipid species were measured using the MRM mode of QqQMS. By applying fatty acyl-based MRM transitions in negative ion mode, lipid species, including structural isomers [e.g., PC 16:0_20:4 (840.6 > 255.3 and 840.6 > 303.2) and PC 18:2_18:2 (840.6 > 279.3)], can be distinguished even when co-eluted by SFC. Although the target lipid species must be determined in advance when using the MRM mode, qualitative information can be obtained by screening lipid species in a reference sample using in-house lipid libraries (approximately 2,500 compounds). Lipid species can be analyzed within a reasonable quantification range of 64.9% to 103.5% by adding an internal standard of each lipid class to normalize the corresponding ionization efficiency³¹. In addition, our previous study indicated that each lipid class could be quantified within a plausible value by summing the constituent lipids of the same class³⁸. A targeted proteomics system was also described in a previous study¹⁴. Based on the complete amino acid sequences of apolipoproteins obtained from the NCBI protein database, trypsin-digested peptides of apolipoproteins were determined using Skyline software³⁰. To quantify proteins using the nano-LC/MS system, it is necessary to prepare isotopomers of the target peptides to normalize for matrix effects caused by co-eluted components (Supplementary Table 3). The number of lipoprotein particles can be calculated using the Avogadro constant because a single apoB-100 molecule is present in individual VLDL and LDL particles. The average number of moles in lipoprotein particles was estimated by normalizing the lipid concentration in the lipoprotein fractions to the number of lipoprotein particles.

Lipid species in blood originate from blood cells or plasma. PCs and SMs are distributed outside the membrane in erythrocyte lipid bilayers. In contrast, PEs and PSs are distributed within the membrane³⁹. To transport oxygen, erythrocytes circulate in the blood approximately 200–300 thousand times over a lifetime of 120 days. Since they come into contact with different cells in the course of their circulation, their properties reflect the status in vivo. Polycythemia is caused by an increase in red blood cells, which leads to thrombus and atherosclerosis⁴⁰. In the present study, some lipid species in erythrocytes in the severe group were significantly changed compared with those in the mild group (Supplementary Fig. 4). In particular, PC and PE changed significantly at 8 and 16 months, respectively. These results indicate that membrane lipids are affected by the progression of coronary atherosclerosis in WHHLMI rabbits.

Lipoproteins also reflect atherosclerotic disease status. After VLDLs are formed from lipid species and apolipoproteins in hepatocytes, lipoproteins transport lipids from the liver to peripheral tissues. In lipoprotein metabolism, TGs and phospholipids in VLDLs are hydrolyzed by lipoprotein and hepatic lipases³², and the VLDLs are then converted to LDLs. WHHL rabbits lack functional LDL receptors, similar to familial hypercholesterolemia homozygotes. ApoB-100 levels are elevated in VLDL, intermediate density lipoprotein (IDL), and LDL in WHHL rabbits³⁷. The delayed clearance of VLDL and its metabolites leads to increased production of LDL, resulting in its accumulation in the circulation⁴¹. ApoE levels are markedly elevated in WHHL rabbits, and both VLDL and IDL are enriched in this apolipoprotein. This increase has been attributed to both reduced clearance and upregulation of apoE synthesis⁴². Prolonged exposure of these lipoproteins to the bloodstream increases their susceptibility to oxidative modifications, particularly in LDL. Oxidatively modified LDL play a key role in the initiation and progression of atherosclerosis. For instance, oxidized LDL and its associated lipid species contribute to the formation of atherosclerotic lesions by promoting endothelial dysfunction, inflammation, and foam cell formation^43,44. In the present study, the concentrations of most lipid classes correlated with the number of VLDL particles at all ages, whereas this trend was reduced for the LDL fractions (Fig. 3c, d). This association was observed regardless of the severity of the coronary atherosclerosis. A possible reason for this is that the lipid components of lipoproteins are modified after LDL particles are formed from VLDL particles by hepatic and lipoprotein lipases. In addition, it is possible that the reduction in the ratio of lipid species by hepatic and/or lipoprotein lipases varies from individual to individual.

Although the apolipoproteins that regulate lipoprotein lipase and LDL receptor ligands did not change (Fig. 2c), the lipid composition of VLDL and LDL particles differed between the severe and mild groups. Instead of apolipoprotein abnormalities, our results revealed that the severity of atherosclerosis in WHHLMI rabbits may be more closely related to changes in the lipid composition of lipoproteins. Among all lipid species, Cer species in serum and lipoprotein particles show a unique behavior. The number of Cer species that significantly changed in LDL particles ranked first at eight months (Fig. 5b), and the ratio of Cer species that changed to those that were detected was highest among all lipid classes. Cer, which functions in intracellular signaling, is composed of a sphingosine backbone and FAs. In hepatic cells, Cers are not only metabolized to complex sphingolipids (e.g., SMs, gangliosides, and glucosylceramides) but are also incorporated into VLDLs⁴⁵. To date, several functions of Cers in the pathogenesis of atherosclerosis have been reported. They induce apoptosis and suppress proliferation of cells involved in vascular endothelial injury^46,47. They are also involved in atherogenesis through LDL aggregation in the arterial wall⁴⁸. Through these mechanisms, Cers have become one of the most extensively studied lipid species and have emerged as reliable biomarkers for a range of cardiovascular conditions⁴⁹. Elevated plasma Cer levels have been associated with cardiovascular and metabolic conditions, including insulin resistance in type 2 diabetes⁵⁰, acute coronary syndrome⁵¹, chronic heart failure⁵², and vulnerable plaque phenotypes detected using intravascular ultrasound imaging⁵³. Furthermore, Cer-based risk scores have been developed and validated to predict ASCVD events, highlighting their clinical significance beyond that of conventional risk factors^54–57. A recent study elucidated the mechanism by which G protein-coupled receptors that signal via the Gq pathway, specifically CYSLTR2 and P2RY6, sense circulating long-chain saturated Cers and exacerbate atherosclerosis⁵⁸. These receptors are expressed in endothelial cells and macrophages. Moreover, plasma Cer levels have been shown to correlate with the severity of coronary artery disease in patients with varying degrees of renal impairment, independent of total cholesterol and TG levels⁵⁸. Considering the concordance between our findings in WHHLMI rabbits and those of existing human studies, this model may provide valuable insights into the mechanisms underlying Cer alterations in pathophysiological conditions. WHHLMI rabbits, which spontaneously develop coronary atherosclerosis of varying severities, are suitable for bridging the gap between lipidomics and clinical application. Except for Cer species, other lipid species, such as CE and PI species, were significantly decreased in LDL particles (Fig. 5a, b). PI promotes reverse cholesterol transport by increasing cholesterol efflux from peripheral tissues to HDL and enhancing its transport to the liver for excretion in bile and feces^59–61. Recently, oxidized PI species derived from PI species have been identified in human oxidized LDL and atherosclerotic plaques⁶². Future studies using WHHLMI rabbits will also elucidate these relationships.

We focused on the time-dependent profiles of apolipoproteins and lipid species that correlate with the severity of coronary atherosclerosis at 20 months. In addition, lipid metabolism in organs that interact with erythrocytes and/or lipoproteins should be characterized to understand the cause of blood alterations. The origin of the altered circulating Cer levels remains unclear. WHHLMI rabbits have reduced levels of functional HDL, which impairs cholesterol efflux from macrophages. Therefore, once modified lipoproteins are taken up by macrophages and deposited in plaques, their lipid components are unlikely to return to circulation. In addition, owing to the lack of LDL receptors, LDL particles accumulate chronically in the plasma, making them susceptible to oxidative modification. Oxidized LDL activates sphingomyelinase, promoting SM hydrolysis and subsequent Cer production. In addition to lipoprotein-related pathways, other tissues may contribute to the circulating Cer levels. A previous study revealed that adipose tissue and liver are major sources of circulating Cers⁶³. The thoracic adipose tissue releases extracellular vesicles enriched in Cers and their derivatives, and circulating Cer levels are positively correlated with the abundance of these vesicles⁶⁴. These findings highlight the importance of fractionating both lipoproteins and extracellular vesicles when investigating the source of circulating Cers. Another potential contributor is de novo Cer biosynthesis during hepatic VLDL secretion. Preclinical models have indicated that the enzymes involved in Cer biosynthesis, including serine palmitoyltransferase, Cer synthase, and dihydroceramide desaturase, are potential therapeutic targets⁶⁵. In this way, the increase in Cer species in VLDL and LDL particles requires further investigation of lipid metabolism and coronary atherosclerosis phenotype at eight months.

Limitations of the present study include the small number of WHHLMI rabbits. Recent lipidomics studies in mice have reported minimal sex-related differences in plasma lipid levels, whereas significant differences have been observed in kidney glycolipids⁶⁶. Although lipidomics data in rabbits are limited, WHHLMI rabbits exhibit plasma lipid concentrations that are 5–50-fold higher than those of Japanese white rabbits because of LDL receptor deficiency¹³. This large difference may mask the relatively minor physiological sex differences in lipid levels. Therefore, any observed sex-related differences in lipid composition are more likely to reflect disease-associated mechanisms, such as hormonal regulation of pathological lipid metabolism, rather than normal physiological variations. Although age showed a continuous trajectory in the PCA, sex did not produce a distinct clustering (Fig. 3a). Estrogen protects against atherosclerosis by regulating cholesterol levels, reducing inflammation, and maintaining endothelial function^67–69. Although sex-based differences were not observed, further investigation may provide valuable insights into estrogen-mediated modulation of lipid metabolism under pathological conditions. This is particularly relevant in blood-based lipidomics, which allows for longitudinal monitoring while providing limited direct mechanistic insights. In such cases, the detection of sex-related differences may provide important information for understanding the disease mechanisms and the role of sex hormones in lipid dysregulation.

Improving the resolution of lipoprotein composition and associated pathophysiological features requires deep lipidomics approaches with enhanced sensitivity and coverage. In the current study, ultracentrifugation was used to fractionate the lipoproteins. This enabled the application of large plasma volumes, which compensated for the limited MS sensitivity. However, lipoprotein particles display considerable compositional heterogeneity that is not fully captured by stepwise fractionation alone. Atherogenic variants, such as oxidized LDL and small dense LDL, differ in their lipid composition and apolipoprotein content. As single-cell omics has advanced our understanding of cellular heterogeneity⁷⁰, profiling individual lipoprotein particles is an ideal approach. However, current technologies remain insufficient to resolve the lipid and apolipoprotein compositions of heterogeneous lipoprotein populations with adequate sensitivity and throughput.

Moreover, coronary plaques are much smaller than aortic plaques, thus posing significant challenges for direct compositional analysis by conventional MS, owing to sensitivity limitations. Advanced lipidomics techniques are required to accurately characterize plaque biology, particularly features related to plaque instability. These should include not only conventional lipid classes but also oxidized phospholipids (e.g., oxidized PCs), which are key components of oxidized LDL and have been implicated in plaque vulnerability. To address these analytical challenges, we have recently developed several lipidomics techniques. Oxidized PCs containing epoxide and hydroxide motifs on polyunsaturated FAs are currently being analyzed in the mouse liver using a structural MS/MS library that includes 465 oxidized PC species⁷¹. This study emphasizes the importance of antioxidant treatment during sample storage and pretreatment, as oxidized lipids are often present at concentrations more than 100-fold lower than their precursor lipids. Although the levels of precursor lipids remain unchanged, artificial oxidation during handling can significantly increase oxidized lipid levels, potentially leading to misleading interpretations⁷¹. Emerging mass fragmentation techniques, such as electron activation dissociation and oxygen attachment dissociation, have enabled the structural characterization of lipids, including the precise positioning of polar functional groups, fatty acyl chains, and double bonds^72,73. In addition, a solid-phase enrichment protocol using a metal oxide-coated silica column was developed to enhance lipid annotation⁷⁴. This method allows for the effective fractionation and recovery of various lipid classes, including neutral lipids, FA metabolites and glycolipids, and phospholipids, through electrostatic and coordination interactions. From the standpoint of both analytical sensitivity and lipidome coverage, the continued development of deep lipidomics techniques is crucial for the detailed characterization of lipoprotein particles and lipid composition of small, potentially unstable coronary plaques.

The absolute quantification of lipidomes using MS remains a challenge. Although our targeted lipidomics techniques provide reliable quantitative values for both individual molecules and total lipid classes^31,38, the calculated concentrations are not compatible with those obtained from other types of assays, such as biochemical analyses or commercial kits in all cases. In principle, normal-phase LC can provide plausible quantification by including internal standards for each lipid class. However, CEs and TGs are poorly retained in polar stationary phases, and annotation at the molecular species level and precise quantification are beyond the scope of these methods. Furthermore, direct comparison of quantitative values across different molecules and classes should be avoided because of inconsistent normalization efficiencies. Another important limitation is the potential for sample oxidation during handling. In this study, antioxidants were not added during sample storage. Although previous studies have indicated that precursor lipids remain largely unaffected by artifactual oxidation, the absence of visibly oxidized lipids does not fully guarantee accuracy in untargeted lipidomics. Therefore, MS-based analysis should be regarded primarily as a highly sensitive and high-resolution platform for screening and profiling the lipidome, including minor species, rather than as a tool for absolute quantification. Alternative analytical approaches are more appropriate for applications requiring strict cross-assay comparability, such as establishing biomarker cutoff values. Accordingly, in this study, we focused on lipoprotein profiling during atherosclerosis development by evaluating the time-course alterations of the same molecules and classes rather than the absolute ratios across different lipid classes. The same principle applies to the targeted proteomics of apolipoproteins.

In conclusion, we characterized some lipid species involved in the progression of coronary atherosclerosis in WHHLMI rabbits. In particular, Cer species in VLDL and LDL particles were associated with the severity of coronary atherosclerosis at an early stage, although apolipoproteins and common risk factors for abnormal lipid metabolism did not change. These findings may contribute to the elucidation of pathological mechanisms and the development of novel therapeutic drugs, as well as to further exploration of the components that are directly responsible for pathogenesis.

Methods

Chemicals and reagents

JIS special grade sodium chloride and sodium bromide, sodium hydroxide solution (1 mol L^‒1), and ethylenediaminetetraacetic acid disodium salt dihydrate were purchased from Wako Pure Chemical Industries, Ltd. (Osaka, Japan). Ammonium acetate was purchased from Sigma-Aldrich (St. Louis, MO). MS grade methanol was purchased from Kanto Chemical Co., Ltd. (Tokyo, Japan), HPLC-grade chloroform was purchased from Kishida Chemical (Osaka, Japan), and HPLC-grade distilled water was purchased from Wako Pure Chemical Industries. All synthetic lipid standards were purchased from Avanti Polar Lipids, Inc. (Alabaster, AL, USA). Carbon dioxide (99.5% grade, Yoshida Sanso Co., Ltd. (Fukuoka, Japan) was used as the mobile phase for SFC. Stable isotope-labeled peptides (¹³C- and ¹⁵N- labeled arginine and lysine: IEIPLPFGGK [¹³C₆, ¹⁵N₂] for apoB-100, VQESLSSYWDSAK [¹³C₆, ¹⁵N₂] for apoC-II, and AGQPWELALGR [¹³C₆, ¹⁵N₄] for apoE) were acquired from Eurofins Genomics (Tokyo, Japan). Stable isotope-labeled peptides (¹³C- and ¹⁵N- labeled arginine and lysine; GWVDAGISSLK [¹³C₆, ¹⁵N₂] for apoC-III) were acquired from Genenet Co. Ltd. (Fukuoka, Japan).

Animals

All animal procedures were approved by the Kobe University Animal Care and Use Committee and were conducted in accordance with the Regulations for Animal Experimentation of Kobe University, the Act on Welfare and Management of Animals (Law No. 105, 1973, revised in 2006), Standards Relating to the Care and Management of Laboratory Animals and Relief of Pain (Notification No. 88, 2006), and Fundamental Guidelines for the Proper Conduct of Animal Experiments and Related Activities in Academic Research Institutions under the Jurisdiction of the Ministry of Education, Culture, Sports, Science and Technology (Notice No.71, 2006). Ten WHHLMI rabbits (males, n = 4) were housed individually in metal cages and fed standard rabbit chow (120 g day^‒1; LRC4, Oriental Yeast Co., Ltd, Tokyo, Japan) and water ad libitum. Room was kept with constant temperature (22 ± 2 °C) and lighting cycle (12 h light/dark). Ten WHHLMI rabbits were sacrificed at 20 months of age after blood sampling.

Pathological analysis of WHHLMI rabbits

Hearts and aortas were obtained from WHHLMI rabbits through exsanguination of the carotid artery under sodium pentobarbital anesthesia. The degree of lesions in ten WHHLMI rabbits, including CSN and aortic surface lesion area, was previously evaluated^28,75. The upper quarter of each heart was sectioned into four tissue blocks: Block I (origin of the left and right coronary arteries), Block II (left septal artery), Block III (left circumflex artery), and Block IV (left anterior descending and right coronary arteries)⁷⁵. Each block was embedded in paraffin. The major extramural coronary arteries were serially and transversely sectioned at 200 μm intervals in Block I and at 500 μm intervals in Blocks II, III, and IV. Atherosclerotic lesions in the left circumflex artery (Block III), a major coronary artery, were analyzed. The sections were stained with Victoria blue-hematoxylin and eosin. Approximately 20 sections were examined in each rabbit (Supplementary Table 1). The degree of coronary atherosclerosis was evaluated using CSN. The average CSN was calculated across all examined sections, including those without atherosclerosis. The frequency of sections with 75‒89% CSN and ≥90% CSN was determined by dividing the number of sections in each range by the total number of sections examined. The severity of the coronary lesions was assessed using the coronary severity score, defined as follows:

$$\begin{array}{c}\left[\mathrm{Coronary}\mathrm{severity}\mathrm{score}\right]=\left[\mathrm{average}\mathrm{CSN}\right]\times (1+[\mathrm{requency}\mathrm{of}\mathrm{sections}\mathrm{with}75-89\%\mathrm{CSN}]\hfill \\ \times 10+[\mathrm{frequency}\mathrm{of}\mathrm{sections}\mathrm{with}>90\%\mathrm{CSN}]\times 20)\hfill \end{array}$$

Based on this score, WHHLMI rabbits were classified into two groups: severe (score > 2.0) and mild (score < 0.5). The extent of atherosclerosis in the aorta was assessed by calculating the ratio of the lesion area to the total luminal surface area.

Lipoprotein fractionation

Blood was collected from the marginal ear vein of WHHLMI rabbits after fasting for > 12 h. Blood samples were collected at 8, 12, 16, and 20 months to monitor lipid alterations in erythrocytes and lipoproteins related to the severity of coronary atherosclerosis. Plasma lipoproteins were fractionated by ultracentrifugation using a stepwise method (VLDL, d < 1.006 g mL^‒1 and LDL, 1.019 g mL^‒1 < d < 1.063 g mL^‒1), according to a previously described method^76,77. Prior to fractionation, a sodium chloride solution (d = 1.006 g mL^‒1) and a sodium chloride/bromide solution (d = 1.182 g mL^‒1) were prepared. Additionally, intermediate-density solutions (d = 1.019 g mL^‒1 and 1.063 g mL^‒1) were prepared by mixing these two solutions. For the isolation of VLDL (d < 1.006 g mL^‒1), 3.2 mL of the sodium chloride solution was slowly added to 1.5 mL of plasma in ultracentrifugation tubes using a Pasteur pipette. After ultracentrifugation at 375,000 g for 5 h at 4 °C, the upper and lower layers were carefully separated with a microslicer and collected in clean tubes (upper VLDL fraction, 1.5 mL; lower fraction, 3.2 mL). The fractions were then subjected to a second ultracentrifugation under the same conditions for further purification. As with the VLDL isolation, LDL was purified by repeating the fractionation steps: first, the density was changed to 1.019 g mL^‒1 to remove fractions with d < 1.019 g mL^‒1 (with LDL and HDL collected from the lower layer), and then adjusted to 1.063 g mL^‒1 to eliminate fractions with d > 1.063 g mL^‒1 (with LDL collected from the upper layer). To achieve the target density of 1.019 g mL^‒1 or 1.063 g mL^‒1, a denser solution (d = 1.182 g mL^‒1) was initially added, followed by the appropriate 1.019 g mL^‒1 or 1.063 g mL^‒1 solution to fill the ultracentrifuge tube to 4.7 mL.

Biochemical analysis

Cholesterol and TG concentrations were measured using a Cholesterol E-Test Assay Kit and Triglyceride E-Test Assay Kit (Wako Pure Chemical Industries, Ltd.), respectively, following the manufacturer’s instructions. To determine the sample volume (containing 50 μg of proteins) for apolipoproteins extraction, protein levels were measured using a Pierce™ BCA Protein Assay Kit and Micro BCA™ Protein Assay Kit (Thermo Fisher Scientific Inc., Waltham, MA) following the manufacturer’s instructions.

Lipids extraction

Lipoprotein fractions were separated for lipid and apolipoprotein analyses. Lipid species in erythrocytes and lipoprotein fractions were extracted using Bligh and Dyer’s method, with minor modifications as described in our previous study^31,78,79. Before lipid extraction, a dodecanoyl- or heptadecanoyl-based synthetic internal standard mixture was added to the samples for lipid quantification. The concentrations of the internal standards spiked into the samples were as follows: 1.6 μM (SM 18:1;O2/17:0 and Cer 18:1;O2/17:0); 8 μM (LPC 17:0); 16 μM (MG 17:0, DG 12:0_12:0, and TG 17:0_17:0_17:0); 130 μM (PE 17:0_17:0); 160 μM (LPE 17:1, PC 17:0_17:0, CE 17:0, and FA 17:0); and 1300 μM (PS 17:0_17:0). Samples were mixed with 30 μL of internal standard mixture and 1,000 μL of methanol/chloroform/water (10:5:3, v/v/v) and extracted using a vortex mixer and ultrasonication for 1 min and 5 min, respectively. The solution was then centrifuged at 16,000 g for 5 min at 4 °C, and 700 μL of the supernatant was transferred to a clean tube. The supernatant was mixed with 235 μL of chloroform and 155 μL of water using a vortex mixer for 1 min. After centrifugation at 16,000 g for 5 min at 4 °C, the lower layer was collected.

Apolipoproteins extraction

Apolipoproteins were extracted in four steps: denaturation (reduction and alkylation) of apolipoproteins, methanol and chloroform precipitation, trypsin digestion, and desalting. In short, 50 μg of protein samples was diluted to 220 μL with lysis buffer (100 mM tris-HCl, pH 8.8, 7 M urea, 2% sodium dodecyl sulfate). Samples were mixed with 1 μL of tris (2–carboxyethyl) phosphine hydrochloride solution (final concentration, 2.5 mM) and incubated for 30 min at 37 °C for reduction of apolipoprotein. In addition, they were mixed with 2.5 μL of iodoacetamide solution (final concentration, 12.5 mM) and reacted for 30 min at room temperature (approximately 25 °C) for alkylation of apolipoprotein. After the reduction and alkylation steps, 1200 μL of cold methanol/chloroform/water (600:150:450, v/v/v) was added to samples, and they were centrifuged at 2000 g for 15 min at 4 °C using a swing rotor. Aliquots of the organic layer were removed, and 450 μL of chloroform was added to them. After they were vortexed at the maximum setting for 1 min, they were centrifuged at 2000 g for 15 min at 4 °C using the swing rotor. The supernatant was removed completely and 100 μL of ammonium bicarbonate (100 mM) was added to them. For digestion of proteins, 1 μL of trypsin (500 ng μL^‒1) were added to samples and incubated for 3 h at 37 °C. After this trypsin digestion, 1 μL of trypsin (500 ng μL^‒1) were added to samples again and incubated for 12 h at 37 °C to prevent autodigestion. Styrenedivinylbenzene was used for desalting of samples and the peptide extract was diluted to a final volume of 10 μL with a mixed solvent (0.1% trifluoroacetic acid, 1% acetonitrile and 98.9% water) for the nano-LC/MS analyses.

Targeted quantitative lipidomics using SFC/MS

Lipid analysis was performed using SFC/MS. SFC-MS was performed using an ACQUITY Ultra Performance Convergence Chromatography (UPC²) system (Waters, Milford, MA) with a Xevo TQ-S micro tandem mass spectrometer (Waters) controlled by MassLynx software version 4.1 (Waters). An HPLC 515 pump (Waters) was used as the make-up pump and controlled manually. This was introduced to enhance ionization efficiency under the initial gradient conditions. The SFC and MS analytical conditions were optimized as described in a previous study³¹. The SFC conditions were as follows: mobile phase, liquefied carbon dioxide (A) and methanol/water (95/5, v/v) with 0.1% (w/v) ammonium acetate (B); make-up solvent, methanol/water (95/5, v/v) with 0.1% (w/v) ammonium acetate; injection volume, 1 μL; flow rate, 1.0 mL min^‒1 (mobile phase) and 0.2 mL min^‒1 (make-up pump); modifier gradient; 1% (B) (1 min), 1–65% (B) (11 min), 65% (B) (6 min), 65–1% (B) (0.1 min), 1% (B) (1.9 min); temperature of column manager, 50 °C; temperature of sample manager, 4 °C; analytical time, 20 min; and column, Torus DEA (3.0 × 100 mm, particle size: sub-1.7 μm, Waters). The MS analysis conditions were as follows: capillary voltage, 3.0 kV; desolvation temperature, 500 °C; cone gas flow rate, 50 L h^‒1; and desolvation gas flow rate, 1000 L h^‒1. The MRM parameters were as follows: limit on number of MRM transitions, 150; dwell time, 1 ms; MS inter-scan and inter-channel delay, 2 ms; and polarity switch inter-scan, 15 ms.

Targeted quantitative proteomics using nano-LC/MS

Apolipoproteins were quantified using nano-LC/MS. Nano-LC-MS was performed using an Ultimate 3000 system (Thermo Fisher Scientific, Waltham, MA, USA) with an LCMS-8060 tandem mass spectrometer (Shimadzu) and controlled using Labsolution software version 5.6 (Shimadzu). Nano-LC and MS analytical conditions were optimized based on our previous study¹⁴. The nano-LC conditions were as follows: mobile phase, water with 0.1% formic acid (A) and acetonitrile with 0.1% formic acid (B); injection volume, 1 μL; flow rate, 250 nL min^‒1; modifier gradient; 0% (B) (7 min), 0–65% (B) (38 min), 65–100% (B) (5 min), 100% (B) (15 min), 100–75% (B) (2 min), 75–0% (B) (8 min), 0% (B) (15 min); temperature of column manager, room temperature (approximately 25 °C); temperature of sample manager, 4 °C; analytical time, 90 min; and column, L-column 2 (0.1 × 150 mm, particle size: 3 μm, CERI, Tokyo, Japan). The MS analysis conditions were as follows: electrospray voltage, 2.0 kV; desolvation line, 250 °C; heat block temperature, 400 °C; drying gas flow, 0 L h^‒1; and gas pressure of collision-induced dissociation, 0.27 MPa. Each peptide (IEIPLPFGGK for apoB-100, VQESLSSYWDSAK for apoC-II, GWVDAGISSLK for apoC-III, and AGQPWELALGR for apoE) was measured to quantify the apolipoproteins. The MRM transitions of each peptide are summarized in Supplementary Table 3.

Data analysis

Serum lipidomics data were obtained from our previous study²⁸. In the present study, we analyzed lipid species in erythrocytes and lipid species and apolipoproteins in lipoprotein fractions. The annotation and peak picking of the lipid species were carried out using MassLynx software version 4.1. Lipidomics data were normalized to the number of erythrocytes and lipoprotein particles. In this study, the quantification type of the SFC/MS system corresponds to Level 2 of the Lipidomics Standards Initiative International Guideline, indicating that the lipid class of the analyte and internal standard is identical, with co-ionization of both analyte and internal standard^80,81. Although absolute comparisons across lipid classes remain challenging, they enable robust and reproducible relative quantification of lipid species across biological samples. Additionally, the nano-LC/MS system enables absolute quantification, as fully prepared ¹³C- and ¹⁵N-labeled internal standards were used in this analysis.

The fragmentation efficiencies of the diacyl phospholipids were normalized to evaluate the lipid composition within their respective classes. Lipid peak areas were obtained using Multi-ChromatoAnalysT software version 1.4.0 (Beforce Inc., Fukuoka, Japan) (https://www.be-force.co.jp/en)⁷⁹. Differences in the peak areas between the two MRM transitions derived from the FA moieties of the same lipid species (such as [FA 16:0–H]^– and [FA 18:2–H]^– in PC 16:0_18:2 detected in the negative ion mode) reflect variations in the fragmentation efficiency. Therefore, correction factors to normalize these efficiencies were calculated using simultaneous equations based on the product ion intensities obtained from the MRM analysis. Correction factors for diacyl lipids, including PC, PE, PI, and DG, were determined for each FA composition. The lipid compositions of the DG and PE classes in human plasma were determined using concentration data from individual lipids available in an online database (SRM1950-DB) at https://srm1950-dta.wishartlab.com³⁴.

Visualization and statistical analysis

The individual lipid profiles of the mild and severe groups were visualized using a volcano plot. Statistical significance was determined using either the Student’s or Welch’s t-test after F-test for variance. Correlations between apoB-100 and lipid classes, and between coronary severity scores and lipid species were evaluated using Pearson’s correlation coefficient. Statistical significance was set at ^*p < 0.05 and ^**p < 0.01.

PCA was performed using the prcomp function in the R programming environment on autoscaled data (mean-centered and variance-scaled). The explained variance was calculated as the ratio of the squared singular values, and the results were visualized using ggplot2. PCA score plots were generated for the first two components, with data points color-coded by age (4, 8, 12, 16, and 20 months) and shaped by sex (male and female). The heatmap was generated using the pheatmap, tidyverse, and openxlsx packages. Data matrices were prepared by excluding metadata columns (age and lipoprotein) and then annotated with metadata for visualization. It was generated without clustering for rows or columns using a custom divergent color scale ranging from blue to red, with white as the midpoint. Lipoprotein types were color-coded (VLDL: cyan; LDL: yellow-green). Network analysis of lipid species was performed using the R packages igraph, tidygraph, and ggraph. First, Pearson correlation coefficients were calculated between each lipid species and the response variable (coronary severity score), and lipids with |R | >0.6 were selected. Next, pairwise correlation coefficients were calculated between the selected lipids, and lipid pairs with |R | >0.6 were connected as edges to construct the network. Nodes represented individual lipid species, with node size corresponding to the average lipid abundance and node color indicating lipid classes. Edge thickness reflects the correlation strength (|R | ) between lipid pairs. The color of the edges was assigned based on the direction of the correlation, with positive correlations in gray and negative correlations in beige. The Kamada-Kawai layout was used to highlight the relationships between lipid species.

Author contributions

H.T., Y.I., T.K., M.M., M.S., and T.B. designed the study, including the sampling points for WHHLMI rabbits and analytical strategies for lipids and apolipoproteins. Y.K. performed time-course sampling of WHHLMI rabbits. H.T. and Y.K. fractionated lipoproteins and erythrocytes. H.T. extracted and analyzed the lipid molecules. H.T., K.N., and K.H. extracted and analyzed the apolipoproteins. H.T. and Y.I. wrote the manuscript. The other authors revised and approved the manuscript.

Data availability

The lipidomics and proteomics MS raw data have been deposited in the MB-POST repository with the data set identifier MPST000014.0 and in the ProteomeXchange Consortium via the jPOST partner repository with the data set identifier PXD061885 (JPST003706), respectively. The numerical source data behind the graphs can be found in the Supplementary information.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s44325-025-00091-5.