Abstract
Aims Common chromosome 9p21 SNPs increase coronary heart disease (CHD) risk, independent of “traditional lipid risk factors”. However, lipids comprise large numbers of structurally-related molecules not measured in traditional risk measurements, and many have inflammatory bioactivities. Here we applied lipidomic and genomic approaches to three model systems, to characterize lipid metabolic changes in common Chr9p21 SNPs which confer ∼30% elevated CHD risk associated with altered expression of ANRIL, a long ncRNA.
Methods and Results Untargeted and targeted lipidomics was applied to plasma samples from Northwick Park Heart Study II (NPHSII) homozygotes for AA or GG in rs10757274. Elevated risk GG correlated with reduced lysophosphospholipids (lysoPLs), lysophosphatidic acids (lysoPA) and autotaxin (ATX). Five other risk SNPs did not show this phenotype. Correlation and network analysis showed that lysoPL-lysoPA interconversion was uncoupled from ATX in GG, indicating metabolic dysregulation. To identify candidate genes, transcriptomic data from shRNA downregulation of ANRIL in HEK293 cells was mined. Significantly-altered expression of several lysoPL/lysoPA metabolising enzymes was found (MBOAT2, PLA2G4C, LPCAT2, ACSL6, PNPLA2, PLBD1, PLPP1, PLPP2 and PLPPR2). Next, vascular smooth muscle cells differentiated from iPSCs of individuals homozygous for Chr9p21 risk SNPs were examined. Here, the presence of risk alleles was associated with altered expression of several lysoPL/lysoPA enzymes. Importantly, for several, deletion of the risk locus fully or partially reversed their expression to non-risk haplotype levels: ACSL3, DGKA, PLA2G2A, LPCAT2, LPL, PLA2G3, PNPLA3, PLA2G12A LIPC, LCAT, PLA2G6, ACSL1, MBOAT2.
Conclusion A Chr9p21 risk SNP associates with complex alterations in immune-bioactive phospholipids and their enzymatic metabolism. Lipid metabolites and genomic pathways associated with CHD pathogenesis in Chr9p21 and ANRIL-associated disease are demonstrated.
One sentence summary Inflammatory phospholipid metabolism defines a cardiovascular disease SNP
Introduction
The association of altered plasma “lipids” with coronary heart disease (CHD) risk has been known for decades, however for some CHD-risk SNPs, there is no association with “traditional lipid measurements”, such as lipoproteins (HDL or LDL) or their constituents: cholesteryl esters (CE) and triglycerides (TG) 1. As a prominent example, the relatively common CDKN2A/2B (rs10757274, A>G) (minor allele frequency = 0.48) SNP on chromosome 9p21 confers ∼30% elevated risk of CHD, but acts independently of traditional lipid risk factors 1. Chr9p21 SNPs, including rs10757274, are believed to alter disease risk through modulation of the long non-coding (lnc)RNA, ANRIL, although both up and downregulation has been associated with risk 2, 3 The ANRIL product is detected in peripheral blood cells, aortic smooth muscle, endothelial cells and heart, and SNPs in Chr9p21 are associated not only with CHD but also numerous cancers 2, 4-6. Cellular studies show that ANRIL lncRNA down-regulates the tumour suppressors CDKN2A/2B by epigenetic regulation, modulating expression of pathways involved in differentiation, apoptosis, matrix remodelling, proliferation, apoptosis, senescence and inflammation 5, 7.
Lipids represent thousands of diverse molecules. However, CHD clinical risk algorithms such as Framingham or QRISK include circulating lipoproteins only 8, 9. Importantly, bioactive lipids that regulate vascular inflammation/proliferation in line with the function of ANRIL and thus maybe directly relevant to Chr9p21-medicated CHD are not included in these measures. Indeed, whether ANRIL mediates its effects via an impact on bioactive lipid signalling has not been examined, and was studied herein using lipidomics.
Here, plasma from a prospective cohort (Northwick Park Heart Study II, NPHSII) which recruited ∼3,000 men aged 50 - 61 years clinically free of CHD in 1990-1991, was analysed using targeted and untargeted lipidomics, followed by validation, metabolic correlation and network analysis 10, 11. Then, gene transcription for lipid metabolic enzymes was mined in data from a cellular ANRIL knockdown study, and from vascular smooth muscle cells differentiated from iPSCs obtained from individuals carrying Chr9p21 risk SNPs12, 13.
Methods
(more details methods are in Supplementary Data)
Patient samples
NPHSII is a prospective CHD study of ∼3000 men 10, 11. Middle-aged men (aged 50–64 yrs) were recruited from 9 general practices in the UK 27-yrs ago. Exclusions included a history of CHD or diabetes. Full information on the cohort is in Supplementary Methods. SNPs were chosen based on known association with altered risk of CHD and described in full in Supplementary Methods. Details of sample sizes, genes, SNPs and average levels of total triacylglycerides (TAG) and total cholesterol determined for these samples, are provided in Table 1. Samples were randomly chosen.
Global lipidomics, informatics and statistics analysis
Lipids were extracted using two consecutive liquid-liquid extractions, first, hexane:isopropanol:acetic acid, then a modified Bligh and Dyer method as outlined in Supplementary Methods 14. Orbitrap datasets were processed using the R version of XCMS (Version 3.4), then using LipidFinder as described in Supplementary Methods 15, 16. This enabled assignment of a putative lipid class to 30 – 50 % of all ions detected. Our approach to statistical analysis of global datasets is described in Supplementary Methods.
Targeted analysis of lysoPLs
Lipids were extracted from plasma using the liquid:liquid extraction method outlined in Supplementary Methods. LC-MS/MS was performed on a Nexera liquid chromatography system (Shimadzu) coupled to API 4000 qTrap mass spectrometer (Sciex) as in Supplementary Methods.
Measurement of ATX
The ATX levels in the plasma were determined using a two-site immunoenzymatic assay with an ATX assay reagent equipped with an automated immunoassay analyzer, AIA-2000 (Tosoh, Tokyo, Japan) as previously described 17.
Measurements of LPA
Quantification of LPA was performed according to a previous method with minor modification 18. Briefly, plasma samples were mixed with a 10-fold volume of methanol containing internal standard (17:0-LPA), then sonicated. After centrifugation at 21,500×g, the supernatants were filtered and subjected to LC-MS/MS. This consisted of a Ultimate3000 HPLC and a TSQ Quantiva triple quadropole mass spectrometer (Thermo Fisher Scientific, San Jose, CA).
Analysis of Affymetrix data from ANRIL down-regulation in cell lines and RNAseq data from iPSC-derived VSMCs
Raw Affymetrix CEL files relating to total transcript expression in HEK 293 cells stimulated with Tetracycline (shRNA ANRIL silenced for 0h, 48h and 96h) were downloaded from the GEO database (accession: GSE111843) and analysed using packages in CRAN and Bioconductor: limma, oligo, ggplot2 12, 19-22, as described in Supplementary Methods. Data from iPSC-derived VSMCs available at GEO (GSE120099) was analysed as described in Supplementary Methods.
Statistics
Statistics for untargeted lipidomics was performed as described in Supplementary Methods. Targeted data are shown as Tukey box plots, * p < 0.05, ** p < 0.01, *** p < 0.005, Mann Whitney U and Student’s t-test were used as described in legends. Correlation analysis was undertaken using Answerminer (https://www.answerminer.com/calculators/correlation-test), using Pearson’s correlation coefficient. To compare the slopes (or “Pearson correlation coefficients”) of regression lines between AA and GG carriers, we used the method described 23. See Supplementary Methods for more details on statistical methods used.
Results
Global lipidomics demonstrates that lysoPLs are reduced in GG plasma versus AA
To capture all lipids (knowns/unknowns), high resolution Orbitrap MS data from long chromatographic separations was analysed using XCMS, then processed for cleanup/assignment to LIPID MAPS categories, using LipidFinder (Figure 1 A,B) 16. Plasma quality was checked through careful comparison with fresh plasma, detailed in Supplementary Data. Most lipid categories were unchanged, however oxidized phospholipids and lysoPCs had elevated somewhat in storage (Supplementary Figures 1-3). This is not unexpected and we include a full discussion of this phenomenon in Supplementary Data. To assess the impact of the rs10757274, A>G SNP, we compared AA (n = 39) with the risk genotype GG (n = 33). The full dataset is provided as Supplementary Data (supplementary data.xlsx, tab 1). Data was analysed first using a Mann Whitney U test, then chromatograms for all features with p<0.075 were manually checked for quality. LipidFinder detected 1878 lipids, with 872 assigned to a category (Figure 1 B). Next, quantile normalization was applied followed by Mann Whitney U test, and then a p-value adjustment using sequential goodness of fit metatest (SGoF) to each subclass24. The SGoF has been shown as especially well suited to small sample sizes when the number of tests is large. This data is shown in volcano plots in Figure 1 C-J, and the p-values are in column M (Supplementary Data.xls, tabs 1,2). Those most affected by genotype were GPLs and unknowns (Figure 1 C-J, Table 2). Following p-value adjustment the number of significantly different lipids was 17, with 7 putatively identified as lysoPC ions and adducts (supplementary data.xlsx, tab 1,2). An additional group of 8 had p-values close to significance at 0.05-0.08. All were reduced in GG plasma. As this method is used as for hypothesis generation only, we next validated our results using gold-standard quantitative targeted methods.
Quantitative targeted lipidomics confirms decreased lysoPLs in the GG samples
The same plasmas were analysed using a targeted fully-quantitative assay for 15 lysoPLs. Of these several lysoPCs significantly decreased, with both lysoPC and lysoPEs all trending towards lower levels in GG (Supplementary Figure 4 A). This was replicated using new samples from NPHSII (n = 47: AA, 49: GG) (Supplementary Figure 4 B). When both datasets were combined (n = 82 – 86/group), all 8 lysoPCs were significantly lower in the GG genotype (Figure 2 A). Thus, lysoPLs are overall suppressed in the GG genotype, with a more robust effect on lysoPCs than lysoPEs.
Significantly altered lysoPLs are not detected in five additional CHD risk-altering SNPs
LipidFinder data was analysed for additional SNPs from the NPHSII cohort, comparing subjects homozygous for the common alleles with subjects homozygous for rare protective alleles for SORT1, LDLR or APOE E2/E2, or rare risk alleles APOA5 or APOE4/E4 (Table 1). For most lysoPLs, levels were not significantly altered, with the exception of one for APOA5 (upregulated, lysoPE(18:1), and one for LDLR (downregulated, lysoPC(18:2)) (Figure 2 B). This indicates that lysoPLs are consistently reduced only in the GG risk SNP rs10757274.
The plasma lysophosphatidic acids(lysoPA)/autotaxin (ATX) axis is dysregulated in the GG group
Next, lysoPL-related metabolites/enzymes were measured. Metabolism of lysoPL to lysoPA in healthy plasma can be mediated by ATX25. Here, we used a targeted LC/MS/MS assay for lysoPAs, and an immunoenzymatic assay for ATX. ATX was significantly decreased (p = 0.026). Based on power calculations (Supplementary Data), an additional set of plasmas was included to increase sample numbers to 95-100 per group for lysoPAs. LC/MS/MS demonstrated overall small reductions, but with several being significantly lower (Figure 2 C,D). Taken with the lysoPL data, this indicates a global suppression of lysoPL/lysoPA/ATX metabolic pathway in the GG group.
Next, correlation analysis was undertaken to determine the contribution of ATX in metabolizing lysoPL to lysoPA. In AA plasmas, ATX showed weakly negative or positive correlations with total lysoPL or lysoPA, respectively (Figure 3 A,B). This agrees with reports that ATX contributes to lysoPL conversion to lysoPA in healthy subjects 25. In contrast, in GG plasmas, these trends were reversed (Figure 3 C,D). Next, we correlated substrates with products (Figure 3 E-H). In the AA group, significant positive correlations were seen for total lysoPA with lysoPL (p = 0.034). Comparing lipids with the same fatty acyl, significant correlation was seen between lysoPA(18:2) and lysoPL(18:2) (p = 0.023) (Figure 3 E,F). This indicates that as the pool of lysoPL increases, the level of lysoPA increases in parallel, consistent with conversion by ATX. This relationship was fully reversed in the GG group, where total lysoPL, lysoPL(18:2) or lysoPL(20:4) were negatively correlated with their corresponding lysoPAs (p = 0.019, 0.054, 0.019 respectively) (Figure 3 G-I). We next analysed correlation slopes for AA versus GG, comparing either lysoPL:lysoPA (Figure 3 E versus G), or lysoPL(18:2):lysoPA(18:2) (Figure 3 F versus H). Both these comparisons revealed significant differences (p = 0.0264 and 0.0029 respectively) 23. These data confirm altered metabolism of lysoPL and lysoPA lipids between genotypes. Specifically, conversion of lysoPL to lysoPA is suppressed in the GG homozygotes.
The direct contribution of ATX to metabolizing lysoPL to lysoPA was next examined by correlating normalized ratios of lysoPC(18:2):lysoPA(18:2) with ATX. In this comparison, we expect that as ATX increases, the ratio of substrate:product will reduce due to their interconversion. Indeed, the For AA plasma, a weak negative correlation was seen (Figure 3 J). In contrast, a significant positive correlation was observed for GG plasma (Figure 4 K). Thus, as ATX increases, a higher ratio of substrate:product was seen in GG, decoupling ATX from metabolizing lysoPL to lysoPA. Comparing the slopes for AA versus GG revealed significant differences based on genotype (p = 0.0157). This further underscores the dysregulation of the lysoPL metabolic pathway in the GG group, suggesting that non-ATX pathways mediate lysoPL to lysoPA conversion. Last, the relative ratios of all lysoPL and lysoPA molecular species were unchanged in the GG versus AA groups (Figure 3 L). Thus, while metabolism of lysoPL/lysoPA by ATX is altered, there was no influence of genotype on molecular composition overall.
Next, a Pearson correlation analysis looking at relationships between individual lipids and ATX was next undertaken using Cytoscape. For thresholds, the classification system of Schober was used 26. Here, we see that there are moderate (r = 0.40-0.69, green) or strong (r = 0.70-1.00 grey) correlations between lipids of the same class, while there are weak (r = 0.10-0.39, red) correlations between different lipid classes (Figure 4 A). Importantly, the key difference in the dataset is that the weak correlations between classes are positive for the AA group, while they are negative for the GG group (Figure 4 A). Overall, this indicates that these lipids behave similarly within AA subjects. In contrast, in GG plasma, while lipid classes still positively correlate within their groups (e.g. lysoPCs correlate strongly with each other), the links between lysoPL and lysoPA are lost. Instead correlations were negative between lysoPE and lysoPA (Figure 4 A). As in Figure 4, ATX weakly positively correlates with lysoPA in the AA group, but instead with lysoPL in the GG group. This analysis reinforces our findings of altered metabolism for lysoPL/lysoPA, but here at the level of individual lipid species.
ANRIL knockdown significantly alters lipid and lysoPL metabolism gene expression
Chr9p21 risk SNPs are believed to act via altering expression of ANRIL, which regulates cell proliferation/senescence in vitro 2, 4, 5. To examine for a functional link with lysoPL/lysoPA metabolism, we analysed the effect of shRNA downregulation of the proximal ANRIL transcripts EU741058 and DQ485454 in HEK 293 cells at 48 hrs and 96 hrs 12. A GO analysis found significant alterations of several lipid pathways by ANRIL, including Regulation of Lipid Metabolic Processes (GO: 0019216), Phospholipid Metabolic Processes (GO:0006644), Cellular Lipid Metabolic Process (GO:0044255) and Lipid Biosynthetic Processes (GO:0008610), for example, Regulation of Lipid Metabolic Processes was 1.9 or 1.88 fold-enriched (FDR < 0.05, Benjamini-Hochberg) respectively at 48 and 96-hrs respectively (Table 3, Supplementary Data.xlsx, tabs 3,4). Thus, large numbers of lipid-associated genes were significantly differentially regulated (Supplementary Data.xls, tabs 5,6).
We next examined the effect of ANRIL knockdown on 48 candidate lysoPL metabolism genes (Supplementary Data.xlsx, tab 7). Of these, 9 were significantly changed at both timepoints, and another 6 at a single timepoint (Table 4). Several were consistent with lowered lysoPL/lysoPA including reduced PNPLA2, PLA2G4C, increased LPCAT2, MBOAT2, ACSL6, PLBD1, PLPP1, PLPP2 and PLPPR2 (Table 4, Scheme 1). Additional relevant genes were regulated, but in the opposing direction, including decreased LPCAT1 and LPCAT3 and increased LPL, PLA2G7, and DGKA (Table 4). ENPP2 (the gene encoding ATX) was significantly increased by ANRIL suppression (Table 4, Scheme 1). This data is displayed in volcano plots of the full Affymetrix dataset (Figure 4 B,C, Supplementary Figure 5). Genes in red represent significantly-different lysoPL metabolizing genes from the lipid GO pathways.
VSMCs generated from iPSCs from Chr9p21 risk haplotypes show altered expression of lysoPL metabolism genes
VSMCs generated by differentiation of iPSCs from humans homozygous for risk haplotypes in Chr9p21 show globally altered transcriptional networks, dysregulated adhesion, contraction and proliferation, with deletion of the risk haplotype rescuing the phenotype 13. Here, we interrogated the RNAseq dataset of mature iPSC derived VSMCs for expression of the same 48 lysoPL metabolism genes. Multivariate analysis using PCA shows clear separation of cell lines containing the risk haplotypes (RRWT) from controls (NNWT) in PC1 (Figure 5 A). When the risk locus was deleted, the resulting RRKO cell lines instead clustered with NNWT and NNKO (Figure 5 A). This indicates that overall expression of lysoPL metabolizing genes is different in risk haplotype cells, but reverts closer to non-risk (NN) on removal of the 9p21 locus. Examination of individual genes revealed 13 that were significantly different between NNWT and RRWT, where removal of the risk locus in RR led to partial or complete rescue: ACSL3, DGKA, PLA2G2A, LPCAT2, LPL, PLA2G3, PNPLA3, PLA2G12A LIPC, LCAT, PLA2G6, ACSL1, MBOAT2 (Figure 5 B, Supplementary Figure 6).
Discussion
Lipidomics MS has been commonly applied to prospective CHD cohorts that contain no genetic information, while conversely GWAS studies have examined associations with traditional “lipid” measures only (e.g. total cholesterol or triglycerides) 27-41. Cohorts are only now starting to examine the association of individual lipid molecular species with specific risk SNPs, and little information on this is yet available. Here, we show that a common Chr9p21 (rs10757274, A>G) CHD-risk SNP is associated with metabolic alterations to the lysoPL/lysoPA/ATX axis (Figures 1-4). This revealed a genotype-specific defect absent in five other GWAS-proven CHD-risk SNPs. Since the action of rs10757274 GG is independent from “traditional lipid” measurements, it represents a different component of the disease, characterized by changes to bioactive signalling phospholipids (PL), rather than circulating storage/energy lipid pools 1. The three nearest genes to chromosome 9p21 are two cyclin-dependent kinase inhibitors CDKN2A and CDKN2b, and the non-coding RNA ANRIL. Compared to AA individuals, GG have almost 50% lower ANRIL transcription in peripheral blood cells 2, 4, 5. However, there are multiple ANRIL isoforms, including short, long and circular, and reports differ on which are specifically increased or decreased in CVD. In this regard, iPSC-derived VSMC lines from humans with risk haplotypes express higher levels of short ANRIL transcripts 13. The ANRIL gene product upregulates metabolic genes in cultured cells, as well as stimulating VSMC proliferation 2, 5, 12. Here, we showed that either shRNA knockdown of two proximal ANRIL transcripts or the presence of risk haplotypes in VSMCs from humans with risk haplotypes lead to multiple changes in PL-related lipid-metabolizing genes (Figure 4, Scheme 1). In the case of VSMCs, removal of the risk locus indicated that this was controlled by the Chr9p21 locus itself (Figure 5 A,B). Collectively, this suggests that Chr9p21 risk alleles may alter lysoPL/lysoPA in humans via ANRIL regulation, providing novel insights into the biology of this important cause of CHD. With the exception of MBOAT2, which was consistently elevated when ANRIL was lower, genes that were significantly altered were different between the in vitro datasets. This is likely a reflection of the complex and incompletely understood role of ANRIL in CVD risk, in line with the conflicting reports on its expression in risk haplotypes and its incompletely characterized multiple isoforms.
Our initial plasma screen comprised a relatively small sample size, while measuring large numbers of “features”. However, multiple comparison testing/correction (as in genomics), is problematic for this type of data for many reasons including: (i) multiple variables can represent the same metabolite, because it is impossible to remove all duplicates ions, (ii) lipidomic/metabolomics datasets are considerably more heterogeneous than genomics, at least in part because co-efficients of variation for MS data are relatively high. Here, we used LipidFinder as a screening tool only, to identify candidate lipids that we then validated using all the approaches outlined above. Next, we applied quantile normalization followed by Mann Whitney U test, and then a p-value adjustment using sequential goodness of fit metatest (SGoF) to each subclass. The SGoF has been shown to be especially well suited to small sample sizes when the number of tests is large24. This provides the level of confirmation required for identifying underlying pathways and mechanisms (rather than biomarkers). A similar approach led to the discovery (repeated now by many others) of the high importance of trimethylamine N-oxide (TMAO) in CHD pathogenesis 42.
Cohort lipidomics can be fraught with pitfalls, and it is critical these are taken into account to enable findings to be interpreted rigorously. Untargeted methods enable large numbers of samples to be analysed, but do not provide quantitation or full identification of lipids. This major issue inhibits cross-cohort data comparisons, and lacks validation. On the other hand, gold standard targeted methods are not generally amenable to large-scale screening. Here, we used untargeted lipidomics as a first screen only, avoiding naming lipids. This hypothesis-generating tool preceded a full replication and validation in the same and additional samples proving the observations were real, not false positives. We also combined two separate sample sets in our targeted approach to increase statistical power, as described43. LysoPCs/PAs can be formed/metabolized by enzymes in plasma, and the rate of this when samples are at room temperature is not trivial44, 45. This means that sample processing and storage are critical, and that it is essential that time-to-freezing is consistent. If not, it will impossible to reliably compare across samples both within and between cohorts. Lack of consistent sample processing is a major issue underlying well-known problems of reproducibility across cohort studies. In NPHSII, and the Bruneck cohort, which reported similar findings of lower lysoPC in subjects who subsequently developed CVD, sample processing was on site and fast, with immediate low temperature freezing following plasma isolation (Manuel Mayr, personal communication). Thus, artefactual generation of these lipids during plasma isolation will have been minimized, and the reduced levels found will either have occurred in vivo, or gradually during long term low temperature storage. Regardless, it is seen that AA and GG plasmas metabolize lysoPL/lysoPA differently, and understanding the underlying biological differences is important. We provide an expanded discussion on lipidomics methodologies and storage considerations in our Supplementary Data file.
LysoPL/lysoPAs were significantly lower in GG plasmas (Figure 2 A,D), and correlation of lipids and subjects revealed that their metabolism was altered in the risk SNP group (Figures 3, 4 A). This finding could indicate reduced involvement of ATX in metabolising lysoPL to lysoPA (consistent with the reduced levels of ATX detected), but it is likely that there are additional factors responsible that are not understood. One interpretation is that as ATX levels increase in this group, another metabolic pathway that either generates lysoPC or removes lysoPA is concurrently increased, and that the outcome is reflective of this overall cumulative change in enzyme activity. Potential candidates could include the following, which were all identified as altered in the HEK293 dataset: PLA2G7, LPCAT3, LPCAT1 (all generate lysoPL), PLPP1, PLPP2, PLPPR2 (all metabolise lysoPA).
Using the HEK293 dataset, we identified several enzymes that metabolize these lipids, and are known to be expressed in leukocytes, platelets, erythrocytes, heart, adipose tissue and plasma. We first focused on ATX (ENPP2), a plasma enzyme that converts primarily unsaturated lysoPC to lysoPA in healthy subjects (Table 4) 46. However, ATX was lower (Figure 2 C), and lysoPC metabolism by ATX appeared suppressed in the GG group (Figure 3 A-K). Furthermore, all lysoPCs were impacted by GG, including both saturated and unsaturated. Conversely, ENPP2 was elevated in vitro in HEK cells. However, the HEK dataset only measures the impact of ANRIL knockdown on basal expression of ENPP2 and it has been reported that inflammatory cytokine induction of ENPP2 is suppressed by 50% in primary human monocyte derived macrophages that carry the 9p21 risk haplotype allele47. The role of ATX in CHD is not understood, and may vary with underlying genetic cause48. Indeed, while it metabolizes lysoPL to lysoPA in health, in acute coronary syndromes other pathways appear to predominate 46. Here, we found significant downregulation of ATX in GG plasma from middle-aged men who are otherwise healthy and without clinically detectable CHD (Figure 2 C). Thus, switching to ATX-independent metabolism may precede cardiovascular events in this risk group.
In healthy subjects, lysoPLs, particularly lysoPC, circulate at relatively high concentrations, where they could be generated by (i) lipases bound to the cell surface of endothelial cells in liver, heart and adipose tissues (LPL, LIPC, LIPG), (ii) Land’s cycle enzymes in circulating blood cells/platelets 25, (iii) lecithin-cholesterol acyl transferase (LCAT) trans-esterification in the liver, or (iv) by remodelling pathways for platelet activating factor (PAF) removal (Scheme 1). In healthy tissue, lipases predominate, but during vascular inflammation the balance may alter. The Land’s cycle involves phospholipase A2 (PLA2) hydrolysis, although the isoforms controlling blood levels are unknown. Candidates include stromal isoforms and cellular or secreted PLA2s from circulating cells and platelets. Also, a role for circulating/platelet PLA1 from platelets in lysoPL formation has also been proposed 49. In HEK cells, downregulation of PNPLA2 (ATGL, a lipase) and PLA2G4C (PLA2Group IVC, a PLA2 strongly expressed in artery and heart) is consistent with our lipidomics findings of decreased lysoPL (Table 4) 50, 51 (Supplementary Data.xlsx, tab 7, Table 4, Scheme 1).
We also tested for upregulation of potential lysoPL removal pathways following ANRIL knockdown. LysoPL can be metabolized by PLBD1 (Phospholipase B Domain Containing 1, expressed in neutrophils), removing the phospholipid headgroup, and this was significantly elevated in GG (Table 4, Scheme 1) 52. LysoPL can also be recycled back into PL pools via Land’s cycle enzymes, and consistent with this, upregulated LPCAT2 (PC-acyl transferase in blood cells) and MBOAT2 (a PE-acyl transferase in neutrophils) were seen (Table 4, Scheme 1) 50, 53, 54. Also, significant upregulation of ACSL6, a long chain acyl-CoA synthetase expressed in leukocytes and erythrocytes, required for fatty acid re-acylation was noted (Table 4, Scheme 1) 50, 55-57. ACSL6 works in concert with LPCAT2 and MBOAT2.
GG plasma showed significantly lower lysoPAs (Figure 2 D). These can be removed by phospholipid phosphatases, including PLPPR2, PLPP1 (expressed in platelets), or PLPP2 (in lung), and all were significantly induced by ANRIL knockdown 58, 59. In summary, our in vitro analysis provided several new candidates for reducing lysoPL/lysoPA in the context of Chr9p21-mediated CHD risk, including MBOAT2, LPCAT2, ACSL6, PNPLA2, PLA2G4C, PLBD1, PLPP1, PLPP2 and PLPPR2. Based on their known cellular localization, potential sources are proposed (Figure 5 C)
LysoPLs have well characterized in vitro bioactivities, through mediating G protein-coupled receptor (GPCR) signalling that causes immune cell migration and apoptosis. This has led them to be proposed as “pro-inflammatory” 49, 60-63. However, most lysoPL is bound to albumin, immunoglobulins and other plasma carrier systems, and levels are already higher than required for mediating GPCR activation 64, 65. Lower levels of lysoPL in GG plasma suggests they are not pro-atherogenic in this case.
In addition to lipid class-specific changes in phospholipids, many significantly-decreased “unknowns” were found, which are currently absent in databases (Figure 1). The plasma lipidome contains large numbers of such species and a significant challenge lies in their structural and biological characterization. The comprehensive list of all lipids detected with fold-change and significance levels is provided (Supplementary data.xlsx, tab 1) as a resource for further mining.
A final question relates to how ANRIL and lysoPLs are functionally connected. PL metabolism is finely tuned during cell proliferation, with higher concentrations of lysoPC and lysoPE detected at G2/M, that then fall dramatically along with concomitant increases in PC/PE due to acylation during progression to G1 66. This provides the PL membranes required to complete the cell cycle. Given ANRIL’s ability to regulate cell proliferation, and observations that silencing ANRIL prevents division and promotes senescence, the lower levels of lysoPLs in plasma may simply reflect altered rates of cell turnover in the vasculature, but this remains to be determined (Figure 5 C,D).
In summary, we reveal a selective association of altered GPL metabolism with CHD risk in a common risk SNP. In support of our findings, lower lysoPCs are associated with CHD factors such as visceral obesity, and a trend towards higher future risk of acute coronary events, although since lysoPA cannot be measured using shotgun or untargeted methods, we have not found other cohort data that includes this lipid as yet 33, 37, 38. In the Bruneck cohort, inclusion of lysoPCs in classifiers improved power for CHD risk prediction, indicating that although the reduction is rather modest, it is clinically significant37. Furthermore, the Malmö cohort reported that CVD development is preceded by reduced levels of lysoPCs67. Like our study, the reductions were rather modest being only around 8% for each lipid. In Malmö and Bruneck, the lipidomics methods were not as highly validated as in this study, thus our new data provides stronger analytical confidence while linking their findings to a specific risk locus and. Specifically, Bruneck applied a shotgun method with precursor/neutral loss scanning, while Malmö used untargeted MS and did not undertake any validation. Given the prevalence of rs10757274 GG in the general population (∼23%), our data may at least in part explain the findings in Bruneck and Malmö, with lower lysoPLs associating with a sub-group with a common SNP. Last, the alterations in multiple GPL regulatory pathways seen on ANRIL silencing, or the presence/removal of the risk locus in vitro, further indicate the involvement of bioactive lipids in this form of vascular disease, and mechanistic studies are warranted. To this end, fresh blood from AA and GG subjects is required to measure plasma and cellular levels of all candidate enzymes and relevant lipids, in order to identify how lysoPC/lysoPA metabolism is altered by the presence of the risk SNP. It would also be interesting to compare AG with AA and GG men, although for this, considerably larger sample numbers would be required than we have available currently in NPHSII.
Sources of Funding
Wellcome Trust (094143/Z/10/Z), British Heart Foundation (RG/12/11/29815) and European Research Council (LipidArrays) to VBO. VBO is a Royal Society Wolfson Research Merit Award Holder and acknowledges funding for LIPID MAPS from Wellcome Trust (203014/Z/16/Z). SEH acknowledges grant RG008/08 from the British Heart Foundation, and the support of the UCLH NIHR BRC.
Author contributions
SM, JH, DW, PR, AE, MA, YZ, AC, CB, JAJ, RA, VT, CH, DS, JAoki, KK conducted experiments and undertook data analysis. SA supervised computational tool design. JAcharia, JC and JM collected and processed clinical samples. SM, VOD and SH conceived the experiments, designed the studies, and drafted the manuscript. All authors edited and approved the manuscript.
Disclosures
The authors declare no competing interests.
Acknowledgements
We gratefully acknowledge expert discussion with Gerhard Liebisch (Regensberg), during revision of the manuscript.