Despite the overt recognition of the taxing effects of obesity on both medical and social programs throughout the world, the estimated 300 million adults currently considered clinically obese in addition to the universal increase in childhood obesity indicates we are still succumbing to this global epidemic. Indeed, the poorly understood gene-environment interactions have revealed the complexity of this metabolic disease; however, with each passing year an increasing number of genetic candidates are discovered that help to further unravel the perturbed metabolism underlying the obese phenotype 12. Recently, phospholipid scramblase (Plscr) 3 was identified as a genetic candidate capable of influencing adipose function and, ultimately, the obese phenotype. Mice deficient in PLSCR3 were found to accumulate lipid in abdominal fat pads and were characterized with insulin resistance, dyslipidemia, and glucose intolerance, classic tell-tale markers for metabolic syndrome 3. While these observations suggest a role for PLSCR3 in adipose function, much work remains if the obese phenotype and the downstream consequences stemming from a dysfunctional PLSCR3 are to be understood.

PLSCR3 is one of four structurally related members (termed PLSCR1 through 4) in the phospholipid scramblase family 4. The first of these plasma membrane proteins to be cloned and characterized (PLSCR1) implicated this protein family in the trans-bilayer migration of membrane phospholipids in platelets, erythrocytes, and other cell types in response to an elevation in intracellular calcium. As such, these proteins were thought to have roles in platelet procoagulent activity, cell injury by complement, and apoptosis. Since their initial discovery, phospholipid scramblases are hypothesized to have a more complex biology than previously thought. Studies aimed at defining the biological functions of Plscr1, the most widely studied member of the phospholipid scramblase family, have demonstrated that it is transcriptionally up-regulated by interferon-α and other factors 56, and that a portion of the newly synthesized PLSCR1 protein can translocate into the cell's nucleus and interact with genomic DNA, suggesting it has a potential role in the regulation of gene expression 78.

The development of murine models deficient in PLSCR proteins provides a means to elucidate the biochemistry underlying Plscr3-mediated obesity. As previously observed with members of a protein family, a degree of redundancy exists in order for an organism to maintain physiological homeostasis and preserve the full gamut of biological functions required for survival 910. With regards to the phospholipid scramblases, adipocytes accumulate neutral lipid in Plscr3-deficient mice; however, Plscr1-deficient animals also have a small increase in adipose lipid and the Plscr1&3-deficient mice have an even greater accumulation of lipid than the Plscr3-null mice 3. This would suggest that, to some extent, a redundancy in the adipose functions of these two proteins may exist.

To begin to unravel the perturbed biochemistry associated with Plscr3 deficiency, we employed comprehensive metabolite screening technologies to determine whether the biological abnormalities stemming from the lack of PLSCR3 protein are reflected in plasma. As metabolites represent a metabolic endpoint of gene and protein function, their analysis provides insight into the cellular function of genetically modified mice 1112. As such, untargeted metabolomics offers a powerful method to further define the obese phenotype in organisms characterized by genetic modifications 1314. Furthermore, an additional inherent advantage of untargeted metabolomics (which can also be extended to alternative functional genomic strategies) versus a more targeted analysis (that is, the lipidome 15) is its ability to generate novel hypotheses through the identification of previously unrecognized signaling pathways 2. As demonstrated in the present manuscript, the characterized metabolites identified in animals lacking Plscr3 suggest a novel molecular link between the chronic low-level inflammation characteristic of an obese state and the heightened downstream risk of cardiovascular disease.

Plasma from male mice of each genotype (n = 4) were obtained, separated into two groups of two, and analyzed in triplicate at two separate run dates. As such, only metabolites that were consistently identified in all four animals of a specific genotype, irrespective of date of analysis, were considered 'true' peaks. Using XCMS software, we were able to confidently examine the data generated in the multiple analyses by first performing nonlinear alignment compensating for minor differences in ion retention times between the runs, and then identifying and matching peaks for further analysis (Figure 1). In this regard, correction of retention times permits the relative metabolite ion intensities to be statistically compared between the various genotypes in order to identify unique and/or shared ions associated with deficiencies in PLSCR proteins. For this to be accomplished with confidence, a number of criteria were used prior to considering an ion as significantly different between the various phenotypes. Initially, an intensity threshold was selected to ensure that ions could be subsequently enriched and analyzed for their accurate mass by electrospray ionization coupled with time of flight analysis (ESI-TOF). Subsequent to this analysis, ions were only deemed significantly different between the various genotypes if they had a p value ≤ 0.01. Furthermore, only those metabolites simultaneously identified in both a single knock-out model and the double knock-out model (that is, KO1 and DKO, or KO3 and DKO) were considered as 'candidate metabolites'. This approach meant that all metabolites discussed in this manuscript were identified in eight mice (each in triplicate liquid chromatography (LC) runs), thereby strengthening the biological relevance and statistical significance of our results. Finally, data generated on the individual ions were manually inspected to validate the changes and identify the most important molecules for future isolation.

All possible comparisons were performed, with the goal of identifying metabolites both unique to a specific genotype and shared between two or more genotypes. XCMS detected over 10,000 peaks across the 48 LC/mass spectrometry (MS) runs performed (corresponding to 16 mice, each run in triplicate; see Additional data file 1 for all raw data); however, to minimize spurious results, we considered only those peaks that were consistently identified in a genotype (that is, found in 12 LC/MS runs). The comparison between wild type (WT) and the three knock-out models did not identify common metabolites amongst the various mutant models, suggesting that Plscr1 and Plscr3 affect the plasma metabolome differently. Pair-wise comparisons (WT and KO1 versus KO3 and DKO, WT and KO3 versus KO1 and DKO, and WT and DKO versus KO1 and KO3) revealed that eliminating Plscr3 gave rise to an increase in the abundance of metabolites common to both KO3 and DKO models. This was not unexpected, as DKO animals have a phenotype similar to that of KO3 animals. The XCMS software identified 19 metabolites that were significantly different from WT and KO1 mice (Additional data file 2), of which 5 were characterized using the Metlin database (Figure 2) 16. While our initial statistical cut-off for significance was arbitrarily set at a p value of 0.01, it is noteworthy to mention that, for all 19 identified metabolites, the p values were less than 0.0005, and the 5 characterized metabolites were of even greater statistical significance (p ≤ 0.00001). It is expected that the 14 currently non-identifiable metabolites, once characterized and incorporated into public databases by the greater metabolomic community, will provide further information into the dysfunctional biology underlying the Plscr3-/- mouse. The five identifiable metabolites, with exact masses (m/z) of 494.3241, 522.3554, 542.3241, 546.3570, and 570.3543, correspond to lysophosphatidylcholine (LPC) species containing palmitoleic (C16:1n-7; PO), oleic (C18:1n-9; OA), dihomo-γ-linolenic (C20:3n-6; DGLA), eicosapentaenoic (C20:5n-3; EPA) and osbond (C22:5n-6; ObA) acids, respectively (Figure 2). Changes in all 19 metabolites were uniquely associated with Plscr3 deficiency, since these metabolites did not vary in the KO1 model. By contrast, we did not observe significant differences in metabolite profiles between the WT and KO1 animals, which was further reinforced by revealing no common, significantly modulated metabolites between the KO1 and DKO models.

Increasing evidence has positioned adipose tissue as not merely a reservoir for lipid storage, but also as a major endocrine and secretory organ. The recognition that obesity is characterized by chronic mild inflammation led to the discovery of factors, termed adipokines, that are released from adipose tissue and critical in regulating such physiological processes as inflammation, lipid metabolism, insulin sensitivity, angiogenesis, and eating behavior 1718. Furthermore, several factors have been shown to regulate atherogenic processes, such as hypertension and vascular remodeling 1920. It is interesting to find that plasma from Plscr3-deficient mice show increases in pro-inflammatory LPC molecules that have been linked to certain pathophysiological conditions, including atherosclerosis, cancer and rheumatoid arthritis 212223. While these bioactive lysophospholipids have not been previously described in obesity, their identification suggests a novel pro-inflammatory class linking obesity and atherosclerosis that merits further examination. The characterization of normal human serum revealed that four LPC metabolites (LPC containing 16:0, 18:0, 18:1, or 18:2 fatty acids) are among the most abundant circulating metabolites found 24. The genetic deletion of murine Plscr3 led to significant changes in only one of these highly abundant LPC species (18:1). Additionally, as changes in the abundance of other lysophospholipid classes were not observed, this indicates that increases in LPC species are specific to Plscr3 deficiency. Although to our knowledge LPC molecules have not been previously examined in obese models (either rodent or human), this finding may be a general characteristic of the obese state and requires confirmation in other models of obesity.

Conformational and positional characterization of the acyl portion of lipid metabolites was not performed; therefore, the discussion herein is based on the previous reports of the most common fatty acids produced in eukaryotes 25. The LPC species found to be most abundant in plasma was oleoyl-LPC (Figure 2). Due to this abundance, we were able to further confirm the presence of this molecule by converting it to the corresponding fatty acid methyl ester, followed by gas chromatography (GC)/MS analysis. After production of the methyl esters, the GC/MS experiment confirmed the presence of the expected fatty acid C18:1 (C₁₈H₃₄O₂) in an LPC molecule. More specifically, the hypothesized molecular ion for the 18:1 methyl ester (m/z 296, M+H) and the fragmentation pattern were found, which matched the correct model spectrum in the NIST 2002 spectral database. Both palmitoleic (C16:1n7) and oleic acids (C18:1n9) are synthesized via stearoyl-CoA desaturase (SCD1), the rate-limiting enzyme regulating the introduction of a cis-double bond in the Δ9 position of its fatty acyl-CoA substrates palmitate (C16:0) and stearate (C18:0). SCD1 has a pivotal role in whole-body lipid metabolism, as exemplified by the finding that Scd1-deficient mice are resistant to diet-induced obesity 26. However, the physiological complexity underlying the obese state suggests that interpreting plasma metabolite profiles has the power to identify biomarkers without indicating their originating tissue source.

Thus, to begin to elucidate the tissue responsible for altering lipid metabolism in Plscr3-/- mice, the expression of a subset of lipogenic genes (those encoding Ppar-γ, Lxr-α, Srebp-1c, Fasn, and Scd1) was examined in the liver and white adipose tissue (WAT) of the various genotypes. While a similar profile for Ppar-γ expression was observed in both tissues - WAT (KO1, 2.0-fold; KO3, 5.5-fold; DKO, 9.4-fold) and liver (KO1, 4.4-fold; KO3, 12.1-fold; DKO, 39.5-fold) - the expression pattern for the remaining genes were tissue-specific. Hepatic gene expression in the three knock-out models indicated similar fold-increases in Lxr-α, Srebp-1c, and Fasn, with the exception of Scd1, which was not significantly changed in the liver (Figure 3a). In contrast, gene expression in WAT revealed marked differences between the various genotypes (Figure 3b). For all genes examined, induction was minimal and less significant in KO1 mice, whereas changes in KO3 and DKO mice were highly significant and of greater importance (p ≤ 0.001). Of particular interest was the expression of Lxr-α and Srebp-1c, transcriptional regulators of Fasn and Scd1. KO1 animals had only a slight 1.7-fold increase in Srebp-1c expression, while KO3 and DKO mice had far more prominent increases of 6.5- and 10.8-fold, respectively. Similar changes were seen with Lxr-α (KO1, 2.1-fold; KO3, 9.8-fold; DKO, 16.8-fold), Fasn (KO1, 2.1-fold; KO3, 3.6-fold; DKO, 4.7-fold) and Scd1 expression (KO1, 3.0-fold; KO3, 5.9-fold; DKO, 4.4-fold). As Plscr3 expression is very low in liver 27 and Plscr1 expression is nearly absent in WAT 3, the transcriptional and lipid metabolite changes observed with Plscr3 deficiency would suggest they originate from the WAT and may be directly related to the SREBP-1c-mediated transcriptional cascade. Indeed, if the plasma lipid profile reflected hepatic transcriptional events, one could hypothesize that all genotypes would have similar changes in plasma lipids; however, since this was not the case, we propose that the differential plasma lipid profile arises from transcriptional events occurring in WAT.

As described above, Plscr3-/- mice have an increase in Δ9-desaturase expression similar to that previously noted in various obese models and animals fed diets promoting weight gain (that is, obesity-inducing diets increased Scd1 expression and activity, while diets reducing adiposity decrease it) 11. The identification of DGLA, EPA, and ObA fatty acid species may reflect additional modulation of adipose Δ5 and Δ6 desaturase enzyme activities, as both enzymes have roles in the metabolism of long-chain polyunsaturated fatty acid (LC-PUFA) species 2829; however, in the absence of adipose metabolite data this hypothesis remains to be proven. It has recently been proposed by Das 30 that defects in Δ5 and Δ6 desaturases may play a role in the development of insulin resistance by reducing the synthesis of beneficial LC-PUFA arachidonic acid (AA) and docosahexaenoic acid (DHA). Plscr3-/- mice are insulin resistant 3 and the increase in 22:5-containing LPC would further support Das's hypothesis, as increases in the abundance of 22:5 fatty acids are considered an indicator of DHA deficiency 31. These findings position Plscr3 as a regulator of adipose lipid metabolism.

Of additional interest, both DGLA and EPA are purported to have anti-inflammatory properties via their conversion into the 1 series and 3 series of prostaglandins, respectively 32. Although the precise contribution of ω-3 and ω-6 LC-PUFA to the inflammatory state has yet to be resolved, both classes are precursors to inflammatory eicosanoids; however, it appears that ω-3 LC-PUFA derived eicosanoids are associated with a less severe inflammatory profile than those derived from ω-6 LC-PUFA 33. Thus, it is conceivable that plasmatic increases in LPC containing DGLA and EPA reflect altered biochemical pathways in WAT eicosanoid metabolism associated with a deficiency in Plscr3; however, no changes in the eicosanoids currently characterized in public databases (Metlin 16, LipidMaps 34) were found in plasma.

Plasma metabolite profiling coupled with analytical software such as XCMS provides a method by which the unique biochemical signatures and functional redundancy of related proteins can be explored. Furthermore, the ability to identify perturbations in previously unrecognized metabolic pathways reinforces the potential of untargeted metabolite profiling for generating hypotheses and new research directions. As demonstrated in the present study, Plscr1 and Plscr3 do not modulate the plasma metabolome in a similar fashion. While no changes were detected in the plasma metabolite profile upon genetic deletion of Plscr1, deletion of Plscr3 clearly modulates the plasma metabolome through the release of pro-inflammatory lipids and other, currently unidentifiable, small molecules. Although profiling plasma metabolites does not definitively unravel the molecular mechanisms underlying lipid accumulation in Plscr3-deficient animals, invaluable clues were provided into the perturbed physiology of these animals and will serve as indicators for future targeted experiments. Not only do LPC molecules present a potential link between obesity, inflammation, and atherogenesis, the acyl composition of LPC suggests that Plscr3-deficient mice may have modified desaturase enzyme activities and eicosanoid metabolism. Indeed, Scd1 and its transcriptional regulators (Lxr-α, Srebp-1c) are significantly up-regulated in the adipose of Plscr3-deficient mice. Furthermore, the accumulation of LC-PUFA precursors (DGLA, EPA, and ObA) in LPC species may suggest deficiencies in the abundance of beneficial AA and DHA fatty acids. To reinforce this notion, decreases in LC-PUFA and increases in monounsaturated species were previously reported in obese Zucker rats 35. It is interesting to note that additional work by Wilson and colleagues 3637 has also demonstrated the ability to discriminate the plasma metabolite profiles of lean and obese 20-week old Zucker (fa/fa) rats using metabolomics coupled with bioinformatics algorithms. While the authors found six metabolites in the positive ion data of an LC/MS analysis to be different between the two rat strains, none of these small molecules were identified 38. Thus, while the untargeted metabolite profiling presented in this study is still in its early stages of development, the ability to identify and correlate metabolites with functionally related protein family members and provide novel and previously unrecognized insight into the perturbed metabolism stemming from dysfunctional genes positions this analytical platform as an attractive means towards understanding the fundamental biochemistry of disease states.

The following additional data are available with the online version of this paper. Additional data file 1 is a table containing the raw dataset from XCMS processing, where 'name' denotes feature or peak name, 'M' denotes m/z, 'T' denotes retention time in units of seconds, 'mzme' denotes median m/z value, 'mzmin' denotes minimum m/z value, 'mzmax' denotes maximum m/z value, 'rtmed' denotes median retention time in seconds, 'rtmin' denotes minimum retention time in seconds, 'rtmax' denotes maximum retention time in seconds, 'WT' denotes wild-type animal intensity area values, 'KO3' denotes Plscr3-/- animal intensity area values, 'KO1' denotes Plscr1-/- animal intensity area values, 'DKO' denotes double knock-out animals animal intensity area values, and the letters a, b, and c following Mouse strain_Mouse Number denote the replicated experimental runs. Additional data file 2 lists the 19 metabolites associated with PLSCR3 deficiency. Additional data file 3 is a table containing sequence information for real-time RT-PCR primers.