Homeostatic and pathogenic roles of GM3 ganglioside molecular species in TLR4 signaling in obesity

Abstract Innate immune signaling via TLR4 plays critical roles in pathogenesis of metabolic disorders, but the contribution of different lipid species to metabolic disorders and inflammatory diseases is less clear. GM3 ganglioside in human serum is composed of a variety of fatty acids, including long‐chain (LCFA) and very‐long‐chain (VLCFA). Analysis of circulating levels of human serum GM3 species from patients at different stages of insulin resistance and chronic inflammation reveals that levels of VLCFA‐GM3 increase significantly in metabolic disorders, while LCFA‐GM3 serum levels decrease. Specific GM3 species also correlates with disease symptoms. VLCFA‐GM3 levels increase in the adipose tissue of obese mice, and this is blocked in TLR4‐mutant mice. In cultured monocytes, GM3 by itself has no effect on TLR4 activation; however, VLCFA‐GM3 synergistically and selectively enhances TLR4 activation by LPS/HMGB1, while LCFA‐GM3 and unsaturated VLCFA‐GM3 suppresses TLR4 activation. GM3 interacts with the extracellular region of TLR4/MD2 complex to modulate dimerization/oligomerization. Ligand‐molecular docking analysis supports that VLCFA‐GM3 and LCFA‐GM3 act as agonist and antagonist of TLR4 activity, respectively, by differentially binding to the hydrophobic pocket of MD2. Our findings suggest that VLCFA‐GM3 is a risk factor for TLR4‐mediated disease progression.

On the other hand, it has been suggested that GM3 on the plasma membrane plays important roles in pathogenesis of metabolic disorders (Inokuchi et al, 2018). GM3 is also a major ganglioside in adipocytes, and its expression is induced by proinflammatory cytokines derived from adipose tissue macrophages (Tagami et al, 2002;Nagafuku et al, 2015;Wentworth et al, 2016). GM3 biosynthesis occurs in Golgi, and it subsequently becomes secreted into extracellular compartment or localized in plasma membrane as a component of membrane microdomains (also called "rafts"), which are signaling platforms comprised of sphingolipids (Lingwood & Simons, 2010). GM3 on plasma membrane affects diffusion kinetics of insulin receptors and regulates signal transduction (Kabayama  A Biosynthetic pathway (schematic) of GM3, from ceramide, and to complex gangliosides. B TLC analysis of ganglioside species in human serum. C Quantification by densitometry of major ganglioside species GM3 and GD1a in human serum. Data expressed as mean AE SD, n = 6. D Detailed structures of GM3 species: sialyllactose head group, sphingoid base (d18:1), typical fatty-acid lengths (LCFA, VLCFA), and acyl-chain modifications (a-hydroxylation, x-9 desaturation). E Quantification by LC-MS/MS analysis of serum GM3 species with differing acyl-chain structures. Total abundance of 10 species was defined as 1. Data expressed as mean AE SD, n = 6.
In this study, we investigated how serum GM3 species, carrying different acyl chains, regulate inflammatory signaling and contribute to onset of metabolic disorders. Here, we demonstrate that GM3 acts as an endogenous TLR4 modulator. VLCFA-GM3 synergistically and selectively augmented TLR4 activation by LPS and HMGB1, and in contrast, LCFA and unsaturated VLCFA-GM3 suppressed TLR4 activation. Serum VLCFA-GM3 increased significantly and LCFA-GM3 decreased sharply in metabolic disorders. Computational approaches using artificial intelligence revealed that specific GM3 species are related to clinical symptoms. VLCFA-GM3 also increased in the adipose tissue of obese mice and the increase was attenuated in TLR4-mutant mice, implying that TLR4 signaling itself is involved in production of VLCFA-GM3. Our findings suggest that serum GM3 plays a role of rheostat for TLR4 signaling, and the increase in VLCFA-GM3 is a risk factor for TLR4-mediated disease progression.

VLCFA-GM3 species are involved in progression of chronic inflammation in metabolic disorders
To elucidate the role of GM3 species in pathophysiology of metabolic disorders, we analyzed expression patterns of serum GM3 species in human subjects (Veillon et al, 2015;Appendix Fig S1A-I). Sera were collected from human subjects, representing five pathological phases: healthy subjects (control, n = 24), visceral fat accumulation (VFA, n = 38) in presymptomatic phase, VFA with dyslipidemia (lipidemia, n = 28), VFA with hyperglycemia (glycemia, n = 15), and VFA with dyslipidemia and hyperglycemia (lipidemia + glycemia, n = 17). Scores of homeostatic model assessment for insulin resistance (HOMA-IR) and serum C-reactive protein (CRP) were evaluated as indicators of insulin resistance and chronic inflammation, respectively. HOMA-IR and CRP displayed significant correlation with each other (Appendix Fig S1J), and a gradual increase in the order: control < VFA < lipidemia < glycemia < lipidemia + glycemia (Appendix Fig S1K and L). These findings indicate that the order of the five phases corresponds to increasing severity of insulin resistance and chronic inflammation.
Early-phase increases in body mass index (BMI) (> 25) or abdominal circumference (> 85 cm) were associated with sharp reduction in LCFA species (Figs 2E and EV1A) and increase in VLCFA species (Figs 2F and EV1B). These findings suggest that increases in VLCFA-GM3 species occur in obesity, and play a role in early pathogenesis of metabolic disorders. In cases of severe obesity (BMI > 30 and/or abdominal circumference > 100 cm) and severe metabolic disorders (lipidemia + glycemia), there was moderate reduction in VLCFA-GM3 species (Figs 2F and EV1B) and significant increase in unsaturated species (Figs 2G and EV1C). These findings indicate that desaturation of VLCFA species occurs after onset of metabolic disorders.
Abundance of a-hydroxy VLCFA-GM3 (h24:0) showed a linear increase along with increases in BMI and abdominal circumference (Figs 2H and EV1D), with strong correlation (Fig EV1E and F). ahydroxy VLCFA-GM3 was also strongly correlated with indicators of insulin resistance and chronic inflammation (ALT, HOMA-IR, CRP) (Figs 2I and J,. In particular, the ratio of h24:0 to 24:0 was much higher in subjects with abnormal CRP value (> 0.3 mg/dl) (Figs 2K and EV1I), indicating considerable involvement of h24:0 in chronic inflammation. Relationships between these GM3 species and pathophysiology of metabolic disorders are summarized schematically in Fig 2L. In steady state, homeostasis is maintained by balance of GM3 species; in presymptomatic and early phases, VLCFA species increase in correlation with chronic inflammation and insulin resistance; in late phases, modifications such as desaturation and a-hydroxylation could occur in VLCFA species.
Data information: Data shown are individual values and mean AE SD, analyzed by two-tailed unpaired t-test with Bonferroni's correction. *P < 0.05, **P < 0.01, and ***P < 0.001 for comparisons between indicated groups.
◀ Figure 3. Self-organization map (SOM) analysis based on relative abundances of serum GM3 species.
A Procedure (schematic) for self-organization map (SOM) analysis, a pattern recognition method using neural-network-type artificial intelligence. Complex patterns of multi-dimensional information (in this case, expression patterns of the ten major GM3 species in human subjects) are mapped onto a 2D surface. Subjects having similar GM3 patterns are located proximal to each other and form several clusters (red arrows), whereas subjects having different GM3 patterns are located distal to each other (blue arrows). B SOM analysis of control and lipidemia subjects based on expression patterns of ten GM3 species. C Mapping of expression levels of ten GM3 species onto SOM in (B). D Identification of sub-clusters having different GM3 patterns based on SOM in (B). E Metabolic pathways for fatty acids: elongation, desaturation, and a-hydroxylation (a-oxidation) (schematic). Sub-clusters identified by SOM analysis are mapped on metabolic pathways. F Heat map analysis for GM3 species and clinical markers of six clusters. Sample sizes: sub-clusters 1-3 (total), n = 22; cluster 4, n = 7; cluster 5, n = 9; cluster 6, n = 12. G Self-organization map (SOM) based on four GM3 species as indicated at bottom. H ROC curve derived from Bayesian regularized neural networks (BRNNs) based on four GM3 species in (G).

VLCFA-GM3 species selectively enhance mouse TLR4/MD-2 signaling
We also investigated the effects of GM3 species on mouse TLR4/ MD-2 (mTLR4/mMD-2). In RAW macrophages, VLCFA species strongly enhanced TNF-a production by TLR4 ligands LPS and HMGB-1 but not by other TLR ligands, similarly to results in human cells (Fig 6A and B). The enhancement was clearly observed in chronic/weak TLR4 activation by low-dose LPS and was saturated in rapid/strong activation by high-dose LPS (Fig 6B).

A
Profiling of bioactivities of serum GM3 species in LPS-mediated monocyte activation (schematic). B, C GM3-mediated enhancement and inhibition of proinflammatory cytokine production from LPS-stimulated monocytes (LPS: 0.06, 0.13, 0.25 ng/ml). TNF-a (B) production and IL-6 (C) production in culture supernatant were measured by ELISA. D, E Co-stimulation of monocytes by LPS plus GM3 species or complex ganglioside species (1.5, 3.0, 4.5 lM). TNF-a (D) production and IL-6 (E) production were shown in heat maps. F Co-stimulation of monocytes by LPS plus GM3 species or precursor GSL species. TNF-a production, IL-6 production, and IL-12/23 production were shown in heat maps. G Inhibitory effect of LCFA and unsaturated VLCFA-GM3 on VLCFA-GM3 species. IL-6 production in culture supernatant was measured by ELISA.
Data information: Data shown are mean AE SD (n = 3), analyzed by Tukey's multiple comparison test. **P < 0.01 for comparisons between indicated groups.
ª 2020 The Authors The EMBO Journal 39: e101732 | 2020 and unsaturated VLCFA-GM3 species moderately enhanced TLR4 activation (Fig 6A), in contrast to results in human cells. These effects were also observed in BMDMs from C3H/HeN (WT TLR4) mice, but not in C3H/HeJ (dominant-negative TLR4, P712H) mice (Fig 6B), indicating that mTLR4 and its downstream signaling pathway are required. Activation patterns of macrophages were 8 of 20 The EMBO Journal 39: e101732 | 2020 ª 2020 The Authors reproduced by overexpression of mTLR4/mMD-2 complex in HEK293T cells (Fig 6C), and NF-jB activity increased progressively associating with acyl-chain length of GM3 species. Addition of soluble mouse CD14 enhanced the synergistic activation by GM3 species, indicating that CD14 facilitate GM3 representation to mTLR4/mMD-2. Among complex gangliosides and precursor GSLs, synergistic activation in the mouse model was highest for VLCFA-GM3 species (Fig 6D and E), similarly to findings in human cells. LCFA-GM3 (16:0, 18:0) induced synergistic activation at physiological concentration to a similar degree as other GM3 species, but they displayed antagonistic effects at higher concentrations ( Fig 6D). Molecular characteristics of GM3 species and effects on TLR4/MD-2 are summarized in Fig 6F and Appendix Fig S3. VLCFA-GM3 and LCFA-GM3 displayed enhancement and inhibition for TLR4/MD-2, respectively, correlating with volume of the hydrophobic moiety. Interestingly, lipid A/IVa is known to show agonistic and antagonistic activities correlating with total number of fatty acids, also corresponding to volume of the hydrophobic moiety (Akashi et al, 2001;Saitoh et al, 2004). Lipid IVa synergizes with LPS or lipid A in lowdose range, but show antagonistic effects in high-dose range (Mueller et al, 2004), similarly to LCFA-GM3. These current and reported findings suggest that GM3 species utilize molecular mechanisms closely similar to lipid A/ IVa in regulating TLR4 activation.
We previously reported that proinflammatory cytokines released from adipose tissue-resident macrophages induce GM3 production in adipocytes (Nagafuku et al, 2015). TLR4 is a key receptor for cytokine productions in adipose tissue (Shi et al, 2006;Suganami et al, 2017), implying that TLR4 activation itself induces increase in VLCFA-GM3. So, we analyzed GM3 species in HFD C3H/HeN and C3H/HeJ mice by LC-MS/MS. HFD in 8-week-old C3H/HeN mice for 8 weeks resulted in increased body weight, blood glucose level, and visceral adipose tissue weight (Fig EV4E). TLC analysis showed moderate increase in total GM3 (Figs 7C and EV4F), and LC-MS/MS analysis revealed notable increases in a-hydroxy VLCFA-GM3 species in visceral adipose tissues of HFD C3H/HeN mice (Fig 7D). Diabetic phenotypes and increased levels of a-hydroxy VLCFA-GM3 species were ameliorated in HFD C3H/HeJ mice (Figs 7D and EV4E), suggesting that TLR4 signaling is partially involved in production of a-hydroxy VLCFA species in obesity. These findings, taken together, suggest that a-hydroxy VLCFA-GM3 increases in both human serum and mouse adipose tissue (Fig 7E), and an interplay between TLR4 and GM3 species results in a feedback loop from TLR4 to GM3 (shown schematically in Fig 7F).

GM3 species recognition by TLR4/MD-2 induces receptor dimerization/ oligomerization
To elucidate the molecular basis of GM3 recognition and signal transduction, we performed structure-based mutation mapping on TLR4/MD-2 complex. Previously reported crystal structures of TLR4/MD-2 complex (Park et al, 2009;Ohto et al, 2012a) indicate that two ligand-binding sites are formed on the dimerization interface between two TLR4/MD-2 units (Fig 8A). MD-2 forms hydrophobic pockets that bind to the acyl-chain moiety of LPS, while TLR4 leucine-rich repeats (LRRs) provide charged amino acids that recognize the hydrophilic head group of LPS (Fig 8B). Lys (K) and Arg (R) residues around the LPS-binding pocket were replaced by Ala (A), because these cationic residues may recognize the sialic acid on GM3 saccharide chain. Mutations of R264, K341, and K362 greatly reduced synergistic hTLR4 activation by GM3 22:0 and ◀ Figure 5. VLCFA-GM3 species synergistically and selectively control human TLR4/MD-2 activation.
Data information: Data shown are mean AE SD (A and B, n = 3; C, E, and F, n = 4; D, n = 6) analyzed by Tukey's multiple comparison test (A, C, and D) or two-tailed unpaired t-test (B). *P < 0.05 and **P < 0.01 for comparison with stimulation by TLR ligand without GM3 species.
10 of 20 The EMBO Journal 39: e101732 | 2020 ª 2020 The Authors The EMBO Journal Hirotaka Kanoh et al partially reduced hTLR4 activation by LPS single stimulation ( Fig 8C). R322, which recognizes a heptulose-phosphate group on LPS oligosaccharide region (Park et al, 2009), contributed weakly to GM3-mediated TLR4 activation (Fig 8C). Mutations of R264A, K341A, and K362A in combination effectively suppressed GM3mediated TLR4 activation (Fig 8D). On the other hand, nickel ion, an allosteric TLR4 ligand (Schmidt et al, 2010), did not display synergistic activation with GM3 22:0 (Appendix Fig S4). We also confirmed that R264A, K341A, and K362A had no effect on nickelmediated hTLR4 activation. These findings indicate that R264, K341, and K362 are required for hTLR4 activation by both LPS and GM3 species and facilitate their synergistic activation, whereas nickel ion does not synergize with GM3 species because its activity is independent of these amino acids.
Data information: Data shown are mean AE SD analyzed by Tukey's multiple comparison test. *P < 0.05, **P < 0.01 for comparisons between indicated groups.
To investigate activation state of TLR4/MD-2 complex induced by GM3, we performed chemical cross-linking and SDS-PAGE analysis of recombinant mTLR4 extracellular domain/mMD-2 complex. Addition of LPS, GM3 18:0, and chemical cross-linker induced large mobility shift of mTLR4/mMD-2 complex and observed molecular weights indicate the presence of dimers and higher-order oligomers (Fig 8H). Previous reports indicate that LPS-mediated signal transduction is initiated by dimerization of TLR4/MD-2 unit (Akashi et al, 2001;Saitoh et al, 2004;Kobayashi et al, 2006). Clustering of TLR4 was observed by fluorescent and electron microscopy after LPS stimulation (Visintin et al, 2003;Triantafilou et al, 2004;Latty et al, 2018), and the signaling was mediated by a left-handed helical oligomer of downstream adaptors consisting death domains (Lin et al, 2010); i.e., receptor oligomerization may provoke further downstream signaling. These previous and current findings, taken together, suggest that GM3 species act as TLR4-selective endogenous modulators to induce receptor dimerization/ oligomerization, and consequently enhance signal transduction leading to chronic inflammation in metabolic disorders.

Molecular docking approach implicates different binding modes of GM3 species modulating TLR4 activation
To figure out how GM3 species enhance and suppress TLR4 activation depending on the acyl-chain structure, we performed a ligandmacromolecular docking study on hTLR4/hMD-2 complex. Binding modes of VLCFA-GM3 (24:0) and LCFA-GM3 (16:0) were sought on the molecular surface around the hydrophobic pocket of hMD-2 and the basic residues of hTLR4. Docking models of hTLR4/hMD-2/ GM3 (24:0 or 16:0) complex are shown in Fig 9A and B. Similarly to LPS and lipid IVa, both GM3 24:0 and 16:0 bound the hydrophobic pocket of hMD-2 via the fatty acid and the sphingoid base (Fig 9C-E). The binding model of GM3 24:0 overlapped closely to Ra-LPS in the crystal structure of reference, and the saccharide chain of GM3 24:0 was surrounded by basic residues of TLR4 (Figs 9C and EV5A-C). The basic residues of TLR4 (K341, K362, and R322), that interact with the saccharide chain of LPS and show conformational changes upon TLR4 activation (Park et al, 2009;Ohto et al, 2012b), were closely associated with the saccharide chain of GM3 24:0. However, R264, a key residue recognizing 4 0phosphate of LPS and triggering TLR4 activation (Park et al, 2009), was far from the saccharide chain of GM3 24:0. These results imply the underlying mechanism of VLCFA-GM3 capability to enhance TLR4 signaling without triggering activation by itself. Since synergistic activation by VLCFA-GM3 was mainly observed in the presence of low-dose LPS or weak TLR4 ligands, VLCFA-GM3 may act as an endogenous LPS mimic without intrinsic activity, which could sensitize TLR4 signaling by decreasing the ligand concentration required for TLR4 activation and increasing dimer/ oligomer formation ( Fig 9F).
Next, binding model of GM3 16:0 was compared to lipid IVa in complex with hMD-2 (Figs 9E and EV5D-F). Lipid IVa shows different binding mode in comparison with LPS, with reverse orientation of 4 0 -phosphate and acyl chains, which may inhibit dimer formation of hTLR4/hMD-2 by presenting hydrophilic groups (phosphate and glucosamine) to the lipophilic dimer interface (Park et al, 2009;Ohto et al, 2012a;Fig 9C and E). Similarly, GM3 16:0 showed opposite binding mode to GM3 24:0, with reverse orientation of the saccharide chain and the acyl chain (Fig 9D), which may interfere and reduce TLR4 activation through presentation of the saccharide chain to the dimer interface (Figs 9E and EV5D-F). On the other hand, lipid IVa and lipid A (LPS-core structure) are known to show same binding orientations on mTLR4/mMD-2 (Ohto et al, 2012a). Binding model of GM3 16:0 on mTLR4/mMD-2 showed almost the same orientation compared with GM3 24:0 (Appendix Fig S5A-C). These comparative analyses implicate the mechanism by which GM3 species can enhance and reduce TLR4 activation in an acylchain-dependent manner.

Discussion
TLR4 signaling plays crucial roles in pathogenesis of obesity and metabolic disorders. This study demonstrated that human TLR4/ MD-2 received positive regulation by VLCFA-/a-hydroxy VLCFA-GM3 and negative regulation by LCFA-/unsaturated VLCFA-GM3 in the presence of LPS and HMGB1. LCFA-GM3 species such as 16:0 consistently inhibited TLR4 activation even in the presence of VLCFA-GM3 species 22:0 or 24:0; 18:0; and 20:0 ( Fig 5I). These findings indicate that GM3 species function as a rheostat for TLR4 signaling (Fig 5J). Increases in VLCFA-/a-hydroxy VLCFA-GM3 species, and decreases in LCFA-GM3 species, were involved in pathogenesis of metabolic disorders via chronic inflammatory processes. Computational approaches revealed that elongation, ahydroxylation, and desaturation of fatty-acid structures of GM3 were related to signatures of disease progression. a-hydroxy VLCFA-GM3 species were also increased in adipose tissue of obese mice. The increase in a-hydroxy VLCFA-GM3 was attenuated by TLR4 mutation, implying a feedback loop from TLR4 activation to GM3 production, analogous to that for free fatty acids (Suganami et al, 2007). GM3 induced dimerization/ oligomerization of TLR4/MD-2, and MD-2 was involved in recognition of the fatty-acid structure of GM3. These findings suggest that GM3 plays an important role in TLR4 signaling, and the increase in VLCFA-GM3 species, showing the strongest synergistic TLR4 activation, is a risk factor for TLR4mediated disease progression.
Measurement of serum GM3 species will potentially allow evaluation of hidden risks of TLR4-signaling-related inflammatory diseases (e.g., inflammatory bowel disease, chronic kidney disease, rheumatoid arthritis, cancer metastasis) via LPS and endogenous ligands such as HMGB1, S100A8/9 (Mrp8/14), and SAA3 (Vogl et al, 2007;Hiratsuka et al, 2008;Harris et al, 2012). Over 20 GM3 species, in addition to the ten major species examined in this study, are present in human serum (Veillon et al, 2015). Moreover, there is increasing evidence for important roles of GSLs in innate immune responses and chronic inflammation (Kondo et al, 2013;Nakayama et al, 2016;Nagata et al, 2017;Nitta et al, 2019). Expression pattern analysis utilizing artificial intelligence will allow us to deal effectively with the complexity and variety of GM3 and other GSL species, and to further elucidate the relationships between particular species and inflammatory diseases.
The detailed mechanism whereby GM3 species are secreted and presented to TLR4/MD-2 complex is currently under study. It is supposed that GM3 is secreted as part of a lipoprotein complex ª 2020 The Authors The EMBO Journal 39: e101732 | 2020 14 of 20 The EMBO Journal 39: e101732 | 2020 ª 2020 The Authors (Senn et al, 1989;Veillon et al, 2015) allowing circulation from the liver to most body tissues, including adipose tissue. TLR4 has been shown to mediate innate immune responses by LDL cholesterol (Stewart et al, 2010). Ceramide 24:0 is preferentially incorporated into LDL cholesterol (Boon et al, 2013). The present study shows that levels of VLCFA-GM3 and non-HDL cholesterol increase together, whereas levels of LCFA-GM3 and HDL cholesterol decrease together, indicating species-selective incorporation into lipoproteins. However, other secretion pathways, such as exosomes and microvesicles (Skotland et al, 2017), may also be involved. Furthermore, a-hydroxylation, mediated by enzymes such as fatty acid-2 hydroxylase (FA2H), may contribute to secretion of GM3 species via reducing hydrophobicity and affecting lipid diffusion (Hama, 2010). In regard to activation mechanisms, our results displayed that CD14 and MD-2 facilitate GM3 to modulate TLR4 signaling. It is possible that CD14, MD-2, and LPS-binding protein take up serum GM3 species and transport them to TLR4, as reported for LPS (Ryu et al, 2017). As shown in docking study, VLCFA-GM3 and LCFA-GM3 may interact with TLR4/MD-2 by utilizing different interaction modes to promote or disrupt dimerization, similarly to lipid A/IVa species and eritoran (a strong antagonist in lipid IVa analogs) (Mullarkey et al, 2003;Kim et al, 2007;Ohto et al, 2007Ohto et al, , 2012aAppendix Fig S3). In particular, molecular features of LCFA-GM3 and unsaturated VLCFA-GM3 resemble those of eritoran: (i) short aliphatic-chain length (C10) in comparison with agonistic lipid A species (C14) and (ii) desaturation (C18:1, x-7) making a 180degree turn of the acyl chain in the hydrophobic pocket of MD-2 (Kim et al, 2007). Mimetic compounds based on lipid A/IVa precursors (diacyl monosaccharide species), carrying less number of fatty acids, show antagonistic effect (Facchini et al, 2018). Thus, less fatty-acid number, shorter acyl-chain length, and desaturation may cooperatively contribute to antagonistic activity by affecting interaction mode. Our findings suggest that GM3 species modulate TLR4 activation by utilizing molecular mechanisms closely related to lipid A/ IVa. Formation of two ligand-binding pockets on the dimerization interface between two TLR4/MD-2 units has been suggested by crystallographic analyses (Park et al, 2009;Ohto et al, 2012a). Reported binding study of TLR4/MD-2 with lipid A suggests that the maximal binding of the agonistic E. coli lipid A was approximately half-fold lower than that of the antagonistic lipid IVa (Akashi et al, 2001;Saitoh et al, 2004); i.e., under physiological conditions, one ligand pocket is occupied by agonist (e.g., LPS) while the other is vacant or occupied by unknown intrinsic ligands. It may allow GM3 species to modulate dimerization efficiency via the second pocket (Fig 9F and G). Future studies are expected to reveal structures of oligomeric TLR4/MD-2 signalosomes complexed with GM3 species. Additionally, it is known that the dimerization and internalization of mTLR4/MD-2 upon acute stimulation by LPS can be analyzed by flow cytometry (Akashi et al, 2003;Zanoni et al, 2011;Tan et al, 2015), which might enable to detect GM3-mediated receptor dynamics directly on the plasma membrane of living cells.
Biosynthesis of the various GM3 species may depend on several enzymes: fatty-acid elongases (Elovls), acyl-CoA desaturases, ceramide synthases (CerSs), and GM3S. Blocking of 16:0-to-18:0 fattyacid elongation in mice by Elovl6 knockout was found to inhibit progression of metabolic disorders through alterations of fatty-acid structures, i.e., increased 16:0 and decreased 18:0-to-24:0 levels (Matsuzaka et al, 2007). Elovl6 deficiency therefore may attenuate increase in VLCFA-GM3 species, and achieve homeostatic balance of acyl-chain structures. On the other hand, increase in LCFA-Cer (16:0) and decreases in VLCFA-Cer (22:0, 24:0) in obese subjects, resulting from imbalance of CerS2/6 expression and inhibition of b-oxidation, were reported to correlate to progression of metabolic disorders (Raichur et al, 2014;Turpin et al, 2014). Our results imply that such imbalances in Cer species might be involved in decreased production of LCFA-GM3 and increased production of VLCFA-GM3 in metabolic disorders. Fatty-acid desaturation was shown to occur in the resolution phase of innate immune response, and to reduce inflammation (Oishi et al, 2017); however, the direct mechanism whereby x-9 mono-unsaturated VLCFA attenuates chronic inflammation is not completely understood. Increased levels of unsaturated GM3 species in severe metabolic disorders may result from desaturation mechanisms after the activation phase. Both elongase and desaturase genes are regulated by SREBP-1, a key transcription factor in lipid signaling whose activation occurs in parallel with that of NF-jB (Matsuzaka et al, 2007;Oishi et al, 2017). We previously reported that proinflammatory cytokines TNF-a and IL-1b induce GM3S expression and GM3 production in adipocytes (Tagami et al, 2002;Nagafuku et al, 2015). In the present study, TLR4 deficiency reduced production of a-hydroxy VLCFA-GM3 (Fig 7D), suggesting the involvement of TLR4 signaling in GM3 production. These previous and current findings indicate that fatty-acid structures and total expression level of GM3 species are controlled by complex, coordinated mechanisms regulated by innate immune signaling, lipid signaling, and other cellular responses.
Moreover, it should be clarified directly in adipose tissue that GM3 species could mediate the adipocyte-macrophage communication in the future study. It would be important to specify the GM3 and other ganglioside species expressed in a specific type of cells, such as macrophages, pre-adipocytes, and differentiated adipocytes, that are mixed in adipose tissue. While pre-adipocytes/adipocytes predominantly express GM3, it is considered that human monocytes and mouse macrophages express GM3 and GM1/GD1a, respectively (Yohe et al, 2001;Tanabe et al, 2009). However, it remains unclear how ganglioside species and their acyl-chain structures are different in a cell-type-specific manner in the intact adipose tissue. To characterize miscellaneous cells in adipose tissue, in vitro enzymatic digestion/fractionation and antibody-based cell sorting are performed generally. On the other hand, our previous report suggested that GM3 expression in adipocytes was regulated by the co-presence of the resident macrophages in adipose tissue (Nagafuku et al, 2015). It has been also known that the activation of GM3 synthase in monocyte/macrophages was easily occurred during culturing in vitro (Gracheva et al, 2007). Therefore, the specific method such as the imaging mass spectrometry for GM3 species should be established in order to analyze GM3 species directly in the intact adipose tissues without in vitro cell manipulation (Sugimoto et al, 2016).
In regard to potential therapeutic approaches, treatment with supplemental LCFA-GM3 16:0 may inhibit systemic and local production of TNF-a, IL-6, and IL-12/23 via TLR4, and in part via TLR2, driven by LPS and HMGB1. On the other hand, VLCFA-GM3 24:0 could act as a booster for immunological adjuvants such as monophosphoryl lipid A species (LA505, LA504) and other synthetic ª 2020 The Authors The EMBO Journal 39: e101732 | 2020 TLR4 ligands (Wang et al, 2016;Chan et al, 2017;Okamoto et al, 2017). Utilization of naturally occurring GM3 species may prevent production of autoantibodies (Bowes et al, 2002).
In conclusion, our findings would help clarify the pathophysiological roles of serum/ adipose GM3 species in TLR4 signaling, and the complex interplay between glycosphingolipid metabolism and innate immune signaling in metabolic disorders.

Vector construction
Vector carrying mouse MD-2 and TLR4 cDNA (pDUO-mMD2/ TLR4) was from InvivoGen (San Diego). cDNA fragments, fused with a KpnI site and one Kozak sequence (ACC) at 5 0 -end and SalI site at 3 0 -end, were amplified by PCR (KOD-Plus-Neo; Toyobo) and inserted separately into pCDNA3 at KpnI and XhoI sites (Invitrogen). A set of vectors for dual luciferase assay, NF-jB reporter gene (pGL3-ELAM; a firefly luciferase gene controlled by NF-jB-dependent promoter of ELAM-1), control reporter gene (pRL-TK; a Renilla luciferase gene controlled by constitutive active promoter of thymidine kinase), and pCDNA3 vectors carrying human MD-2 and TLR4 cDNA were previously described (Muta & Takeshige, 2001;Fujimoto et al, 2004). Reporter vectors for AP-1 and ISRE were purchased (Promega; Australia). Site-directed mutagenesis was performed according to the manufacturer's protocol of QuikChange (Agilent; Santa Clara, CA, USA) with minor modifications.

Data availability
The mass spec data of the GM3 species in human serum are available at the database GlycoPOST (https://glycopost.glycosmos.org). The accession number is GPST000057.
Expanded View for this article is available online.