Metabolomics reveals increased isoleukotoxin diol (12,13-DHOME) in human plasma after acute Intralipid infusion.

Intralipid is a fat emulsion that is regularly infused into humans and animals. Despite its routine use, Intralipid infusion can cause serious adverse reactions, including immunosuppression. Intralipid is a complex mix of proteins, lipids, and other small molecules, and the effect of its infusion on the human plasma metabolome is unknown. We hypothesized that untargeted metabolomics of human plasma after an Intralipid infusion would reveal novel insights into its effects. We infused Intralipid and saline into 10 healthy men in a double-blind, placebo-controlled experiment and used GC/MS, LC/MS, and NMR to profile the small-molecule composition of their plasma before and after infusion. Multivariate statistical analysis of the 40 resulting plasma samples revealed that after Intralipid infusion, a less-well-characterized pathway of linoleic acid metabolism had resulted in the appearance of (9Z)-12,13-dihydroxyoctadec-9-enoic acid (12,13-DHOME, P < 10−3), a leukotoxin that has powerful physiological effects and is known to inhibit the neutrophil respiratory burst. Intralipid infusion caused increased plasma 12,13-DHOME. Given that 12,13-DHOME is known to directly affect neutrophil function, we conclude that untargeted metabolomics may have revealed a hitherto-unknown mechanism of intralipid-induced immunosuppression.

biochemistry was performed using a clinical pathology system (SYNCHRON LX System, Beckman Coulter). For metabolomics analysis, 1 ml aliquots of plasma were separated at 4°C using a methanol-water-chloroform 1:1:1 (v/v/v) extraction. The water and chloroform fractions were evaporated to dryness in a vacuum centrifuge and stored at Ϫ 80°C. Just prior to analysis, the lyophilized water fraction was reconstituted in either 500 l of HPLC-grade water (for mass spectrometry) or aqueous NMR buffer [100% D 2 O + 2 mM 4,4-dimethyl-4-silapentane-1-sulfonic acid (DSS)].

GC-MS
Lyophilized chloroform fractions were reconstituted in 20 l chloroform, vibrated for 15 s, then removed with a micropipette into a 150 l glass insert inside a Waters 2 ml HPLC vial. Twenty microliters of BSTFA were added; the sample was mixed and then heated to 60°C for 15 min. These samples were then analyzed by GC/MS on a Varian 3800 GC coupled to a Bruker 300 triple quadrupole mass spectrometer, using helium as carrier gas at 1.2 ml/min in constant-fl ow mode. The column was a Varian "Factor Four" VF-5ms (30 m × 0.25 mm internal diameter and 0.25 m fi lm). Injections of 1 l were made using a Varian CP-8400 autosampler and a Varian 1177 split/splitless injector in split mode using a 10:1 split. The injector temperature was 260°C. The column oven was programmed at 100°C to 150°C at 30 degrees per min, then to 290°C at 8°C per min. The ion source was held at 220°C, and the transfer line at 290°C.
Full-scan MS data were acquired using electron ionization over the range m / z 35 to 500 in 0.2 s. Selected ion channels for the characteristic ions at m / z 285 (14:0), 311 (16:1), 313 (16:0), 335 (18:3), 337 (18:2), 339 (18:1), 341 (18:0), and 361 (20:4) with dwell times of 20 ms per channel were also included as a precaution in case of any weak full-scan data. Peaks were identifi ed by the use of reference standards. Data were quantifi ed on the total ion current (TIC) for each FA TMS ester peak, with the exception of linolenic acid, which partially coeluted with oleic acid under these conditions. For this minor FA, the extracted ion chromatogram at m / z 335 was measured, and this was scaled up to the full TIC value after examining a pure spectrum of linolenic acid to determine this correction factor. All FAs were then expressed as their percentage of the total FAs determined. Data were processed using Bruker Workstation software.

LC/MS
LC/MS analysis was conducted using a hybrid linear trap quadrupole/Orbitrap high-resolution mass spectrometer (Thermo Fisher Scientifi c; Bremen, Germany). The samples were separated using a Waters 2690 separations module employing a Waters Nova-Pak 4.0 m C18 column (3.91 × 150 mm) at a fl ow rate of 0.8 ml/min -1 using HPLC-grade solvents (0.1% formic acid in water for buffer A and 0.1% formic acid in methanol for buffer B). A 10 l aliquot of each sample was injected onto the column, and compounds were eluted over a step gradient of 10-50% buffer B over 3 min, then 50-80% buffer B over 8 min, and 80-100% buffer B over 1 min. After holding at 100% buffer B for 2 min, the column was reequilibrated in 10% buffer B for 0.5 min. The column was operated at ambient temperature (20°C), and the samples were maintained at 10°C.
Centroid mass spectra were acquired in the mass/charge ( m/z ) range of 50-1,000 at a target resolution of 30,000, operating according to the following parameters: capillary temperature of 300°C; sheath gas and auxiliary gas fl ow rates set to 30 au and 5 au, respectively. A capillary voltage of 7 V was used for positiveion acquisition and Ϫ 44 V was used for negative-ion acquisition. Raw MS fi les were exported and analyzed offl ine using XCMS ( 15 ) running in the R environment. its occasional toxicity may be more obscure than is currently thought. By extension, the chemical complexity of Intralipid may explain some of the contradictions in the literature with regard to effects of acute increases in circulating lipids (and lipid metabolites); for example, see recent work from this laboratory ( 14 ). Given that untargeted small-molecule profi ling of human plasma or serum in response to Intralipid infusion has, to our knowledge, never been carried out, the complex interaction between Intralipid and physiology, both in the clinic and elsewhere, remains unclear.
The aim of this study was to measure changes in the human plasma metabolome in response to a standard Intralipid infusion, by using untargeted metabolomic profi ling. We used GC/MS, high-resolution LC/MS, and 1 H-NMR and aimed to profi le novel compounds that were signifi cantly increased in human plasma due either to the infusion or to secondary metabolism of the infusate. Our aim was to shed light on the possible contribution of these small molecules to both the beneficial, as well as the adverse, effects of Intralipid infusion in humans.

Subjects
Ten healthy men were recruited from the community by local advertisement. The subjects' ages, weights, heights, and body mass index (BMI) were measured during an initial assessment visit. Subjects were deemed eligible for participation if they were aged over 18 years, had a normal BMI (18.5-25), were not on any cardiac or metabolic medications, and had no adverse reaction to egg, nuts, soy, heparin, or Intralipid. They were screened twice for possible complicating conditions (for example coronary heart disease, diabetes mellitus, aortic valve stenosis, and diseases of the liver, pancreas, or kidney). Subjects were informed of the nature and possible risks associated with each experiment and were required to complete both a health screening questionnaire and a statement of informed consent. All protocols were approved by the human research ethics committee of the University of Queensland and conformed to the Declaration of Helsinki.

Study design and sample collection
The study was a randomized, double-blind, placebo-controlled experiment. Subjects attended the laboratory on two separate occasions separated by approximately 1 week (alternate randomization was implemented during the second visit). On each visit, the subjects arrived in the morning following an overnight fast. After at least 10 min supine rest in a temperature-controlled room, an intravenous cannula was inserted in the cephalic vein of each arm; the right for infusion and the left to obtain blood samples. A 20 ml baseline blood sample was taken, immediately after which a saline infusion was initiated (90 ml/h). After a 500 U bolus of heparin, subjects were randomized to one of two groups: 20% Intralipid or saline. Both were infused at 90 ml/h and coinfused with heparin. After 60 min of infusion, another blood sample was taken.

Clinical biochemistry and metabolomics sample preparation
Blood sampling yielded a total of 40 samples. Samples were collected in lithium heparin (gel) tubes for clinical biochemistry and EDTA tubes for subsequent metabolomic analysis. Clinical confi rmed using 2-dimensional spectroscopy and/or statistical correlation spectroscopy. As part of our in-house standard operating procedure, we use an annotation confi dence scoring system similar to that recommended by the Metabolomics Standards Initiative and others ( 20 ).

Data analysis and statistics
For NMR data, post-Intralipid and postsaline spectra were compared using projection onto latent spaces discriminant analysis (PLSDA), using the SIMPLS algorithm as implemented in PLS Toolbox 6.5 (Eigenvector Research, Inc.; Wenatchee, WA). Prior to analysis, data were mean-centered and orthogonal signal-corrected (for ease of interpretation). To protect against overfi tting, we performed permutation testing ( 21 ) as implemented in PLS Toolbox. A PLSDA model was considered signifi cant if the estimated P value was <.05. However, PLSDA performed poorly when analyzing the LC/MS data (probably due to a relative lack of collinearity); therefore, we instead performed repeated-measures ANOVA on each feature/peak with correction of P values using the Benjamini and Hochberg False Discovery Rate correction ( 22 ). Features were accepted as significant if the corrected P value was <.05. Clinical biochemistry measurements and FA profi les were compared (Intralipid vs. saline) using a repeated-measures ANOVA, as implemented in SPSS 18.0 (IBM Corporation; Armonk, NY).
Data processing and analysis were conducted using R 2.14 or higher (http://cran.r-project.org/), Matlab R2011b (including the statistics toolbox) (Mathworks; Natick, MA), SPSS Statistics 18 or above (IBM Corporation), Microsoft Excel, Sirius 2 [for isotope pattern matching ( 23 )] and MZmine 2 [for offl ine visualization ( 24 )], all running under Mac OS 10.6 or greater. All Matlab and R scripts are available from the corresponding author on request.

Clinical biochemistry
Subjects' triglycerides and FAs were within normal ranges at baseline ( Table 1 ). As expected, the infusion of Intralipid resulted in a signifi cant increase in triglycerides. An increase in circulating free FAs was also seen both postsaline and post-Intralipid, with the increase postsaline most probably due to the action of heparin on lipoprotein lipase. There was no change in total cholesterol in either group, although there was a signifi cant decrease in VLDL postsaline and post-Intralipid and a decrease in LDL post-Intralipid only.

FA profi ling (GC/MS)
The 3-to 4-fold increase in circulating free FAs was accompanied by a signifi cant shift in the FA profi le ( Table 2 ). The most distinctive feature was a dramatic rise in the contribution to the total plasma FA pool from linoleic acid (from 12% to 35%), a major constituent of Intralipid. There was also a 4-fold increase in the contribution from linolenic acid, from ‫ف‬ 0.5% to >2.0%. The percentage contribution of other FAs either diminished or stayed the same, although this should be taken in the context of the large increase in total free FAs. Thus, although the contribution from oleic acid was unchanged, oleic acid concentration itself must have risen substantially for this to be the case (subsequently confi rmed by LC/MS).

NMR spectral acquisition and processing
One-dimensional 1 H nuclear Overhauser effect spectroscopy (NOESY) spectra were acquired from the reconstituted water fractions using an Agilent/Varian Inova 400 MHz spectrometer (Varian Medical Systems; Palo Alto, CA). Spectra were acquired with a 90° pulse, pulse length 9.7 s with a sweep width of 4,299 ppm at 25°C. A total of 128 transients were collected into 32,000 datum points. The water signal was set to the center of the transmitter offset, and a presaturation pulse of 2 s was applied. Presaturation occurred during the relaxation delay and also during the mixing time (100 ms) of the NOESY pulse sequence.
The resulting free induction decays were Fourier transformed, line broadened (1 Hz), and baseline corrected in NMR Manager 12.0 (Advanced Chemistry Development; Toronto, Canada). All chemical shifts were referenced relative to DSS at 0 ppm. The spectra were normalized and bucketed using probabilistic-quotient normalization ( 16 ) and adaptive intelligent binning ( 17 ), respectively; both routines were customwritten in Matlab. Peaks due to EDTA were excluded from further analysis ( 18 ).

Metabolite identifi cation
This experiment presented unusual challenges with regard to feature annotation, given that the small molecules comprising the metabolic profi le could have come from any of four sources: human metabolism, soy metabolism (from the soybean oil), chicken metabolism (due to the egg yolk content of Intralipid), and possible synthetic chemicals (including plasticizers). To aid in distinguishing human metabolites from those compounds that were components of Intralipid (and thus directly infused), we added 30 mM Intralipid directly to a sample of whole blood, treated it identically to the experimental samples, and used this as a control. Peaks that were not present in the control sample, but only in the postinfusion sample, were likely to be the products of secondary (human) metabolism.
For LC/MS, features were putatively annotated by searching public databases by accurate m/z alone. Once candidates had been selected, we used any orthogonal data that were available (including isotope patterns and retention times) for either exclusion or increased confi dence. For example, the positive identifi cation of (3R)-3-hydroxybutanoic acid in the 1 H-NMR data was used to confi rm the annotation of the corresponding LC/MS chromatograms. Furthermore, we used retention times to exclude putative annotations, even though their molecular mass was consistent with the formula. Our guiding principle was to not annotate rather than mis -annotate. Finally, a number of compounds of interest were further characterized using tandem mass spectrometry (MS/MS). These compounds [monobenzyl phthalate, dihydroxy-stearic acid, (3R)-3-hydroxybutanoic acid (3HB), 12, 13-dihydroxy-9Z-octadecenoic acid (12,13-DHOME), 9S-hydroperoxy-10E,12Z-octadecadienoic acid (9(S)-HPODE), 12R-hydroxy-9Z-octadecenoic acid (Ricinoleic acid), and glycerol] were fragmented, and the resulting daughter ions were compared with library spectra. For example, the [M-H] Ϫ ion that was believed to correspond to 12,13-DHOME was fragmented using collisioninduced dissociation at 25 eV; the resultant diagnostic masses at both 183.14 and 129.09 were entirely consistent with an equivalent spectrum for 12,13-DHOME in the METLIN database. When library MS/MS spectra were not available, fragments were compared with results generated using the in silico fragmentation prediction tool, MetFrag ( 19 ). However, no important compounds were annotated in this way.
For 1 H-NMR, features were putatively annotated from a custom database based on chemical shift only. Annotation was then The ketoacid (3R)-3-hydroxybutanoic acid (3HB) was signifi cantly raised post-Intralipid, thus independently confi rming its identifi cation as a compound of interest by 1 H-NMR (see below). Also putatively annotated were several fl avones, a lysophospholipid, and some unusual FAs such as ricinoleic acid and sebacic acid ( Table 3 ).

NMR spectroscopy
A two-latent-variable PLSDA model was fi tted to the postintervention 1 H-NMR data and accounted for 99% of the between-class variance ( Fig. 1 ). The degree of fi t was confi rmed by permutation testing as being unlikely to be due to chance ( P < .05). The fi rst latent variable showed clear separation between the post-Intralipid and postsaline samples, and alone accounted for 94% of the between-class variation. Latent variable 1 was dominated by a pair of features, corresponding to a doublet in the NMR spectra at 1.20 ppm. These peaks were subsequently identifi ed using correlation spectroscopy ( 25 ) as 3HB. Thus, the NMR data provided further confi rmation of the LC/MS data that 3HB was raised in the post-Intralipid samples compared with the postsaline samples.

DISCUSSION
To study the effects of Intralipid infusion on the human plasma metabolome, we profi led the plasma of 10 healthy men, using GC/MS, high-resolution LC/MS, and 1 H-NMR, before and after infusion with either Intralipid or saline. As expected, Intralipid infusion caused an increase in circulating triglycerides and nonesterifi ed FAs. However, we also identifi ed over 50 novel chemical and biochemical

LC/MS
A total of 6,914 features of interest in the mass spectra were identifi ed using XCMS: 4,254 features from spectra acquired in positive-ion mode and a further 2,660 features in negative-ion mode. Of these, 101 features were significantly altered by Intralipid infusion ( P < .05 after the Benjamini and Hochberg False Discovery Rate correction). These 101 features were hand-checked for artifacts (for example, Fourier transform "shoulders" from centroiding), which were removed. After isotopes and adducts were also removed, 52 features remained. Of the 52, 24 (46%) could be identifi ed through the use of public databases and subsequent MS/MS experiments. Of these 24, three were identifi ed as synthetics (monobenzyl phthalate, benzyl butyl phthalate, and polyethylene glycol) and excluded from further analysis. Table 3 is a complete list of annotated features. A spreadsheet of all the signifi cant LC/MS features is available in the supplementary materials.
Of the 35 annotated features (in both positive-and negative-ion mode), some were FAs that had been released from the infused triacylglycerols by lipoprotein lipase, including linoleic acid, linolenic acid, and oleic acid. Several more features were due to metabolites of linoleic acid (the most-abundant FA in Intralipid), including hydroxy-linoleic acid, 12,13-dihydroxy-9Z-octadecenoic acid (12,13-DHOME), and 9S-hydroperoxy-10E,12Z-octadecadienoic acid (9(S)-HPODE). We also identifi ed a chromatogram at m/z = 295.229 that may have corresponded to 12 (13)-EpOME (the 12,13-DHOME precursor). The signal at this mass and retention time was signifi cantly higher after Intralipid infusion (uncorrected P = .0048), although the Benjamini and Hochberg correction ( P = .0157) meant that this signal was not identifi ed in our initial, unbiased analysis. and their diols that are thought to have toxic effects (discussed below). Although a few of these features were only putatively annotated, their copresence on the same metabolic pathway(s) ( Fig. 2 ) signifi cantly increased our confidence in these annotations. Indeed, such an approach (using existing knowledge about metabolic pathways as a bioinformatics tool for metabolome annotation) has already been suggested ( 27,28 ). Of these metabolites, 12,13-DHOME (isoleukotoxin diol) ( 29 ) ( Fig. 3 ) is of particular note. 12,13-DHOME is produced in humans from linoleic acid, via the protoxin 12(13)-EpOME (isoleukotoxin) by soluble epoxy hydrolase [sEH features that were increased by Intralipid infusion. The signifi cance of these is discussed below.
IVFEs have been regularly infused into both patients and healthy humans in a range of settings for more than 50 years. Clinically, they are most commonly used for parenteral feeding, either on their own or as a component of total parenteral nutrition. They have also recently found a novel application as a drug delivery vector. In this case, they are used as a solvent for drugs that are poorly soluble in water ( 26 ). Finally, infusion of IVFE is an effective therapy for accidental cardiac poisoning by local anesthetic ( 3 ). Many biomedical researchers also infuse IVFEs into healthy subjects (human and animal) as an experimental model of increased plasma triacylglycerol and FA concentration. Given that chronically elevated plasma lipid concentrations (dyslipidemia) are associated with an increased risk of cardiovascular disease and increased mortality rates in humans, experiments using IVFEs have played an important role in advancing our understanding of the effect of acute elevations of plasma lipid.
Yet despite its seeming ubiquitousness, IVFE infusion can and does cause adverse reactions. These include liver disease, pancreatitis, hypertriglyceridemia, and immunosuppression, with infants being particularly susceptible. Although these adverse reactions are rare, they can be catastrophic when they do occur. Furthermore, given that the chemical profi le of many IVFEs is complex (due to their biological origin), the exact mechanism whereby adverse events occur remains unclear.
Therefore we used metabolomics methods to profi le the plasma of 10 healthy men, before and after infusion with either saline or one of the most-common IVFEs, Intralipid. We identifi ed a range of compounds that were previously not known to be increased in response to Intralipid infusion, including several linoleic acid epoxides + was most signifi cant, then P for the most probable structure is given). b Classifi cation is correct, but exact structure was not determined with certainty. Most probably, structure is in brackets. c [M±H] ion was also detected. d [M+Na] + ion was also detected. mechanisms and appropriate prevention. Any mechanism that involves linoleic acid, the major FA constituent of Intralipid, is attractive. The more commonly considered route of linoleic acid metabolism is via the 20-carbon FA arachidonic acid to the eicosanoids, a class of lipids (including prostaglandins, leukotrienes, and thromboxanes) that are key mediators of infl ammatory and immune function ( 13 ). Thus, the prevailing hypothesis is that Intralipid infusion leads to a spike in eicosanoid concentrations with attendant effects on infl ammatory status and immune response. Yet data supporting this hypothesis are inconsistent (for a discussion, see Ref. 4 ).
In addition to its role as a PPAR ligand, 12,13-DHOME is also a potential toxin. The leukotoxin diols (including 12,13-DHOME) can exert a range of pathophysiological effects in mammals including inhibition of mitochondrial function ( 32 ) and increased oxidative stress ( 33 ), whereas their leukotoxin precursors have been implicated in cases of acute respiratory distress syndrome in burn patients ( 34,35 ) and cardio-pulmonary toxicity (36)(37)(38). Although it was once thought that sEH acted to render the leukotoxins safe by converting them to their diols, it now appears that the diols (including 12,13-DHOME) are the morepotent cytotoxins ( 29 ). Thus, inhibition of sEH ameliorates many of the toxic effects of the leukotoxins ( 39 ), whereas 12,13-DHOME adversely affects the electrophysiology of cells including oligodendrocytes ( 40 ) and cardiac myocytes ( 41 ). In addition, both 12(13)-and 9(10)-EpOME can be metabolized to the THF-diols ( Fig. 2 ), themselves (together with the leukotoxin diols) capable of acting as endocrine disruptors in rats ( 42,43 ).
Given that a compromised immune system is a significant adverse effect of Intralipid infusion, particularly for the very young and the critically ill, both of whom receive Intralipid infusions and both of whom are especially vulnerable to infections, much interest is focused on potential  our other fi ndings (that Intralipid infusion increases the concentration of a number of linoleic acid metabolites that could potentially have a detrimental effect on cardiac function), an attractive hypothesis arises, that the effect of Intralipid on, for example, cardiac function would depend on the metabolic balance between the production of positive inotropes (such as 3HB) and negative inotropes (such as leukotoxin diols). More work in this area is clearly warranted.

CONCLUSIONS
We infused Intralipid into 10 healthy men and studied the changes in their plasma metabolome using LC/MS and 1 H-NMR. We found that several metabolites of linoleic acid were signifi cantly raised after Intralipid infusion. Of these, 12,13-DHOME (isoleukotoxin diol) is known to have adverse effects on neutrophil function. Thus, we hypothesize that one of the known adverse effects of Intralipid infusion, immunosuppression, may in part be mediated by raised 12,13-DHOME. sEH to the protoxin 12(13)-EpOME and the potentially toxic 12(13)-DHOME, may directly inhibit neutrophil function and thus immune response. This mechanism raises the possibility for use of sEH inhibitors (e.g., see Ref. 45 ) as a potential therapy for patients suffering adverse effects (or a prophylactic for those at risk), thus adding prevention of Intralipid-induced toxicity to the already-impressive list of potential clinical applications of sEH inhibitors ( 39 ).
Additional analysis of the control samples (blood + 30 mM Intralipid) showed that no 12,13-DHOME was present in the Intralipid itself. However, although our Intralipid was stored in glass bottles, Intralipid is often stored in oxygen-permeable bags for extended periods of time in clinical environments, which may well lead to spontaneous oxidation of linoleic acid and the presence of 12,13-DHOME and other epoxides and epoxide diols prior to infusion. Further studies should establish whether Intralipid that has been stored under different conditions, and for different durations, contains leukotoxins or their diols. One potential limitation of our work is that only a single duration and infusion rate of Intralipid was studied. It is unclear whether lower doses (or different durations of infusion) may have reduced the fl ux into 12,13-DHOME, and future studies should also address this question.
Finally, although our study shows that a sharp increase in circulating linoleic acid (due to Intralipid infusion) leads to formation of leukotoxins and their diols, it remains a possibility that a sudden increase in dietary intake of linoleic acid could have similar effects. There is evidence that humans evolved with a relatively low dietary linoleic acid intake (approximately 1:1 with the omega-6 FA ␣ -linolenic acid) ( 46 ). Yet this is no longer the case, and a modern Western diet contains prodigious amounts of linoleic acid from ingredients such as vegetable oil that were not components of our legacy diet ( 47 ). This may mean that leukotoxins and their diols are now produced ubiquitously at low levels in many humans; future research should address this important question.
Our fi nding that Intralipid infusion acutely increases 3HB is consistent with existing theory and earlier data in rats ( 48 ). The accumulation of 3HB suggests that mitochondrial FA transport and ␤ -oxidation in the liver were rapid enough to exceed the rate at which acetyl-CoA was being used by the TCA cycle. This leads to an oversupply of acetyl-CoA and a redirection of acetyl-CoA through acetoacetyl CoA thiolase (EC:2.3.3.10) into ketogenesis. Although perhaps rather obvious, the accumulation of 3HB during Intralipid infusion may yet have profound physiological implications. Ketones are an unusual metabolic fuel in that they increase the free energy available from ATP hydrolysis above normal values ( 49 ). Thus, acute infusion of 3HB into Langendorff-perfused rat hearts leads to a 25% increase in hydraulic work effi ciency of the heart ( 50 ) (36% when combined with insulin). Furthermore, very recent work shows that feeding mice a ketone ester diet affects both brown fat ( 51 ) and brain metabolism ( 52 ). Thus, the effects of Intralipid infusion that were previously ascribed to acute increases in lipids may in fact be due to increased ketone concentration. In the context of