Lipidomic QTL in Diversity Outbred mice identifies a novel function for α/β hydrolase domain 2 (Abhd2) as an enzyme that metabolizes phosphatidylcholine and cardiolipin

We and others have previously shown that genetic association can be used to make causal connections between gene loci and small molecules measured by mass spectrometry in the bloodstream and in tissues. We identified a locus on mouse chromosome 7 where several phospholipids in liver showed strong genetic association to distinct gene loci. In this study, we integrated gene expression data with genetic association data to identify a single gene at the chromosome 7 locus as the driver of the phospholipid phenotypes. The gene encodes α/β-hydrolase domain 2 (Abhd2), one of 23 members of the ABHD gene family. We validated this observation by measuring lipids in a mouse with a whole-body deletion of Abhd2. The Abhd2KO mice had a significant increase in liver levels of phosphatidylcholine and phosphatidylethanolamine. Unexpectedly, we also found a decrease in two key mitochondrial lipids, cardiolipin and phosphatidylglycerol, in male Abhd2KO mice. These data suggest that Abhd2 plays a role in the synthesis, turnover, or remodeling of liver phospholipids.


Introduction
Lipids play a variety of roles in physiology, including providing structure, signaling and as fuel sources. Disruptions to lipid metabolism can lead to disease states such as obesity [1,2], insulin resistance [3,4], cardiovascular disease [5,6], and hepatic steatosis [7,8]. Manipulations to lipid composition in plasma, tissues, and organelles can have a profound impact on disease susceptibility. For example, alterations in the fatty acid compositions of lipids in the endoplasmic reticulum (ER) have been shown to affect obesity-associated ER stress and to improve glucose metabolism in a leptin-deficient mouse model of obesity [9]. Improvements in detection methods and their sensitivity, such as untargeted lipidomics, have allowed for discovery of previously undefined roles of lipids in physiology. Within the past decade, a new class of lipids (fatty acid esters of hydroxy fatty acids, FAHFAs) have been discovered [10]. For example, the identification of FAHFAs as a novel bioactive lipid class has opened a new field of study into their roles in normal physiology and metabolic disease [11][12][13].
Commensurate with the diversity of lipids is the diversity of enzymes that metabolize lipids. One substantial challenge is discovering the in vivo substrates of lipid metabolizing enzymes and the enzymes responsible for synthesis and turnover of newly discovered lipids.
We have used genetics to assist us in establishing a causal link between enzymes and their substrates. When we perform lipidomic surveys in the context of a segregating population, we can identify loci where specific lipid species are genetically associated with loci harboring genes that encode highly plausible candidate enzymes responsible for the metabolism of the lipids. In a prior study, we showed that the substrate and product of an enzyme in glycosphingolipid metabolism mapped to a locus containing that enzyme [14]. This was a proof-of-principle that genetics could be used to de-orphanize lipid metabolism enzymes.
The same study identified several ABHD members as modulators of lipid classes [14]. In validation experiments, ABHD1 and ABHD3 overexpression revealed distinct specificity for lipid classes and acyl chain lengths. The ABHD family of such enzymes (α/β -hydrolase domain) has 23 known members, which are characterized by a α/β -hydrolase fold and a catalytic serine hydrolase domain [15,16]. ABHD6 is the most characterized lipase in this family, with a wide variety of physiological roles including adipose biology, islet insulin secretion, and cold tolerance [17][18][19][20]. ABHD3, another lipase, was shown to selectively modulate phospholipids with C14 acyl chain lengths [21]. The biological roles of many ABHD family members are still being discovered. Here, we incorporate murine liver untargeted, mass spectrometrybased lipidomics and quantitative trait loci (QTL) genetics to identify α/β-hydrolase domain 2 (Abhd2) as a novel driver of hepatic phospholipids.

Identification of ABHD2 as novel driver of liver phosphatidylcholine
In a recent genetic screen of circulating and hepatic lipids in Diversity Outbred (DO) mice, we identified a quantitative trait locus (QTL) for multiple phospholipids phosphatidylcholine (PC) and phosphatidylethanolamine (PE) on chromosome 7 at~79 Mbp [22]. In parallel, we performed RNA-sequencing to survey the liver transcriptome in the same DO mice that were used for the lipidomic survey, enabling us to identify expression QTL (eQTL) for all genes. We found a strong association of the abundance of the Abhd2 mRNA with SNPs located near the Abhd2 gene (a cis-eQTL) with a LOD of 65. This QTL co-mapped with the phospholipid QTL on chromosome 7 (Fig 1A).
DO mice segregate alleles from eight founder strains. We can identify the contribution of each allele to a given phenotype and display the allele effect patterns. The allele effect patterns for the phospholipids and the Abhd2 eQTL were similar, partitioning the founder haplotypes into two subgroups: CAST and WSB versus B6, A/J, NOD and 129 (Fig 1B). However, the directionality of the haplotype separation was different for the phospholipids and Abhd2 expression. Whereas alleles derived from CAST and WSB were associated with high expression of Abhd2, the same alleles were associated with lower abundance of phospholipids (Fig 1B). Thus, the phospholipids and Abhd2 expression show shared but inverted genetic architecture. This inverse pattern is indicative of a "substrate" signature, suggesting that Abhd2 participates in the phospholipid's degradation.
Next, we identified the SNPs most strongly associated with the phospholipids and the expression of Abhd2. The QTL for PC-20:4 peaks at~79.2 Mbp and includes a block of SNPs with strongest association, which span from~79.2 to~79.4 Mbp on chromosome 7 (Fig 1C). The gene for Abhd2 is located~79.3 Mbp, right under the SNPs with strongest association to PC-20:4. The SNP association profile for the Abhd2 cis-eQTL was the same as that for PC-20:4, suggesting a common genetic architecture for the lipids and Abhd2 expression. There are 67 protein-coding and non-coding genes that are located between 78.2 and 80.2 Mbp on Chr 7 (S1 Table).
We next used mediation analysis to identify a causal gene driver from among the genes present at the phospholipid QTL. In mediation analysis, the QTL for a lipid is conditioned on the expression of all other genes, including those at the locus to which the lipid maps. If the genetic signal of the lipid QTL decreases upon conditioning of the expression level of a specific gene, that gene becomes a strong candidate as a driver for the lipid QTL. We focused on the QTL for PC-20:4, as this demonstrated the strongest genetic signal (Fig 1A). Mediation of the PC-20:4 QTL against the expression of Abhd2 in liver resulted in a large drop in the LOD score for the PC 20:4 QTL (Fig 1D). To extend these observations, we asked if Abhd2 is a strong driver for all phospholipids mapping to the chromosome 7 QTL. For the seven phospholipids with a QTL to the Abhd2 gene locus, mediation against Abhd2 expression resulted in the largest drop in the LOD scores (S1 Fig). In summary, the inverse allele effects for the phospholipid versus Abhd2 expression profiles strongly suggest that Abhd2 functions as a negative driver of the hepatic phospholipid QTL on chromosome 7.

Experimental validation of Abhd2 as a driver of liver phospholipids
To determine if Abhd2 is a key driver of liver phospholipids, we obtained a whole-body knockout of Abhd2 from Dr. Polina Lishko at UC Berkeley [23,24]. Wildtype (WT) and Abhd2 knockout (Abhd2 KO ) mice were maintained on the same Western diet (WD), high in fat and sucrose, that was provided to DO mice used for the lipidomic genetic screen [14].
To experimentally validate the genetic prediction that Abhd2 is a driver of liver phospholipids (PC and PE), we performed mass spectrometry (MS)-based lipidomics on liver tissue from male and female WT and Abhd2 KO mice. A total of 583 unique lipid species were quantified (S2 and S3 Tables), including 67 and 50 PC and PE lipids, respectively. Fig 2 highlights the liver lipids that were the most differentially abundant between WT and Abhd2 KO mice. Female Abhd2 KO mice had 21 liver lipids decreased and 9 lipids increased (Fig 2A), whereas male Abhd2 KO mice showed 44 and 16 liver lipids decreased and increased, respectively (Fig 2B). Consistent with the prediction from the genetic screen, the PC and PE species that mapped to the chromosome 7 QTL were significantly increased in liver from both male and female Abhd2 KO mice (Fig 2C).
In addition to PC and PE, other lipids were significantly altered in the liver of Abhd2 KO mice. For example, several species of cardiolipin (CL) (Fig 2D) and phosphatidylglycerol (PG) (S2A Fig) were significantly reduced in liver from male, but not female, Abhd2 KO mice. CL and PG are synthesized in mitochondria [25] and play important roles in mitochondrial function [26]. To determine if the decrease in CL and PG levels in Abhd2 KO males reflect a change in mitochondrial number, we performed quantitative PCR for several mitochondrial-encoded genes. In both male and female Abhd2 KO mice, the expression of eight mitochondrial-encoded These results suggest that lower levels of hepatic CL and PG in male Abhd2 KO mice are not the consequence of reduced mitochondrial number. It is therefore more likely that ABHD2 plays a key role in the metabolism of these two mitochondrial lipids.
To provide additional support for ABHD2 in regulating hepatic PG and CL levels, we asked if there was genetic association for CL and PG lipid species in liver among DO mice. We identified several QTL for both lipids, including a hotspot on chromosome 3 at~46 Mbp where several CL species co-mapped (S4 Table). CL-16.0/18.1/16.0/18.1 yielded the strongest genetic signal on chromosome 3, with a LOD of~12, along with possible secondary QTL on chromosomes 7 and 13 ( Fig 3A). Interestingly, the gene Abhd18, which is relatively uncharacterized but has been localized to mitochondria [27], is physically located at the CL QTL on chromosome 3, raising the possibility that Abhd18 and Abhd2 work in concert to regulate CL levels. While no CL species mapped to the Abhd2 gene locus on chromosome 7, conditioning CL on PC-20.4/22.6 as an additive covariate resulted in CL acquiring a QTL to the Abhd2 locus (Fig 3A). This QTL on chromosome 7 of CL adjusted by PC demonstrates an allele pattern that is like that of the cis-eQTL for Abhd2, and the inverse of the PC QTL (Fig 3B), consistent with CL being a downstream product of ABHD2-dependent metabolism of PC. Similar results were observed for two PG lipids; when conditioned on PC-20.4/22.6, QTL were acquired to the Abhd2 gene locus (S5 Table). Changes in fatty acyl composition (number of carbons and degree of saturation) have been associated with differential response to metabolic stressors [28,29]. Therefore, we evaluated the composition of the acyl chains in PC, PG, and CL lipids in WT and Abhd2 KO mice (S2C- S2E Fig). Both PC and PG lipid classes were equally represented by acyl chain lengths of C16 and C18; in CLs, however, C18 comprised more than 95% of the acyl chains (S2C- S2E Fig).
PCs were primarily composed of saturated fatty acids, PGs had similar monounsaturated and saturated fatty acyl chains (~43% and 50%, respectively), while 80% of CL fatty acyl chains contained two double bonds (S2C- S2E Fig). Acyl chain length and degree of saturation for PC, PG, and CL species were not different in Abhd2 KO mice. Taken together, these results suggest that ABHD2 is not involved in specific alteration of the acyl chain composition of phospholipids.

Physiological characterization of Abhd2 KO mice
While the increase in hepatic phospholipids we observed in the Abhd2 KO mouse confirms the predictions from the genetic screen that Abhd2 is a negative driver of these lipids, it does not inform us about the physiological role of ABHD2. To gain a better understanding of this, we performed a series of physiological measurements in WT and Abhd2 KO mice.
WT and Abhd2 KO  To evaluate a role for Abhd2 deletion on broad metabolic pathways, we performed an oral glucose tolerance test (oGTT) to assess whole-body insulin signaling and glucose homeostasis, a β 3 -adrenergic receptor agonist tolerance test (β 3 TT) to examine differences in adipose lipolysis and glucose metabolism, and a fast/re-feed (FRF) paradigm to probe liver lipolysis/lipogenesis pathways.
During the oGTT, no differences in plasma glucose, insulin, or C-peptide levels were observed for male or female WT vs Abhd2 KO  Another member of the ABHD family of enzymes, ABHD6, has been shown to have a direct effect on islet insulin secretion by hydrolyzing monacylglycerols, inhibiting MUNC13-1 action and thereby regulating insulin granule release [20]. To directly evaluate the effect of ABHD2 on pancreatic β-cell function, we determined insulin secretion from cultured islets isolated from WT and Abhd2 KO mice. Insulin secretion in response to varying glucose concentrations or monoacylglycerol (2-arachidonoylglycerol or 1-palmitoylglycerol) was the same for WT and Abhd2 KO mice (S8 Fig).
Given that hepatic phospholipids have been shown to play a major role in lipoprotein metabolism and cholesterol homeostasis [30][31][32][33], we measured circulating total cholesterol and triglycerides (TG) in WT and Abhd2 KO mice. Total cholesterol and TG were not different in Abhd2 KO mice (S9A and S9B Fig). To assess whole-body cholesterol metabolism, we measured biliary and hepatic cholesterol content. These remained unchanged in Abhd2 KO   Taken together, while our data supports Abhd2 as a driver of several hepatic phospholipid and cardiolipin (in male) species, we were unable to link these changes to differences in serum lipoproteins, suggesting that the role of ABHD2 in phospholipid metabolism is confined to intracellular lipids.

Discussion
Genetic diversity plays a pivotal role in lipid metabolism and homeostasis. By leveraging genetic diversity of murine populations, it is possible to define novel drivers of physiological traits, including lipid classes.
Through untargeted MS-based lipidomics in the context of a genetic screen, our study is the first to nominate and validate Abhd2 as a genetic driver of hepatic phosphatidylcholine and phosphatidylethanolamine. Phospholipid species (PC and PE) that mapped to chromosome 7 were increased in livers of knockout mice (both sexes), following the substrate signature prediction of our genetic screen. By integrating lipidomics and transcriptomics, we show how a mouse genetic screen can be used to identify novel drivers of hepatic lipids.
Abhd2 has been previously characterized as a monoacylglycerol lipase with potent effects on male fertility [24] and ovulation in female mice [23]. In sperm, ABHD2 is activated by progesterone and cleaves monoacylglycerols (1-arachadonoylglycerol and 2-arachadonoylglycerol) to remove the inhibition of the CatSper calcium channel, thereby allowing for sperm activation. In a gene-trap mouse model of age-related emphysema, loss of Abhd2 resulted in decreased PC levels in bronchoalveolar lavage [34]. These Abhd2-deficient mice had increased lung macrophage infiltration and inflammatory markers and spontaneously developed emphysema with aging. It is interesting that their study showed a decrease in PC lipids with loss of Abhd2, whereas PCs increased in livers of our whole-body Abhd2 KO mice, perhaps highlighting tissue-specific roles of ABHD2. Nevertheless, Abhd2 appears to have a causative role in PC species homeostasis. Our study is the first to demonstrate an in vivo role for Abhd2 in phospholipid regulation in non-reproductive tissues.
An unexpected finding was a decrease in cardiolipins and phosphatidylglycerols in male Abhd2 KO mice. Cardiolipins comprise~20% of the inner mitochondrial membrane, whereas phosphatidylglycerols reside in the outer mitochondrial membrane [26]. To explore a genetic association between PC and CL or PG, we performed QTL analyses in which the PC lipid showing strongest association to the Abhd2 gene locus (PC-20.4/22.6) was used as an additive covariate when mapping CL or PG. This QTL analysis yielded an intriguing result: CL and PG acquired QTL at the Abhd2 locus with an inverted allele signature to that for the PC. This inverted allele signature is also indicative of a substrate signature, where an increase in PC is associated with a decrease in PG and CL. Thus, ABHD2, through its effect on PC, may indirectly play a role in the synthesis of CL species.
One hypothesis for ABHD2's effect on CL biosynthesis is through the role of an acyltransferase. ABHD2 contains two enzymatic motifs: the canonical serine hydrolase motif and the highly conserved HxxxxD acyltransferase motif between H120 and D125. Synthesis of CL involves a transfer of a fatty acyl chain from PC or PE phospholipids to monolysocardiolipin (MLCL) to form mature CL species. Four MLCL species were detected in our liver samples (S3 Table). In males, there was a 2.5-fold reduction in one MLCL species (MLCL-56:6) in Abhd2 KO mice. If ABHD2 affected mature CL synthesis through a direct fatty acyl chain transfer to MLCL, an increase in MLCL species would be expected. Therefore, the reduction in MLCL indicates ABHD2's role is likely upstream of mature CL synthesis. Since PG is also required for CL synthesis, it's also possible that the reduction in CL concentrations is secondary to alterations in PG concentrations [25]. In our initial QTL analyses of all liver lipids, we identified a CL hotspot on chromosome 3 at~46 Mbp, which includes the ABHD enzyme, Abhd18. Recently, ABHD18 was shown to reside in the mitochondria [27]; however, its mechanism has not been well characterized. In the STRING protein-protein association network database (string-db.org), ABHD2 and ABHD18 are predicted to have an interaction, although this has not been experimentally validated [35]. It is possible that ABHD2 mediates utilization of PC or its acyl chains in the synthesis of CL and PG, or that it interacts with another mitochondrial enzyme, such as ABHD18, to effect these changes.
It is important to note that changes to mitochondrial lipids were only observed in male Abhd2 KO mice, whereas the increase in PC and PE phospholipids occurred in both sexes. Progesterone-induced activation of ABHD2 is required for its lipid cleavage function and regulating ovulation in females [23,24]; however, the effect of male sex hormones on ABHD2 has not been demonstrated. In a study of cerebral cortex development, a perinatal testosterone spike in male mice drove mitochondrial lipid composition and maturation [36]. It is possible that ABHD2 is required for testosterone-dependent regulation of mitochondrial lipid synthesis or maturation.
Reduced abundance of PG and CL lipids may indicate a reduction in total mitochondrial number or a defect in the inner mitochondrial membrane leading to altered metabolic function. As a surrogate for mitochondrial number, we measured expression of key mitochondrial genes by qPCR but did not see a sex-specific or genotype effect. Thus, the decrease in CL and PG does not appear to be due to a reduction in mitochondrial number but does not rule out altered mitochondrial function in liver from Abhd2 KO mice.
The monoacylglycerol lipase, Abhd6, has also been shown to modulate mitochondrial lipid metabolism [37,38]. However, the changes in lipid class concentrations were in the opposite direction of the Abhd2 lipids. Loss of Abhd6 results in an increase in liver PG, which was attributed to defective degradation of lysophosphatidylglycerol (LPG) [37]. Another group later showed increased plasma concentrations of bis(monoacylglycerol)phosphate (BMP) in mice lacking Abhd6 and in humans with a loss-of-function mutation in ABHD6 [38]. Both BMP and CL synthesis require PG as a precursor [39]; therefore, it is possible that the reduction of PG and CL content in the Abhd2 KO livers may reflect alterations in one or both of these pathways. Recently, the Abhd2 locus was linked to age-related macular degeneration (AMD) through a human GWAS of mitochondrial variants [40]. Alterations to mitochondrial function and lipid oxidation have been implicated in AMD disease progression [41,42]. We did not assess phenotypes related to vision; however, it would be intriguing to determine a role for Abhd2 in mitochondrial function and AMD disease risk. We observed a reduction in CL in livers of Abhd2 KO mice; thus, we are tempted to speculate that Abhd2 deficiency may contribute to macular degeneration through its effects on CL.
Loss of Abhd2 has been previously shown to regulate vascular smooth muscle migration and induce blood vessel intima hyperplasia after a cuff experiment in a mouse model [43]. The same group showed an increase in macrophage ABHD2 expression abundance in vulnerable plaques in humans [44] but no mechanism of action was determined.
In human genome-wide association studies [45], there is a significant region on chromosome 15 associated with coronary artery disease (CAD). This locus sits between two genes: ABHD2 and MFGE8. Soubeyrand et al. showed that deletion of the intergenic locus results in a marked increase in MFGE8 expression but did not affect the expression of ABHD2 [46]. Knockdown of MFGE8 in coronary smooth muscle cell and monocytes inhibited proliferation, indicating MFGE8 as the causal gene for CAD-associated at this locus [46]. Splice variants of MFGE8 have been associated with reduced risk of atherosclerosis in FinnGen, a large Finnish biobank study [47]. However, an in vivo role for MFGE8 has not been established. In our genetic screen, hepatic expression of Mfge8 did not significantly correlate with hepatic lipids or plasma lipoproteins. We did not observe a difference in plasma lipoproteins with Abhd2 deletion. We did not assess any indicators of vascular smooth muscle physiology or blood pressure. Thus, ABHD2 is likely not the causative gene at the CAD-associated region on chromosome 15 in human GWAS.
All studies were completed in whole-body knockout mice, allowing us to definitively identify Abhd2 as the driver of the observed phenotypes. For this reason, one limitation of our study is that we cannot speak to tissue-specific roles of Abhd2 on metabolic stress and disease risk. It will be important for future studies to carefully consider tissue-specific knockout models to further refine our understanding of Abhd2's role in physiology. Similarly, we did not assess effects of Abhd2 at the subcellular level. Disruptions in phospholipid homeostasis could lead to subcellular dysfunction, such as those seen in lysosomal storage disorders (LSD). We did not observe sphingomyelin or cholesterol accumulation in liver nor evidence of muscle wasting, which are often hallmarks of LSD. However, we did not perform microscopy of lysosomes to directly assess alterations to morphological features and our untargeted lipidomics did not annotate bis(monoacylglycerol)phosphates, the key lipids of lysosomes. Additionally, we did not expand on ABHD2's known mechanisms, namely that it operates as a lipase to cleave monoacylglycerols [24] and also contains an acyltransferase motif [15]. Instead, we identified new substrates for ABHD2 and its potential roles in phospholipid homeostasis.
With biochemical approaches alone, it is challenging to discover novel candidate substrates for known enzymes. Through integration of gene expression data with untargeted, mass-spectrometry lipidomics, we identified a hepatic phospholipid hotspot on chromosome 7 and nominated Abhd2 as a novel driver of PC, PE, and cardiolipin. Using a whole-body knockout mouse model, we validated Abhd2 as the causative gene for several PC and PE lipids, and CLs, precisely as predicted by the QTL analysis. Our study demonstrates the power of metabolite QTL analysis to discover novel candidate substrates for enzymes.

Ethics statement
All animal work was approved by the Institutional Animal Care and Use Committee at University of Wisconsin-Madison under protocol #A005821.

Mouse genetic screen to nominate novel drivers of hepatic lipid metabolism
Details of the mouse genetic screen has been previously described [22]. Briefly, 500 Diversity Outbred (DO) mice were obtained from Jackson Laboratories (Bar Harbor, ME) and maintained on a high-fat, high-sucrose diet (TD.08811, Envigo, Madison, WI) for 16 weeks. For this study, livers from 384 mice (191 female, 193 male) were collected for transcriptomics and untargeted mass spectrometry-based lipidomics. Mapping of gene expression and phenotypes were performed to identify quantitative trait loci (QTL) and nominate candidate drivers for individual lipid species using the GRCm38 genome build and Ensembl 75 for gene annotation. Genome scans were completed with R/qtl2 software [48], using sex and wave as additive covariates. To investigate genetic associations between mitochondrial lipid classes and phosphatidylcholines mapping to chromosome 7, genome QTL scans were performed with sex, wave, and PC-20:4/22:6 as additive covariates. A logarithm of odds (LOD) greater than 6.0 was used as the threshold for identifying suggestive QTL and LOD greater than 7.5 identified significant QTL. As previously described, LOD thresholds were defined through permutation testing to establish a genome-wide family-wide error rate (FWER) for genome-wide QTL [14,49]. Mediation analysis to establish causality was performed by regression of the target phenotype on the locus genotype to establish the direct genetic effect [50]. Next, we included the candidate gene expression as a covariate in the regression. If the phenotype-genotype association was no longer significant in the conditional regression model, we considered the gene to be a mediator of the genetic effect on the target locus.

Abhd2 mouse housing and maintenance
Whole-body Abhd2 heterozygous mice, generated on a C57BL/6N background [23], were a kind gift of Dr. Polina Lishko at University of California-Berkley. All animal work was approved by the Institutional Animal Care and Use Committee at University of Wisconsin-Madison under protocol #A005821. Heterozygous mice were backcrossed with C57BL/6J mice and bred to produce knockout mice and wild-type littermate controls. All mice were housed at the University of Wisconsin-Madison animal facilities with standard 12-hour light/dark cycles. Animals were weaned and provided a high-fat, high-sucrose diet (TD.08811, Envigo, Madison, WI) and water ad libitum. At 23-25 weeks of age, mice were euthanized by carbon dioxide asphyxiation and exsanguinated by cardiac puncture. Whole blood was collected with EDTA, centrifuged at 10,000xg for 10 minutes at 4˚C and plasma separated. Tissues were collected, snap frozen in liquid nitrogen, and stored at -80˚C until assay.

in vivo physiologic measurements
At 6, 10 and 14 weeks of age, mice were fasted four hours and blood collected by retro-orbital bleed for measurement of plasma glucose (#23-666-286, FisherScientific), insulin (#SRI-13K, MilliporeSigma) and triglycerides (#TR22421, ThermoFisher). At age 16 weeks, mice were subjected to a 24-hour fast and 6-hour refeed to assess hepatic lipid storage during energy deficits. Body weights and whole blood were collected at 0, 24, and 30 hours and plasma measured for non-esterified fatty acids (NEFA) using the Wako Linearity Set (#999-34691, #995-34791, # 991-34891, # 993-35191, FisherScientific). in vivo insulin action was assessed at 18 weeks of age by an oral glucose tolerance test as previously described [22]. Mice were fasted for four hours and given a 2 g/kg BW glucose dose by oral gavage. Blood was collected by retroorbital eye bleed and assayed for glucose, insulin, and c-peptide concentrations. β 3 -adrenergic receptor agonist tolerance tests (β 3 TT) were performed at 20 weeks of age on four-hour fasted mice. Mice were dosed with 1 mg/kg BW of CL-316,243 by i.p. injection. Blood, collected by retroorbital eye bleed, was assayed for glucose, non-esterified fatty acids, glycerol, and insulin content.

Liver lipidomics
Frozen tissues were sectioned to 10mg on dry ice and added to phosphate buffered saline (PBS) and methanol containing internal stable isotope metabolomics standards (S6 Table). Tissues were mechanically homogenized (Qiagen TissueLyser) for 5 minutes at maximum frequency (30.0 Hz/s). 20μL of homogenate was removed for protein quantification (Pierce BCA Protein Assay Kit). Samples were mixed with methyl tertiary-butyl ether (MTBE), vortexed, centrifuged, and supernatant was transferred into new tube. Original samples were reextracted with MTBE: Methanol: dd-H2O (10:3:2.5), vortexed, centrifuged, and supernatant was transferred into tubes containing the first extraction's supernatant. Samples were evaporated in a speed-vac and then resuspended with isopropyl alcohol: acetonitrile: dd-H 2 O (8:2:2). Samples were then vortexed and centrifuged before transferring supernatant to glass vials (Agilent Technologies). Samples were analyzed by liquid chromatography-tandem mass spectrometry (LC-MS) with a 6545 UPLC-QToF mass spectrometer for non-targeted lipidomics. Results from LC-MS experiments were collected using Agilent Mass Hunter Workstation and analyzed using the software package Agilent Mass Hunter Quant B.07.00. Lipid species were quantified based on exact mass and fragmentation patterns and verified by lipid standards. Mass spectrometry was performed at the Metabolomics Core Facility at the University of Utah. Mass spectrometry equipment was obtained through NCRR Shared Instrumentation Grant 1S10OD016232-01, 1S10OD018210-01A1 and 1S10OD021505-01.

Liver, bile, and plasma cholesterol
Total cholesterol in undiluted plasma and bile was assessed with Infinity Cholesterol reagent (TR13421, Thermo Scientific, Waltham, MA) and concentrations determined by a standard curve. Liver cholesterol was extracted by homogenizing 50 mg of tissue in a TissueLyser with 1 mL chloroform:isopropanol:IGEPAL CA-630 (7:11:0.1). The organic phase was collected and dried at 50˚C. Dried lipids were resuspended in 200μL cholesterol assay buffer (MAK043, Millipore Sigma, St. Louis, MO) and total cholesterol determined following manufacturer's protocol.
To analyze lipoprotein size distributions, plasma was analyzed using a Superose 6 10-300GL column and size-exclusion fast protein liquid chromatography (FPLC). Fractions were assayed for total cholesterol and triglycerides as previously described [51].

RT-PCR for mitochondrial genes
For mitochondrial gene analyses, DNA was isolated from liver samples (n = 5/sex/genotype) with an overnight incubation in proteinase K. Isolated DNA was dried and resuspended in ultrapure water for qPCR analysis. Mitochondrial gene expression (primers in S7 Table) were normalized to the nuclear cystic fibrosis transmembrane conductance receptor (Cftr) and fold-change calculated using the 2 -ΔΔCt method.

Western blot analysis
Tissues were lysed in RIPA buffer and total protein determined by Pierce BCA assay (#23225, ThermoFisher Scientific) to ensure equal loading. Samples (15-30 ug) were heat inactivated with 4X Laemmli dye containing 4% 2-mercaptoethanol at 70˚C for 10 minutes and run on 7.5% tris-glycine gels following standard protocols. PVDF membranes were stained for total protein with 0.1% ponceau S in 5% acetic acid, and then probed for the protein of interest. For blotting of FPLC-separated plasma lipoprotein fractions, 25 μL of each fraction was incubated with 4X Laemmli dye containing 4% 2-mercaptoethanol at 70˚C for 10 minutes and probed for protein as described above. Primary and secondary antibodies are listed in S8 Table.

Statistical analyses
Statistical analysis of in vivo mouse data and tissue assays were performed by ANOVA followed by Tukey's post-hoc analysis. Lipidomics data were analyzed using MetaboAnalystR [52]: liver lipid concentrations (pmol lipid/mg liver) were log 10 -transformed, normalized by Pareto scaling, and then fold change calculated. Unless noted, data are presented as mean ± standard error. Differences were considered significant at p<0.05.

Statement of data availability
Raw data for Abhd2 expression in 500 DO and Abhd2 KO mouse phenotyping data are provided as supporting information. DO liver lipidomic data has been previously published [14]. Following a 24-hr fast, female mice averaged a 1.7 ± 0.9 gm weight loss and an average 1.1 ± 0.1 gm weight gain following the 6-hour refeed period and were not different for Abhd2 KO versus WT mice (A). Plasma NEFAs, measured before and after prolonged fast, were similar between genotypes (B). Male mice lost 2.5 ± 0.2 gm with prolonged fasting and regained 0.6 ± 0.1 gm following refeeding and were not different between genotypes (C). Plasma NEFAs of male mice during the fast/refeed protocol did not differ by genotype (D).