Diastereomer-specific quantification of bioactive hexosylceramides from bacteria and mammals[S]

Mammals synthesize, cell-type specifically, the diastereomeric hexosylceramides, β-galactosylceramide (GalCer) and β-glucosylceramide (GlcCer), which are involved in several diseases, such as sphingolipidosis, diabetes, chronic kidney diseases, or cancer. In contrast, Bacteroides fragilis, a member of the human gut microbiome, and the marine sponge, Agelas mauritianus, produce α-GalCer, one of the most potent stimulators for invariant natural killer T cells. To dissect the contribution of these individual stereoisomers to pathologies, we established a novel hydrophilic interaction chromatography-based LC-MS2 method and separated (R > 1.5) corresponding diastereomers from each other, independent of their lipid anchors. Testing various bacterial and mammalian samples, we could separate, identify (including the lipid anchor composition), and quantify endogenous β-GlcCer, β-GalCer, and α-GalCer isomers without additional derivatization steps. Thereby, we show a selective decrease of β-GlcCers versus β-GalCers in cell-specific models of GlcCer synthase-deficiency and an increase of specific β-GlcCers due to loss of β-glucoceramidase 2 activity. Vice versa, β-GalCer increased specifically when cerebroside sulfotransferase (Gal3st1) was deleted. We further confirm β-GalCer as substrate of globotriaosylceramide synthase for galabiaosylceramide synthesis and identify additional members of the human gut microbiome to contain immunogenic α-GalCers. Finally, this method is shown to separate corresponding hexosylsphingosine standards, promoting its applicability in further investigations.

Mammals produce a large variety of glycosphingolipid (GSL) structures, based on the initial synthesis of either glucocerebrosides [the term is used here only for -glucosylceramides (GlcCers)] or cerebrosides [the term is used here only for -galactosylceramides (GalCers)] by -GlcCer synthase (UGCG) or -GalCer synthase (UGT8A), respectively (1)(2)(3). Due to the Golgi and ER location of their synthesis and the topology of the corresponding enzymes, -GlcCer and -GalCer are in contrast to more complex GSLs also located on the cytosolic side of cell membranes (4)(5)(6), by that widening their terrain of biological action to cytosolic binding partners. Whereas glucocerebrosides appear to be omnipresent in almost all cell types of the mammalian body, cerebrosides have mainly been associated to brain, especially myelin structures, and to kidneys (7). Likewise, systemic deficiency of -GlcCer-based GSLs due to mutation of Ugcg is embryonically lethal at E6.5 (8) and organ-specific functions of -GlcCer, including its derivatives, have been addressed with cell-specific deletion of the Ugcg-gene function (9)(10)(11)(12)(13)(14)(15)(16)(17)(18)(19). In contrast, mice with a systemic deletion of Ugt8a pass embryogenesis. However, Ugt8a-deficiency results in severe demyelination and motor dysfunction before most of the mice die by the age of 5 weeks (20,21).
The catabolism of GSLs includes the lysosomal degradation of -GlcCer and -GalCer by -glucosylceramidase (GBA) and galactosylceramidase (GALC) with the help of activator proteins (22,23). Mutations of GBA and of GALC resulting in substantial loss of corresponding enzyme activities cause Gaucher disease and Krabbe disease, respectively (24)(25)(26)(27)(28). Symptoms of defective cerebroside catabolism caused by GALC mutations are motor deterioration, cognitive decline, and the presence of numerous, often multinucleated, globoid cells. The lysosomal storage of glucocerebroside especially affects phagocytic cells ("Gaucher cells") and goes along with hepatosplenomegaly, which is currently treated with enzyme replacement and/or substrate reduction (29)(30)(31)(32). Some patients on enzyme replacement therapy have been reported to develop type 2 diabetes or metabolic syndrome, malignancies, and central nervous system disorders (33), which underlines the biological activity of glucocerebroside and its metabolites. Likewise, increased levels of glucocerebroside correlate multidrug resistance in several tumor cell lines and tumor specimens (34,35). In general, sphingolipid metabolism defects further promote complications like cardiovascular events via oxidant stress in chronic kidney diseases (36) and downregulation of GSL synthesis appears to help reverse symptoms/phenotypes (37), including polycystic kidney disease (38).
Enzymes catalyzing the production of corresponding -anomers in mammals have not been described so far, although recently it has been proposed that, at very low levels, -GalCer is endogenously present and necessary for invariant natural killer T (iNKT) cell homeostasis (39). Furthermore, -GalCer, which was described initially in the marine sponge, Agelas mauritianus (40), is a highly potent activator of iNKT cells, when presented via CD1d (structurally related to the major histocompatibility complex proteins) by antigen-presenting cells (41). Currently, the potential of -GalCer-stimulated iNKT cells in tumor treatment is under investigation (42)(43)(44)(45). The natural stimulation of iNKT cells via CD1-presented -GalCer may derive from members of the gut microbiome, such as Bacteroides fragilis, which was shown to produces -GalCer and thereby probably influences host immune homeostasis (46). Although this is, so far, the only strain investigated to produce -GalCer, the majority of bacteroides strains were described to produce precursor Cers and Cer phospholipids (47).
Differentiation of -GlcCer and -GalCer from complex biological samples has, so far, been achieved by normal phase TLC in the presence of borate (48), by normal phase HPLC with (49) or without derivatization (50,51), and quite recently by isocratic or zwitterionic hydrophilic interaction chromatography (HILIC)-coupled MS 2 (52,53). On the contrary, discrimination of -GalCer from -GalCer with high sensitivity from complex samples, so far, is based on TLC-immuno-overlays with antibodies specific for the -glycosidic linkage (39).

Mice
Tissue was isolated from euthanized mice of recently published mouse strains: Gba2 Pax8Cre and (Ugcg (16), and fatty acid 2-hydroxylase (Fa2h) / (56). All animal procedures were approved and performed in accordance with federal laws.

Bacteroides culture
According to a previous publication (57), Bacteroides were grown anaerobically on Columbia agar plates with 5% sheep blood and Bifidobacterium on MRS medium plates containing L-cysteine-HCl (0.5 g/l) for 48 h. Lactobacillus was grown on MRS medium plates in an aerobic overnight culture. One culture plate was harvested and transferred in a tube containing 200 l methanol.

Lipid extraction from mouse tissue
Mouse tissue was homogenized in water on ice using an Ultra Turrax T25 and homogenates were lyophilized. GSLs were isolated as previously described (58). Briefly, dried tissue samples were extracted three times with mixtures of chloroform/methanol/ water. Supernatants were pooled, dried, and subjected to methanolic mild alkaline hydrolysis (0.1 M potassium hydroxide) for 2 h at 37°C to eliminate glycerophospholipids and subsequently were neutralized with acetic acid. Saponified extracts were then desalted using reverse phase C-18 columns.

Lipid extraction from bacteria
Dried bacteria samples were homogenized in methanol using the Quiagen TissueLyser II. One stainless steel bead (5 mm) per tube was agitated for 2 min at 25 Hz. Lysates were extracted in 1 ml chloroform/methanol (2/1) at 37°C for 15 min with occasional sonication. Lipid-containing supernatants were collected following centrifugation at 2,000 g for 10 min and pellets were then re-extracted once with chloroform/methanol/water (10/10/1) and once more with chloroform/methanol/water (30/60/8). Supernatants were pooled and dried.

Separation of -GlcCer, -GalCer, and -GalCer by HILIC
The diastereomeric GSLs, -GlcCer, -GalCer, -GlcCer (not shown), and -GalCer are not distinguishable by their fragmentation patterns in triple quadrupole-based MS 2 ( Fig. 1A, B) (59). Furthermore, they are not separated by C18-reversed phase mode with LC (Fig. 1C). The latter is predictable, as the structural differences of these glycolipids are within the hydrophilic moiety and not within the hydrophobic part, which interacts with the reversed phase. With normal-phase high-performance TLC (HPTLC), all four diastereomeric HexCer standards can be separated, if the HPTLC plate is impregnated prior to separation with borate (39,48) and if they contain identical Cer anchors (supplemental Fig. S1). To quantitatively differentiate all four isomers in biological samples, chromatography, therefore, has to be combined with information about the Cer anchor. Therefore, we set up a method for HILIC coupled MS 2 . Initially, we used solvent gradient systems starting with 95-97% acetonitrile, but did not observe separation of our diastereomeric HexCer mixtures on HILIC columns used in our laboratory (data not shown). Consequently, we increased the hydrophobicity of the starting solvent (solvent A): acetonitrile was exchanged by propionitrile and 2% of water by 2% of 2-butanol (because methanol in solvent A widened the peaks). In a gradient of increasing polarity, solvent A was replaced successively by increasing proportions of solvent B (97% methanol, 2% 2-butanol, 1% water) in a nonlinear fashion with increasing steepness. As a result, -GlcCer eluted first followed by -GalCer, -GlcCer, and finally -GalCer, all of them containing the identical Cer anchor, Cer(d18:1/24:1). Importantly, -GlcCer and -GalCer (R = 1.7), as well as -GalCer and -GalCer (R = 2.3), separated by baseline (R  1.5, Fig. 2A).

Comparison of the fatty acid chain length distribution in -GlcCer with that in -GalCers of various mouse tissues
Applying the HILIC-MS 2 -based separation of -GlcCer and -GalCer, we determined the distribution of both compounds in various mouse tissues and, by that, confirming enrichment of -GalCer mainly in brain (97%) and kidney (42%), and additionally revealing its abundance in lymph nodes (85%) and auricles (59%) (Fig. 3). Furthermore, results confirmed enrichment of nervonic acid within brain cerebrosides, whereas corresponding glucocerebrosides were enriched in stearic acid. The latter is the main acyl chain of neuronal gangliosides. Although not always as clear cut, the acyl chain length distribution of cerebrosides and glucocerebrosides was not identical in any tissue investigated.

The method confirms pathological changes of -GlcCer and -GalCer in mouse organs
Gba2 catalyzes the hydrolysis of -GlcCer to glucose and Cer at the cytosolic side of intracellular membranes (66). Mice lacking Gba2 activity were reported with 2- to 3-fold increased levels of -GlcCer in liver (67,68). Analysis of HexCers from liver, kidney, and intestine of WT and Gba2deficient mice revealed an increase of -GlcCers, specifically, in Gba2-deficient mice, but not corresponding -GalCers (Fig. 4A-C). Interestingly, the increase of intestinal -GlcCers was restricted to the rather minor NS-type, whereas the AS-type, NP-type, and dominant AP-type of -GlcCers were not affected. Furthermore, HILIC-MS 2 signals corresponding to -GlcCers, but not to -GalCers, were significantly reduced in liver and kidney samples of mice with a hepatocyte-specific and renal proximal tubulusspecific deficiency of the UGCG (Ugcg f/f AlbCre and Ugcg f/f Pax8Cre), respectively (Fig. 4A, D). However, mainly NS-type -GlcCers were decreased and AS-type -GlcCers even increased a little bit in hepatocyte-Ugcg-deficient liver samples. Vice versa, -GalCer accumulated more than 10-fold in kidneys of CST-deficient mice due to a block in further conversion to sulfatides (Fig. 4D). Here, -GlcCer concentrations did not change. Likewise, analysis of male kidney cortex samples from Gb3S-deficient mice revealed a more than 2-fold increase of -GalCers, which was almost exclusively due to an increase of NS-type cerebrosides (Fig. 4E). Levels of -GlcCers were also not affected by this model. We expected increased levels of cerebrosides (-GalCer) in kidneys of this model, as murine Gb3S also catalyzes the conversion of -GalCer into galabiaosylceramide (Ga 2 Cer/Gal4Gal1Cer/GaOse 2 Cer) (55), a galaseries GSL enriched in male kidneys (69); altogether, these data supported the correct association of chromatographic peaks to cerebrosides (-GalCers) and glucocerebrosides (-GlcCers), respectively.

AS-type cerebrosides with 2R-hydroxyacyl chains are separated from those with 2S-hydroxyacyl chains
Cerebrosides from brain are enriched in cerebrosides of the AS-type (70), which contain 2R-hydroxy fatty acids (71) and depend on FA2H (56,72), which specifically incorporates acyl chains with an -hydroxy group in Rconfiguration (73). Here, we compared the retention times of AS-type cerebrosides with corresponding synthetic standards containing a 2-hydroxyacyl chain with either R-or S-stereochemistry. The -GalCer standard with a 2R-hydroxystearoyl chain separated from the corresponding compound with 2S-hydroxy configuration and eluted together with corresponding AS-type cerebrosides from murine brain about 0.38 ± 0.02 min later than the 2S-isomer (Fig. 5). By that, cerebrosides with 2S-hydroxy fatty acids eluted significantly earlier (0.27 ± 0.02 min) than AS-type glucocerebrosides (-GlcCer) with 2R-configuration at the -hydroxy group and, thus, could be distinguished by this HILIC method. In contrast to brain, stomach contains AS-type -GlcCer. Normal levels depend on synthesis by FA2H as corresponding enzyme deficiency (Fig. 5H) reduces the signal to about one tenth. Due to the specificity of the enzyme (73), this dependence implies the -hydroxy acyl group of AS--GlcCers from stomach to be of 2R-configuration.

The -GalCer of B. fragilis is identified by the HILIC-MS 2 method
B. fragilis has been reported to produce Cers and -GalCers containing an iso-or anteiso-branched C17-, C18-, and C19-sphinganin and a (R)-hydroxylated and iso-branched C17-fatty acid (46,74). Here, we observed -GalCers from B. fragilis to elute 0.03 min before our calculated retention times (t R = 4.00-4.03 min) for corresponding AS-type -GalCers, but definitely beyond AS-type -GalCers (t R = 3.92-3.95 min,  = 0.05-0.06 min). This slight forward shift may be due to the hydroxy group of the acyl chain, which is, for B. fragilis, in the 3()-position, but not in the 2()-position, of the fatty acid. This shift was consistent for all three compounds reported by B. fragilis (Fig. 6A, B).

Corresponding -GalCers are common to further bacteria of the human gut microbiome
In addition to B. fragilis, we analyzed lipid extracts of nine further bacteria of the human gut microbiome. In P. copri and B. vulgatus, we detected the identical HexCer peaks as for the N--hydroxy acyl sphingosine (BS)-type -GalCers of B. fragilis (Fig. 6C). However, overall levels of these compounds were at least 100-fold lower than in B. fragilis, respectively. The other seven species investigated did not contain quantifiable amounts of -GalCer. Fig. 3. Relative distribution of -GlcCer and -GalCer in various mouse organs. NS-, AS-, NP-, and AP-HexCers were determined, which contained a C18-sphingoid base and N-bound fatty acids with the chain length C16 up to C26 (as annotated). Special structures with ultralong acyl chains as they occur in epidermis (ultra-long omega hydroxyl fatty acids) or in male germ cells (ultra-long polyunsaturated fatty acids) as well as with different sphingoid bases (e.g., C20-sphingosine in brain or kidney papillae or C17-sphingosine in epidermis) were not considered in this study. Note, nervonic acid is mainly incorporated into brain cerebrosides, whereas corresponding glucocerebrosides are enriched in stearic acid, which is the typical acyl chain of neuronal gangliosides.
Nevertheless, B. ovatus, B. thetaiotaomicron, and B. caccae contained signals for -GalCers, which were above the LOD in both the HILIC and the reversed phase LC method (supplemental Table S5).

Diastereomeric hexosylsphingosines elute differently from the HILIC column
Finally, we addressed the elution behavior of hexosylsphingosines (HexSphs) with standards for glucopsychosine (-GlcSph), psychosine (-GalSph), and -GalSph, all with a C18-sphingosine as long chain base. Although not baseline separated, the -GlcSph eluted before -GalSph (R = 1.0) and -GalSph came off the column last (R = 0.7) in the same order as the corresponding HexCers (Fig. 7). Hence, this method may also be useful to identify the type of HexSph present.

DISCUSSION
Lipidomics is a fast growing field of interest impacting more and more medical research. Often all-in-one analysis is demanded and high throughput methods like direct infusion into high resolution mass spectrometers with MS all -scans and subsequent data processing software try to fulfill these requests (75). These MS-only techniques, however do not distinguish stereoisomeric compounds with qualitatively identical fragmentation behaviors. Separation and quantification of such stereoisomers therefore requires combination of MS with orthogonal methods, such as chromatography or ion mobility. LC, especially, is a widespread technique, often coupled to MS to increase specificity and sensitivity of detection.
Here, we present a HILIC-based method, which enables the separation of and -anomers of GlcCer and GalCer comprising identical Cer anchors, which was not possible with reversed phase LC. HexCers are amphipathic molecules with their structural differences buried in the polar head group, but not the hydrophobic anchor, the latter interacting with the stationary phase in reversed phase LC and the former in HILIC. Coupling to MS 2 allows identification of differences in Cer anchor composition to unambiguously ascribe NS-, AS-, NP-, and AP-type HexCer-derived HILIC-peaks of biological samples to either GlcCer or GalCer with either or -glycosidic linkage. The method can be used to analyze complex biological lipid extracts, as demonstrated by the application to various mouse organs and bacteria. The correct assignment of peaks was supported by measuring a selective increase or decrease of -GlcCer over -GalCer in various tissues with GBA2 or UGCG deficiencies, respectively. Loss of GBA2 activity leads to the accumulation of the substrate -GlcCer. In contrast, the lack of UGCG activity either in hepatocytes or in renal tubular cells omitted corresponding production of -GlcCer, thereby reducing the overall -GlcCer concentration of liver or kidney, as expected (11,16,67,68,76). Further confirmation was obtained by detecting a selective increase of -GalCer over -GlcCer in organs lacking activity of either CST or Gb3S, for both of which -GalCer is the substrate to produce either sulfatide SM4s or galabiaosylceramide, respectively (16,55,77).
Interestingly, however, the levels of not all -GlcCer species went up with Gba2-deficiency and the levels of AS-type -GlcCers remained unchanged. There are two possible explanations, which are: i) GBA2 may not be expressed in cells that express the AS-type; or ii) the AS-type -GlcCer resides in lipid layers with no access to the cytosolic activity of GBA2. Whereas loss of UGCG activity in renal tubular epithelial cells caused a decrease of NS- and AS-type -GlcCers in kidney, the hepatocyte-specific loss of UGCG activity, causing NS-types to go down, even led to an increase of minor AS-type -GlcCers. Because tissues are multi celltype structures, the loss of enzyme activity in one specific cell type will always lead only to a decrease of a product and not to complete disappearance, if this product is synthesized by several cell types of the respective organ. An increase of a subspecies may imply either an increased relative abundance of the corresponding synthesizing cell type in mutant tissue or a reactive increased production or decreased catabolism of these compounds within the corresponding cells.
The method on top separates AS-type cerebrosides containing the -hydroxy group of the acyl chain either in 2R-or 2S-configuration. Both configurations may appear in biological samples and AS-type cerebrosides of the brain were reported to contain basically the 2R-configuration (71), which is confirmed with this HILIC-MS 2 method.
Screening different mouse tissues with this method, we further confirmed high abundance of cerebrosides in brain and kidney, whereas most other tissues, including liver and spleen, are dominated by the presence of glucocerebrosides. In compliance with a report that AS-type -GalCer from brain contain 2R-hydroxy fatty acids, the AS-HexCer(d18:1/h18:0) from phrenosin and from mouse brain migrate together with the -GalCer[d18:1/(2R)h18:0] standard. Note, the decrease of -GalCer(d18:1/h24:0) with 2R-hydroxy configuration in Fa2h / stomach.
Two exceptions were identified: auricles and, especially, lymph nodes are relatively abundant in cerebrosides, as compared with glucocerebrosides, and further investigations need to address from which cell types they originate.
As this method delivers data for individual species, it allows detection of differences in Cer anchor compositions between cerebrosides and glucocerebrosides or following of changes in these compositions upon aging, activation, or disease progression. Here, we plotted the relative acyl chain length distributions of cerebrosides against those of glucocerebrosides found in WT mouse tissues. As expected, brain cerebrosides are rich in nervonic acid, a major component of myelin sheaths (60,78,79). In contrast, the small amount of brain glucocerebrosides is synthesized mainly with stearic acid. This points to neuronal cell origin, as these cells convert this compound to complex gangliosides, such as GM1a, GD1a, GD1b, or GT1b, all of which are known to basically contain stearic acid in their lipid anchor (80,81). Likewise, nervonic acid was the major fatty acid incorporated into cerebrosides of colon, small intestine, thymus, and lymph nodes, whereas glucocerebrosides of these organs were enriched with the shorter palmitic acid. In summary, this points to different cell types or differentiation stages of cells forming either cerebrosides or glucocerebrosides and expressing different pattern of Cer synthases, which remains to be investigated. It may be noted that this investigation did not include HexCers with other sphingoid bases than C18-sphingosine and C18-phytosphingosine or acyl chains shorter than C16 or longer than C26, excluding, thereby, compounds such as those with C20-sphingosine that are described for aging brain (82) or renal papillae (65) or the well-known sphingolipids with ultra-long acyl chains (>C26) specific for epidermis or male germ cells (83).
In contrast to the afore-discussed cerebrosides and glucocerebrosides, iNKT cells are activated when recognizing GalCer with an -glycosidic linkage presented on the cell surface receptor, CD1d, of antigen-presenting cells (41,84). One natural source of this compound in contact with the human body is B. fragilis, a bacterial member of the human gut microbiome (46). Here, we show detection of -GalCers from B. fragilis by HILIC-MS 2 with unique retention times, different from the -glycosidic mammalian counterparts. How many other bacteria of the human gut microbiome may also express these immune stimulatory compounds has to be elucidated, as well as their potential role in colitis, when the intestinal barrier gets leaky for bacteria or in sepsis. To elucidate the use of HILIC-MS 2 for screening of bacteria that express -GalCer, we analyzed an initial set of 10 bacteria of the human gut microbiome. At least two other bacteria were detected with quantifiable amounts of these compounds and another three strains delivered signals corresponding to -GalCer, which were above the lower detection limit. Although concentration of -GalCer in all five of these strains was at least 100-fold lower than in B. fragilis, the results show that there might be a remarkably high number of bacteria present in our gut, which is capable of synthesizing the immunogenic -anomeric variant of GalCer. These results underline the power of the method to screen for such compounds and further support the idea that -GalCer is a natural CD1dligand for iNKT cell activation.
Very low amounts of endogenously produced -GalCer (and eventually -GlcCer) have also been discussed to contribute to successive iNKT cell maturation (39). In this context, endolysosomal catabolism of -GalCer has been discussed to regulate levels of -GalCer on antigenpresenting cells. Cerebrosidase and GBA are specific for -glycosylceramides and seem not to turn over -GalCer. Catabolism rather appears to be initiated by acidic ceramidase releasing the lyso-compound, -GalSph, which, in contrast to -GalCer, is a substrate for -galactosidase. Either pharmacological inhibition of these enzymes or defective activity in Fabry disease thereby induces a proinflammatory phenotype of iNKT cells (39,(85)(86)(87). Under pathological conditions of Krabbe disease, metachromatic leukodystrophy, or Gaucher disease, when either cerebrosidase or GBA is dysfunctional, significant amounts of -GalSph or -GlcSph accumulate due to turnover by acidic ceramidase, and these compounds contribute to the pathology of the disease (25,26,88,89). Here, significantly different retention times were achieved for the three types of HexSphs, making the HILIC-MS 2 method a tool to identify the type of HexSph appearing and to study concentration changes of these cell-toxic or immune-stimulatory compounds.