pubmed.ncbi.nlm.nih.gov

A large-scale genome-lipid association map guides lipid identification - PubMed

A large-scale genome-lipid association map guides lipid identification

Vanessa Linke et al. Nat Metab. 2020 Oct.

Abstract

Despite the crucial roles of lipids in metabolism, we are still at the early stages of comprehensively annotating lipid species and their genetic basis. Mass spectrometry-based discovery lipidomics offers the potential to globally survey lipids and their relative abundances in various biological samples. To discover the genetics of lipid features obtained through high-resolution liquid chromatography-tandem mass spectrometry, we analysed liver and plasma from 384 diversity outbred mice, and quantified 3,283 molecular features. These features were mapped to 5,622 lipid quantitative trait loci and compiled into a public web resource termed LipidGenie. The data are cross-referenced to the human genome and offer a bridge between genetic associations in humans and mice. Harnessing this resource, we used genome-lipid association data as an additional aid to identify a number of lipids, for example gangliosides through their association with B4galnt1, and found evidence for a group of sex-specific phosphatidylcholines through their shared locus. Finally, LipidGenie's ability to query either mass or gene-centric terms suggests acyl-chain-specific functions for proteins of the ABHD family.

PubMed Disclaimer

Conflict of interest statement

Competing Interest Statement

The authors declare that they have no conflicts of interest with the contents of this article.

Figures

Extended Data Fig. 1
Extended Data Fig. 1. Identified lipids and unidentified features occupy characteristic regions in the m/z vs. RT space

a, In plasma, we quantified 1,721 lipidomic features, 621 of which were identified, and b, In liver, we quantified 1,562 lipidomic features, 615 of which were identified. Abbreviations: m/z (mass-to-charge), RT (retention time).

Extended Data Fig. 2
Extended Data Fig. 2. Lipid profiling and subsequent QTL mapping reveals clusters of associated lipids

a, Lipid class distribution of all 1,721 plasma and b, 1,562 liver lipidomic features. c, 1,405 plasma and d, 1,190 lipid features showed at least one QTL with an LOD > 6 as displayed in a Manhattan plot (n = 3,353 and 2,269 total QTL, respectively). Hierarchical clustering of these features against the 69,005 markers on the mouse genome, resulted in clustering of lipid class based on hotspots at the genetic level. Abbreviations: Chr (chromosome), DO (diversity outbred), QTL (quantitative trait loci), LOD (logarithm of odds).

Extended Data Fig. 3
Extended Data Fig. 3. Apoa2 as the candidate gene at the largest lipid hotspot

a, 255 plasma (black) features mapping to the apoa2 locus on chromosome 1 share an allele effect pattern with upregulation in the 129 allele, while 2 mapping liver features (white) do not share the pattern (based on hierarchical clustering on allele effects, with a Euclidean distance cutoff of h = 1.5). b, The allele effect is exemplary replicated in an independent experiment of founder strain plasma CE(18:2) levels (n = 4 for each sex and strain, boxplots are defined with the first and third quartiles (25th and 75th percentile) for lower and upper hinges, 1.5x interquartile range for the length of the whiskers, center line at median (50% quantile)). c, The same pattern was not visible in previously reported Apoa2 liver protein and RNA allele effects. Abbreviations: CE (cholesteryl ester), FS (founder strain).

Extended Data Fig. 4
Extended Data Fig. 4. B4galnt1 as the candidate gene at the hotspot with the largest LOD

a, The selection of B4galnt1 as the candidate gene for the chromosome 10:127 Mbp locus was corroborated by NOD-specific allele effects in previously reported liver eQTL and b, pQTL. c, The allele effect patterns of the later as gangliosides identified features mapping to the B4galnt1 locus could further be validated in an independent experiment of founder strain mice (exemplar GM3 pattern, n = 4 for each sex and strain, boxplots are defined with the first and third quartiles (25th and 75th percentile) for lower and upper hinges, 1.5x interquartile range for the length of the whiskers, center line at median (50% quantile)). Abbreviations: FS (founder strain), Mbp (megabase pair).

Extended Data Fig. 5
Extended Data Fig. 5. Allele effects characterize genome-lipid hotspots

a, Hierarchical clustering of allele effects at Chr 6:91 Mbp resulted in 21 features with matching A/J down effect (main cluster featuring the six B6 male specific features (red) after row-scaling and Ward clustering, cutoff at h=5). b, Consistently, the pattern of male >> female was observed for each of the FS except for A/J as visible in the example for m/z 1130 (n = 4 for each sex and strain, boxplots are defined with the first and third quartiles (25th and 75th percentile) for lower and upper hinges, 1.5x interquartile range for the length of the whiskers, center line at median (50% quantile).) c, Hierarchical clustering of allele effects at Chr 5:31 Mbp locus resulted in 10 features with matching B6 and NZO up effect (main cluster featuring LysoPC 14:0 (turquoise) after row-scaling and Ward clustering, cutoff at h=8). d, This pattern could be replicated in the FS (n = 4 for each sex and strain, boxplots are defined with the first and third quartiles (25th and 75th percentile) for lower and upper hinges, 1.5x interquartile range for the length of the whiskers, center line at median (50% quantile)), as shown for LysoPC 14:0, as well as e, in opposite directionality in a liver eQTL. f, Hierarchical clustering of allele effects at Chr 7:79 Mbp locus resulted in 8 features with matching WSB down effect (main cluster featuring PUFA-containing phospholipids (turquoise) after row-scaling and Ward clustering, cutoff at h=2.5). g, The mapping phospholipids contained polyunsaturated fatty acids such as 20:4 and 22:6. h-i, Abhd2 liver RNA and protein allele effects matched with an opposite WSB high effect. Abbreviations: DO (diversity outbred), FS (founder strain), Chr (chromosome), Mbp (megabase pair), PC (phosphatidylcholine), PUFA (polyunsaturated fatty acid).

Extended Data Fig. 6
Extended Data Fig. 6. Overexpressing ABHD1 and ABHD3 results in distinct phospholipid signature

a, Experimental design of the validation experiment featuring three technical and four biological replicates of Hepa1-6 cells either untransfected (CTL), transfected with a His-tag GFP control (GFP), or transfected with MYC-tagged ABHD1 or ABHD3. b, Western blot of Hepa1-6 overexpression of ABHD1 and ABHD3. Shown is an overlay of membrane and ECL blot for MYC-tag. c, Heatmap of top 49 features from discovery lipidomics experiment with p < 0.05 (ANOVA, Fisher’s LSD post-hoc). Features were sum-normalized and log2-transformed. Hierarchical clustering (Ward clustering, Euclidean distance) shows two clusters with opposite fold changes distinguishing between ABHD1 and ABHD3 and the GFP control.

Extended Data Fig. 7
Extended Data Fig. 7. Lipid class abbreviations and identifications with respective adduct types

As searched for in LipiDex databases (see Methods).

Figure 1.
Figure 1.. LC-MS/MSL lipidomics and QTL mapping as ways to lipid identification.

a, A modified MTBE lipid extraction was performed on plasma and liver from 64 FS and 384 DO mice. b-d, Lipid extracts were analyzed by LC-MS/MS. Identifications were obtained through LipiDex based on retention time window (b), exact mass (c), retention time window and tandem mass fragmentation (d). e, Quantitative values over large dynamic ranges for both identified and unidentified features were obtained. f, All lipidomic features (identified and unidentified) were then mapped onto the mouse genome via QTL mapping, revealing genomic position and founder strain allele effect pattern as results for each QTL. This additional information enabled identification of otherwise unidentified features. Abbreviations: MTBE (methyl-tert-butyl ether), FS (founder strains), DO (diversity outbred), LC-MS/MS (liquid chromatography - tandem mass spectrometry), QTL (quantitative trait loci), m/z (mass-to-charge), RT (retention time), LOD (logarithm of odds).

Figure 2.
Figure 2.. Large scale lipid quantitative profiling and subsequent QTL mapping reveals hotspots of associated lipids.

a, In plasma, we quantified 1,721 lipidomic features, 621 of which were identified, and in liver, we quantified 1,562 lipidomic features, 615 of which were identified. Hierarchical clustering of all 3,283 lipidomic features’ intensities by the 384 DO mice resulted in distinct clustering by lipid class, notably across tissue type. b, When mapped onto the mouse genome, 1,405 plasma and 1,190 liver features showed at least one QTL with an LOD > 6 as displayed in a Manhattan plot (n = 3,353 + 2,269 = 5,622 total QTL). A number of lipid hotspots are shared by identified lipids and unidentified features (e. g. at Apoa2), while others only appear among the unidentified features (e. g. at B4galnt1). Abbreviations: MTBE (methyl-tert-butyl ether), Chr (chromosome), DO (diversity outbred), ESI (electrospray ionization), LC-MS/MS (liquid chromatography - tandem mass spectrometry), QTL (quantitative trait loci), m/z (mass-to-charge), RT (retention time), LOD (logarithm of odds).

Figure 3.
Figure 3.. Co-mapping of lipids at the Apoa2 locus facilitated identification of additional cholesteryl esters.

a, One lipid hotspot on chromosome 1 at 171 Mbp is shared by 255 plasma lipid features co-mapping with a common 129 high allele effect (Extended Data Figure S3a). b, The candidate gene at this locus is Apoa2, which encodes for apolipoprotein II, which is carried on HDL cholesterol particles along with c, a variety of lipid classes, mostly phospho- and sphingolipids, which mapped to the locus. d, When plotting all 255 Apoa2-specific lipid features in the m/z-RT plane, a group of unidentified features sharing the RT region with CEs stood out. e, Notably, all six CEs show their primary QTL at this locus, as visible from their individual LOD plots. f, Subsets of the unidentified features could subsequently be identified as CE-related features, including heterodimers, cholesterol-adducts and in-source fragments. Abbreviations: HDL (high-density lipoprotein), CE (cholesteryl ester), QTL (quantitative trait loci), LOD (logarithm of odds), m/z (mass-to-charge), RT (retention time), FA (fatty acid), PC (phosphatidylcholine), PE (phosphatidylethanolamine), PI (phosphatidylinositol), SM (sphingomyelin), Cer (ceramide), AC (acylcarnitine), RKMD (referenced Kendrick mass defect).

Figure 4.
Figure 4.. Lipid features mapping to B4galnt1 lead to identification of GM2 and GM3 gangliosides.

a, A hotspot of solely unidentified features with exceptionally strong correlation was composed of 11 liver and 15 plasma features mapping to chromosome 10:127 Mbp with b, a similar NOD-driver allele pattern (two main clusters from hierarchical clustering, row-scaled, Euclidean cutoff of h=2.5). Two groups of lipid features (circles vs. triangles) emerged as distinct in strength of LOD (a), directionality of allele effect (b), and m/z space (c). d, The candidate gene B4galnt1 pointed us to the putative identifications of GM3 (circles) and GM2 (triangles) gangliosides, which were confirmed by e, spectral matching with a human GM3 standard. f, Secondary QTL for these gangliosides, as exemplary shown for GM2 d18:1_22:0, mapped to eight additional candidate genes within 4 Mbp of the 15 total ganglioside hotpots that were previously linked to ganglioside metabolism. g, The various candidate genes influencing GM3 and GM2 levels span well-known enzymes (e. g. B3galt4) but also include indirect affectors including Cog2 and Slc9a6. Abbreviations: NOD (non-obese diabetic mouse strain NOD/ShiLtJ), Mbp (megabase pair), LOD (logarithm of odds), QTL (quantitative trait loci), Chr (chromosome), m/z (mass-to-charge), RT (retention time), Glc (glucose), GalNAc (N-acetylgalactosamine), Gal (galactose), SM (sphingomyelin), Cer (ceramide), NGNA (N-glycolylneuraminic acid), NANA (N-acetylneuraminic acid).

Figure 5.
Figure 5.. Web resource LipidGenie guides exploration of genome-lipid connections.

a, We quantified 2,558 features in B6 plasma (n=4 for each sex). 254 features were sex-specific (FC > 1.0, p < 0.05, non-paired, two-sided Student’s t-test). Precursor m/z (±10 ppm) matching to our DO database provided genetic information for ⅓ of the otherwise unidentified features. b, Six male-specific unidentified features (red) share a QTL on Chr 6:91 Mbp with a common A/J down effect (Extended Data Figure S5a). c, The features further clustered in m/z-RT space. d-g, Targeted fragmentation spectra (exemplary spectra for two species ([M+H]+ m/z 1156 and 1158) in positive (MS2) and negative (MS2 and MS3) mode) exhibited signals consistent with a lipid class built of a PC headgroup and three FAs. h, The DO database further confirmed LysoPC 14:0 mapping to Abhd160 with j, an enrichment of FA 14:0 containing lipids k, We compare Hepa1-6 overexpressing ABHD1 and ABHD3 versus a control overexpressing GFP (n=12 for each, 4 biological x 3 technical replicates, boxplots are defined with first and third quartiles for lower and upper hinges, 1.5x interquartile range for the length of the whiskers, center line at median). The boxplots show absolute FC of each mutant over GFP by lipid class; the dashed line is at FC=0.4 l, The lowest (negative) FC for both is observed for LysoPC 14:0; isomers are summed. m, All 14:0 containing PCs exhibit a negative FC for ABHD1 and ABHD3 mutants consistently, while 18:0 containing species are showing opposing positive FC. Plotted are sum-normalized, log2-transformed FC means with error bars representing 95% confidence interval, significance indicated by * (p < 0.05), ** (p < 0.01), *** (p < 0.001) of non-paired two-sided Student’s t-test, equal variance, n=12 for each, details in source data. Abbreviations: DO (diversity outbred), FC (fold change), m/z (mass-to-charge), QTL (quantitative trait loci), Chr (chromosome), Mbp (megabase pair), SNP (single nucleotide polymorphism), PC (phosphatidylcholine), F/M (female-to-male), RT (retention time), FA (fatty acid), PE (phosphatidylethanolamine), GFP (green fluorescent protein)

Similar articles

Cited by

References

    1. Han X. Lipidomics for studying metabolism. Nat. Rev. Endocrinol 12, 668–679 (2016). - PubMed
    1. Yang L. et al. Recent advances in lipidomics for disease research. J. Sep. Sci 39, 38–50 (2016). - PubMed
    1. Kind T. et al. LipidBlast in silico tandem mass spectrometry database for lipid identification. Nat. Methods 10, 755–758 (2013). - PMC - PubMed
    1. Gross RW & Han X. Lipidomics at the Interface of Structure and Function in Systems Biology. Chemistry & Biology vol. 18 284–291 (2011). - PMC - PubMed
    1. Cajka T & Fiehn O. Comprehensive analysis of lipids in biological systems by liquid chromatography-mass spectrometry. Trends Analyt. Chem 61, 192–206 (2014). - PMC - PubMed

Publication types

MeSH terms

Substances