**Elusive Roles of the Di**ff**erent Ceramidases in Human Health, Pathophysiology, and Tissue Regeneration**

**Carolina Duarte 1,\*, Juliet Akkaoui 1, Chiaki Yamada 1, Anny Ho 1, Cungui Mao 2,3 and Alexandru Movila 1,4,\***


Received: 18 April 2020; Accepted: 27 May 2020; Published: 2 June 2020

**Abstract:** Ceramide and sphingosine are important interconvertible sphingolipid metabolites which govern various signaling pathways related to different aspects of cell survival and senescence. The conversion of ceramide into sphingosine is mediated by ceramidases. Altogether, five human ceramidases—named acid ceramidase, neutral ceramidase, alkaline ceramidase 1, alkaline ceramidase 2, and alkaline ceramidase 3—have been identified as having maximal activities in acidic, neutral, and alkaline environments, respectively. All five ceramidases have received increased attention for their implications in various diseases, including cancer, Alzheimer's disease, and Farber disease. Furthermore, the potential anti-inflammatory and anti-apoptotic effects of ceramidases in host cells exposed to pathogenic bacteria and viruses have also been demonstrated. While ceramidases have been a subject of study in recent decades, our knowledge of their pathophysiology remains limited. Thus, this review provides a critical evaluation and interpretive analysis of existing literature on the role of acid, neutral, and alkaline ceramidases in relation to human health and various diseases, including cancer, neurodegenerative diseases, and infectious diseases. In addition, the essential impact of ceramidases on tissue regeneration, as well as their usefulness in enzyme replacement therapy, is also discussed.

**Keywords:** ceramides; ceramidases; inflammation; neurodegenerative diseases; infectious diseases

#### **1. Introduction**

Ceramides are bioactive sphingolipids responsible for cell apoptosis, senescence, and autophagy [1]. They are the precursors of other bioactive sphingolipids, including sphingosine (SPH), sphingosine-1-phosphate (S1P), and ceramide-1-phosphate, which play specific roles in signal transduction pathways (Figure 1) [1].

**Figure 1.** Role of ceramidases in ceramide metabolism. Ceramide in mammalian cells may be generated: (1) via the de novo synthesis pathway, which begins with the condensation of L-serine and Palmitoyl-CoA); (2) by the hydrolysis of sphingomyelin and glucosylceramide; or (3) from the dephosphorylation of ceramide-1-phosphate. Ceramidase is an enzyme that cleaves fatty acids from ceramide, producing sphingosine. Sphingosine may then be phosphorylated by a sphingosine kinase to form sphingosine-1-phosphate. SMase—sphingomyelinase; GCS—glucosylceramide synthase.

The de novo anabolic pathway for the biosynthesis of ceramide begins with the condensation of the amino acid, L-serine, and palmitoyl-CoA, producing 3-ketosphingonine. Then, 3-ketosphingonine is quickly converted into dihydrosphingosine (dhSPH) by 3-ketosphinganine reductase. The subsequent acylation of dhSPH by (dihydro)ceramide synthases gives rise to dihydroceramides. Finally, the removal of two hydrogens from a fatty acid chain of the dihydroceramides by the enzyme desaturase results in the formation of ceramides (Figure 1) [2]. Ceramides can be further hydrolyzed into sphingosine (SPH) and free fatty acids by ceramidases (Figure 1). SPH is the most common sphingolipid base molecule in mammalian cells and is the precursor of S1P [1].

The bioactive lipid mediator S1P is involved in cell proliferation, differentiation, and survival, whilst ceramides and SPH mediate cell death [1,2]. Notably, SPH is exclusively generated from the catabolism of ceramides by ceramidases [2]. Ceramidases control the balance between S1P and ceramides/SPH concentration, which leads to either cell survival or cell death [1,3]. Hence, a ceramidase-based enzyme replacement therapy that simultaneously achieves ceramide reduction and SPH elevation has been recently examined [4]. This therapeutic approach intends to reduce the negative pathophysiological impact of cell death mediated by ceramides [4]. To date, five human ceramidases have been identified and classified according to their optimal pH for catalytic activity: one acid ceramidase (ACDase) encoded by the gene ASAH1, one neutral ceramidase (NCDase) encoded by ASAH2, and three alkaline ceramidases (ALKCDase) encoded by the genes ACER1, ACER2 and ACER3 [3]. ACDase is the most widely studied ceramidase, and its effects on pathophysiology are variable. While NCDase and ALKCDases 1–3 have been the subject of studies in recent decades, our knowledge about their roles in pathophysiology remains limited. Thus, the aim of this review was to summarize the most recent knowledge of the biology of ceramidases and their role in the pathology of various common human diseases, including cancer, diabetes, and neurodegenerative diseases

(Figure 2). In addition, the roles of the ceramidases in infectious diseases, tissue regeneration, and healing were also addressed.

**Figure 2.** Pathological consequences of ceramidase dysregulation in mammalian cells which occur upon a loss or a gain of function.

#### **2. Characteristics and Regulatory Pathways of Ceramidases**

#### *2.1. General Characteristics*

#### 2.1.1. Acid Ceramidase

ACDase (ASAH1) is synthesized from a 53–55 kDa polypeptide precursor which is proteolytically processed into the enzyme's 13 kDa α-subunit and 30 kDa β-subunit inside the lysosomes [5,6]. ACDase is a lipid hydrolase found in the lysosomal compartment of cells which catalyzes the hydrolysis of C6:0–C18:0 ceramides to SPH [6]. It is ubiquitously expressed in human tissues, with a particularly high expression in the heart and kidneys and is known for its role in senescence and apoptosis [7].

#### 2.1.2. Neutral Ceramidase

NCDase (ASAH2) is synthesized as 118 kDa and 142 kDa isoforms in humans and as a 96 kDa molecule in mice [6]. NCDase is a transmembrane glycoprotein that is highly expressed in the human intestinal system and uses various ceramides and dihydroceramides as a substrate, with a reported preference for C16:0 and C18:0 ceramides [8,9]. It is localized to different cellular compartments, including the plasma membrane of cells; regulates the conversion of ceramide into SPH and S1P; and is important for the metabolism of dietary sphingolipids [10].

#### 2.1.3. Alkaline Ceramidase

ALKCDases 1–3 (*ACER*1, ACER2, ACER3) are the smallest proteins among the ceramidases, with molecular weights of 31–31.6 kDa [6,11]. ALKCDases are predominantly located in the Golgi

complex and endoplasmic reticulum and play a role in cell differentiation [11]. ACER1 hydrolyses C20:0-C24:0 ceramides [9]. It is predominantly expressed by skin cells and is involved in their differentiation, as well as in the viability of hair follicle stem cells [6,12–14]. ACER2 hydrolyses ceramides and dihydroceramides C18:1 and C20:1 [9]. It is upregulated during DNA damage and induces programmed cell death through an SPH-dependent pathway [15]. ACER3 hydrolyses ceramides, dihydroceramides, and phytoceramides with long unsaturated acyl chains [9]. It has been described as a seven-transmembrane protein, much like the adipocyte receptor (ADIPOR), and is associated with cytokine upregulation [16].

This brief overview of the general characteristics of ceramidases indicates that they have been classified according to their optimal pH. However, ceramidases also differ in molecular weight and expression patterns. Importantly, all three groups of ceramidases have a specific ceramide affinity and reported cellular functions. It is important to highlight that the ALKCDases, although classified together, differ in ceramide affinity and function. Moreover, ALKCDase-3, like NCDase, is a transmembrane protein and not a soluble enzyme. Thus, further consideration should be given to the classification of ceramidases and, particularly, the ALKCDases.

#### *2.2. Regulatory Pathways*

#### 2.2.1. Acid Ceramidase

The activation of ACDase induces a pro-survival state, while its inhibition leads to cell death through a variety of apoptotic pathways mediated by caspases (CASP), poly (adp-ribose) polymerase (PARP), or cathepsins (CTS) [10,17–25]. Cathepsin B and Cathepsin D are activated during ceramide-induced apoptosis but are inhibited by ACDase activity [19,25]. Interestingly, the downregulation of Cathepsin B by ACDase increases ACDase's own activation, triggering a feedback mechanism through which ACDase prolongs its own activation through Cathepsin B inhibition [10]. Additionally, ACDase activity can be regulated by Ceramide Synthase 6 (CerS6) [26]. CerS6 increases the levels of C16:0, which, in turn, activate ACDase through JNK-AP1-dependent mechanisms. However, this same mechanism mediates the inhibition of the gene expression of NCDase and ALKCDases in colorectal adenocarcinoma [26].

An age-dependent inhibition of ACDase leads to ceramide accumulation, an increase in oxidative stress, and the death of retinal cells and erythrocytes [27,28]. By contrast, it was reported that kidney cells collected from aged mice show an elevated expression of *Asah1* mRNA compared to that of young mice [29]. Thus, this published evidence suggests a tissue-specific ACDase activity in relation to cellular senescence and aging.

#### 2.2.2. Neutral Ceramidase

The activity and gene expression of NCDase have been linked to cell-cycle arrest and growth regulation [30]. Biochemically, NCDase is a lipid amidase with a mechanistic similarity to a bacterial NCDase [8]. NCDase activates nitric oxide (NO), the WNT/β-catenin pathway, caspase apoptotic pathways, and autophagosomal activity in vivo and is associated with mitochondrial integrity [31–34]. Its gene expression and activity are regulated by c-Jun/AP-1 signaling, NO, all-trans retinoic acid, and ultra-violet radiation [35,36].

#### 2.2.3. Alkaline Ceramidase

The ALKCDases 1–3 are regulated through markedly different mechanisms. ACER 1 is upregulated by extracellular calcium, through which it contributes to the regulation of cell differentiation and growth arrest [37]. Meanwhile, ACER2 is induced by p53 and activates p38 MAPK and AP-1 signaling to mediate DNA damage response, autophagy, and apoptosis [15,38,39]. ACER3 is associated with the AKT/BAX pathway and activates the S1P phosphorylation of AKT through S1PR2 and PI3K in cancer cells [40,41].

#### **3. Association of Ceramidase Gene Mutations with Human Inheritable Diseases**

#### *3.1. Acid Ceramidase*

#### 3.1.1. Farber Lipogranulomatosis (FRBRL)

FRBRL is an autosomal recessive lysosomal disorder with a broad spectrum of phenotypes caused by 16 identified mutations of *Asah1* [5]. It is characterized by a substantial neurologic deficit, subcutaneous nodules, progressive arthritis with joint deformities, laryngeal hoarseness, and an accumulation of storage-laden CD68<sup>+</sup> macroglia/macrophages in white matter, periventricular zones, and meninges of the brain [42]. Animal models of ACDase deletion present hematopoietic organ hypertrophy, characterized by a foamy macrophage infiltration and increased myeloid progenitor colonies [43]. These myeloid progenitor colonies are comprised of cells that can develop normally when treated with ACDase [43]. Additionally, ACDase deletion causes an impaired airway resistance, elastance, and compliance; reduced blood oxygenation; lung edema; and increased immune cell infiltration of the lungs by foamy macrophages and neutrophils [44]. Furthermore, an increased vascular permeability of the lungs, heart, thymus, liver and spleen, as well as neurologic problems, including decreased deambulation, anxiety, and impaired motor coordination, are also observed [42]. These neurologic problems are caused by abnormal sphingolipid profiles in the brain and CD68+ microglia [42].

Changes in the *Asah1* gene expression in FRBRL patients result in the upregulation of the inflammatory cytokines interleukin 4 (IL-4), IL-6, tumor necrosis factor alpha (TNFα), and macrophage colony stimulating factor (M-CSF) in addition to the angiogenic marker, vascular endothelial growth factor (VEGF) [42,45]. Likewise, the expressions of the chemo-attractants, monocyte chemotactic protein-1 (MCP-1), and interferon gamma-induced protein 10 (IP10), are inversely correlated with the level of ACDase activity [45]. These mediators of inflammation, angiogenesis, and insulin resistance may be associated with the immune cell infiltration found in the organ tissues of FRBRL animal models [45]. MCP-1 deletion can partially rescue FRBRL phenotypes by improving organomegaly, blood cell counts, and liver and lung damage by inflammatory infiltrates, as well as the behavioral and neurologic aspects of the disease. However, hematopoiesis is not improved [46]. Similarly, the overexpression of MCP-1, IP-10, and IL-6 can be partially corrected by hematopoietic stem cell transplants [45]. Moreover, treatment with ACDase induces a dose-dependent decrease in hematopoietic organ weight, macrophage infiltration, and *MCP-1* expression, as well as increased expression of Collagen Type 2 *(Col2),* aggrecan, and Sox-9 by chondrocytes [4].

#### 3.1.2. Spinal Muscular Atrophy with Progressive Myoclonic Epilepsy (SMA-PME)

SMA-PME is a rare autosomal recessive disorder that is frequently associated with FRBRL and is caused by two identified mutations of *Asah1* [5]. This disorder is characterized by motor neuron disease and progressive myoclonic epilepsy, with a variable occurrence of sensorineural hearing loss, action tremor, cognitive dysfunction, and cerebral/cerebellar atrophy. Patients with SMA-PME present a 70–95% reduction in ACDase activity, a low ACDase/β-galactosidase ratio, and increased creatine kinase levels [47]. Additionally, the muscle atrophy associated with SMA can be accompanied by cyclooxygenase deficiency [48].

#### 3.1.3. Intrauterine Growth Restrictions (IUGR)

The consequence of ACDase gene overexpression during gestation and its therapeutic effect on associated genetic disorders has also been described. IUGR can result from the TGFβ/ALK5-mediated overexpression of *Asah1* mRNA and increased ACDase activity, which upregulates SPH but not S1P concentrations during pregnancy [49]. S1P is not upregulated at the same rate as SPH in IUGR due to the inactivation of SPH kinase 1 through the ALK1-SMAD1/5 pathway [49].

This suggests that ACDase may induce embryonic cell death through SPH rather than affect embryonic cell proliferation and differentiation through S1P in IUGR.

#### 3.1.4. Krabbe Disease

Globoid cell leukodystrophy, or Krabbe disease, is a congenital disorder caused by mutations in the galactosylceramidase gene, *GALC*, and is characterized by psychomotor regression, muscular hypertonia, muscular spasticity, truncal hypotonia, irritability, seizures, and nystagmus [50]. This disorder is caused by an accumulation of psychosine, a by-product of the deacylation of GALC by ACDase [51]. Therefore, the inhibition of ACDase activity, as observed in FRBRL or after treatment with ACDase inhibitors, can rescue the Krabbe Disease phenotype by preventing psychosine accumulation [51].

#### *3.2. Neutral ceramidase*

There are no reports demonstrating the association of the point genetic mutations of ASAH2 with inheritable diseases in humans. It is important to mention that *Asah2*−/<sup>−</sup> mice are viable and appear without severe defects [52].

#### *3.3. Alkaline Ceramidase*

#### Progressive Leukodystrophy

Progressive leukodystrophy is a group of disorders that affect the white matter of the brain and can occur as a consequence of ACER3 deficiency [53]. This condition is caused by a loss of function mutation in p.E33G which inactivates the catalytic activity of ACER3 and leads to an accumulation of sphingolipids in the blood [53]. The clinical phenotype associated with ACER3 mutations is caused by incorrect central nervous system myelination due to abnormal levels of ceramides in the brain [16]. While the study reporting the loss of function mutation in p.E33G did not report a sphingolipid accumulation or pattern in the brain, it is reasonable to assume that sphingolipid accumulation due to ACER3 inactivation results in abnormal sphingolipid patterns in the brain.

#### **4. Role of Ceramidase Activity in Human Non-heritable Diseases**

#### *4.1. Role of Ceramidases in Cancer Pathology*

The overexpression of ceramidases have been identified in various cancer cell types, and growing evidence suggests that they can be considered molecular markers and/or therapeutic targets for cancer [54] (Figure 3).

**Figure 3.** Overexpression of acid (ACDase), neutral (NCDase), and alkaline (ALKCDase) ceramidases in specific types of cancer.

#### 4.1.1. Acid Ceramidase

ASAH1 has been identified in cancer cells and is associated with radiotherapy/chemotherapyresistant tumors [22,55–57], metastatic cell lines [58], and estrogen/progesterone/androgen receptor-positive cells [59,60]. While ACDase gene overexpression has been identified in low-survival-rate colorectal adenocarcinoma and glioblastoma [57,61], it has also been observed in node-negative melanoma and breast cancer [59,62], which makes it a questionable marker for the aggressiveness or invasiveness of the disease. The *ASAH1* mRNA expression in cancer cells can be increased by radiotherapy, thereby generating resistance [56]. Likewise, the overexpression of *ASAH1* can be driven by the oncogene microphthalmia-associated transcription factor (MITF) [63]. ACDase activity is increased by the androgen receptor activation by dihydrotestosterone in prostate cancer, leading to decreased C16:0 levels and reduced cell apoptosis [60]. Incidentally, the ACDase activity is significantly more upregulated than the *ASAH1* expression in melanoma cells [62]. This may suggest that gene expression alone should not be the determining factor in the use of ACDase as a marker for cancer; ACDase activity should also be assessed.

Multiple molecular mechanisms by which ACDase activation regulates cancer development and progression have been identified. For instance, drug resistance in leukemia is mediated by the ACDase activation of the drug transporter molecule ATP-Binding Cassette, Subfamily B, Member 1 (ABCB1), through nuclear factor kappa B (NF-κB) [64], whilst leukemic cancer cell survival is increased by the ACDase-mediated upregulation of the myeloid cell leukemia sequence 1 (MCL-1) [10]. Furthermore, cancer cell necrosis is mediated by ACDase gene overexpression in polynuclear giant cancer cells that undergo asymmetric cell division [65]. ACDase also regulates cancer cell motility through the activation of the ITGαVβ5/FAK signaling cascade [63]. Additionally, the significant roles of ACDase in angiogenesis, chronic inflammation, and tumorigenesis may contribute to cancer development and progression [66,67]. ACDase affects multiple factors in cancer pathogenicity, which adds to the complexity of the enzyme in the diagnosis and treatment of the disease.

A variety of ACDase inhibitors have been developed and successfully tested in different cancer cell types. ACDase deletion blocks the cell cycle at G1/S, promotes senescence through the β-Galactosidase/MITF pathway, induces apoptosis, reduces tumorigenesis, increases growth arrest, and decreases malignancy [68]. It was demonstrated that the activity of ACDase was significantly inhibited by Carmofur [55,58], LCL521 [19,65,69,70], Ceranib2 [24,71–73], N-oleocylethanolamine (NOE) [22,57], ARN14988 [74], LCL204 [10,64], Monascus Purperus (MP) [18], Hesperetin (Hst) [17], Hesperetine-7-O-acetate (HTA) [17], Silibinin [20], Curcumin [23], and Sanguinarine [21,75], leading to an increased accumulation of intracellular ceramide and apoptosis in various types of cancer cells, including glioblastoma; squamous cell carcinoma; acute myeloid leukemia; colorectal adenocarcinoma; and breast, prostate, lung, gastric, and kidney cancer. Furthermore, Carmofur, NOE, LCL521, and Ceranib2 have been used in combination with chemotherapeutic drugs or photodynamic therapy to either overcome cancer cell resistance to treatment, increase cell sensitivity to specific drugs, or increase the overall effectiveness of cancer cell apoptosis [22,55,58,70,72,73]. Ceranib2 treatment leads to an abnormal cell and mitochondria morphology and decreases the ability of cells to cluster [24,74]. It activates PARP and CASP3/7/8/9-mediated cell apoptosis; increases the expression of the pro-apoptotic markers BID, BCL2-Associated Agonist Of Cell Death (*BAD*), and BCL2-Associated X Protein (BAX); and decreases the expression of anti-apoptotic protein B-Cell Cll/Lymphoma 2 (BCL-2) [72,73]. Furthermore, MP, Hst, HTA, Curcumin, and Sanguinarine activate the apoptotic pathways dependent on Casp3/9 or reactive oxygen species (ROS) [17,18,21,23,75]. Sanguinarine induces peroxide-dependent ceramide generation and the inhibition of the AKT activation pathway [75]. NOE and LCL204 induce PARPand Casp3-mediated apoptosis [10,22], whereas LCL521 increases C16:0 levels, autophagosome accumulation, ER stress, and Cathepsin B- or Cathepsin D-mediated apoptosis [19]. Altogether, ACDase inhibitors are effective promoters of cancer cell death through different apoptotic pathways and have been shown to affect not only apoptosis but also cancer treatment resistance and cancer cell adhesion.

#### 4.1.2. Neutral Ceramidase

An elevated gene expression of NCDase has been identified in both the plasma membrane and Golgi apparatus of colorectal cancer (CRC) cells, where its overexpression inhibits ceramide C6-mediated cell death [8]. Meanwhile, its deletion induces caspase and autophagosome-mediated apoptosis in the presence of C6 [32]. NCDase regulates CRC cell proliferation through the WNT/β-catenin pathway and by increasing the accumulation of SPH and S1P [31,32]. NCDase inhibition may affect cell-to-cell adhesion by reducing the β-catenin levels through AKT phosphorylation and, subsequently, GSK3β activation [31]. It also significantly reduces Azoxymethane-induced colon carcinogenesis by inhibiting aberrant crypt foci formation and transformation [32]. NCDase inhibition does not affect non-cancerous cell function, which makes it a suitable target for colon cancer therapy [32].

We can conclude that NCDase inhibition, like that of ACDase, activates apoptosis and affects adhesion in cancer cells. In addition, it may be a contributing factor in cancerous transformation.

#### 4.1.3. Alkaline Ceramidase

ALKCDases can also affect cancer development and treatment. ACER2 is upregulated by the tumor suppressor gene p53 [38,39]. It was demonstrated that a moderate upregulation of ACER2 increases the levels of SPH and S1P and inhibits cell cycle arrest and senescence. However, when overexpressed, ACER2 mediates programmed cell death, autophagy, and apoptosis through ROS [15,38,39]. ACER2 also contributes to the effects of ionizing radiation treatment [39]. It also increases the phosphorylation of Ezrin-radixin-moesin through intracellular S1P production, hereby inactivating this group of proteins that regulate cell shape and motility and have been associated with cancer progression and metastasis [76]. ACER3 is expressed in low-survival hepatocellular carcinomas and acute myeloid leukemia [40,41]. It induces the S1P phosphorylation of AKT through the S1P receptor 2 and PI3K and

inhibits the AKT/BAX apoptotic pathway in cancer cells [40,41]. Therefore, the inhibition of ACER3 reduces cell growth and increases cancer cell apoptosis.

These published observations indicate that ALKCDases are also associated with the regulation of cancer cell apoptosis. However, the observations of ACER2 overexpression reflect molecular effects contrary to those expected of ceramidases. Nonetheless, ALKCDases are associated with drug resistance and cancer metastasis, like the previously described ACDase.

#### *4.2. Role of Ceramidases in the Onset of Age-Related Neurodegenerative Diseases*

Ceramidases are involved in myelin and fatty acid metabolism and are associated with changes in the brain during aging [77]. For instance, ACER3 is upregulated with age and leads to a decrease in the brain levels of C18:0 and C18:1 ceramide, and its deletion results in purkinje cell degeneration and impaired motor coordination and balance in mice [78]. It has been reported that the overexpression of ACDase has implications for the onset and progression of neurodegenerative diseases, including Alzheimer's disease (AD) and Gaucher disease. Furthermore, treatment with ACDase inhibitors can control AD and Gaucher Disease, as well as Type IV Mucolipidosis.

#### Acid Ceramidase

AD is a multifactorial, highly heterogeneous, and complex disorder that affects the memory and cognitive functions of patients to the extent that they are completely dependent upon nursing care. It is now estimated that nearly 35.6 million patients are affected by AD worldwide and that about 4.6 million new cases are added each year, causing enormous societal and economic burdens, with the estimated cost reaching \$1 trillion/year [79]. AD is caused by an accumulation of derivates from the amyloid precursor protein (APP), which can be modulated by the ATP-binding cassette transporter-2 (ABCA2). ABCA2 is a phospholipid transporter which increases the transcription of APP by activating the ACDase-mediated production of SPH [80]. Furthermore, ACDase inhibition by Ceranib 1 decreases SPH concentration and, subsequently, APP production in ABCA2-overexpressing cells [80].

Gaucher disease is a disorder caused by a loss of function mutations in the glucocerebrosidase (GCase)-encoding gene, Gba1. In a GCase deficiency, the breakdown of glucocylceramide (GlcCer) into ceramide and glucose by GCase is replaced by the ACDase deacylation of GlcCer into glucocylSPH (Glc-Sph), a cytotoxic compound [80]. The inhibition of ACDase by Carmofur corrects the lipid abnormalities in the GCase deficiency by reducing the accumulation of Glc-Sph [81,82]. GBA1 mutations are also a risk factor for Parkinson's disease, a neurodegenerative disorder characterized by Lewy body inclusions containing α-synuclein. Treatment with ACDase inhibitors decreases the accumulation of α-synuclein in cases of GBA1 mutation [81]. Similar lipid patterns are observed in the optic nerves of glaucoma patients, where Asah1 and Asah2 genes are overexpressed, but non-lysosomal GCase-GBA2 is inhibited, resulting in a lower total lipid content and significantly higher concentrations of Glc-Sph [83].

Type IV Mucolipidosis is a neurodegenerative disease caused by a loss-of-function mutation of human transient receptor potential-mucolipin-1 (TRPML-1). Treatment with the ACDase inhibitor, carmofur, induces the activity of TRPML-1 tunnels by increasing the SPH concentration in kidney cells and acting as a mediator of lysosome fusion and trafficking in multivesicular bodies, which can potentially compensate for the loss of function of TRPML-1 [84].

#### *4.3. Role of Ceramidases in Cardio-Pulmonary Disease*

Elevated levels of ceramide are known to be correlated with adverse cardiac events, whereas SPH has been shown to increase intracellular NO levels and maintain the mitochondrial integrity of the cardiovascular system [33]. Conversely, increased blood S1P levels are associated with the pathogenesis of inflammatory and cardiovascular diseases [85]. Hence, an association between ceramidase and cardiopulmonary events is expected.

#### 4.3.1. Acid Ceramidase

The inhibition of ACDase activity is associated with cystic fibrosis (CF), which is caused by a dysregulation of the epithelial fluid transport in the lungs, resulting in a sticky dry mucous accumulation [86]. In CF, β1-Integrins are ectopically expressed in the luminal pole of epithelial cells and downregulate ACDase, leading to an increased ceramide accumulation. However, treatment with recombinant ACDase internalizes the β-Integrins and regulates ceramide accumulation, rescuing the CF phenotype [86].

#### 4.3.2. Neutral Ceramidase

NCDase is inhibited in coronary artery disease vessels. NCDase and ADIPOR mediate the NO-dependent flow-induced dilation (FID) through S1P. Meanwhile, NCDase inhibition induces the damaging peroxide-dependent FID [33]. In addition, the inhibition of NCDase also leads to mitochondrial dysfunction in diabetic hearts through a lactocylceramide accumulation [87].

#### 4.3.3. Alkaline Ceramidase

A high expression of ALKCDase genes, particularly ACER2, has been observed in cardiac tissue during hypoxia, where it plays a protective role [88]. However, an overexpression of ACER2 has been associated with chronic obstructive pulmonary disease (COPD) [89]. ACER2 inhibition contributes to a reduction in the circulating S1P and its analogue, dhS1P, as well as their precursors, SPH and dhSPH, in hematopoietic cells and reduces the concentration of dhS1P in the lungs [85,88].

These data indicate that ACDase and ALKCDase are increased in CF and COPD, respectively, whereas NCDase is decreased in coronary artery disease.

#### *4.4. Role of Ceramidases in Metabolic Disorders*

#### 4.4.1. Acid Ceramidase

Multiple factors are involved in the onset and progression of metabolic disease, including the activities of ceramidases. Genetic variations of *ASAH1* have been associated with exercise tolerance and skeletal/cardiac muscle adaptation to exercise, which can condition adherence to physical activity regimens necessary for a healthy lifestyle, thereby increasing the individual risk of metabolic diseases [90]. After onset, metabolic disorders affect the physiology of the cardiovascular system, kidneys, and liver. Hyperglycemia inhibits the Unc51-Like Autophagy-Activating Kinase 1 (ULK1) phosphorylation in aortic endothelial cells, which leads to a dysregulation of autophagy and atherogenesis. However, ACDase activity can increase the phosphorylation of ULK1 and restore its function even in nutrient-rich conditions, thus preventing atherogenesis [91]. In addition, obesity-induced kidney damage is caused by hyperglycemic conditions that stimulate the NLR Family Pyrin Domain-Containing 3 (NLRP3) inflammasomes to release IL-1β in podocytes, but the treatment of podocytes with ACDase decreases the NLPR3-induced cytokine release through extracellular vesicles [92]. ACDase reduces the activity of Pannexin-1 (Panx1), a transmembrane channel glycoprotein that activates NLRP3 through S1P accumulation [93]. Animal models of ACDase deficiency show significant damage to the liver and change to lipid profiles and metabolism, including hepatomegaly with higher serum levels of aspartate, aminotransferase, alanine aminotransferase, and alkaline phosphatase and decreased levels of free fatty acids, triglycerides, and cholesterol [25]. The inducible liver-specific overexpression of ACDase in the Alb-AC transgenic mice, results in significantly reduced C16:0 ceramide in the liver and improved total body glucose homeostasis and insulin sensitivity under a high-fat diet [94]. However, aberrant ACDase overexpression in very low-density lipoprotein (VLDL) deficiency may result in non-alcoholic fatty liver disease, which can be normalized by supplementation with Vitamin E [95]. Adipocyte-specific ACDase overexpression improves glucose metabolism by white adipose tissue, reverses insulin resistance, reduces lipid accumulation in the liver, and reduces adipose inflammation and fibrosis [94]. This could be due to

an ACDase-mediated activation of the adiponectin receptor that triggers an AMP-dependent kinase pathway, which subsequently inhibits adipogenesis and induces fatty acid oxidation [96].

#### 4.4.2. Neutral Ceramidase

Palmitate is a precursor of palmitoyl-CoA, a thioester used in the de novo biosynthesis of ceramide that is associated with pancreaticβ-cell apoptosis and insulin resistance. Palmitate inhibits NCDase gene expression and activity in pancreatic β cells, which, in turn, exacerbates apoptosis through ceramide accumulation [97]. Pancreatic β cells secrete NCDase via exosomes that reduce palmitate-induced ROS and act as a protective mechanism against free fatty acid-induced apoptosis [98,99]. An overexpression of NCDase inhibits palmitate-induced apoptosis and may be a therapeutic target for type 2 diabetes mellitus and lipotoxicity [97]. Furthermore, Asah2 is one of the four genes related to sphingolipid metabolism that are deregulated in animal models of type 3 maturity-onset diabetes of the young [100]. This pathology is characterized by increased ceramide and SPH levels as well as hypochromic microcytic anemia, with abnormally-shaped and osmotically fragile red blood cells characterized by an accumulation of SPH [100].

#### 4.4.3. Alkaline Ceramidase

Non-alcoholic fatty liver disease is associated with an increased expression of ACER3, which reduces the accumulation of C18:1-ceramide in the liver [101]. Acer3 deletion reduces inflammation, fibrosis, oxidative stress, and apoptosis of hepatocytes through a palmitic acid-induced increase in C18:1-ceramide [101].

Altogether, we can conclude that the holistic beneficial effects of ACDase in metabolic disease have been demonstrated. ACDase activity controls atherogenesis, kidney damage, and liver damage, while improving glucose and lipid metabolism. In addition, NCDase also appears to improve metabolic conditions via a protective effect on pancreatic β cells, while ALKCDase3 mediates liver damage.

#### **5. Role of Ceramidase Activity in Infectious Diseases**

#### *5.1. Role of Ceramidase Activity in Bacterial Infection*

#### 5.1.1. Acid Ceramidase

Ceramidases have been identified as contributors to bacterial infection and mediators of the immune response and inflammation. The α-toxin released by *Staphylococcus aureus* inhibits ACDase gene expression, causing decreased levels of SPH that contribute to bacterial infection susceptibility [102]. Moreover, this mechanism further increases the risk of infection by *S. aureus* and *Pseudomonas aeruginosa* in already ACDase- and SPH-deficient CF patients [86,102]. *Porphyromonas gingivalis*, an etiological factor for periodontitis, downregulates ACDase in periodontal tissues, thereby increasing its own apoptotic potential and inhibiting the host's inflammatory response [103]. The inhibition of ACDase by bacteria increases host cell apoptosis and reduces the production of inflammatory cytokines, such as TNF-α, IL-1β, IL-6, and IL-17A, which delay the immune response [67,103]. Conversely, ACDase overexpression upregulates the inflammatory cytokines involved in the recruitment of neutrophils and macrophages, as demonstrated in ulcerative colitis, where ACDase mediates the associated histopathological characteristics of the disease [67].

#### 5.1.2. Neutral Ceramidase

Ceramide accumulation is increased after burn injuries and may be associated with bacterial infections that frequently lead to death. NCDase treatment protects against *Pseudomonas aeruginosa* infection after burn injuries by controlling ceramide accumulation and inducing the accumulation of SPH, which directly kills bacteria [104].

#### 5.1.3. Alkaline Ceramidase

Bacterial lipopolysaccharides may also downregulate the expression and activity of *Acer3* and increase C18:1 ceramide accumulation in mice [105]. A loss of *Acer3* expression leads to the production of pro-inflammatory IL-1β, IL-6, IL-23α, and TNF-α cytokines from peritoneal macrophages, bone mononuclear cells, and colonic epithelial cells isolated from Acer3-/- mice [105]. Overall, the bacterial species inhibit ceramidase activity to reduce the concentration of SPH in the host cells, which results in a reduced immune response.

#### *5.2. Role of Ceramidase Activity in Viral Infection*

Viruses have the potential to spread among individuals, resulting in epidemics that cause loss of human life and heavy burdens to healthcare systems [106]. The influenza, Ebola, and Zika epidemics are recent examples of the effects of broad viral infection and of the mechanisms by which viruses can be studied and controlled [106–108]. A recent mutation of the coronavirus, named SARS-CoV-2, has caused a pandemic of unprecedented magnitude. This virus has a lower mortality rate but is exponentially more contagious than the closely related SARS-CoV and MERS-CoV [109]. However, our knowledge of the potential role of host ceramidases in viral pathology remains elusive. It was reported that the overall inhibition of ceramidase activity in host peripheral blood lymphocytes using Ceranib 1 and Ceranib 2 significantly reduces the replication of the rhinovirus and measles virus, respectively [110,111]. Furthermore, the inhibition of ACDase activity in macrophages significantly increases the propagation of herpes simplex virus-1, which, in turn, elevates the mortality rate in Asah1−/<sup>−</sup> mice [112].

Collectively, these published observations indicate that ceramidases may have an important antiviral effector role that should further studied.

#### **6. Role of Ceramidases in Tissue Regeneration and Healing**

Ceramidases are expressed in epithelial cells and fibroblasts and may be involved in their response through S1P [12,103,113]. However, only a limited number of studies have demonstrated the effects of these enzymes in tissue regeneration and healing.

#### *6.1. Acid Ceramidase*

ACDase activity contributes to physiological processes involving collagen turnover. ASAH1 is associated with familial keloid healing and is overexpressed in keloid scar tissue and hypertrophic scars caused by excessive collagen deposition during epidermal healing [114]. In the liver, hepatic stellate cells (HSC) are activated during normal wound healing but can, after multiple activations, cause hepatic fibrosis. However, the inhibition of ACDase by tricyclic antidepressants leads to ceramide accumulation, which inactivates HSCs and prevents hepatic fibrosis [115]. In vivo studies have also demonstrated a positive effect of ACDase in chondrocyte differentiation. In cartilage replacement therapy, pre-treatment with ACDase induces chondrocyte proliferation, the production of glycosaminoglycan, the expression of *COL2*, the adhesion of chondrocytes to a scaffold, a reduced resorption after implantation, and an improved differentiation to cartilage [116]. Furthermore, a variation of FRBRL characterized by peripheral osteolysis not associated with MMP-2 and MMP-14 was found, suggesting the involvement of ASAH1 in bone remodeling [117].

#### *6.2. Neutral Ceramidase*

Various studies have focused on the use of exosomes for tissue repair and regeneration [118,119]. The results of a recent study indicated that hepatocyte exosomes show significant NCDase activity and promote hepatocyte proliferation in vitro and liver regeneration in vivo [118]. This suggests a role of NCDase in tissue regeneration. The ceramidase has also been identified as an antagonist of cell necrosis

caused by 2DG/AA-dependent ceramide accumulation and mitochondrial damage [34]. Furthermore, NCDase increases autophagy and protects cells from ER stress-mediated cell death [34].

#### *6.3. Alkaline Ceramidase*

ACER1 inhibition leads to abnormal hair, alopecia, hyperproliferation, inflammation, an abnormal differentiation of the epidermis, sebaceous gland abnormalities, and infundibulum expansion, as well as an increased trans-epidermal water loss and hypermetabolism with an associated reduction in fat content during aging [12]. Its inhibition gradually depletes the number of hair follicle stems and causes alopecia through decreased hair follicle activity [14]. The specific mechanisms through which these effects of ALKCDase occur are still not detailed in the literature. However, its expression has been associated with keratinocyte growth arrest and differentiation [13]. Altogether, these data suggest that ACDase is involved in collagen matrix metabolism, whereas NCDase and ALKCDase appear to affect tissue regeneration and healing through their anti-apoptotic effects.

#### **7. Conclusions**

Ceramidases (acid, neutral, alkaline) are key enzymes that maintain the intracellular homeostasis of ceramide/SPH and are critical regulators of signals that tilt the balance between cell survival and death. Various studies have demonstrated the involvement and potential therapeutic role of these enzymes in a diverse set of common human diseases, including bacterial-induced infectious diseases, neurodegenerative diseases, cancer, diabetes, and others (Figure 4). Therefore, the clinical applicability of studies examining the versatility of the effects of ceramidases in health and disease deserves further examination.

**Figure 4.** Potential therapeutic targeting of acid (ACDase), neutral (NCDase), and alkaline (ALKCDase) ceramidases. \*Unspecified class of ALKCDase [80].

**Funding:** This research was funded by NIH, grant number AG064003, DE027153, and DE028699 (A.M), and NIH Research supplements to Promote Diversity in Health-Related Research (C.D., J.A.).

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Review* **Sphingosine-1-Phosphate Metabolism in the Regulation of Obesity**/**Type 2 Diabetes**

#### **Jeanne Guitton 1, Cécile L. Bandet 2,3, Mohamed L. Mariko 1, Sophie Tan-Chen 2,3, Olivier Bourron 2,3,4, Yacir Benomar 1, Eric Hajduch 2,3 and Hervé Le Stun**ff **1,\***


Received: 9 June 2020; Accepted: 7 July 2020; Published: 13 July 2020

**Abstract:** Obesity is a pathophysiological condition where excess free fatty acids (FFA) target and promote the dysfunctioning of insulin sensitive tissues and of pancreatic β cells. This leads to the dysregulation of glucose homeostasis, which culminates in the onset of type 2 diabetes (T2D). FFA, which accumulate in these tissues, are metabolized as lipid derivatives such as ceramide, and the ectopic accumulation of the latter has been shown to lead to lipotoxicity. Ceramide is an active lipid that inhibits the insulin signaling pathway as well as inducing pancreatic β cell death. In mammals, ceramide is a key lipid intermediate for sphingolipid metabolism as is sphingosine-1-phosphate (S1P). S1P levels have also been associated with the development of obesity and T2D. In this review, the current knowledge on S1P metabolism in regulating insulin signaling in pancreatic β cell fate and in the regulation of feeding by the hypothalamus in the context of obesity and T2D is summarized. It demonstrates that S1P can display opposite effects on insulin sensitive tissues and pancreatic β cells, which depends on its origin or its degradation pathway.

**Keywords:** Sphingosine-1-phosphate; obesity; type 2 diabetes; insulin resistance; pancreatic β cell fate; hypothalamus

#### **1. Introduction**

Obesity is a major public health problem, which results in over nutrition that leads to a net-positive energy balance characterized by the storage of excess fat in the subcutaneous and visceral adipose tissues, as well as in ectopic tissues, including skeletal muscles, liver, and pancreatic β cells [1]. In physiological conditions, ingested lipids are usually used as an energy source by most organisms and can be substituted by carbohydrates for ATP production, based on acute changes in nutrient availability and energy requirements [2]. However, in pathophysiological conditions, adipose tissue lipid metabolism becomes dysfunctional, which leads to increased delivery of fatty acids to other peripheral tissues [3]. Increased free fatty acids (FFA) produced from adipose tissue as well as secretion of hormones, cytokines, and pro-inflammatory markers, which are directly linked to obesity, induce reduced glucose uptake in muscle cells and increased hepatic glucose production. These metabolic dysfunctions lead to a glucose overflow in the circulation, which culminates in glucose intolerance and the installation of type 2 diabetes (T2D) [4].

T2D is a serious metabolic condition due to the insufficient secretion of insulin by pancreatic β cells, and due to an inefficient response from the body to secreted insulin. Diabetes is one of the fastest growing global health emergencies of the 21st century. In 2019, the world prevalence of diabetes was estimated as 463 million people, and this number is projected to reach 578 million by 2030, and 700 million by 2045. T2D is also the most common form of diabetes and accounts for 90% of the disease worldwide [5]. T2D is most commonly observed in older adults but is increasingly seen in children and younger adults due to the rise of obesity, physical inactivity, and inappropriate diet.

High levels of circulating FFA are known to induce not only insulin resistance, but also defects in the insulin secretory capacity of β cells, as well as in insulin gene expression [6,7]. The nature of FFA, that is, its degree of saturation and carbon chain length, is one of the critical factors involved in the induction of lipotoxicity, such as inhibition of insulin secretion, β cell apoptosis, and insulin resistance [8,9]. Non-adipose tissue accumulated FFAs are metabolized into lipid derivatives such as ceramides, which, in turn, lead to lipotoxicity in these tissues [10]. In mammalian cells, ceramides are key lipids of sphingolipid metabolism and are widely distributed in cell membranes where they play a crucial structural role. It also has important functions in intracellular signaling, regulation of growth, proliferation, cell migration, apoptosis, and differentiation [11–13]. Ceramides consist of a sphingoid long chain base to which a fatty acid is attached via an amide bond. In the context of fatty acid overload, ceramide is mainly produced *de novo* in the endoplasmic reticulum (ER), through different enzymatic reactions [14,15]. It is now clearly established that *de novo* synthetized ceramides are among the most active lipid second messengers, which inhibits the function of some key proteins of the insulin signaling pathway [16,17] and stimulates pancreatic β cell death [18]. Apart from its structural and signaling functions, ceramide is a central lipid intermediate in sphingolipid metabolism. It is a precursor for other bioactive sphingolipids, from complex glycosphingolipids or sphingomyelin to more "simple" lipids such as ceramide-1-phosphate, sphingosine, and sphingosine-1-phosphate (S1P) [12].

The most important site of S1P production is the plasma membrane where sphingomyelin is metabolized into ceramide by sphingomyelinase, then S1P is produced by cooperating two enzyme families, namely ceramidases, and sphingosine kinases (SphK) (Figure 1) [19]. Contrary to most sphingolipids, S1P does not possess a structural function, but is a potent signal mediator that modulates multiple cellular functions important for health and diseases [14]. The multimodal actions of S1P can be explained by the fact that the sphingolipid, on the one hand, directly modulates intracellular functions, and, on the other hand, acts as a ligand of G protein-coupled receptors (GPCR) after secretion into the extracellular environment, transported by ApoM-containing high density lipoproteins (HDL) or albumin, to exert either autocrine and/or paracrine functions [19].

In many cellular and animal models, both ceramide and S1P display opposite effects. This is well documented in cancer cells where ceramide stimulates apoptosis, whereas S1P promotes cell survival [11]. While the role of ceramide in the development of muscle insulin resistance is now well established [16], the relationship of S1P with insulin resistance and T2D still remains controversial in some tissues. Elevation of tissue and plasma S1P levels has been recognized as a critical feature of both human and rodent obesity [20], which suggests that S1P metabolism could be involved in the onset of T2D, or that its regulation is an adaptive process in the presence of high levels of circulating lipids. Thus, this review describes the role played by S1P metabolism in the development of obesity/T2D by analysing the enzymes regulating both its tissue and circulating levels in insulin resistance of peripheral tissues and pancreatic β cell fate.

**Figure 1.** Sphingosine-1-phosphate metabolism in mammals. Sphingolipid *de novo* synthesis is initiated in the endoplasmic reticulum (ER), starting by the condensation of serine and palmitoyl-coA followed by a cascade of enzymatic reactions to produce ceramide. In the ER, ceramide is deacylated by neutral CDase into sphingosine. Sphingosine is phosphorylated to produce S1P by SphK1/2. Produced S1P can be either dephosphorylated back to sphingosine by ER resident SPPs, or irreversibly transformed into hexadecenal and phosphoethanolamine by S1P lyase. Ceramide is transported to the Golgi apparatus to be transformed into SM, which will reach the plasma membrane. In the plasma membrane, SM can be transformed into ceramide through the action of SMases. Ceramide will then be deacylated by acidic CDase to give sphingosine that will be phosphorylated into S1P by SphK1. Produced S1P can be dephosphorylated by ecto-LPPs. S1P can also be secreted through ABC, SPNS2, and MFSD2B transporters in extracellular space to activate S1P receptors. Extracellular S1P can also be transported by either albumin or ApoM/HDL. The latter can activate S1P receptors. SM can be endocytosed to be recycled into ceramide and sphingosine inside lysosomes. SphK2 can catalyze S1P production in the mitochondria and the nucleus. ABC: ATP-binding cassette. CDase: ceramidase. ER: endoplasmic reticulum. HDL: high density lipoproteins. MFSD2B: Major Facilitator Superfamily Domain Containing 2B. S1P: sphingosine-1-phosphate. SM: sphingomyelin. SMase: sphingomyelinase. SphK: sphingosine kinase. S1P1-5: S1P receptor 1 to 5. SPNS2: Spinster homolog 2. SPP: Sphingosine-1-phosphate phosphohydrolase.

#### **2. S1P Metabolism in Mammals**

#### *2.1. S1P Synthesis*

S1P is produced by deacylation of ceramide by ceramidases to give sphingosine. Subsequently, sphingosine kinases (SphK) are responsible for the phosphorylation of sphingosine, which results in the formation of S1P (Figure 1). As to the anabolic pathway of S1P, two isoforms of sphingosine kinases (SphK) have been discovered, called SphK1 and SphK2. Both are widely expressed [21]. Compared to SphK1, SphK2 possesses 240 additional amino acids in its N-terminal region corresponding to a nuclear export sequence [22]. Although they have similar sequences, these enzymes differ in their intracellular localization, regulation, level of tissue expression, and, therefore, in their functions [23], especially in sphingolipid metabolism and, thus, the level of ceramide [24].

While SphK1 resides in the cytosol, SphK2 is localized in the nucleus, the inner mitochondrial membrane, and the endoplasmic reticulum (ER) (Figure 1). Under basal conditions, SphK1 is mostly present in the cytoplasm. SphK1 catalytic activity increases from 1.5 to 4-fold as it translocates to

the plasma membrane upon stimulation. Both translocation and activity are regulated not only by the phosphorylation of SphK1 Ser225 residue by extracellular signal-regulated kinases (ERK1/2) [25], but also by anionic lipids (phosphatidylserine and phosphatidic acid) and Ca2+/calmodulin [22].

SphK2 can also be phosphorylated by ERK1/2, but the exact phosphorylation site remains unclear, as Ser351 and/or Thr578 residues may be involved [25]. As SphK2 is localized in the nucleus, it can directly interact and form a complex with H3 histone and histone deacetylases 1 and 2 (HDAC1/2) in the promotor of transcriptional regulator c-fos and dependent kinase inhibitor p21 genes, where it enhances local histone H3 acetylation and transcription [26]. Synthetized S1P by SphK2 binds to and inhibits both HDAC1 and HDAC2, which suggests that nucleus-generated S1P via SphK2 influences the dynamic balance of histone acetylation and, thus, the epigenetic modulation of specific target genes [27]. In addition, when produced in the mitochondria by SphK2, S1P regulates prohibitin 2 (PHB2) function, which is a highly conserved protein that regulates mitochondrial homeostasis [28].

According to Maceyka et al., SphK1 and SphK2 display opposite functions in sphingolipid metabolism in the regulation of ceramide biosynthesis. Indeed, in HEK293 cells, specific down-regulation of SphK2 reduced conversion of sphingosine into ceramide in the recycling pathway and, conversely, down-regulation of SphK1 increased it [24]. This difference could be linked to a potent dialogue between SphK2 and the S1P phosphatase 1 (SPP1) that favors the conversion of S1P into ceramide [29] (see below Section 2.2).

#### *2.2. S1P Recycling and Degradation*

S1P can be quickly and irreversibly degraded by the endoplasmic reticulum resident enzyme S1P lyase (SPL), which cleaves the C2-C3 bond of S1P to generate two products: hexadecenal (palmitaldehyde) and phosphoethanolamine [30] (Figure 1). Both products can then be transferred as glycerol substrates and phospholipid substrates in the glycerophospholipid pathway [31]. Phosphoethanolamine will be used for the synthesis of phosphatidylethanolamine and hexadecenal will be used for reloading the palmitoyl-CoA pool [31].

Alternatively, S1P can also be reversibly dephosphorylated by several phosphohydrolases to regenerate sphingosine. The first lipid phosphohydrolases involved are lipid phosphate phosphohydrolases (LPPs) (Figure 1). They belong to the superfamily of lipid phosphatases that includes three isoforms characterized in mammals: LPP1, LPP2, and LPP3. LPPs are membrane-associated, magnesium-independent and N-ethylmaleimide-insensitive enzymes [29]. Their active sites are located on the outer surface of plasma membranes or at the lumenal surface of internal membranes (Golgi and endosomes) [32]. LPP2 resides intracellularly, whereas LPP1 and LPP3 are mainly localized at the plasma membrane and function as ecto-enzymes, while degrading lipid phosphate substrates such as S1P as well as lysophosphatidic acid in the extracellular space [33].

S1P can also be dephosphorylated by two specific S1P phosphohydrolases called SPP1 and SPP2 (Figure 1). These two mammalian isoforms are differentially expressed-sphingoid base-specific phosphatases localized in the ER. SPP1 regulates the salvage of sphingosine for the synthesis of ceramide in the ER (rescue pathway) [33], and it has been shown that SPP1 overexpression induces ceramide accumulation in the ER, which suggests that dephosphorylation of S1P is a limiting step for the recycling pathway [33]. A regulatory role in the recycling pathway for SPP2 has not yet been demonstrated, but its expression was increased during the inflammatory response [33]. In addition, it was reported that both SPP1 and SPP2 were also involved in ER stress-induced-autophagy [34] and proliferation [35].

#### *2.3. S1P Transport*

Contrary to most sphingolipids, S1P does not possess any structural function, but is a potent signal mediator that affects multiple cellular functions important for health and diseases. The multitude of different S1P-mediated actions is linked to its capacity to be secreted by various cells and tissues. To exert its extracellular functions, intracellularly generated S1P is transported across the plasma

membrane. Since S1P is too hydrophilic to simply diffuse through the membrane, it is exported by specific ATP-binding cassette (ABC) transporters or the spinster homolog 2 (SPNS2) transporter, which is a member of non-ATP-dependent organic ion transporter family [36]. In the erythrocyte, S1P was recently shown to be secreted through the protein MFSD2B [37]. Once outside the cell, S1P can either bind to albumin [38], or ApoM [39]. Approximately 35% of plasma S1P is bound to albumin and 65% to ApoM, which is found on a small percentage (~5%) of high density lipoprotein (HDL) particles [40]. S1P has a four-times longer half-life when bound to ApoM/HDL than to albumin, as seen when tested in vivo (15 min) and in vitro (30 min) under albumin binding conditions [41,42]). This suggests that the binding of S1P to HDL prevents its degradation. ApoM/HDL-bound S1P has been proposed as a primary contributor to the vasoprotective properties of HDLs [43], and S1P has also been shown to be a key component in the anti-atherogenic properties of HDL [44]. However, S1P-bound albumin has been suggested to represent a reservoir for free S1P [39].

#### *2.4. S1P Receptors*

As an extracellular second messenger, S1P is a high-affinity ligand (Kd from 2 to 63 nM) of a family of five GPCRs, termed S1P1-5 [45]. Receptor-bound S1P induces a wide range of physiological responses such as proliferation, migration, inhibition of apoptosis, formation of actin stress fibers, stimulation of adherent junctions, and enhanced extracellular matrix assembly [46]. S1P1–3 are ubiquitously expressed throughout tissues, whereas S1P4 is predominantly expressed in the immune system, and S1P5 is expressed in the central nervous system and the spleen [27]. S1P receptor activation on different cell types depends on specific G protein coupling. S1P1 couples exclusively with the inhibitory G protein alpha subunit (Gαi), whereas S1P2 and S1P3 bind to Gαi, Gαq, and Gα13, while S1P4 and S1P5 couple to both Gαi and Gα13 [47]. Following ligand binding and subsequent activation, the α subunit of the heterotrimeric G protein is released and interacts with various downstream effectors (see review [48] for more information).

#### **3. S1P Metabolism and Insulin Action: Muscle, Liver, and Adipose Tissue**

Since the early 2000s, several studies have looked for the potential role of S1P in mediating insulin action in insulin-sensitive tissues such as liver, skeletal muscle, and adipose tissue.

#### *3.1. Liver*

Liver is a major organ for glucose and lipid metabolism, and it has been known for many years that lipid accumulation is linked to the development of insulin resistance and constitutes the first stage of non-alcoholic fatty liver diseases (NAFLD) [49]. Several studies have shown that the SphK1/S1P axis can control the insulin response in the liver. One such study highlighted that hepatic SphK1 expression increased in animals under lipid overload induced by a high-fat diet (HFD) [40]. This increased expression of SphK1 was also found in the liver of human patients displaying NAFLD [40].

Several other studies have also highlighted a positive action of the SphK1/S1P axis on glucose metabolism in hepatocytes. A pioneer study performed in human hepatocytes showed that tumor necrosis factor α (TNFα), which is a cytokine involved in inflammation [50], was unable to activate the NFκB pathway-induced apoptosis, but rather activated the pro-survival SphK1 pathway [51]. The authors also emphasized that TNFα protected hepatocytes from apoptosis by activating SphK1 upstream of the PI3K/Akt pathway [51]. An increase in Akt phosphorylation was observed after 5 min of treatment with exogenous S1P (without insulin), which suggests that a relationship between the SphK/S1P axis and the PI3K/Akt pathway exists in hepatocytes [51]. Similar results were observed by treating primary rat hepatocytes with exogenous S1P [52]. In addition, treatment of a human liver cell line (LO2 cells) with S1P induced an increase in glucose uptake [53]. SphK1 overexpression also induced glucose uptake in the absence of insulin in hepatocellular carcinoma [54]. Conversely, inhibition of SphK1 reduced glucose uptake in the presence or absence of insulin in the same cell line [54]. The authors extended their observation in vivo by injecting an adenoviral vector containing

the human SphK1 cDNA in diabetic KK/Ay mice [54]. Under these conditions, transfected diabetic KK/Ay mice displayed a decrease in basal glycemia and a better glucose tolerance compared to control animals [54]. In parallel, they observed a decrease in total cholesterol, triglycerides, and low density lipoproteins as well as an increase in circulating HDL in SphK1-transfected animals compared to control animals [54].

All of these parameters demonstrated that SphK1 overexpression in diabetic animals improved glucose homeostasis at the systemic level. The authors also assessed the hepatic insulin response in these animals, and they showed that both Akt and GSK3 phosphorylation levels were increased in animals overexpressing SphK1 when compared to control animals (Figure 2) [54].

**Figure 2.** Role of S1P metabolism on insulin in peripheral tissues in response to palmitate. In hepatocytes, palmitate increases intracellular S1P content through SphK1/2 activities. According to studies, produced S1P seems to exert a direct positive action on insulin signaling, or a negative action by stimulating its S1P2 receptors. In muscle cells, palmitate increased intracellular S1P through SphK1 activity, which favors Akt activation, glucose uptake, and glycogen synthesis in response to insulin. In adipocytes, palmitate increases intracellular S1P to inhibit Akt activation in response to insulin. Produced S1P also favors expression of pro-inflammatory cytokines that will contribute to inhibit Akt activity. CDase: ceramidase. GLUT4: glucose transporter 4. IRS: insulin receptor substrate. PI3K: phosphatidylinositol-3-kinase. SphK: sphingosine kinase. S1P2: S1P receptor 2.

These results were confirmed in another study showing that increased mouse hepatic S1P following the overexpression of acid sphingomyelinase (ASM) favored hepatic Akt phosphorylation as well as improved glucose tolerance [55]. When SphK1 expression was reduced in these animals, increased Akt phosphorylation was no longer observed (Figure 2) [55]. In the same study, these observations were confirmed in vitro by treating isolated primary hepatocytes with exogenous S1P, which induced an increase in Akt phosphorylation in the absence of insulin [55].

Contrary to SphK1 expression, incubation of primary mouse hepatocytes with palmitate did not induce any increase in SphK2 expression [56]. Nonetheless, a hepatic role of SphK2-produced S1P in regulating glucose metabolism was investigated by Lee et al. [56]. Endoplasmic reticulum (ER) stress is known to participate in the development of insulin resistance in liver, mainly by promoting the accumulation of lipids in the liver, by directly blocking insulin signaling, and by modifying the expression of key enzymes of gluconeogenesis or lipolysis [49]. Lee et al. found that ER stress transcriptionally up-regulated SphK2 in liver [56]. Overexpression of SphK2 in the AML12

hepatocyte cell line induced an increase in S1P concentration, which was associated with increased Akt phosphorylation in the absence of insulin [57]. In addition, SphK2 overexpression induced a decrease of some sphingolipid species (C16-ceramide, C18-ceramide, C18:1-ceramide, C16-sphingomyelin, C18-sphingomyelin) as well as a decrease in cholesterol and hepatic triglyceride concentration [57]. As observed previously in SphK1-overexpressing animals, hepatic SphK2 overexpression induced an improvement in insulin sensitivity, an increase in hepatic Akt phosphorylation, and, therefore, an improvement in glucose tolerance of these animals when fed an HFD (Figure 2) [57].

Although these studies demonstrate that the hepatic SphKs/S1P axis positively regulates liver insulin response and carbohydrate metabolism under lipotoxic conditions, some other studies showed the opposite and gave a deleterious role to this axis of hepatic insulin signaling. One study reported that S1P inhibited insulin signaling in the liver both in vitro and in vivo [58]. As already described above, Fayyaz et al. showed that, after palmitate treatment, concentrations of intra- and extracellular S1P were increased in primary rat hepatocytes [58]. However, they also observed generated S1P counteracted insulin signaling [58]. The negative role of S1P on insulin signaling in rat or human hepatocytes with exogenous S1P was counteracted in the presence of JTE-013, which is an S1P2 antagonist. This suggests that S1P inhibited the insulin signal through the activation of S1P2 receptor [58]. These observations were extended in vivo Diabetic New Zealand obese (NZO) mice were treated with JTE-013 for seven days before being sacrificed. Both an increase in liver Akt phosphorylation and a decrease in basal glycaemia were observed [58]. Overall, this study demonstrates that palmitate-produced S1P stimulates S1P2 to impair hepatocyte insulin signaling (Figure 2).

In addition, other studies have also highlighted a relationship between liver S1P levels and hepatic lipid accumulation. Mouse hepatic overexpression of ASM has been shown to increase hepatic triglyceride content, which was blunted by SphK1 deletion [55]. This suggests a potent role of SphK1 in steatosis. SphK1 knock-out (KO) mice fed an HFD for 24 weeks displayed an increase in circulating triglycerides compared to wild-type (WT) animals fed the same diet [59]. By contrast, mice displaying a liver-specific overexpression of SphK1 via the use of an adeno-associated-viral (AAV) 8, whose tropism is specific of the liver [60], exhibited reduced hepatic triglyceride levels (steatosis) without affecting glucose metabolism on a low-fat diet [60]. However, no impact of increasing SphK activity on hepatic lipid content or glucose metabolism was observed in HFD fed mice [60]. The discrepancies between these studies could arise from animal models, which use enzyme overexpression (i.e., ASM and SphK1). Lastly, a study showed that SphK1 expression increased in hepatic steatosis and that SphK1 KO mice were protected against hepatic steatosis induced by HFD [56], which supports the idea that endogenous S1P/SphK1 axis could be a major promoter of lipid accumulation in liver (steatosis) [61]. In contrast, SphK2 KO mice fed with HFD showed an increase in hepatic lipid accumulation, which supports the idea that this isoform protected mice from steatosis [62].

Overall, the effect of the SphK/S1P axis on liver glucose metabolism remains not completely solved. Most of the genetic approaches used, to either overexpress or invalidate SphK1 (and SphK2), showed a positive action of the SphK/S1P axis on hepatic insulin response. However, these studies were carried out at the level of the whole organism and, thus, were not liver-specific. It is, therefore, possible that, in addition to hepatic S1P, circulating S1P coming from other tissues could also affect hepatic homeostasis. It has already been shown that hepatic S1P could be secreted to regulate macrophage chemotaxis [63]. The divergent effect of hepatic S1P could also be related to the specific activation of S1P receptors [58] and will require more exploration as to their role in liver homeostasis during obesity. Moreover, it also remains to determine how S1P signals could move from the beneficial effect through insulin signaling to the dysregulation of lipid homeostasis (steatosis). Only one clinical study has shown SphK1 expression increase in liver biopsies from patients with steatosis compared to healthy lean people, which supports the notion that SphK1/S1P axis could play a role in the onset of these diseases [64]. However, whether the localization of increased SphK1 in the human liver is specific just to hepatocyte, as well as its role, still remain unknown. Therefore, future work and analysis will be required to translate data obtained in cell/mouse to those in humans.

#### *3.2. Muscle*

Muscles constitute 40% of the body weight and are responsible for 40–75% of the glucose uptake in response to insulin in the postprandial period [65]. They are, therefore, major tissues toward the regulation of carbohydrate homeostasis within the body. Compared to liver, few studies have looked for the role of S1P on glucose metabolism in muscle, but, unlike liver, it seems that they all demonstrate a positive action of this lipid.

Saturated fatty acid (palmitate) induced an increase in SphK1 expression as well as an increase in S1P concentrations in a muscle cells line (C2C12 myotubes) [66], and in mouse primary myotubes [67]. It is important to note that no increase in SphK1 expression was observed in response to unsaturated fatty acids such as oleate [66]. These data were confirmed in vivo where a 2.5-fold increase in SphK1 expression was observed in skeletal muscles of mice fed an HFD compared to control mice (Figure 2) [66]. The addition of exogenous S1P on C2C12 myotubes increased basal Akt phosphorylation, which led to a concomitant increase in glucose uptake [68].

In vivo studies also reported a positive role of SphK1/S1P on insulin signaling in muscle. SphK1-overexpressing mice displayed increased SphK activity in skeletal muscle, and when fed a HFD, skeletal muscle and whole-body insulin sensitivity were improved in these mice compared with control mice fed the same diet [69]. In addition, animals overexpressing SphK1 fed an HFD for six weeks displayed better muscle Akt phosphorylation and were more glucose-tolerant and more sensitive to insulin than wild type animals (Figure 2). However, although skeletal muscles show an increase in SphK1 overexpression, it cannot be excluded that other untested tissues could also overexpress SphK1 and, thus, participate with the observed phenotype.

To complicate the picture, Bruce et al. showed that SphK1 overexpression induced a decrease in muscle ceramide concentration [69]. Considering the importance of this sphingolipid species in the development of insulin resistance [66,67], it remains difficult to ascertain if the observed phenotypes were linked to a decrease in ceramide content or rather from an independent action of S1P. Likewise, both Bruce et al. [70] and Kendall et al. [71] showed that administration of FTY720, which is an S1P analogue that downregulates all S1PR expressions except for S1P2 [72], to animals fed an HFD induced a better muscle insulin signaling as well as a better glucose tolerance compared to animals receiving vehicle only. However, FTY720, which has also been shown as a potent inhibitor of CerS [73], inhibited ceramide production in mice under HFD [70]. These data suggest that the insulin sensitizer effect of FTY720 was associated with a decrease of ceramide levels in muscle rather than an antagonist action on S1P receptors.

Altogether, even if all studies reported a positive role of the SphK1/S1P axis on muscle insulin signaling, and, consequently, on the systemic glucose metabolism, no specific muscle approach was performed. Thus, this possibly hid some cross-talk mediated by S1P between muscles and other S1P producing tissues such as liver, adipose tissue, or even immune cells. In addition, no study, so far, has investigated the role of SphK2-produced S1P in this tissue, nor shown the opposite roles of SphK1 and SphK2 [24]. However, it would be interesting to study the role of the latter on insulin signaling in muscle. It would also be important to explore the role of S1P catabolism and S1P signaling through its receptors in muscle homeostasis. To date, clinical data demonstrating the role of S1P metabolism in regulating muscle insulin resistance in man are lacking and will, therefore, require extensive study.

#### *3.3. Adipose Tissue*

Adipose tissue (AT), in addition to its storage functions, is an endocrine tissue that secretes several adipokines and chemokines [74]. AT also participates in the development of insulin resistance when it is in a state of inflammation known as "low-grade" [75]. In addition, when maximum AT storage capacities are reached, excess lipids are then stored in peripheral tissues, which causes insulin resistance or apoptosis in these various tissues [16]. Homeostasis of adipose tissue is, therefore, important for maintaining sensitivity to systemic insulin [76].

Expression of SphK1, but not SphK2, has been reported to be increased in subcutaneous adipose tissue from *ob*/*ob* mice compared to wild type mice [77]. Similar results were observed in epididymal adipose tissue and isolated mature adipocytes from mice fed an HFD compared to animals on a low-fat diet [78]. Similar profiles were also reported in human inflamed subcutaneous AT compared to less inflamed AT [58]. Concentrations of S1P are also increased in subcutaneous AT from obese patients compared to those from lean people [79].

One study reported a positive role of S1P on insulin signaling in AT. Administration of the S1P analogue FTY720improvedinsulin sensitivityin animals fed an HFD [71]. Immune cellinfiltrationis known to play an important role in insulin resistance [80], and it was found that FTY720 decreased lymphocyte and macrophage infiltration in TA of this mice, likely through its lymphopenic properties [81]. This phenomenon contributes to improving insulin sensitivity in mice. However, another study demonstrated the opposite results. It showed that, in HFD-fed mice, SphK1 deficiency increased adipogenic markers such as adiponectin and the anti-inflammatory cytokine IL-10, but reduced adipose tissue macrophage recruitment as well as pro-inflammatory molecules TNFα and IL-6 (Figure 2). These changes were associated with a better insulin response in the AT and improved insulin sensitivity and glucose tolerance (Figure 2) [78]. Obesity was found to increase SphK1 expression in AT macrophages of both M1 and M2 phenotypes [82]. Elevated SphK1 expression in AT macrophages was associated with the reduction of endoplasmic reticulum stress related genes, which suggests that Sphk1 promotes AT macrophage survival [82].

Overall, these few studies indicate that SphK1/S1P axis leans towards a pro-inflammatory and negative action on AT insulin signaling. However, extracellular S1P through S1P receptors may have the opposite effect [71]. Therefore, analysis of the role of other S1P metabolic enzymes in adipose tissue homeostasis will be necessary to confirm this tendency. It will also be important to decipher whether differences in S1P function exist between AT distributions (visceral vs. subcutaneous) known to play a distinct role in obesity. Although SphK1 expression is increased in the adipose tissue from obese patients [78], no clinical study has, so far, described the functional role of S1P in human adipocyte insulin resistance.

#### **4. S1P Metabolism and Pancreatic** β **Cell Fate**

Pancreatic β cells secrete insulin in response to glucose and various hormones to maintain glycaemia and, therefore, regulate glucose homeostasis. However, obesity is associated with deleterious effects of elevated fatty acid levels on pancreatic β cell function and survival. Excessive fatty acids leads to the loss of β cell insulin secretory responsiveness and β cell death by apoptosis, which favors induction of chronic hyperglycemia [18]. Sphingolipids and, in particular, ceramide have been shown to play a central role in pancreatic β-cell apoptosis induced by palmitate [18]. More recently, S1P has also been implicated in mediating β-cell function and viability with a specific role for its metabolizing enzymes.

In 2005, a pioneering study characterized the SphK/S1P axis in rat pancreatic β cells and in INS-1 cells [83]. This study was followed by numerous others that focused on the SphK/S1P axis involvement in β-cell secretory function. Hasan et al. reported for the first time that SphK1 activity was important for insulin synthesis and secretion [84]. The knock-down of SphK1 expression in pancreatic β INS-1 cells resulted in both lowered glucose-stimulated insulin secretion (GSIS) and insulin content associated with decreased insulin gene expression. Conversely, SphK1 overexpression restored both insulin synthesis and secretion [84]. In contrast, pancreatic β MIN6 cells exposed to high glucose concentrations displayed an increase in S1P levels due to SphK2 activity, which is concomitant with higher insulin secretion. In addition, inhibition of S1P production through SphK2 KO in MIN6 cells resulted in the abolition of GSIS [85]. Overall, these data suggest that S1P synthesis through both SphK1 and SphK2 could be positively involved in regulating insulin secretion (Figure 3). However, this conclusion still needs in vivo and in vitro exploration of GSIS in mice KO for either SphK1 or SphK2.

**Figure 3.** Role of S1P metabolism on pancreatic β cell fate. In pancreatic β cells, cytokines, such as IL1β increase SphK1 expression and repress SPL expression. This contributes to the increase of intracellular S1P content and apoptosis. In contrast, extracellular ApoM/HDL-bound S1P y antagonizes apoptosis induced by IL1β. In pancreatic β cells, palmitate increases the expression of both SphK1 and2. SphK1 activation represses palmitate-induced pancreatic β cell apoptosis, whereas SphK2 activation promotes apoptosis. SphK1-produced S1P can be secreted and stimulates S1P2 to promote apoptosis. High glucose levels could activate both SphK1 and 2, which contribute to the secretion of insulin. CDase: ceramidase. SPL: S1P lyase. SphK: sphingosine kinase. S1P2: S1P receptor 2.

The SphK/S1P axis was shown to be stimulated by cytokines in rat pancreatic β cells and INS-1 cells (Figure 3) [83], which suggests a potential role in the pathological response to cytokines observed during low-grade inflammation induced by obesity. Later on, Hahn et al. showed that cytokines decreased SPL expression in pancreatic β cells, whereas overexpression of SPL protected them against cytokine toxicity (Figure 3) [86], which comforts a pathological role of intracellular S1P metabolism of pancreatic β cells in diabetes. In contrast, Laychock et al. showed that exogenous S1P counteracted pancreatic β cell apoptosis induced by cytokines [87], which suggests a divergent role of cellular S1P from circulating S1P (Figure 3). Supporting this notion, Rütti et al. found that HDL, known to be enriched in S1P through its binding to apoM, also counteracted pancreatic β cell apoptosis induced by cytokines (Figure 3) [88].

Although the SphK/S1P axis appears to regulate β-cell induced-apoptosis induced by cytokines, the circulating levels are increased by obesity, whether it is implicated in β-cell apoptosis induced by free fatty acids still remains unknown. Palmitate increased not only ceramide but also S1P levels, through SphK1 up-regulation in pancreatic β INS-1 cells (Figure 3) [89]. Japtok et al. also demonstrated that palmitate increased S1P levels in pancreatic β MIN6 cells, which were released in the extracellular medium (Figure 3) [90]. Apoptosis was abrogated in INS-1 cells over-expressing SphK1 [89]. Similarly, either S1P supplementation or SphK1 overexpression in palmitate-treated INS-1 or MIN6 cells prevented cell death (Figure 3) [59]. Conversely, dominant negative expression of SphK1 in these cell lines enhanced palmitate-induced apoptosis [59]. The protective role of SphK1 was independent of S1P receptors, but was mediated by decreasing formation of pro-apoptotic ceramides induced by palmitate [89]. In addition, endoplasmic reticulum-targeted SphK1 also partially inhibited apoptosis induced by lipotoxicity, which suggests a specific localization for the anti-apoptotic action of S1P [89]. Nevertheless, JTE-013, which is an antagonist of S1P2, partially counteracted pancreatic β-cell apoptosis and the reduced proliferation induced by palmitate [89,90], which suggests that the

S1P produced could determine pancreatic β-cell fate under lipotoxicity by interacting with specific receptors (Figure 3). Overall, these studies reported a survival and protective role of both intracellular S1P and its enzyme SphK1 against palmitate-induced β-cell apoptosis [59,90].

In addition, one study discovered that HFD-fed SphK1 KO mice displayed a reduction in β cell size, number, and mass associated with increased β cell apoptosis compared to WT HFD-fed mice, which all favor the installation of glucose intolerance [59]. These data indicated that in vivo SphK1 deficiency predisposes mice to T2D-onset by promoting pancreatic β cell death under lipotoxic conditions [59]. SphK1 has been shown to interact with SKIP (SPHK1-interacting protein) and that SKIP overexpression in NIH 3T3 fibroblasts reduces SphK1 activity and interferes with its biological functions [91]. In another study, SKIP-deficient mice improved glucose tolerance by increasing insulin and GLP-1 secretion [92], which suggests that SKIP deficiency in mice allow SphK1 to better regulate glucose tolerance. However, it remains to be established whether SKIP is acting only at the level of intestinal L cells or on the pancreatic β cell since it is already known that islet-derived GLP-1 is necessary for glucose-stimulated insulin secretion [93]. Not surprisingly, a consensus on the role of the SphK1/S1P axis in β-cells has not been reached. Although the above studies demonstrated a beneficial role of SphK1 on glucose homeostasis and β cell function, another study showed that SphK1 KO mice were protected from HFD-induced glucose intolerance due to a reduced adipocyte pro-inflammatory response, which suggests a negative role of SphK1/S1P axis on regulating glucose homeostasis [78].

In contrast, a negative role of the SphK2/S1P axis was observed on β-cell fate. SphK2 expression KO reversed palmitate-induced cell death, whereas SphK2 overexpression promoted cell death under lipotoxic conditions in both INS-1 and MIN6 cells (Figure 3) [94]. In fact, lipotoxicity induced the shuttling of SphK2 from the nucleus to the cytoplasm, where it led to mitochondrial apoptosis [94]. SphK2 KO diabetic mice under HFD significantly improved their diabetic phenotypes [94], which suggests that, contrary to SphK1, SphK2 exerts a major role in promoting lipotoxicity-induced apoptosis of β cells [94]. Mice with a deletion of the S1P phosphohydrylolase SPP2 exhibited glucose intolerance due to a defect in the adaptation of pancreatic β cell mass, which supports the idea that the rise of endogenous S1P regulated by SphK2 and SPP2 can promote β cell lipotoxicity [35].

Overall, the opposed functions on β cell survival between both SphKs could be explained by expression differences observed in pathophysiological situations but more likely by differences in produced S1P subcellular localization. Nevertheless, it remains crucial to determine the potent role of S1P receptors in pancreatic β cell fate during obesity. To date, there are no clinical studies available describing a potential role of S1P in human islets in the context of obesity or T2D.

#### **5. S1P and the Hypothalamic Regulation of Body Weight and Energy Homeostasis**

The hypothalamus is a key brain area that plays a crucial role in regulating energy metabolism. It consists of several nuclei including the arcuate nucleus (ARC), ventromedial (VMH), dorsomedial (DMH), lateral (LH), and paraventricular (PVH) hypothalamus, which interact functionally to coordinate adaptive physiological responses controlling feeding behavior and energy expenditure. This process involves the integration of metabolic, endocrine, and neural signals from the periphery and autonomic circuitries that encode information about energy availability and energy reserve in the body [95–98].

Growing evidence suggests that hypothalamic lipid sensing plays a key role in controlling food intake, fat deposition, and energy balance [99,100], and that its dysregulation could lead to the development of obesity and T2D. Recent investigations reported that S1P is involved in the hypothalamic control of energy homeostasis [101]. Precisely, the intracerebroventricular (ICV) administration of S1P decreased food intake and increased energy expenditure [101]. Conversely, selective disruption of S1P1 in the mediobasal hypothalamus (MBH) induced the opposite effects [101]. At the molecular level, S1P exerted its effects by activating S1P1, which is highly expressed in key hypothalamic nuclei, ARC, VMH, and DMH, which controls feeding, particularly in the anorectic pro-opiomelanocortin (POMC) neurons of the ARC. Altogether, these findings identified S1P/S1P1/JAK2/STAT3 as a new regulatory pathway that plays a crucial role in the hypothalamic control of energy homeostasis and body weight gain. A positive correlation between plasma S1P and body fat percentage exists [20,102], as rodent models of obesity also exhibited an increased hypothalamic S1P/S1P1/STAT3 signaling [101,103]. From a therapeutic point of view, the ICV injection of S1P or the S1P1 agonist, SEW2871, induced anorexigenic effects, and prevented the development of obesity and associated metabolic dysfunctions [101].

In the context of an HFD-induced obesity, and, as it has already been observed in the AT (see section on adipose tissue), inflammatory processes also occur in the brain [104]. HFD triggers brain inflammation, notably in the hypothalamus, by activating microglia and astrocytes, which results in reactive gliosis, production of pro-inflammatory cytokines such as IL1β and TNF, and the development of neuronal inflammation [105]. This contributes to the deregulation of hypothalamic control of energy homeostasis, which promotes the development of obesity and associated metabolic disorders [106–108]. Emerging evidence suggests a pivotal role of S1P metabolism and S1P-mediated signaling in the development of neuro-inflammation. It was shown that S1P was able to induce astrocytes activation [109,110] and increase the inflammatory response of activated microglia, which results in reactive gliosis and the upregulation of pro-inflammatory cytokines [111,112]. Additionally, S1P modulated neuro-inflammation by regulating the infiltration of peripheral immune cells in the central nervous system [113,114]. Considering the emerging importance of S1P metabolism in neuro-inflammation, further studies will be required to determine the role of S1P signaling in the early onset of hypothalamic inflammation and gliosis in the context of diet-induced obesity.

The role of hypothalamic S1P in the regulation of obesity and dysregulation of glucose homeostasis is actually an emerging area. Thus, future studies will be important to determine the role of S1P metabolism and signaling at the level of the hypothalamus in the context of obesity and T2D. This will constitute an important step toward identifying new targets for therapeutic intervention in obesity and obesity-related metabolic disorders.

#### **6. Conclusions and Perspectives**

The elevation of S1P levels in tissues and plasma has been associated with obesity, which suggests that S1P metabolism could be negatively or positively linked to this pathology and to the onset of T2D. Many studies performed in the last decade suggest that S1P metabolism plays a positive role in insulin signaling in peripheral tissues, which points to an adaptive role of S1P to counteract the installation of insulin resistance in muscle, adipose tissue, and liver [54,55,57,69]. However, it should be noted that some studies argue for a causative role of S1P metabolism in insulin resistance in the liver and in adipose tissue [58,78]. S1P metabolism has also been linked to pancreatic β-cell fate during obesity or T2D with opposite roles of S1P produced by SphK1 or SphK2 on pancreatic β cell apoptosis [59,94]. Moreover, the specific role of the SphK1/S1P axis in glucose homeostasis will require further attention since studies reveal divergent phenotypes of SphK1 KO mice under HFD [59,78].

These discrepancies in the results may be linked to the fact that cellular S1P levels are fine-tuned by a concerted regulation of S1P synthetizing enzymes (SphK) and S1P degradation enzymes (SPL and SPPs). The other reason could come from the intrinsic nature of S1P, which is both an intracellular mediator and a circulating bioactive lipid. This supports the idea that S1P could act not only intracellularly but also as an endocrine or paracrine signal through its secretion to regulate the insulin response in distant organs as well as in the pancreatic β-cell fate. Plasma apoM/HDL-bound S1P has been shown to regulate brown adipose tissue activity in the context of obesity [115] and also pancreatic β-cell survival induced by cytokines [88].

In conclusion, more work is required to understand the role of the enzymes involved in S1P metabolism/signaling, especially of the catabolizing enzymes SPL and SPPs in the development of obesity and diabetes. It will also be important to determine the role of its transporters and its receptors. Due to the duality of actions of S1P (intracellular and extracellular), the development of tissue-specific disruption of S1P metabolism enzyme genes in mice would also help us understand the divergent

roles of S1P observed in whole KO models used to date. This would be crucial before the modulation of S1P metabolism can be considered as a potential therapeutic target for treating obesity/T2D.

**Author Contributions:** Conceived the idea: H.L.S. Wrote the manuscript: J.G., C.L.B., M.L.M., S.T.-C., O.B., Y.B., E.H. and H.L.S. Figure preparation: S.T.-C. and J.G. Critically reviewed the manuscript and figures: C.L.B., S.T.-C., J.G., E.H. and H.L.S. All authors approved the final manuscript. All authors have read and agreed to the published version of the manuscript.

**Funding:** The authors would like to acknowledge the support by the Centre National de la Recherche Scientifique (CNRS), the Institut National de la Recherche Médicale (INSERM), the Sorbonne Université, and the Université Paris Saclay. H.L.S. is funded by the Société Francophone de Diabétologie (SFD). E.H. is funded by the Fondation de France. J.G. is funded by a scholarship from the French Research Ministry/University Paris Saclay.

**Acknowledgments:** We are grateful to Froogh Darakhshan-Hajduch (Anglais Pour Vous, Melun, France) for professional editing of this review.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Review* **Sphingolipids in Type 1 Diabetes: Focus on Beta-Cells**

#### **Ewa Gurgul-Convey**

Institute of Clinical Biochemistry, Hannover Medical School, Carl-Neuberg-Str.1, 30625 Hannover, Germany; Gurgul-Convey.Ewa@mh-hannover.de

Received: 30 June 2020; Accepted: 3 August 2020; Published: 4 August 2020

**Abstract:** Type 1 diabetes (T1DM) is a chronic autoimmune disease, with a strong genetic background, leading to a gradual loss of pancreatic beta-cells, which secrete insulin and control glucose homeostasis. Patients with T1DM require life-long substitution with insulin and are at high risk for development of severe secondary complications. The incidence of T1DM has been continuously growing in the last decades, indicating an important contribution of environmental factors. Accumulating data indicates that sphingolipids may be crucially involved in T1DM development. The serum lipidome of T1DM patients is characterized by significantly altered sphingolipid composition compared to nondiabetic, healthy probands. Recently, several polymorphisms in the genes encoding the enzymatic machinery for sphingolipid production have been identified in T1DM individuals. Evidence gained from studies in rodent islets and beta-cells exposed to cytokines indicates dysregulation of the sphingolipid biosynthetic pathway and impaired function of several sphingolipids. Moreover, a number of glycosphingolipids have been suggested to act as beta-cell autoantigens. Studies in animal models of autoimmune diabetes, such as the Non Obese Diabetic (NOD) mouse and the LEW.1AR1-iddm (IDDM) rat, indicate a crucial role of sphingolipids in immune cell trafficking, islet infiltration and diabetes development. In this review, the up-to-date status on the findings about sphingolipids in T1DM will be provided, the under-investigated research areas will be identified and perspectives for future studies will be given.

**Keywords:** type 1 diabetes; beta-cells; islets; insulin; cytokines; sphingolipids; S1P; animal models

#### **1. Introduction**

Sphingolipids (SLs) are a diverse family of lipid molecules playing a pivotal role in a number of autoimmune and inflammatory disorders [1–4]. The role of SLs in glucose homeostasis and insulin sensitivity is relatively well described in the context of metabolic-syndrome related type 2 diabetes (T2DM) [4–12]. In contrast, the importance of SLs in the beta-cell demise during autoimmune type 1 diabetes (T1DM) development has been so far less frequently addressed. Interestingly, a number of new investigations suggest that dietary fats and lipid metabolism may be considered as triggers that could induce or sensitize the autoimmunity onset in T1DM [13]. Polymorphisms in several genes encoding proteins involved in the SL pathway were recently linked to T1DM overt [14]. Moreover, profound changes in SL serum profiles upon autoimmunity development were detected in T1DM patients [14–20]. The last years have revealed that the enzymatic machinery and the system of receptors and transporters for bioactive SLs are significantly affected in pancreatic beta-cells by proinflammatory cytokines that are released from immune cells infiltrating islets [21]. SLs may be useful biomarkers for T1DM development [17]. In vitro studies of cytokine toxicity using genetically modified beta-cells, naturally occurring SLs and their analogues suggest that alterations of beta-cell SLs may affect insulin secretory capacity and beta-cell fate during T1DM development.

In this review various aspects of sphingolipid action and effects of the major proinflammatory cytokines on the SL pathway in pancreatic beta-cells will be discussed. Next, the engagement of SLs in the autoimmune reaction against beta-cells during T1DM development will be addressed. The present

status of SL studies in animal models of autoimmune diabetes and an update on findings in T1DM patients will be summarized. Finally, perspectives, which should drive future research in the context of SLs and T1DM, will be presented.

#### **2. Overview of Mechanisms of Beta-Cell Destruction in T1DM**

Type 1 diabetes mellitus (T1DM) is an autoimmune disease with a strong genetic background, affecting millions of people worldwide, mostly in their childhood or early adolescence [22,23]. The incidence of T1DM has been significantly increasing in the last decades, similarly to other autoimmune diseases, indicating an important role of environmental factors [22,24]. During T1DM development pancreatic beta-cells are gradually destroyed due to an autoimmune reaction [22,25–29]. Beta-cells produce and supply our body with insulin, the most important anabolic hormone controlling blood glucose levels. The factors triggering the activation of immune cells, T-cells and macrophages, in T1DM remain unclear. It is speculated that certain food components (such as cow milk proteins or gluten), vitamin D3 deficiency, viral infections (e.g., enterovirus) and most recently saturated fats may trigger this response [13,22,26,30,31] (summarized in Figure 1). T1DM patients require a life-long substitution with insulin and are prone to severe secondary complications, such as cardiovascular dysfunction, nephropathy or retinopathy [22].

Immune cell activation is accompanied by dynamic changes in the gene expression of proteins related to inflammation and secretion of inflammatory mediators [26,27,30,32,33] (Figure 1). The infiltrate consists of a mixture of CD4+ and CD8+ T-cells, B-cells, macrophages and Natural Killer (NK) cells [31]. The activated immune cells infiltrating pancreatic islets convey their beta-cell directed cytotoxic effects via cell–cell interactions and through secretion of proinflammatory mediators. So far the research has focused predominantly on the three cytokines, namely IL-1β, TNFα and IFNγ, and their role in beta-cells is meanwhile relatively well characterized [25,27–31,33,34]. The intracellular effects of proinflammatory cytokines in beta-cells are pleiotropic and involve the activation of several transcription factors (NFκB, AP-1, Nova and others), disturbances of mRNA splicing, changes of the expression, activity and post-translational modifications of proteins [27,30,32,35]. Moreover, in beta-cells isolated from T1DM donors the upregulated mRNA and protein expression of Class I and Class II HLA were detected [36,37]. Several molecular processes have been associated with beta-cell death in T1DM including mitochondrial and endoplasmic reticulum (ER) stress, the generation of reactive oxygen species (ROS) and nitric oxide (NO•), the synthesis of inflammatory mediators, activation of immunoproteasome, no-go and nonsense-mediated RNA decay, dysregulation of calcium homeostasis and altered autophagy as well as the activation of the perforin-granzyme system [21,25,30,32,38–46] (summarized in Figure 1).

Beta-cells are particularly vulnerable to oxidative stress due to an imbalance in the enzymatic capacity responsible for generation and detoxification of ROS [40,42,44,45]. The superoxide dismutases, which dismutate superoxide radical anions to hydrogen peroxide (H2O2), are well expressed in the cytosolic (CuZnSOD) and in the mitochondrial compartment (MnSOD). This is in contrast to the enzymes catalyzing the detoxification of H2O2 to water, namely glutathione peroxidase (Gpx) and particularly catalase (Cat), which are very weakly expressed. Proinflammatory cytokines potentiate this imbalance specifically in the mitochondrial compartment through the induction of MnSOD expression and activity [40,44]. The increased generation rate of H2O2 together with the lack of an appropriate H2O2 detoxification system and a parallel increase of intracellular NO• concentration are the perfect prerequisite for the formation of highly toxic hydroxyl radicals. This leads to intramitochondrial oxidative damage, caspase-9-dependent apoptosis induction, disturbed autophagy and beta-cell death [40,42]. Moreover, cytokines activate the ER stress response, leading to accumulation of unfolded/misfolded proteins, dysregulation of ER calcium homeostasis and induction of the proapoptotic transcription factor CHOP [47]. Additionally, cytokines induce alternative splicing and immunoproteasome activation in beta-cells, which may contribute to neoantigen formation, and aggravate the autoimmune reaction [47]. Several studies showed also that the vulnerability of

beta-cells to proinflammatory cytokines is associated with an imbalance of the enzymatic capacity responsible for the generation of pro- and anti-inflammatory eicosanoids [39,48–50]. Our recent data indicate that MCPIP1 (monocyte chemotactic protein-induced protein 1), a strong regulator of the inflammatory response, which acts as a specific RNase, is strongly induced by proinflammatory cytokines in clonal beta-cells and upon diabetes development in the animal model of human T1DM [43]. The transcriptomic analysis of beta-cells from T1DM individuals suggests a significantly increased expression of the gene encoding MCPIP1 [37] (GEO Bioproject PRJNA497610 HLA Class II analysis of human pancreatic beta-cells). Our recent studies revealed that MCPIP1 regulates the beta-cell response to cytokine toxicity and the fine-tuning of its expression is essential for the beta-cell fate [43]. Interestingly, our preliminary observations indicate that a number of the sphingolipid pathway enzymes might be affected by this specific RNase, a mechanism that could contribute to cytokine effects on beta-cell sphingolipidome. Finally, proinflammatory cytokines dysregulate the beta-cell function by shutting down insulin biosynthesis and by disrupting glucose-stimulated insulin secretion (GSIS) [21,41,51–53].

**Figure 1.** Model of cytokine-mediated beta-cell death in T1DM. In genetically predisposed individuals various environmental factors trigger the autoimmune response aimed at pancreatic beta-cells. Environmental triggers lead to beta-cell stress and release of autoantigens, which are processed and presented by antigen-presenting cells (APC). This leads to T-cell and macrophage (MΦ) activation. Consequently proinflammatory cytokines and radicals (NO•, nitric oxide and O2 •−, superoxide anion radicals) as well as perforin-granzyme mediators are released in the vicinity of beta-cells. Proinflammatory cytokines potentiate autoimmune reaction by stimulation of CD8+ and CD4+ T-cells. Activated immune cells interact with beta-cells via FasL-Fas and also via HLAI/II-TCR systems. The action of proinflammatory cytokines requires the binding and activation of cytokine receptors (R) on beta-cells. This accelerates the multifaceted stress response and induces inflammation in beta-cells. The aggravation of the autoimmunity is achieved by biosynthesis and release of inflammatory mediators from beta-cells. Beta-cells are particularly vulnerable to the stress response and inflammation due to their weak antioxidative and anti-inflammatory defense status. Cytokines induce reactive oxygen species (ROS) and reactive nitrogen species (RNS) formation in beta-cells. Both defects in the immune response and vulnerability of beta-cells participate in the execution of beta-cell demise during T1DM development (more details in text).

#### **3. Biosynthesis of Sphingolipids**

Sphingolipids are composed of a polar head group and two nonpolar tails. The core of SLs is the long-chain aliphatic amino alcohol sphingosine, which is O-linked to a polar head group (e.g., ethanolamine, serine or choline) and N-linked to an acyl group of various fatty acids. SLs are the most structurally diverse lipid family, considering the hundreds of possible head groups, dozens of long-chain bases, and many fatty acids, that can be used as building blocks. The maintenance of the so-called SL rheostat is crucial for the normal function of cells and cell survival, and is compromised by a network of enzymatic reactions depicted in Figure 2. Consequently, this promotes site-specific effects and downstream targets of various SLs.

**Figure 2.** Transcriptomic data from islets and qRT-PCR results from beta-cell lines suggest that beta-cells express all genes regulating the SL pathway. The exact mRNA and protein expression level of various enzymatic components of the SL pathway in human beta-cells still needs to be characterized. Enzymes: ALDH2A3, fatty aldehyde desaturase A3, CD, ceramidase, CERK, ceramide kinase, CPP, ceramide 1-phosphate phosphatase, CerS, ceramide synthase, Des, sphingolipid-delta-4-desaturase, DGAT2, diacylglycerol acyltrasferase-2, GCS, GlcCer synthase, GalCerS, GalCer synthase, GCase, glycosidase, KSR, 3-keto sphingosine reductase, LPP, lysophospholipid phosphatase, SMS, sphingomyelin synthase; SMase, sphingomyelinase, SK, sphingosine kinase, SPP, sphingosine 1-phosphate phosphatase, SPT, serine palmitoyltransferase, SPL, sphingosine 1-phosphate lyase. Biomolecules: Cer, ceramide, C1P, ceramide 1-phosphate, dhCer, dihydroceramide, GlycoSLs, glycosphingolipids, GlyceroSLs, glycerosphingolipids, 3-keto dhSph, 3-keto dihydro sphingosine, Ser, serine, SM, sphingomyelin, Sph, sphingosine, S1P, sphingosine 1-phosphate, PalCoA, palmitoyl-coenzyme A.

The sphingolipid pathway starts with the de novo biosynthesis of 3-ketosphinganine in the ER from L-serine and palmitoyl-CoA in the rate-limiting reaction catalyzed by serine-palmitoyltransferase (SPT; Figure 2) [3,12,54,55]. SPT can use acyl-CoAs in the range of C12 to C18, but the most typically used is the C18-CoA. Besides serine, alanine and glycine may also be used under specific circumstances, resulting in the formation of a variety of atypical sphingoid bases, which represent approximately 15% of SLs in human plasma [56]. De novo synthesis of SLs is upregulated by substrate availability and can

be downregulated by mammalian orosomucoid-like (ORMDL) proteins, which form a complex with SPT and inhibit its activity [57].

In the next step of the SL biosynthesis, 3-ketosphinganine undergoes a reduction to dihydrosphingosine by 3-keto-sphinganine reductase (KSR) that is thereafter N-acylated by the action of one of six ceramide synthases (CerS1-6) to form dihydroceramide in ER [3,54]. Each of CerS types uses specific acyl chains, with saturated or monounsaturated fatty acids (C14-C26). Dihydroceramides are then dehydrogenated to ceramides by dihydroceramide desaturase (DES) [54] (Figure 2). Ceramides are the central part of the SL pathway and can be used in numerous reactions (Figure 2). Cer can be utilized for formation of (i) sphingosine, (ii) sphingomyelin, (iii) complex glycosphingolipids, (iv) ceramide 1-phosphate (C1P) or (v) acylceramide [3,54] (Figure 2).

Ceramide is deacylated by ceramidases (CDs) to form sphingosine [3,54]. This reaction can take place in different subcellular organelles and is catalyzed by distinct subtypes of CDs [3]. Sphingosine is either reutilized for Cer generation by CerS or can be phosphorylated by one of two sphingosine kinases (SK1 and SK2) to form a potent bioactive SL called sphingosine 1-phosphate (S1P; Figure 2). SK1 localizes to plasma membrane and ER, while SK2 was found in the cytosol, ER, mitochondria and nucleus [58–60]. This differential spatial location of two SK isoforms defines the specialized and contradictory effects of each isoforms [58,59]. Cytosolic S1P can be dephosphorylated by ER-localized S1P phosphatases (SPP1 and SPP2), while at the cell surface S1P dephosphorylation is catalyzed by lipid phosphate phosphatases (LPP1-3). These reversible reactions are crucial for SL homeostasis within the cells, and enable refilling of sphingosine into the SL pathway [5].

The final step of SL metabolism is the irreversible degradation of S1P catalyzed by S1P lyase (SPL), an enzyme localized in ER with the catalytic center facing the cytosol (Figure 2) [3,61]. S1P is degraded to hexadecenal and phosphoethanolamine, which are intermediates in the phospholipid biosynthesis pathway. The potentially toxic, when accumulated in high amounts especially within the nucleus [62,63], hexadecenal is under normal conditions effectively metabolized by the fatty acid dehydrogenase ALDH3A2 to hexadecenoate [64]. Hexadecenoate can be further utilized for palmitoyl-CoA generation within the glycerolipid pathway [64]. Additionally to ER, SPL was also found by proteomics approaches to be present in the mitochondrial associated membranes (MAMs), the membrane crossroads of mitochondria and ER, which is involved in cell signaling and metabolite exchange [65]. These observations suggest a potential participation of SPL in unique functions of mitochondria and MAMs [66].

Ceramide can be transported to other subcellular locations by vesicular transport or with help of the ceramide transfer protein CERT [3,54]. In mammalian cells CERT function is regulated by SL levels through the PKD-dependent phosphorylation [67]. In the trans-Golgi Cer can be converted to sphingomyelin by sphingomyelin synthase (Figure 2) (SMS1) [3,54]. Another isoform of SMS, namely SMS2, is localized in plasma membrane [3]. Sphingomyelin can be metabolized back to Cer by sphingomyelinases (SMase), which are found in plasma membrane, the Golgi apparatus and in mitochondria-associated membranes (Figure 2) [3,54]. In the cis-Golgi Cer is converted by glucosylceramide synthase (GCS) to glucosylceramide (GluCer; Figure 2) [68]. Galactosylceramide (GalCer) is produced by GalCer synthase (Figure 2) [68]. The formation of complex glycosphingolipids (glycoSLs) requires the transport of GluCer to trans-Golgi network by four-phosphate adaptor protein 2 (FAPP2) that also regulates trafficking of vesicles from Golgi to plasma membrane [3]. Lactosylceramide (LacCer) is then produced by LacCer synthase transferring galactose from UDP-galactose to GluCer [68]. The more complex glycoSLs all use LacCer as a common backbone. The generation of multiple glycoSL species takes place by the action of specific enzymes catalyzing the stepwise addition of sugar monomers, which branch to form complex chains [68]. Complex glycoSLs are then transported from the Golgi to their target location, which is typically the plasma membrane [68]. Sphingomyelin and glycoSLs have been shown to be transported in a vesicle-dependent manner [3,68]. Both SM and glycoSLs undergo lysosomal degradation by glycosidases and acid SMase, respectively, that are responsible for the removal of head groups to form ceramides (Figure 2) [68].

Alternatively, ceramide can be also phosphorylated to ceramide 1-phosphate (C1P), a bioactive SL, by ceramide kinase (CERK) in the Golgi [3,69] (Figure 2). C1P may be converted back into ceramide by the action of one of LPPs or by C1P phosphatase (CPP) [3,69–71]. The irreversible reaction converting Cer to acylceramide (acylCer) is catalyzed by diacylglycerol acyltransferase-2 (DGAT2) [3] and is used to store Cer in lipid droplets [72] (Figure 2).

#### **4. Overview of Major Functions of Bioactive Sphingolipids**

Sphingolipids are constituents of cell membranes (about 50–60% of the membrane content) and important bioactive molecules acting as second messengers. They form lipid microdomains, important for cell signaling and participating in cell–cell interactions and absorption of extracellular lipids (e.g., fatty acids). SLs were shown to regulate calcium homeostasis, ROS formation, histone acetylation and activation of numerous transcription factors. Pathophysiological conditions result in alterations of the ratio between the structural and bioactive SL content, a phenomenon that has been shown to contribute to the regulation of the inflammatory response. Proinflammatory cytokines such as IL-1β and TNFα as well as several other stimuli (Fas ligand, phorbol esters, heat shock, oxidative stress and various chemotherapeutics) are well described disruptors of the SL homeostasis [3,54].

The best studied bioactive SLs are ceramide, sphingosine, S1P, C1P, as well as sulfatide. While ceramide and sphingosine are traditionally believed to exert proapoptotic signals, the action of S1P and C1P is largely context and tissue-dependent, with pro- and antiapoptotic activities described in various tissues. The intracellular concentrations of ceramides, sphingosine and S1P differ by an order of magnitude, with the lowest content of the least one. Therefore a small change in ceramide content might substantially impact S1P concentration. In the following sections the most important functions of major bioactive SLs in various cell types, which could be of relevance for beta-cell pathophysiology in T1DM, is briefly summarized. For a more detailed review on SLs in different tissues and disorders the reader is asked to refer to several excellent reviews addressing these topics [2,3,69].

#### *4.1. Ceramide*

Ceramide is implicated in a variety of cellular processes such as apoptosis, cell growth arrest, differentiation, cellular senescence, cell migration and adhesion [3,54,73]. Many proinflammatory mediators and oxidant agents have been shown to stimulate Cer generation, upon them T1DM-relevant cytokines (IL-1β, TNFα and IFNγ). The best described role of Cer is the regulation of apoptosis [3]. The main mechanism involved in Cer-induced apoptosis is the disruption of mitochondrial integrity and function [74,75]. The induction of mitochondrial apoptosis by ceramides was elegantly demonstrated by means of mitoCERT (CERT equipped with a MOM anchor), enabling Cer flow from ER to mitochondria. The translocation of ER ceramides to mitochondria resulted in the initiation of apoptosis [74]. Elevated levels of Cer were shown to control MOMP (mitochondrial outer membrane permeabilization) and thereby cytochrome c release, by fostering BAX/BAK oligomerization and formation of ceramide channels. Formation of Cer channels promotes the egress of proteins from the intermembrane space and ROS generation [74,75]. Cer contributes also to the suppression of mitochondrial electron transport chain and decreased ATP formation upon inflammation [76]. Recent studies identified the voltage-dependent anion channels VDAC1 and VDAC2 as mitochondrial ceramide binding proteins [77] and reported the expression of CerS within the mitochondrial compartment [77]. Moreover, the recent identification of a protein p17/PERMIT mediating ER-mitochondria tracking of newly translated CerS1 through MAMs to the mitochondrial outer membrane has been shown to be an event necessary for mitophagy induction [78]. CerS1/C18 ceramide generation was reported to foster LC3BII targeting of autophagolysosomes to mitochondria. Overexpression of CerS2, resulting in the formation of very long chain ceramides was also associated with mitophagy activation and mitochondrial dysfunction [78]. Furthermore, intramitochondrial accumulation of C16 generated by CerS6 was reported to induce mitochondrial fission and mitophagy (excellently reviewed by [73]). Additionally to these mitochondria-related effects, Cer was shown to regulate various signaling

pathways (e.g., AKT/PKB or JNK) and to modulate different kinases (e.g., protein kinase C, PKC or protein phosphates PP1 and PP2A) [3]. Finally, elevated levels of Cer in the plasma membrane may increase membrane rigidity thereby stabilizing lipid rafts, which was shown to foster signal transduction [3,54].

#### *4.2. Sphingosine*

Similarly to Cer, sphingosine is thought to play an important role in cell cycle arrest and apoptosis [2,3]. Sphingosine was shown to inhibit the activity of PKC [2]. Moreover, sphingosine can bind to the anti-apoptotic protein 14-3-3 and inhibit its action by phosphorylation of the dimer interface [79]. Furthermore the ANP32a and ANP32b proteins have been shown to be targeted by sphingosine, and this is a mechanism by which sphingosine conveys its inhibitory effect on PP2A [80].

#### *4.3. Sphingosine 1 Phosphate*

Shingosine-1 phosphate is present in plasma (0.2–1 μM) and its concentration varies under different metabolic and pathological conditions. S1P is mostly bound to apolipoprotein M (ApoM, 50–70%) and albumin (ca. 30%), and recent reports point to an important role of HDL-bound S1P, especially in context of protection against cardiovascular disorders [81,82]. The main sources of blood S1P are erythrocytes and platelets [83]. The intracellular concentrations of S1P are much lower (nM, possibly due to a higher activity of enzymes involved in its turnover) and undergo severe changes under proinflammatory conditions. In most cell types S1P inhibits apoptosis, fosters cell proliferation and metabolism, and was shown to play a crucial role in immune cell trafficking and differentiation [1,2,84,85]. The deficiency of SPL promotes tumor growth and survival [86–88]. In neurons [89–95] and pancreatic beta-cells [21] intracellular S1P is involved in cytotoxic effects (see below). The effects of S1P are mediated by cell surface receptors and by activation of intracellular targets [1–3]. Five S1P receptors (S1PR1-S1PR5) were identified. They all belong to G protein-coupled receptors and are characterized by distinct mechanisms of action depending on the G subunit involved. Therefore the outcome of the receptor-dependent S1P action depends on the tissue-specific expression profile of S1PRs and its regulation upon various conditions [96]. Several proteins have been linked to cellular S1P export. Upon them the ABC transporter family and the Spns2 transporter are the best described [1,2,96]. Their cell type-specific expression determines S1P function in various tissues. The activation of S1PR2 was shown to stimulate cell-surface integrins and fibronectin matrix [96]. S1P has been shown to mediate the effect of cytokines on COX2 activation and PGE2 production [97]. Additionally to the classical receptors-dependent mediated S1P effects, the activation of intracellular target-dependent pathways participate in overall S1P action in cells. The zwitterionic head group of S1P, which is sufficiently water-soluble, enables S1P to flow between membranes and cellular organelles. Many molecules have been proposed as intracellular targets of S1P, among them the CIAP2 (cellular inhibitor of apoptosis 2), CerS2 and TRAF2 (TNF receptor associated factor 2) proteins [1–3,98]. Of note, the direct involvement of S1P in the regulation of TRAF2 activation was questioned by Etemadi and coworkers [99]. Moreover the SK2-mediated formation of S1P in the nucleus has been associated with changes in the epigenetic signature of cells due to the ability of S1P to directly bind and inhibit the histone deacetylases HDAC1 and HDAC2 [100,101]. Finally, many studies reported calcium as a possible second messenger in the intracellular action of S1P [21,89,93].

#### *4.4. Ceramide-1 Phosphate*

The function of ceramide 1-phosphate, similarly to S1P, is largely tissue and cell-context specific, with anti- and proinflammatory properties [70,102]. The enzymatic activity of CERK was shown to be upregulated by IL-1β, macrophage colony stimulating factors (M-CSF), calcium ionophore A23187, tyrosine-kinase pathway and agonists of nuclear peroxisome proliferator activated receptors (PPARs) [69]. C1P is believed to have strong mitogenic and antiapoptotic activities [69].

#### *4.5. Sulfatide*

Sulfatide (3-O-sulfogalactosylceramide) is a glycoSL that is synthesized from Cer by two transferases (ceramide galactosyltransferase and cerebroside sulfotransferase). Sulfatide is a multifunctional bioactive SL, which was demonstrated to play an important role in the nervous system, pancreas, immune system, as well as bacterial and virus infections (for more details please refer to [103]. Sulfatide is present predominantly in the nervous system and pancreatic islets [104]. The enzyme arylsulfatase A (ASA) specifically degrades it [103], though an ASA-independent degradation of sulfatide was also described. ASA activity requires the help of saposin B, which liberates sulfatide from membranes and makes it accessible for ASA [103]. Sulfatide is localized mainly in the Golgi apparatus, cellular membrane, and lysosomes; in pancreatic beta-cells it was found in insulin granules [103]. Moreover, sulfatide can activate binding of laminin to integrins and was described as a major L-selectin ligand, playing an essential role in monocyte infiltration in some organs [103].

#### **5. E**ff**ects of Proinflammatory Cytokines on the Sphingolipid Pathway Enzymes in Pancreatic Beta-Cells**

Most data available regarding the sphingolipid pathway in pancreatic beta-cells has been gained from murine and rodent models (RINm5F, INSE cells, MIN6 cell lines and mouse models). The extensive qRT-PCR analysis revealed the presence of all enzymes involved in the SL pathway in rodent beta-cells [21]. Pancreatic beta-cells are characterized by an imbalance of the enzymatic capacity for S1P formation (SK1 and SK2) and degradation (SPL) [21], which seems to play a particularly important role for cytokine toxicity (see Section 7).

Although a growing number of studies describing changes in the SL content upon T1DM development in humans are available, the SL rheostat of human beta-cells has not yet been characterized. Moreover detailed data about the expression pattern of the SL pathway enzymes in beta-cells in animal models of T1DM and in the human pancreas from T1DM individuals is still missing.

#### *5.1. De Novo Synthesis*

The enzymes catalyzing the first steps of the SL biosynthesis are particularly sensitive to cytokine action in rat insulin-secreting INS1E cells [21]. The subunits of SPT are affected differentially, namely the long chain 1 and long chain 2 subunits are strongly increased in response to proinflammatory cytokines (IL-1β, TNFα and IFNγ), while no changes in the expression level of the short chain subunit were detected [21]. Cytokines significantly stimulate the expression of DES in rat INS1E cells [21]. Interestingly, data from human islet transcriptome revealed that ORMDL3 is expressed in human islets and undergoes upregulation upon 24 h-exposure to a mixture of IL-1β and IFNγ [32]. Whether this expression can be accounted to beta-cells, what the contribution of TNFα on the expression of ORMDL3 is and if beta-cells express also other isoforms of ORMDL proteins requires further investigations. The ORMDL3 gene polymorphism was described in the recent studies involving T1DM individuals [14,105].

Recently a new mechanism regulating the expression of SPT has been described [106]. Using genome-wide CRISPR/Cas9 screening it was demonstrated that AHR (aryl hydrocarbon receptor) binds and activates the gene promoter of SPT [106]. Moreover, tissues from AHR KO mice were characterized by reduced expression of several other key genes in the SL biosynthetic pathway and decreased the SL content [106]. Interestingly, AHR is a transcription factor that has been recently shown to link the diet and gut microbiome alterations with islet autoimmunity [107], yielding the possibility that the observed activation of AHR in islets during T1DM development could stimulate the SPT expression and increase de novo SL biosynthesis in beta-cells. Whether this is indeed the case still needs to be experimentally proven.

#### *5.2. Ceramide Metabolism Regulation*

The best studied part of the SL biosynthesis pathway in beta-cells is the formation of Cer. IL-1β was the first cytokine shown to activate the sphingomyelin/ceramide pathway in rat insulin-producing RINm5F cells [108]. The studies revealed that a short exposure to IL-1β (2–5 min) could induce Cer and diacylglycerol (DAG) generation with a parallel rapid decrease of sphingomyelin content, indicating the activation of sphingomyelinase [108]. Another report evaluated the sphingomyelin hydrolysis in response to IL-1β exposure in rat and mouse islets as well as in RINm5F cells and failed to detect a significant effect of IL-1β within the time-frame of the experiment [109]. The elevation of Cer content in beta-cells in response to TNFα was demonstrated in mouse insulin-secreting MIN6 cells and it was suggested to be involved in TNFα-mediated apoptosis [110]. These observations were not confirmed in another insulin-secreting mouse cell line, the β-TC3 cells, in which cytokines (IL-1β, TNFα and IFNγ) failed to stimulate Cer formation [111]. The sphingomyelin content was around 50% decreased in β-TC3 cell line, similarly to RINm5F cells, however the authors showed that the kinetics of sphingomyelin hydrolysis within 4 h after cytokine addition did not differ between control and cytokine-treated cells [111]. The reasons for the contradictory findings could be related to methodological limitations of these early studies, time-specific effects of cytokine action and/or caused by a quick turnover of SLs in beta-cells.

More recent studies, involving modern mass spectrometry technology, confirmed an induction of neutral SMase and a parallel activation of iPLA2β leading to Cer accumulation in rat insulin-secreting INS1 cells exposed to proinflammatory cytokines (IL-1β + IFNγ) [112]. Our studies in INS1E cells revealed that also the mRNA expression of acid SMase is upregulated in response to a mixture of cytokines (IL-1β, TNFα and IFNγ) [21]. Beta-cell specific overexpression of iPLA2β in a mouse model (RIP-iPLA2β-Tg mice) results in upregulation of nSMase mRNA and protein expression, followed by decreased sphingomyelins with a parallel increase of Cer content [113]. This was accompanied by ER stress, mitochondrial damage and caspase-3 activation [112,113].

Ceramide content can also be upregulated by stimulation of Cer production by CerS. We showed that in beta-cells proinflammatory cytokines upregulate the mRNA expression of various types of CerS [21], strongly supporting a notion that under T1DM conditions an intra-beta-cell ceramide formation might be indeed induced. The most prevalent CerS isoforms in rodent beta-cells are CerS2 and CerS5/6, followed by CerS1 ([21], and Gurgul-Convey, unpublished). These isoforms are characterized by distinct subcellular localization (ER vs. mitochondria), as well as differences in the length of generated ceramides. Owning the role of ER stress and mitochondrial damage in cytokine toxicity to beta-cells, the cytokine-mediated induction of CerS expression may significantly contribute to toxic effects of cytokines to beta-cells. An increased Cer generation may particularly promote cytokine-induced mitochondria damage in beta-cells. The recent findings from the Brüning group strongly indicate such a scenario [114]. Using the CerS6 deficient mouse model the authors demonstrated that CerS6-derived C16 Cer, in contrast to CerS5-derived Cer, could promote mitochondrial fission and insulin resistance in obesity [114]. Using the CerS6 KO model exposed to STZ-mediated islet autoimmunity could enable in the future to assess the importance of mitochondrial Cer formation in the development of T1DM.

The involvement of mitochondrial ceramide in cytokine toxicity is indirectly suggested also by our observations in SPL-overexpressing INS1E cells [21]. Recently, we showed that the overexpression of SPL in beta-cells protects from cytokine toxicity by prevention of the ER-calcium leak, decreased mitochondrial calcium levels and promoting the expression of mitochondrial chaperones [21]. SPL deficiency was associated with increased Cer levels in other cell types [115]. Though the effects of SPL overexpression on Cer content in beta-cells still need to be analyzed, it seems plausible that the observed beta-cell protective effects of SPL overexpression may be at least partially related to decreased Cer levels, particularly in the mitochondrial compartment.

Another interesting aspect of mitochondrial Cer formation in the context of beta-cell fate in T1DM could involve sirtuins. Sirtuins are NAD+-dependent histone/protein deacetylases consisting of several subtypes that are present in various subcellular compartments and are involved in the regulation of oxidative stress through direct deacetylation of transcription factors controlling antioxidant genes [116]. Additionally, sirtuins are believed to be important regulators of glucose homeostasis, insulin secretion and mitochondrial biogenesis. Interestingly, the mitochondrial SIRT3 has been recently shown to regulate Cer generation in the brain in response to ischemia/reperfusion [76]. So far the role of altered expression of sirtuins, on mitochondrial formation of Cer in beta-cells has not been investigated.

Finally, cytokines were reported to induce the activity of aCD in INS1 cells and primary rat islets [117]. The pharmacological inhibition of aCD by NOE resulted in enhanced cytokine toxicity in INS1 cells [118]. In contrast to neutral CD [118], no changes on the mRNA and protein levels of aCD in INS1 cells and rat islets treated with cytokines were detected for up to 4 h [117]. We extended this study to a more mature beta-cell line, namely INS1E, and analyzed the mRNA expression pattern of acid and neutral CDs under longer time periods [21]. Our measurements revealed that a mixture of three major proinflammatory cytokines induce the upregulation of both nCD and aCD in INS1Ecells [21].

#### *5.3. S1P Metabolism, Receptors and Transporters*

We observed similar effects of high concentrations of IL-1β (as prevails in the first stage of islet autoimmunity) and of a mixture of the three major diabetogenic cytokines IL-1β, TNFα and IFNγ (as it occurs in the advance stage of insulitis) on the mRNA expression level of enzymes involved in S1P metabolism, indicating a crucial role of IL-1β [21]. This was in contrast to distinct effects of the acute (6 h) versus prolonged (24 h) cytokine exposure on the gene expression of enzymes involved in the SL metabolism, S1P receptors and S1P transporters [21].

Early reports identified four types of S1P receptors (S1P1, S1P2, S1P3 and S1P4) in mouse and rat islets, and in INS1 cells [119]. Recently we demonstrated the mRNA expression of S1PR2, 3 and 5, of which S1PR3 was the predominant subtype, in INS1E beta-cells [21]. S1PR2 and S1PR3 are coupled predominantly to Gq and activate phospholipase C (PLC) to induce Ca2<sup>+</sup> mobilization through the production of inositol 1,4,5-trisphosphate [120,121], and induce activation of MAPK kinases [122]. S1PR5 was shown to interact with Gα subunits [123], which inhibit PLC activity. Beta-cells are equipped with various S1P transporters [21]. The Abca1 transporter is expressed at the highest level, followed by Abcc1 and a nearly 100-fold lower expression of Spns2 [21]. Acute exposure to proinflammatory cytokines results in a strong downregulation of S1PR3 (70% decrease), partially compensated by a mild upregulation of S1PR2. S1PR5 remains unchanged under such conditions [21]. Upon longer incubation with cytokines, the expression of all S1P receptors undergoes upregulation [21]. Similar observations were made in the case of S1P transporter mRNA expression [21]. The impact of these alterations of S1P receptor and transporter system in beta-cells under acute and chronic cytokine exposure on beta-cell fate has not yet been investigated, but it could provide interesting insights into the role of S1P in beta-cell vulnerability to cytokines.

Proinflammatory cytokines enhance the activity of SK in rat beta-cells [124,125]. Our gene expression data suggest that SK2 is the main isoform expressed in insulin-secreting cells and proinflammatory cytokines increase its expression [21]. Cytokines downregulate the expression of SPL in INS1E cells and rat islets, while they enhance the expression of SPP2 [21]. These observations suggest an increased rate of sphingosine generation in beta-cells upon cytokine exposure. This could foster increased Cer formation and/or accumulation in mitochondria or elevate the generation rate of S1P in mitochondria, nucleus und other specific locations. Since SPP has not yet been shown to localize in mitochondria and nucleus, the cytokine-induced S1P formation in these two subcellular compartments is likely unlimited in contrast to other compartments with a parallel cytokine-mediated overexpression of SPP. Thus cytokine action could affect subcellular SL gradients in beta-cells with high local Cer, sphingosine and/or S1P concentrations upon cytokine exposure.

Finally, the cytokine-mediated downregulation of SPL, which we observed in INS1E cells, does not coincidence with increased S1P concentration (Gurgul-Convey, unpublished). This observation further suggests a shift of the SL pathway to sphingosine/Cer formation in beta-cells exposed to cytokines. Such a phenomenon was observed in the Charcot–Marie–Tooth phenotype in humans that is characterized by SPL deficiency and was shown to be associated with elevated Cer levels [126,127].

How these changes on the expression level of enzymes of the SL pathway translate to the sphingolipid composition of beta-cells exposed to cytokines should be further investigated, since rearrangements of the beta-cell SL profile may have potentially remarkable consequences on the susceptibility of beta-cells towards proinflammatory cytokines. This could include alterations of the expression pattern of HLA Class I and II, acceleration of cytokine signaling or disturbed cell–cell interactions. Additionally an interesting question arises whether intracellular S1P may participate in the epigenetic regulation of genes relevant for beta-cell vulnerability to the autoimmune reaction in T1DM. Further studies are needed to characterize the changes of the SL enzymatic machinery in human beta-cells before and after T1DM onset.

#### **6. E**ff**ects of Bioactive Sphingolipids on the Beta-Cell Function in T1DM**

Multiple studies have demonstrated that various SLs may regulate the beta-cell secretory capacity [4,8,9,128–130]. These effects are conveyed by the activation of cell surface receptors, a regulation of ion channels or intersection with insulin production and folding. The abundance of sphingomyelin patches on beta-cell surface was additionally reported to modulate the insulin secretory capacity [131].

The inhibitory effects of Cer and its analogues on insulin production and secretion are well described [111,130,132,133]. So far no data is available on the effects of intracellular Cer production topology on GSIS disturbances in cytokine-treated beta-cells. In contrast to Cer, extracellular S1P is a potent stimulator of insulin secretion [21,129]. Upon the activation of its receptors, S1P potentiates GSIS, most likely by induction of cAMP generation [21,125]. Deletion of SK1 in INS1E cells results in defective insulin gene expression, lower insulin content and GSIS [134]. SK2 KO in MIN6 cells leads to higher GSIS, also in the presence of low glucose [129]. Our studies showed that depletion of intracellular S1P content by overexpression of SPL does not affect GSIS in the absence of proinflammatory cytokines [21]. SPL overexpression was however capable to partially protect against proinflammatory cytokine-mediated GSIS inhibition [21]. While the exact mechanism underlying this protective effect of SPL overexpression needs further investigation, we observed an increased protein expression of mitochondrial chaperones, which play an important role in ATP synthesis (mimitin and prohibitin 2) [21]. Earlier studies revealed that these mitochondrial proteins are essential for a proper GSIS [42,135]. siRNA-mediated inhibition of Phb2 in SPL-overexpressing beta-cells resulted in a partial loss of SPL-mediated protection [21]. Additionally, we observed an increased expression of BIP and Sec61a in ER of SPL-overexpressing INS1E cells that correlated with improved calcium homeostasis [21]. BIP and Sec61a cooperate to prevent a calcium leak from ER [136]. Therefore their enhanced expression in SPL-overexpressing INS1E cells could prevent cytokine-mediated disruption of calcium homeostasis. Moreover, improved BIP expression in SPL-overexpressing INS1E cells may help to secure unbiased insulin folding in cytokine-treated cells.

In the first hours of IL-1β action, which is not associated with cytotoxicity, a temporary potentiation of GSIS is observed [137]. Could cytokine-induced S1P generation be used by beta-cells for promoting this effect? In contrast, the acute phase of cytokine toxicity is accompanied by dysregulation of GSIS [21,51–53]. In this acute phase of cytokine toxicity the expression of SK2, S1P transporters and receptors expression is upregulated while SPL expression is downregulated. This is likely to result in decreased intracellular levels of S1P, due to a parallel high expression of SPP and CerS [21]. Since SK2 is not expressed on the plasma membrane, the SK2-derived S1P is not expected to be efficiently transported inside-out and activate its receptors on the beta-cell surface. Therefore it seems that the increased expression of S1P transport and the signaling system in beta-cells could serve as an adaptation strategy to make up for cytokine-mediated inhibition of GSIS by S1P-mediated cAMP generation and its potentiating effect on GSIS. Further studies are needed to describe the effects of cytokines on the SL content upon an acute phase of cytokine toxicity and the chronic incubation. The results may help to understand the natural history of disturbances of beta-cell function in early and advance stages of cytokine toxicity and could enable designing preventive strategies.

Another twist to SL effects on beta-cell function was added by demonstration that synthaxin 4 (Stx4) is required for aSMase activity [138]. Stx4 is a plasma membrane–localized exocytosis protein that is crucial for GSIS [139]. Interestingly, Stx4 is a T1DM candidate protein [140]. Could therefore Stx-4-mediated changes in the beta-cell sphingomyelin content be involved in beta-cell dysfunction upon cytokine exposure?

Probably the best studied complex glycoSL in context of insulin production and secretion is sulfatide [103]. Interestingly, in T1DM patients the content of sulfatide was recently shown to be significantly reduced [14]. In pancreatic beta-cells sulfatide is present at the surface membrane of and in secretory granules [141], with a predominantly expressed C16:0 isoform [14]. Sulfatide is believed to promote proinsulin folding and to serve as a molecular chaperone for insulin [142]. The content of sulfatide in insulin granules decreases with rising metabolic activity of beta-cells [14]. Sulfatide is secreted together with insulin, facilitates rapid insulin monomerization and is crucial for insulin crystal preservation [142]. Moreover, sulfatide is required for normal insulin secretion through the activation of ATP-sensitive potassium ion channels and stimulation of calcium ion-dependent exocytosis [143]. In vivo administration of Zucker rats with C16:0 sulfatide resulted in significantly elevated GSIS without effects on glucose tolerance [144].

These results suggest that the place of origin and the type of SLs might be crucial for the final outcome of particular SLs on the beta-cell function. Further studies are needed to characterize which types of SLs are generated upon exposure to diabetogenic cytokines and how they influence insulin biosynthesis and beta-cell secretory capacity under T1DM conditions. The use of islets isolated from SK1, SK2 and SPL KO mouse models and exposed to cytokines will be useful in this context.

#### **7. E**ff**ects of Bioactive Sphingolipids on Beta-Cell Fate in T1DM**

Accumulative data indicates that SLs may be crucially involved in beta-cell death during T1DM development. Though the exact underpinning mechanisms remain unclear, evidence indicates that two elements may be of particular importance, namely the changes of the SL pattern of beta-cells and alterations of SL profiles in islet surroundings.

Toxic effects of Cer and sphingosine on beta-cells are well documented [4,5,110,111,118,130,133,145]. As discussed in the Section 5, proinflammatory cytokines were found to induce Cer formation, which was associated with apoptosis. Interestingly, overexpression of Stx4, which was shown to stimulate aSMase activity, reduces the chemokine ligand gene expression in beta-cells and protects beta-cells against cytokine toxicity [139]. Stx4 is a T1DM candidate gene [140]. Therefore a question arises whether the Stx-4-mediated changes in the beta-cell SL rheostat could be a link to islet autoimmunity by affecting cell membrane composition and thereby reducing the chemokine ligand presence on the beta-cell surface.

The role of S1P in cytokine toxicity to beta-cells has been extensively studied in the last decade. First it was demonstrated that extracellularly added S1P protects insulin-secreting INS1 cells and rat islets against cytokine toxicity [124,125]. The cytokine-mediated TUNEL staining, cytochrome c release and caspase-3 activation were reduced after treatment with nM concentrations of S1P. The S1P receptor antagonist BML-241 blocked this protective effect. Beneficial effects of S1P against cytokine toxicity were not associated with decreased cytokine-mediated iNOS expression or NO generation. This observation is particularly interesting in the context of human beta-cells, in which cytokines fail to induce the iNOS pathway and are nevertheless toxic [42,146]. How would extracellular S1P protect the beta-cell against cytokine toxicity? Laychock and colleagues showed that exposure of beta-cells to S1P leads to the stimulation of PLC activity, indicating the activation of Gq subunit of S1PRs in beta-cells [125]. Our own studies revealed that exposure of INS1E cells to S1P results in a rise of cAMP generation [21], extending the earlier observations that S1PR2 activation induces cAMP generation in other cell types [147–150]. Since cAMP conveys its cytoprotective effects in beta-cells via multiple mechanisms, including PKA activation and regulation of calcium homeostasis [40,151], it would be

worth evaluating these pathways in detail by means of specific inhibitors and/or genetic modifications of S1P metabolism.

Could the addition of S1P to the culture medium for islets isolated for transplantation be implemented as a preservation method? Against this procedure points the fact that extracellular S1P was reported to increase insulin secretion, also in the absence of hyperglycemia [21]. Such a scenario could lead to depletion of beta-cell insulin capacity and lead to metabolic stress of isolated islets, making islets bathed in a S1P-containing medium less useful for transplantation. This is in contrast to another bioactive lipid compound, namely prostacyclin and its analogues, which also stimulate cAMP generation in beta-cells, however do not potentiate insulin secretion in the presence of low-glucose culture medium [151]. Furthermore, S1P has been shown to be implicated in islet allograft survival [152,153]. Fingolimod, a S1PR modulator, was demonstrated to enable long-term survival of islet allografts due to its effects on immune cell trafficking [152,153].

Interestingly, the action of intracellularly generated S1P in cytokine-treated beta-cells seems to be opposite to that of extracellular S1P [21]. Our data strongly indicates that intracellularly generated S1P participates in acute cytokine toxicity to beta-cells [21]. We observed an intermediate mRNA expression level of SPL in rodent beta-cells and islets as compared to other tissues [21]. This was downregulated in response to cytokines [21]. Overexpression of SPL protected insulin-secreting cells against cytokine-induced apoptosis [21]. SPL overexpression was accompanied by maintenance of calcium homeostasis, which is strongly impaired by the action of proinflammatory cytokines in beta-cells [21]. Additionally, SPL-overexpressing INS1E cells were protected against cytokine-mediated ER stress, as evident by a significant inhibition of CHOP expression after incubation with cytokines. Moreover, SPL overexpression reduced cytokine-mediated inhibition of cell proliferation and ATP content [21]. These protective effects were independent from the NFκB-iNOS pathway [21]. Furthermore, SPL overexpression provided protection against cytokine toxicity though it failed to downregulate cytokine-induced ROS generation. As mentioned above, we detected a higher expression of various ER and mitochondrial chaperones in SPL overexpressing INS1E cells, indicating that changes in intracellular S1P concentrations may indeed epigenetically regulate gene expression in beta-cells, like in other cell types [100,101]. Importantly, siRNA-mediated suppression of SPL expression resulted in opposite effects to those observed in SPL overexpressing INS1E cells [21]. Interestingly, though SPL overexpression has been reported to be implicated in toxic effects of hexadecenal accumulation in various cell types [62,63], in beta-cells SPL overexpression provided protective effects. The possible explanation to this phenomenon could be that beta-cells are rich in the enzyme responsible for hexadecenal detoxification, namely ALDH3A2 [21], enabling prevention of hexadecenal accumulation and toxic effects of SPL-overexpression in beta-cells. It will be important to evaluate the expression and activity of SPL in beta-cells chronically exposed to cytokines, and to investigate the effects of double-transfection approaches including SK1/SK2 and SPL to determine the role of intracellular S1P in more detail.

The involvement of intracellularly generated S1P in cytotoxic effects of cytokines in beta-cells seems to be very similar to the role of intracellular S1P in neurons as described in elegant studies by the Van Echten-Deckert group [89–95]. Interestingly, pancreatic beta-cells and neurons share multiple common features, though derived from distinct germ layers [154]. It is speculated that the endocrine and nervous systems developed from a common evolutionary ancestor [155,156]. Moreover, *Drosophila*-insulin-like peptides (Dilps), which are synthesized by neurons in flies, were shown to regulate energy metabolism similarly to mammalian insulin [157]. The way beta-cells biosynthesize and store insulin, and answer to external stimuli by insulin secretion mimics very closely the way neurons store and release neurotransmitters [154,156]. Many studies showed that beta-cells and neurons are characterized by similar gene expression patterns and spliceosome activity [158] (for more information please refer to [154]). It is therefore not surprising that beta-cells and neurons may also share similar sensitivity to intracellular S1P.

The mRNA expression of both SPP1 and SPP2 was found to be induced in beta-cells under acute exposure to cytokines [21]; what the impact of a chronic exposure to cytokines on SPP expression is or whether the expression/activation of other phosphatases would be of importance for beta-cell fate should be addressed in the future. Though exogenous Cer was shown to disrupt mitochondrial function [145] and induction of SMase was shown to be accompanied by increased Cer content, mitochondria damage and apoptosis [112], there is no direct evidence linking Cer accumulation in mitochondria to cytokine-mediated beta-cell apoptosis. Finally, several SL-enzyme knockout mouse models have been successfully characterized in context of T2DM [159–161] and many of them show a phenotype that could be interesting for T1DM research. For example mice lacking SPP2 display defective beta-cell proliferation, reduced islet mass and ER stress activation in beta-cells [159]. Exposure of such animal models to STZ, generation of beta-cell specific SPL knockout and knockin mouse models or development of SL-enzymes knockouts in Non Obese Diabetic (NOD) mice, will advance our understanding of T1DM development mechanisms.

Another interesting aspect of cytokine action on the SL pathway is the secretion of nCD via exosomes. Interestingly the low nontoxic cytokine concentrations [38,162] were shown to stimulate nCD, in a similar manner to high concentrations [163]. The lack of toxic effects of low-dose cytokine exposure correlated with a release of neutral ceramidase via exosomes [163] and the presence of nCD containing exosomes prevented apoptosis in INS1 cells incubated with high concentrations of cytokines [163]. The generation and secretion of S1P via nCD-rich exosomes was responsible for the activation of S1PR2 and the observed antiapoptotic effect. This study indicates that beta-cells may activate protective mechanisms at the beginning of the inflammatory response within islets, to prevent further damage caused by high concentrations of cytokines. Once this axis fails, the beta-cell apoptosis and destruction are accelerated.

Overall, this data indicates that proinflammatory cytokines may impact SL generation differentially upon acute and chronic exposure that could have potentially pronounced consequences for beta-cell viability and vulnerability to autoimmune insult. Future measurements of SL species and their distribution in cytokine-treated beta-cells should help to understand how these observed effects influence beta-cell vulnerability to cytokines.

#### **8. Sphingolipids in Animal Models of Autoimmune Diabetes and Human T1DM**

In this chapter the SL changes occurring in animal models of autoimmune diabetes and in human T1DM and their role for islet autoimmunity and for beta-cell survival will be addressed. As discussed above proinflammatory cytokines that are secreted by activated immune cells infiltrating islets during T1DM development were shown to affect the SL metabolic pathway of beta-cells. Changes in the SL content were associated with beta-cell dysfunction and death. Recent studies demonstrated an altered SL profiles in the blood of T1DM patients [14–16,18,130,164]. What are the consequences of these altered SL serum profiles on pancreatic beta-cell function and fate during T1DM development? Is the SL concentration around the islets altered? Is the SL composition of beta-cell distinct in T1DM individuals comparing with nondiabetic, healthy subjects?

#### *8.1. Animal Models*

While these questions remain at present open, the studies with the S1P receptor modulator Fingolimod/Gilenya (FTY720) may shed some light on this topic. Fingolimod was approved for the treatment of multiple sclerosis in over 40 countries [165]. Its effects were analyzed in multiple animal models of autoimmune diabetes, such as the NOD mouse, STZ-induced autoimmune diabetes in mice and in the rat model of human T1DM, the LEW.1AR1-iddm (IDDM) rat [166–168]. In all these animal models fingolimod treatment was reported to improve glycemia and prevent infiltration of islets. Long-term treatment with FTY720 proved to prevent the diabetes onset in IDDM rats and in NOD mice by reducing immune cell infiltration and cytokine-mediated beta-cell destruction [152,153,166–168].

The IDDM (LEW.1AR1-iddm) rat develops spontaneously autoimmune diabetes [169]. Pancreatic islets from fingolimod-treated IDDM rats are characterized by a well preserved architecture and dense insulin immunostaining [167]. This goes along with prevention of T-cell islet infiltration and reduced expression of proinflammatory cytokines in immune cells [167]. The remaining weak macrophage infiltration observed in the minority of islets in fingolimod-treated IDDM rats correlates with the increased beta-cell expression of MCP1, a well-known attractant for macrophages [167]. Interestingly, the infiltrating macrophages in fingolimod-treated IDDM rats do not express and release IL-1β and TNFα even after a prolonged time post fingolimod treatment, explaining the lack of beta-cell demise in these islets. The study by Jörns and colleagues was the first to show that fingolimod treatment may result in decreased expression of proinflammatory cytokines. Why did macrophages remain proinflammatory cytokine negative in response to the fingolimod treatment? Jörns and colleagues observed that the treatment with fingolimod leads to increased production of anti-inflammatory cytokines IL-4 and IL-10 [167], which leads to the activation of an anti-inflammatory M2 phenotype [170,171]. IL-4 was also shown to protect beta-cells against cytokine toxicity [38,172–174]. Thus the phenomenon of fingolimod-mediated macrophage phenotype rearrangements opens a new intriguing research area in the S1P biology.

Further insights into the role of SLs in T1DM development were gained in a well-established mouse model of autoimmune diabetes, the NOD mouse. The NOD mouse is characterized by a genetic susceptibility to autoimmune diabetes and spontaneously develops autoimmune reaction against pancreatic beta-cells. Environmental factors have also been shown to affect the diabetes incidence in this mouse model [175]. A continuous supplementation of L-serine, a precursor of the SL biosynthesis, was shown to reduce diabetes incidence and insulitis score in female NOD mice [176]. The authors observed significant changes in serum SLs, failed however to detect any significant effects on the pancreatic SL content. Such effects of L-serine should not, however, be excluded since the authors did not perform experiments in purified beta-cells. Administration of fingolimod prolonged the survival rate of islet allografts in diabetic mice [152,153]. Another study showed that the serum phospholipid and triglyceride composition might be associated with progression to T1DM both in NOD mice and humans [164]. The authors found that young female NOD mice who later progress to autoimmune diabetes exhibited the same lipidomic pattern as prediabetic children [164], confirming that the NOD mouse is a valuable model to study the role of SLs in T1DM development. Moreover, NOD thymocytes were found to be characterized by a lower level of S1PR1 and a decreased SPL mRNA and protein expression comparing with healthy mice [177]. These changes were suggested to participate in the T-cell migratory abnormalities observed in NOD mice during diabetes development [177]. S1P was shown to reduce CD4+ T-cell activation in NOD mice and to prevent vascular complications [178].

Administration of glycoSLs has been widely used to assess their effects on islet autoimmunity in NOD mice. For instance, ganglioside GM1 decreased the rate of islet infiltration, attenuated production of proinflammatory cytokines (IL-1β, TNFα and IFNγ) and increased the level of NGF in islets [177]. The protective effects of sulfatide against autoimmunity have been recognized already many years ago when Buschard and colleagues demonstrated that the treatment with sulfatide or its precursor GalCer prevents diabetes in NOD mice [179]. Sulfatide was reported to increase the population of CD3+CD25+ regulatory T-cells, while decreasing production of proinflammatory cytokines (for details refer to the excellent review [104]). These protective effects of GalCer and sulfatide against autoimmune reaction during T1DM development rely on their effects on NKT) cells. In T1DM individuals NKT cells are less frequent and display deficient IL-4 responses [180]. Similar observations were made in NOD mice [181]. Interestingly, α-GalCer was shown to activate NKT cells and prevent the onset and recurrence of T1DM in NOD mice [181]. Moreover, a sphingosine truncated derivative of α-GalCer, OCH, was reported to prevent insulitis and diabetes development in NOD mice more efficiently than its precursor, probably by enhancing the activity of NKT cells to produce IL-10 [182]. Rhost and coworkers observed that a fraction of NOD mice develop autoantibody reactivity to sulfatide, though they failed to demonstrate that sulfatide treatment reduces the diabetes incidence under the treatment

scheme they undertook [183]. Using fenofibrate, which activates the sulfatide biosynthesis, Holm and colleagues were able to completely prevent diabetes development in NOD mice [14]. In the follow-up study they demonstrated that fenofibrate selectively elevates the pancreatic content of very-long-chain SLs in NOD mice and reduces the incidence of diabetes by around 50% [184]. Moreover they showed that fenofibrate treatment leads to remodeling of pancreatic lipidome with increased amount of lysoglycerophospholipids [184]. NOD mice treated with fenofibrate were characterized by more stable blood glucose and improved glucose tolerance [184]. Sulfatide was additionally shown to inhibit insulitis and to prevent diabetes in NOD mice by blockage of L-selectin [185]. Furthermore, the C16:0 isoform of sulfatide was reported to downregulate the production of proinflammatory cytokines [186]. In vitro experiments revealed that sulfatide has also the ability to reduce caspase-3/7-dependent apoptosis caused by exposure of insulin-secreting cells to IL-1β, IFNγ and TNFα [187].

Finally, in two additional animal models of autoimmune diabetes (STZ-induced diabetic rats and Akita diabetic mice) the serum concentration of S1P was shown to be significantly elevated in comparison to control animals [17], raising a question whether following the S1P content in blood could serve as a biomarker to track the disease progress.

#### *8.2. Human T1DM*

The incidence of T1DM is rising in the last decades in western countries; this suggests an important role of environmental factors in the pathogenesis of this disease. Accumulating evidence points to significant changes in serum metabolome proceeding T1DM development [14,15,18,130,164]. An increased risk of T1DM was described in response to prenatal exposure to perfluoroalkyl substances that modulate neonatal serum phospholipids [188]. Several phospholipids, particularly sphingomyelins and specific phosphocholines, which were shown to be significantly lower in the serum of children who later progress to T1DM [18] were downregulated by perfluoroalkyl substances [188]. Holm and colleagues identified polymorphisms in eight genes encoding proteins involved in the SL metabolism that contribute to the genetic predisposition to T1DM [14]. Single-nucleotide polymorphisms (SNPs) in the chromosome 17q12-q21 region with the gene coding for ORMDL3 were linked with T1DM [14,105]. Importantly, the level of these polymorphisms correlated with the degree of islet autoimmunity in patients with recent onset T1DM [14].

Interestingly alternations of S1P levels have been shown to control IL-17 production in human T-cells [189]. IL-17 arises as a crucial player in the autoimmunity development in T1DM [190,191]. Therefore a control of S1P levels and of the activation of S1PRs might represent an interesting intervention option that should be tested in the future.

With the respect to S1P action, it is important to mention that the cardioprotective effects of HDL have been shown to depend on its chaperoning function for S1P [81,82]. Dyslipidemia is typical in patients with long-term T1DM, and associates with an increased risk of cardiovascular events. How sphingolipids, or particularly S1P, contribute to the progression of diabetic complications in T1DM-patients requires further investigations.

Furthermore, an altered SL content in immune cells, peripheral blood mononuclear cells (PBMC) as well as in serum were reported in multiple studies with T1DM individuals [14,15,18,130,164]. The changes in the sphingomyelin content were defined as a new hallmark of progression to T1DM. Moreover sulfatide levels in newly onset of T1DM patients were found significantly lower than in healthy children [14]. Studies performed in the newborn infants, who later in life progress to T1DM, indicate that their lipidomic profiles are distinct from those of healthy infants [19,20]. The recently published clinical case report study by the Buschard group demonstrated that the treatment with fenofibrate (160 mg daily) initiated seven days after T1DM diagnosis resulted in a fast decline of insulin dose and long-term insulin-independency [192].

The ceramide pathways were found to be specifically associated with T1DM progression [18]. In T1DM progressors a lower content of L-serine was described, in line with the protective effects of L-serine administration against islet autoimmunity observed in the NOD mouse [176]. This was

in contrast to glycoSLs such as GlcCer, LacCer and GalCer, which were significantly upregulated in T1DM patients. The authors concluded that these changes in the SL blood cell content could contribute to immune dysfunction in children, who later progress to T1DM [18].

Finally, an intriguing observation about a protective role of C1P on insulin signaling in diabetic kidney [193] might also apply to the phenomenon of insulin resistance in T1DM [194]. SMPDL3b (sphingomyelin phosphodiesterase acid-like 3b) is a lipid raft enzyme that regulates plasma membrane fluidity. Its enhanced expression was observed in diabetic kidney disease and was shown to affect the production of SLs resulting in decreased C1P content [193]. The SMPDL3b overexpression impaired insulin signaling by interfering with insulin receptor isoforms binding to caveolin-1 in the plasma membrane, which was rescued by supplementation with exogenous C1P [193]. Moreover S1P was shown to counteract insulin signaling in beta-cells through the activation of S1PR2 [128]. These studies are raising the question whether a lipid-based therapy might be an option for treatment of severe T1DM complications. Further studies involving human beta-cells, genetically modified beta-cell lines, and various T1DM animal models are needed to assess the role of C1P in T1DM development and beta-cell fate.

#### **9. Sphingolipids as Autoantigens and Biomarkers in T1DM**

The progression to T1DM is monitored mainly by evaluation of serological biomarkers (autoantibodies). The positivity of autoantibodies against beta-cells, the age of seroconversion and the positivity for multiple autoantibodies are predictors and major risk factors for the development of T1DM [195–197]. The primary islet autoantibodies are autoantibodies against insulin (IAA), insulinoma-associated antigen-2 (IA-2), glutamic acid decarboxylase (GAD), zinc-transporter 8 (ZnT8) and islet cell antibodies (ICA) [195–197]. These autoantibodies may appear at any age, but the peak of the first islet autoantibody is usually before the age of 3 years [195–198]. Though the vast majority of autoantibodies related to T1DM recognize peptide antigens, antibodies against lipid antigens have also been described [17]. It has been shown that around 60% of sera from children with T1DM react against antigens composed of lysophospholipids [199], with many epitopes directed against gangliosides and sulfatide.

Gangliosides are sialic acid containing glycolipids, which are formed of Cer and an oligosaccharide chain. They are associated with the plasma membrane and in some cell types, including pancreatic islet cells are also present in the cytosol membranes like those of secretory granules. They play an important role in cell–cell interactions. Gangliosides are targets of a variety of anti-islet autoantibodies (please refer to the excellent review [200]). Early studies in the STZ-mouse model of autoimmune diabetes revealed that administration of a mixture of gangliosides (Cronassial, 150 mg/kg body wt., 21% GM1, 40% GD1a, 16% GD1b, 19% GT1b and 5% others) can dampen islet inflammation, but is not able to stop beta-cell destruction [201,202]. Dotta and colleagues were the first to identify GM2-1 in islets that was a target for IgG islet cell autoantibodies [203]. The content of GM2-1 was significantly overexpressed in NOD mice and the antibodies against this ganglioside were found to strongly correlate with T1DM progression in relatives of T1DM individuals [203,204]. Moreover GM2-1 was found to co-localize with insulin granule, similar to other autoantigens. GT3, GD3 and GM2-1 have been shown to be associated with severity of autoimmune reaction in T1DM [204]. Finally, in a case-control analysis of the Croatian population an association of B4GALNT1 gene variations (an enzyme involved in the biosynthesis of GM2 and GD2) with T1DM was detected [205], a phenomenon confirmed by Holm and colleagues in a larger cohort study of T1DM individuals [14].

Interestingly, recent studies in brain cancer have shown that an ER ATP-dependent chaperone GRP94 regulates the ratio of GM2-GM3 gangliosides [206]. The GM2-activator protein, which is a cofactor of beta-hexosaminidase responsible for GM2 hydrolysis to GM3, was shown to be a client for GRP94. Recent studies from the Marzec group have described a reduced expression of GRP94 in beta-cells from T2DM individuals [207] and its role in the inducible proteasome activation-mediated proinsulin degradation [208]. The GRP94 deficiency could disable the proper activity of the GM2-activator protein and thus prevent GM2 hydrolysis to GM3, linking cytokine effects with increased GM2 expression and islet autoimmunity in T1DM.

Moreover, GM1 and GD1a gangliosides have been shown to modulate inflammatory effects of LPS [209] by prevention of TLR4 translocation into lipid rafts. Interestingly, the activation of TLR4 and TLR3 was shown to be activated in viral infections in islets of T1DM patients, as well as in animal models of autoimmune diabetes and beta-cell lines [210–213]. Thus, the ganglioside pattern in beta-cells may be of crucial importance for beta-cell susceptibility to viral infections that are considered as one of major triggers of T1DM development [31,36,210–218]. The generation of tools for influencing ganglioside pattern in islets may represent an interesting new possibility to protect beta-cells from cytokine- and inflammation-mediated toxicity.

Concerning the role of sulfatide in T1DM development, this glycoSL and its precursor GalCer were shown to be ligands for CD14, the macrophage scavenger receptor, in a subset of beta-cells [219]. Anti-sulfatide antibodies have been detected in prediabetic and newly diagnosed T1DM patients [220, 221] (excellently reviewed by Buschard [104]). Anti-sulfatide antibodies were shown to reduce insulin secretion and exocytosis from beta-cells [143].

Finally, sphingomyelin patches on plasma membrane act as epitopes for IC2, a monoclonal antibody that specifically recognizes the surface of beta-cells [222]. This raises the possibility that cell surface sphingomyelin pattern might be involved in the autoimmune reaction directed against beta-cells.

#### **10. Conclusions and Perspectives**

The unique beta-cell sphingolipid rheostat and limitations of self-regulation upon T1DM development might serve as one of the major underlying mechanisms involved in beta-cell dysfunction and death in T1DM, similarly to neurodegenerative disorders. Further studies, involving the modern techniques to track the SL flow and de novo SL biosynthesis (such as [223,224]) as well as SL analogues and inhibitors of the SL pathway enzymes, should enable identification of novel SL-related pathways and targets that are engaged in human beta-cell susceptibility to proinflammatory cytokines during T1DM development. The availability of the excellent model human beta-cell line, EndoC-βH1 beta-cells [225,226], for in vitro studies will drive the progress of studies on the role of SLs in T1DM in upcoming years. The impact of beta-cell S1P homeostasis and SPL needs further investigations in beta-cell specific SL enzyme KO models. Based on so-far obtained data/observations, it is expected that SLs will be likely shown to affect all major functions of beta-cells and will provide crucial insights into the activation of islet autoimmunity (illustrated in Figure 3).

**Figure 3.** The possible involvement of sphingolipids in beta-cell biology during T1DM development. Rearrangements of SLs in response to the action of proinflammatory cytokines that are released by activated immune cells likely participate in islet autoimmunity, beta-cell dysfunction and death by multiple mechanisms.

Certainly, upcoming years will bring the full characterization of the effects of cytokines on SLs of beta-cells leading to exciting new insights into mechanisms underlying cytokine-mediated beta-cell death, which shall show us how to help the vulnerable beta-cell facing T1DM.

**Funding:** This research received no external funding.

**Conflicts of Interest:** The author declares no conflict of interest.

#### **References**


© 2020 by the author. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Review* **New Insights into the Role of Sphingolipid Metabolism in Melanoma**

**Lorry Carrié 1,**†**, Mathieu Virazels 1,**†**, Carine Dufau 1,**†**, Anne Montfort 1, Thierry Levade 1,2, Bruno Ségui 1,**‡ **and Nathalie Andrieu-Abadie 1,\*,**‡


Received: 17 July 2020; Accepted: 24 August 2020; Published: 26 August 2020

**Abstract:** Cutaneous melanoma is a deadly skin cancer whose aggressiveness is directly linked to its metastatic potency. Despite remarkable breakthroughs in term of treatments with the emergence of targeted therapy and immunotherapy, the prognosis for metastatic patients remains uncertain mainly because of resistances. Better understanding the mechanisms responsible for melanoma progression is therefore essential to uncover new therapeutic targets. Interestingly, the sphingolipid metabolism is dysregulated in melanoma and is associated with melanoma progression and resistance to treatment. This review summarises the impact of the sphingolipid metabolism on melanoma from the initiation to metastatic dissemination with emphasis on melanoma plasticity, immune responses and resistance to treatments.

**Keywords:** cancer; ceramide; gangliosides; immunotherapy; metastasis; phenotype switching; sphingosine 1-phosphate

#### **1. Introduction**

Cutaneous melanoma is a skin cancer whose incidence is increasing significantly worldwide (Figure 1). Even though melanoma is not frequent, accounting for less than 5% of skin cancers, it can be very aggressive and causes more than 75% of all skin cancer deaths [1] (Figure 1). Despite significant improvement of treatment strategies in the last decade, owing both to the emergence of BRAF- or MEK-targeted therapies and checkpoint blockade immunotherapies (i.e., anti-cytotoxic T-lymphocyte-associated antigen-4 (CTLA4) and anti-programmed cell death-1 (PD-1)), the prognosis for patients with metastatic melanoma remains uncertain, predominantly due to treatment failures and recurrences [2]. Fortunately, melanoma is usually curable by excisional surgery if detected at an early stage, with a high five-year survival rate [3,4]. Thus, a better understanding of melanoma progression processes, before dissemination, is a major public health issue in order to discern new therapeutic targets.

**Figure 1.** Cutaneous melanoma: the 19th most common cancer worldwide. Estimated age-standardised incidence rates of cutaneous melanoma in the most affected countries in 2018, for both sexes and all ages. Data from the International Agency for Research on Cancer (World Health Organisation).

Melanoma arises from melanocytes, i.e., melanin-producing neural crest-derived cells, which are located at the junction between the epidermis and the dermis [5]. The initial stage of melanomagenesis corresponds to a radial-growth phase (RGP), in which melanoma cells invade laterally but stay confined into the epidermis. This stage is followed by a vertical-growth phase (VGP), in which melanoma cells invade the dermis and are able to reach blood vessels. Then, an extravasation stage, corresponding to the release of melanoma cells from the blood circulation into new tissues, leads to the formation of metastatic niches [6–8]. Melanomagenesis requires at least two key events. The first is activating mutations in oncogenes such as *BRAF* or *NRAS*. BRAF mutations are found in 50% of melanoma patients (Figure 2) and the V600E mutation accounts for approximately 75% of all BRAF mutations detected in cutaneous melanoma [9]. Conversely, the most common NRAS mutations, i.e., Q61R and Q61K, affect about 20% of melanoma patients [10]. Since BRAF and NRAS mutations are mutually exclusives, these driver mutations lead to constitutive activation of the mitogen-activated protein kinase (MAPK) pathway and aberrant cancer cell proliferation in approximately 70% of patients (Figure 2) [11]. The second event is illustrated by the loss of expression of key tumour suppressor genes such as *PTEN* or cell cycle checkpoint regulators such as *CDKN2A,* which occur in 13% and 24% of patients, respectively (Figure 2). These genetic changes can bypass oncogene-induced senescence (OIS) processes and cause the immortalisation of tumour cells [4,7,12] (Figure 2).


**Figure 2.** *BRAF*, *NRAS*, *PTEN* and *CDKN2A* are the most frequently mutated genes in cutaneous melanoma. Mutation rate, genetic alteration (**a**) and mutual exclusivity (**b**) for *BRAF*, *NRAS*, *PTEN* and *CDKN2A* mutations observed in 1635 samples from 1584 patients included in 12 studies analysed on cBioportal for cancer genomics (https://www.cbioportal.org).

Metabolic reprogramming is also crucial for melanomagenesis. Indeed, a shift from mitochondrial oxidative phosphorylation to cytoplasmic anaerobic glycolysis, known as Warburg effect, is required for metastatic dissemination [13]. The present review focuses on alterations in the metabolism of sphingolipids (SL). Interestingly, several key enzymes of the glycolytic pathway can be severely affected by changes in SL metabolism in melanoma. For instance, C16-ceramide, which is the major long-chain ceramide in melanocytes and melanoma cells, impairs pyruvate kinase, hexokinase and LDH activities, consequently altering cellular glycolysis and inhibiting melanoma progression [14].

How modulations of the SL metabolism affect dermatologic diseases have long been studied [15] and accumulating evidence demonstrates the presence of alterations in the ceramide metabolism in melanoma cells. This article aims at providing a comprehensive overview of the effects dysregulations of the SL metabolism have on melanomagenesis, melanoma progression, immunity and resistance to treatment, especially linked to the phenotype switching.

#### **2. Alterations of Sphingolipid Metabolism in Melanoma**

The main dysregulation affecting the SL metabolism in melanoma cells is a trend toward a reduction of ceramide, which promotes cell death (for review, see [16]). This is associated with changes in the expression and/or activity of a number of enzymes as well as the accumulation of tumour-promoting metabolites, which include sphingosine 1-phosphate (S1P) and gangliosides (Figure 3).

**Figure 3.** Multiple dysregulations of sphingolipid metabolism in melanoma. SL metabolites or SL-metabolising enzymes whose levels or expression are altered in melanoma, are mentioned. Decreases are indicated in blue and increases in red. AC, acid ceramidase; Cer, ceramide; CERS, ceramide synthase; CoA, coenzyme A; DihydroCer, dihydroceramide; GalCer, galactosylceramide; GC, glucosylceramidase; GlcCer, glucosylceramide; LacCer, lactosylceramide; S, sphingosine; S1P, sphingosine 1-phosphate; SM, sphingomyelin; SMases, sphingomyelinases; SMS, sphingomyelin synthase; SPC, sphingosylphosphorylcholine; SphK, sphingosine kinase; SPL, S1P lyase; ULCFA, ultralong chain fatty acids.

The impact of SL metabolism dysregulations in melanoma cell lines and/or patients is summarised in Table 1. For instance, a low expression of the ceramide synthase 6 (CerS6) in melanoma cells is related to malignant behaviours as demonstrated in WM35, WM451 and SKMEL28 human melanoma cell lines [17]. In addition, acid ceramidase (AC), encoded by the *ASAH1* gene, which hydrolyses ceramide into sphingosine, is expressed at high levels in melanocytes and proliferative melanoma cells, as observed in vitro as well as in biopsies from patients with stage II melanoma [18]. *ASAH1* expression was: (i) higher in human melanoma cell lines exhibiting a proliferative phenotype as compared to invasive ones; and (ii) reduced at the invasive front on tumour specimens from melanoma patients [19]. Sphingosine kinase 1 (SphK1), which phosphorylates sphingosine to produce sphingosine-1-phosphate (S1P), also shows increased expression and/or activity in melanoma cells compared to melanocytes, not only in human and murine cell lines [20,21] but also in human biopsies [22]. Collectively, these findings suggest a shift of the S1P-ceramide balance towards S1P production in melanoma cells. In accordance, the expression of *SGPL1* gene, encoding for S1P lyase (SPL), is downregulated in melanoma cell lines when compared to adult or juvenile melanocytes, suggesting that *SGPL1* might be downregulated during melanomagenesis [23].

Ceramide can also be converted into more complex SL, such as gangliosides. For instance, monoand disialoganglioside levels are very high in human melanoma cells and tissues, especially GD3 [24]. Interestingly, levels of this latter ganglioside were correlated to the expression of the c-Yes tyrosine kinase whose activity is known to increase in melanoma cells as compared to melanocytes [25] and whose inhibition reduced the malignant potential of GD3<sup>+</sup> melanoma cells only [26].

Alterations of the ceramide metabolism also include changes in the expression of sphingomyelin synthase 1 (SMS1), which is encoded by the *SGMS1* gene and catalyses the transformation of ceramide into sphingomyelin (SM) [27,28]. *SGMS1* is expressed at low levels in most of the human melanoma biopsies and low *SGMS1* expression is associated with a worse prognosis in metastatic melanoma patients. Of interest, a weak expression of *SMS1* was shown not to be associated with an intracellular accumulation of ceramide, most likely due to its conversion into glucosylceramide (GlcCer) through GlcCer synthase (GCS). Consequently, 6 out of 10 human melanoma cell lines exhibited higher levels of GlcCer than SM [29]. Moreover, SM can also be transformed into sphingosylphosphorylcholine (SPC) by a yet uncharacterised SM deacylase and SPC has been shown to stimulate regulators of melanomagenesis such as extracellular signal-regulated kinases (ERK), microphthalmia-associated transcription factor (MITF) and Akt/mTOR [30–33].




**Table 1.** *Cont*.

Finally, the expression of acid sphingomyelinase (A-SMase), which hydrolyses SM into ceramide, has been shown to be higher in benign nevi than in primary melanomas, and further reduced in the lymph-node metastases [34]. Moreover, a lower expression/activity of A-SMase was observed in hyper-pigmented murine and human melanomas as compared to the hypo-pigmented ones, suggesting an inverse correlation between A-SMase expression/activity and melanin content. In accordance, exogenous C2-ceramide decreased melanin content in melanocytes [35].

#### **3. Role of the Sphingolipid Metabolism in Melanomagenesis**

#### *3.1. Do SL Metabolism Alterations Increase the Risk to Develop Melanoma?*

A genome-wide association study has identified the 1q21.3 chromosomal region, containing LASS2 gene that encodes ceramide synthase 2 (CerS2), as a locus predisposing to cutaneous melanoma [36]. These observations suggest the involvement of some genetic determinants of the ceramide metabolism in melanoma predisposition. Interestingly, a defect in glucosylceramidase 1 (GBA1) resulting in Gaucher inborn disorder (GD), owing to a defect in GlcCer hydrolysis into ceramide, is associated with an increased risk of malignancies, including melanoma [37,38]. Indeed, accumulation of GlcCer as well as glucosylsphingosine, which arises from the cleavage of excess GlcCer by AC [39,40], occurs in macrophages and could severely alter the immune and inflammatory responses. This could create a favourable microenvironment to promote melanomagenesis (for review see [38]).

Another hypothetical link between glucosylceramidase (GC) and melanoma development is autophagy, which can be either cell protective, promoting their survival, or lethal, via induction of a programmed-cell death mechanism [41,42]. Defective autophagy has been reported in models of GC or saposin C deficiency [43]. Accumulation of GlcCer was associated to autophagy dysfunction in a drosophila model of GD that lacked the two fly GBA1 orthologues [44]. In accordance, GlcCer accumulation was also associated with autophagy impairment and defective autophagosome-lysosome fusion, resulting in autophagosome accumulation in induced pluripotent stem cells (iPSCs) derived from patients with GD [45]. Moreover, an hyperactivation of the autophagic inhibitor mTOR and a downregulation of the master regulator of lysosome function TFEB were reported in human neuroglioma cells treated with the GC inhibitor conduritol B epoxide [46]. Numerous studies have shown that impaired autophagy can favour melanoma development. Indeed, the activation of mTOR was associated with poor prognosis in melanoma patients [47]. In addition, ERK-induced TFEB phosphorylation impaired expression of autophagy-lysosome target genes in BRAF-mutated melanoma, which elicited the formation of TGF-β-dependent metastases [48].

All these findings indicate a possible link between GC deficiency and melanomagenesis, that may result from an altered immune response or disturbed autophagy. The underlying mechanisms remain, however, to be determined.

#### *3.2. Sphingolipid Metabolism Modulates Melanoma Cell Proliferation and Survival*

Transformation of normal melanocytes into melanoma cells is mediated by the activation of growth stimulatory pathways, typically leading to cellular proliferation as well as the inactivation of apoptotic and tumour suppressor pathways. The RAS-RAF-MEK-ERK pathway is one of the most important signalling pathways involved in melanoma cell growth and survival [49,50]. A constitutive activation of BRAF, mostly due to the substitution of valine by glutamic acid at position 600 (also known as V600E), affects about 50% of melanoma patients [9,51–53] (Figure 2). Therapies targeting the BRAF V600E mutation help advanced melanoma patients live longer [54,55]. Moreover, co-administration of BRAF (vemurafenib) and MEK (cobimetinib) inhibitors improves the progression-free survival [56] and extends the five-year overall survival by ~40% [57]. Unfortunately, most patients, including those who experience an initial tumour regression, exhibit disease progression within 6–8 months following the initiation of targeted therapy [58].

Multiple lines of evidence indicate that some SL-metabolising enzymes regulate melanoma cell proliferation and survival. First, SphK1 expression and activity are induced by ERK1/2 and AKT in numerous mammalian cells [59–61], including melanoma cells [20,21]. Moreover, SphK1 knockdown by siRNA decreased anchorage-dependent and -independent growth of human melanoma cells [20]. Similarly, targeting SphK1 using shRNA in B16F10 [62] or Yumm 1.7 [22] murine melanoma cells reduced tumour growth in syngeneic mice. Accordingly, the SphK1 inhibitor SKI-I, which increases the intracellular ceramide levels and decreases S1P levels in melanoma cells, resulted in a cell cycle arrest between G2/M and S phases as well as increased apoptotic cell death, caspase-3 activation and nuclear accumulation of cleaved PARP [20]. The intraperitoneal administration of SKI-I in mice harbouring melanoma also decreased tumour growth [20,22]. Consistently, the growth of B16F10 tumours is impaired in SphK1−/<sup>−</sup> mice as compared to wild-type animals [21].

S1P, which is mainly produced by SphK1 in melanoma cells, conveys oncogenic signals as an intracellular second messenger via a ligation of a family of G-protein coupled receptors (S1P1-5) expressed both on the malignant and their neighbouring cells [63]. We recently demonstrated that the melanoma cell-autonomous survival in response to the BRAF inhibitor vemurafenib is mediated by S1P1 and S1P3 [64].

Moreover, AC, which is expressed at high levels in proliferative melanoma cells, may also contribute to melanoma cell proliferation and survival. Indeed, AC inhibition by siRNA dramatically reduced the number of 501mel melanoma cells, as shown using short-term cell growth and colony formation assays [19]. Similarly, CRISPR/Cas9-mediated AC ablation in A375 melanoma cells blocked G1/S cell cycle progression, promoted senescence and apoptosis, resulting in reduced cell growth. These cells were unable to form spheroids and showed a lower replication rate as well as a decreased in their invasive capacity compared to controls. Mechanistically, AC ablation resulted in the accumulation of the saturated C14-, C16- and C18-ceramides and is accompanied by the down-regulation of MYC, CDK1, CHK1 and AKT [65]. In accordance, the inhibition of AC activity with a chemically stable AC inhibitor, named compound ARN14988, sensitised proliferative melanoma cells to the cytotoxic actions of various anti-tumour agents [18]. In line with these observations, we previously reported that the cytotoxic action of dacarbazine was accompanied with AC proteolysis in human melanoma cells [66]. Of interest, confocal immunofluorescence analyses revealed the nuclear localisation of AC in normal melanocytes, a phenomenon not observed in melanoma cells, suggesting that AC could activate proliferation pathways only in tumour cells [18].

By reducing ceramide levels, AC, in concert with SphK1, favours melanoma cell proliferation. This was confirmed using short-chain C2-ceramide, which was reported to inhibit AKT and ERK activation as well as proliferation in Malme-3M melanoma cells [67]. In addition, the GCS inhibitor PDMP, which increases intracellular C16-ceramide levels, inhibited cell proliferation, migration and invasion of WM35 and WM451 human melanoma cells. The effect of PDMP was associated with the inhibition of key enzymes from the glycolysis pathway including the pyruvate kinase, hexokinase and lactic acid dehydrogenase. Strikingly, the treatment of melanoma cells with exogenous C16-ceramide neither altered melanoma cell growth nor migration and invasion. In contrast, exogenous C16-ceramide was shown to promote glycolysis. This opposite effect could be explained by the reduction of endogenous C16-ceramide levels, which was induced by exogenous C16-ceramide treatment [14].

Finally, GCS, which catalyses the first committed step in the synthesis of most glycosphingolipids, i.e., the transfer of glucose to ceramide to form GlcCer, is also able to control tumorigenic capability of melanoma cells. Indeed, antisense oligonucleotide targeting the Ugcg gene, encoding GCS, reduced tumorigenicity of MEB4 murine melanoma cells [68]. Similarly, the inhibitor of GCS, OGT2378, inhibited MEB4 melanoma tumour growth in a syngeneic, orthotopic murine model [69]. In agreement with these findings, we previously showed that overexpression of GBA2 in melanoma cells, an enzyme able to degrade GlcCer into ceramide, reduced tumour cell growth both in vitro and in vivo by triggering ER stress-induced apoptosis [70]. Of note, GBA2 gene is downregulated in melanoma cells as compared to melanocytes [70]. Altogether, these observations demonstrate that the transformation of ceramide into GlcCer facilitates melanoma cell proliferation.

Interestingly, sialic acid-containing glycosphingolipids, i.e., gangliosides, can also regulate melanoma cell proliferation. First, treatment of SKMEL-28 melanoma cells with the anti-GD3 antibody R24 reduced their growth in vitro and decreased their tumorigenicity when injected in immunodeficient mice [71]. Second, Furukawa et al. demonstrated that GD3 increased the proliferation of GD3 synthase-overexpressing melanoma cells [72]. In these settings, GD3 mediated the convergence of several pro-tumoral signals, including those induced by hepatocyte growth factor (HGF) and the receptor tyrosine kinase c-MET, notably promoting cell proliferation [73].

Altogether, these data illustrate that SL metabolism alterations, which redirect ceramide metabolism towards S1P or GD3 production, can promote melanoma cell proliferation and survival in response to drugs.

#### **4. Role of the Sphingolipid Metabolism in Melanoma Progression**

#### *4.1. SL Metabolism Regulates Melanoma Cell Adhesion*

Cell junctions, which are crucial for the communication between neighbouring cells and with the extracellular environment, can be divided into three major classes: anchoring junctions (including adherens junctions, desmosomes, hemidesmosomes and focal adhesions), tight junctions and gap junctions. In the epidermis, cadherins are the major adhesion molecules, especially involved in the composition of desmosomes and adherens junctions [74], whereas integrins are the major component of hemidesmosomes [75]. Among cadherins, E-cadherin mediates the adhesion between melanocytes and keratinocytes allowing keratinocytes to control cell growth and dendricity of melanocytes [76]. E-cadherin expression is lost in melanoma cells during the first steps of tumour progression [77]. Interestingly, when E-cadherin expression is restored, keratinocytes recover control of melanoma cells thus preventing tumour progression [78].

Here, we review studies documenting the role of the SL metabolism in the control of the expression of adhesion molecules as well as melanoma cell adhesion capacity. Previous reports have shown that E-cadherin loss was observed in SphK1-overexpressing cancer cells [79,80]. S1P-induced E-cadherin downregulation could be mediated by S1P2 and S1P3, as shown in alveolar epithelial cells [81] and lung fibroblasts [82]. This phenomenon could be indirect in melanoma cells as the SphK1/S1P pathway is able to stimulate TGF-β1 production [62], which may trans-activate S1P2 and S1P3 [82]. Interestingly, overexpression of S1P2, but not S1P1, in B16F10 melanoma cells resulted in the inhibition of the small GTPase Rac activity as well as tumour progression in mice [83]. Importantly, Rac is crucial to create E-cadherin-dependent cell-cell contacts [84].

Moreover, downregulation of AC in melanoma cells induced E-cadherin loss and, inversely, increased expression of the epithelial-mesenchymal transition (EMT)-associated protein TWIST1, which is in accordance with a more aggressive phenotype [19].

Finally, GD3 was shown to favour the recruitment of integrins through glycolipid-enriched microdomains in GD3 synthase-overexpressing melanoma cells. Under these conditions, melanoma cell adhesion to the extracellular matrix (ECM) was increased [85]. Similarly, Ohmi et al. demonstrated that cell adhesion increased in GD2 synthase-overexpressing melanoma cells as compared control cells [86].

Thus, SL alterations appear to impact on melanoma cell adhesion, particularly through E-cadherin loss, which promotes melanoma progression.

#### *4.2. SL Metabolism as a Determinant of Melanoma Plasticity*

To colonise distant organs, tumour cells need, besides losing their cell junctions, to acquire invasive capacities. In skin cancers, EMT plays a key role in this process. This fundamental mechanism allows epithelial cells to gain mesenchymal features, increasing their migration and invasion abilities (for review, see [87]). However, unlike other skin cancers, melanoma does not arise from epithelial cells but from neural crest-derived melanocytes. For this reason, EMT stricto sensu cannot be considered in melanoma progression. Nevertheless, an EMT-like phenomenon has been described, in which melanoma cells can dynamically and reversibly switch between a proliferative and an invasive state; this is known as "phenotype switching". Indeed, the microarray analysis of DNA from different human melanoma cell lines allowed Hoek et al. to determine a transcriptional signature representative of metastatic cell behaviour. The authors indeed demonstrated that MITF is one crucial actor in this switch, particularly in maintaining the proliferative state [88].

MITF represents a melanocytic lineage-specific transcription factor that regulates melanocyte differentiation, function and survival as well as melanoma progression [89]. MITF regulates pigment cell-specific transcription of genes encoding melanogenic enzymes such as TYR, DCT and TYRP1, as well as proteins involved in melanosome formation and maturation such as Melan-A, Premelanosome Protein and G Protein-Coupled Receptor 143. As a matter of fact, MITF expression and activity are modulated by a range of activators and suppressors operating at transcriptional, post-transcriptional and post-translational levels (for review, see [90]). MITF function has been tightly connected to melanoma cell plasticity. It is now well accepted that melanoma cells expressing moderate to high levels of MITF proliferate rapidly and are poorly invasive, whereas melanoma cells characterised by low MITF levels grow more slowly and are more invasive. Thereby, low levels of MITF correlate with a worse prognosis for melanoma patients [19,89,91].

Interestingly, numerous studies showed that the SL metabolism regulates MITF expression. Firstly, A-SMase expression has been shown to induce ERK-mediated MITF degradation by the proteasome. Therefore, the loss of A-SMase observed during melanoma progression accounts for the upregulation of MITF as well as for some of its downstream targets CDK2, Bcl-2 and c-MET [34]. Secondly, AC ablation by the CRISPR-Cas9 technology, which is associated with the accumulation of long-chain saturated ceramides, led to a strong downregulation of MITF expression in human A375 melanoma cells, reducing their ability to form cancer-initiating cells and to undergo self-renewal [65]. Furthermore, exogenous addition of C2-ceramide was shown to reduce MITF expression in human melanocytes [35]. Reciprocally, we demonstrated that MITF expression increased in AC-overexpressing melanoma cells. However, at variance with Lai et al., we observed that melanoma cells expressing AC at high levels displayed a proliferative phenotype as compared to cells with low expression of AC that exhibited high mobility and gain of mesenchymal features [19]. Moreover, using a ChIP-Seq database, we identified AC as a new target of MITF, demonstrating that MITF and AC are part of a positive feedback loop.

SL metabolism could also modulate MITF levels by acting on signalling pathways known to regulate its expression in melanoma cells. For instance, canonical Wnt signalling through the Wnt/β-catenin pathway is a critical activator of MITF expression in melanoma cells [92], and deactivation correlates with a higher metastatic potential [88]. Interestingly, exogenous sphingosine has been shown to reduce nuclear and cytosolic β-catenin expression in SW480 and T84 colon cancer cells [93]. In accordance, pharmacological inhibition of SphK1 with SKI-II was associated to a decreased β-catenin expression in human hepatoma carcinoma [94]. As expected, FTY720, a sphingosine analogue known to inhibit SphK1 [95], led to the reduction of β-catenin as well as MITF expression in melanocytes [96].

Moreover, a switch in EMT-associated transcription factors (EMT-TFs) occurs in melanoma and drives tumour progression. This dynamic network includes Snail, Zeb and Twist families, which are major repressors of epithelial genes, and, conversely, major activators of mesenchymal genes (for review, see [87]). In particular, a reduced expression of ZEB2 and SNAIL2 in favour of an increased expression in ZEB1 and TWIST1 was linked to E-cadherin loss, increased invasion properties and poor clinical outcomes in human melanoma [97]. The EMT-TFs switch was associated with a reduction of MITF expression. Indeed, ZEB1 and TWIST1 have been shown to downregulate MITF whereas SNAIL2 or ZEB2 induce MITF expression, demonstrating that these EMT-TFs act as key players in melanoma phenotype switching.

Recent studies identified a strong connection between SL metabolism and EMT-TFs. Indeed, reduced expression of CerS6, which decreased the levels of intracellular C16-ceramide, was associated with an increased expression of SNAIL2 in SW480 colon cancer cells [98]. Hence, by controlling the expression of EMT-TFs or by altering plasma membrane fluidity, C16-ceramide could affect cancer cell motility [99]. Another SL-metabolising enzyme could also account for the effects of ceramide on EMT-TFs. Indeed, SMS2, which produces SM from ceramide, seems to stimulate the expression of mesenchymal markers and enhance migration and invasion of MCF-7 and MDA-MB-231 breast cancer cell lines. Interestingly, SMS2 expression was higher in metastatic breast cancer than in non-metastatic tumours. Mechanistically, SMS2 was shown to activate the canonical TGFβ/SMAD signalling pathway leading to the expression of its downstream target Snail [100]. In addition, previous studies have demonstrated that some ganglioside-metabolising enzymes are connected with EMT-FTs and gangliosides play a critical role in EMT [101]. For instance, ZEB1 was reported to be a direct regulator of the GM3 synthase gene (St3gal5) in mammary epithelial NM18 cells. ZEB1 also impaired the expression of miR-200a, a microRNA targeting the 3- UTR GM3 synthase mRNA. Knockdown of GM3 synthase partly mimicked the effects of ZEB1 inhibition, leading to increased expression of cell junction components such as E-cadherin as well as intercellular adhesion [102]. Moreover, overexpressing TWIST or SNAIL1 in transformed human mammary epithelial cells enhanced the expression of GD3 synthase [103]. GD3 synthase knockdown reduced breast cancer stem cell-associated properties and completely abrogated tumour formation in vivo. In accordance, other studies have shown that ceramide glycosylation by GCS was enhanced in breast [104] and colon [105] cancer stem cells and GCS inhibition significantly decreased the expression of ZEB1 and β-catenin [105]. Whether GCS, a ganglioside-metabolising enzyme, CerS6 or SMS2 modulates EMT-TFs in melanoma cells remains to be evaluated.

Furthermore, numerous studies have shown that TGF-β is a strong promoter of EMT in many tumours [106,107], including melanoma [88]. TGF-β signalling inhibits MITF expression through PAX3 repression and GLI2 activation. Many studies reported that the SphK1/S1P and the TGF-β signalling are interconnected. Indeed, through its binding to S1P receptors, S1P was shown to promote TGF-β receptor trans-activation leading to Smad phosphorylation and cell migration [108–110]. Additionally, in different cancers including melanoma, S1P was reported to increase TGF-β expression and secretion [62,111,112]. Inversely, TGF-β was able to increase SphK expression and activity, which were essential to control the effects of TGF-β on extracellular matrix remodelling, cell migration and invasion [113,114].

Finally, TEAD transcription factors (TEADs) were identified as key regulators of the invasive state in melanoma [115]. TEADs need coactivators such as YAP and TAZ, which are known effectors of the Hippo pathway. This pathway has been shown to modulate Wnt and TGF-β signalling and confer pro-invasive properties in melanoma [116]. Strikingly, multiple studies have identified S1P as an activator of YAP through S1P2 signalling [117–119]. Similarly, a recent study demonstrated that inhibition of SphK1 using the PF-543 inhibitor could inhibit TGFβ-induced activation of YAP [120]. As anticipated from its close structural similarity with S1P, SPC also regulated the Hippo pathway via S1P2, in a rather unclear manner as it could both inhibit and activate YAP [121].

To summarise, tight connections have been reported between SL metabolism and key players of the phenotype switching in melanoma including transcription factors such as MITF, EMT-TFs, TEADs and fundamental signalling pathways such as Wnt, TGF-β and Hippo. These observations further highlight the importance of the SL metabolism in melanoma progression.

#### *4.3. SL Metabolism as a Major Regulator of Melanoma Aggressiveness*

As discussed above, melanoma aggressiveness depends on the balance between its proliferative potential and migratory/invasive properties. S1P was reported either to activate or to inhibit melanoma cell migration depending on S1P receptor subtypes. Indeed, whereas S1P inhibited cell migration, with the concomitant inhibition of Rac and stimulation of RhoA, in S1P2-expressing B16F10 cells, it stimulated cell migration of S1P1-overexpressing cells, demonstrating a receptor subtype-specific action of S1P on melanoma cells [83]. The inhibitory effects of S1P were reversed by the S1P2-selective antagonist JTE013, which stimulated Rac and migration of B16F10 cells overexpressing either S1P1 or S1P3 [122]. Similar results were obtained in B16F10 cells treated with SPC instead of S1P [122].

In addition, AC overexpression in melanoma cells decreased tumour cell motility, whereas AC silencing had the opposite effect, as was observed for MITF [19]. We recently demonstrated that low AC expression was associated to increased FAK phosphorylation and relocation at focal adhesions instead of cytoplasm. This phenomenon led to increased expression of integrin β5 (ITGβ5) and integrin αV (ITGαV), which play a critical roles in the migratory and invasive capacity of cancer cells. As a result, the melanoma invasive behaviour induced by AC inhibition was reduced using an ITGαVβ5 blocking antibody [19].

Another study reported that lung metastases were reduced in A-SMase-deficient mice injected with B16-F10 melanoma cells. Treating B16F10 cells with exogenous A-SMase or C16-ceramide before inoculation restored lung metastatic lesions in A-SMase-deficient mice. Mechanistically, melanoma cells were shown to activate A-SMase in platelets, leading to ceramide production, which favoured the clustering and activation of α5β1 integrins at the surface of melanoma cells and therefore tumour cell adhesion in the lungs [123]. As the expression of melanoma A-SMase negatively correlates with tumour aggressiveness [34], one could speculate that A-SMase expression acts as a key factor that controls melanoma cell invasion and adhesion into the metastatic niches.

Gangliosides also likely contribute to melanoma cell dissemination. As a matter of fact, the level of GM3, which has been described as one of the major gangliosides in melanoma [124], increased in murine metastatic melanoma [125], suggesting a role for GM3 in tumour aggressiveness. Indeed, the addition of GM3 to B16LuF1 melanoma cells, i.e., B16-melanoma cells of lower metastatic potential to lungs, increased their dissemination capacity once injected in mice [126]. Liu et al. also described that de-N-acetyl GM3 (d-GM3), a derivative of ganglioside GM3, was mainly found in metastatic melanomas but not in benign nevi or most primary melanomas. d-GM3 expressing melanoma cells possess increased migratory and invasive capacities as compared to melanoma cells lacking d-GM3. Mechanistically, d-GM3 stimulated MMP-2 expression via the urokinase-like plasminogen activator (uPA) receptor [127].

Some studies indicated that GD3 also stimulates melanoma cell invasion. Indeed, human melanoma GD3-positive cells showed a markedly increased cell invasion potential as compared to GD3-negative cells. The invasive activity induced by GD3 was shown to be mediated by p130Cas or paxillin, two components of the focal adhesion cytoskeleton [72]. More recently, Ohmi et al. compared the effect of GD3 and GD2 on melanoma progression. Using GD3-high or GD2-high melanoma cells, obtained by overexpressing the respective glycosyltransferases involved in their production, they demonstrated that GD2 enhanced the adhesion properties of melanoma cells, while GD3 stimulated their invasive capacities. These findings led the authors to propose that GD2 would rather act at the primary and metastatic sites in order to promote cell proliferation and dissemination, while GD3 would favour melanoma cell invasion in order to reach a metastatic niche [86].

Finally, emerging literature indicates that tumour exosomes actively participate in tumour invasiveness and favour the formation of pre-metastatic niches in various types of cancer, including melanoma [128–130]. Exosomes are small extracellular vesicles (EVs) that originate from the fusion of multivesicular bodies with the plasma membrane and convey their cargo towards target cells. They carry transmembrane and cytosolic proteins, DNA and small RNAs [131]. In vitro, melanoma-derived exosomes were shown to promote the EMT-like processes in primary melanocytes. This effect occurred in an autocrine/paracrine fashion and was mediated by the microRNA Let-7i [132]. In mice, B16-F10-derived exosomes demonstrated preferential homing to lymph nodes and facilitated the seeding of intravenously injected parental cells [133]. In particular, Peinado et al. showed that, through the receptor c-MET, B16-F10-derived exosomes can educate bone marrow-derived cells, promoting angiogenesis, vascular leakiness, the growth of primary tumours and metastasis [128].

It is well known that ceramide, generated by the neutral SMase2 (nSMase2) on the cytosolic leaflet of endosomal membranes, is involved in the budding of exosomes [134]. Mechanistically, the cone-shaped structure of ceramide could induce spontaneous negative curvature by creating an area difference between the membrane leaflets [135]. Moreover, a decrease in the activity of nSMase2 induced by the GW4869 compound, resulted in the reduced release of exosomal miRNAs [136]. By controlling exosomal miRNA secretion, nSMase2 is able to promote angiogenesis as well as metastasis [137]. Furthermore, Kajimoto et al. also showed that S1P, produced by SphK2 but not Sphk1, can regulate the cargo content in exosomes [138,139] probably through the Gβγ subunit of Gi proteins coupled with S1P1 [140]. Whether nSMase2- and/or SphK2-dependent exosome formation modulates melanoma progression remains to be investigated.

All these findings clearly establish a close relationship between SL metabolism and melanoma invasion and suggest that SL metabolism could be therapeutically targeted in order to improve the outcome of melanoma patients.

#### **5. Role of SL Metabolism in the Immune Response to Melanoma**

Melanoma cells harbour an aberrant antigenic profile, which allows for an anti-tumour immune response [141]. Despite their high immunogenicity, melanoma cells eventually evade the immune system, grow and metastasise [142,143]. A growing body of evidence in the literature indicates that SLs regulate various immune processes. Thus, deciphering the role of SL metabolism in melanoma immune escape is of great clinical interest.

#### *5.1. S1P in Lymphocyte Tra*ffi*c and Di*ff*erentiation*

Lymphocytes sense S1P concentration via S1P1 [144,145] allowing their egress from the thymus and lymph nodes to peripheral tissues [146]. S1P1 expression is modulated cyclically during lymphocyte traffic, depending on the local S1P concentration: it is downregulated in the blood, upregulated in secondary lymphoid organs (SLO) and downregulated again in the lymph [147]. CD69, an early activation marker on lymphocyte surface, induces S1P1 internalisation and degradation [148], sequestering lymphocytes in SLO [149] and peripheral tissues [150]. S1P1 downregulation is necessary to establish a long-term memory in the skin [151,152] and CD69 is one of the markers (with CD103) for tissue-resident memory cells (TRM) [153], which play a critical role in melanoma immunosurveillance [154–156].

Drouillard et al. proposed that the S1P1/S1P2 ratio dictates the migration of T cells, as S1P2 inhibited the chemo-attraction of peripheral T cells [157]. Sic et al. reported that human B cells also migrate towards S1P in an S1P1-dependent manner that is inhibited by CD69 expression [158]. Interestingly, egress of natural killer (NK) cells from the bone marrow and SLO is mediated by the expression of S1P5 [159,160], which is regulated by T-box transcription factor TBX21 [161]. Increased S1P5 expression as well as downregulation of CXCR4 during NK differentiation is necessary for their egress from the bone marrow [162].

SphK activity and S1P1 expression were shown to mediate differentiation of CD4<sup>+</sup> T cells to Th1 cells and inhibit induced Treg (iTreg) generation [163]. In accordance, in T cell-specific S1p1-transgenic mice, S1P1 oriented the differentiation of CD4<sup>+</sup> towards the Th1 lineage when antigen-activated. Moreover, S1P1 overexpression impaired the maintenance of Foxp3 expression in naïve TGF-β-treated CD4<sup>+</sup> T cells. The differentiation of naïve CD4<sup>+</sup> T cells towards Th1 or iTreg appeared to be reciprocal, driven by the S1P1-mTOR axis, and dependent on the SphK activity as demonstrated with the SphK

inhibitors N,N,-dimethylsphingosine (DMS) and SKI. Similarly, CD4<sup>+</sup> T cells deficient for Sphk1 showed a lesser Foxp3 expression when cultured with IL-2 and TGF-β [164].

#### *5.2. S1P Impairs the Immune Response in Melanoma*

We reported that melanoma SphK1 plays a key role in the recruitment and phenotypic switch of TAM notably promoting their commitment to a pro-tumoral M2-like phenotype [62]. Moreover, we recently showed that high SphK1 expression in melanoma cells was associated with shorter progression-free and overall survivals in melanoma patients treated with anti-PD-1-based immunotherapy. In mice, SphK1 knockdown in melanoma tumours potently reduced the production of a number of immunosuppressive cytokines including TGF-β [22,62], limiting Treg tumour infiltration. Under these conditions, the response of melanoma cells to anti-PD-1 or anti-CTLA-4-based immunotherapy highly increased [22]. Interestingly, Chakraborty et al. also reported that tumour-infiltrating lymphocytes display higher Sphk1 expression as compared to splenocytes in B16-F10-bearing mice [164]. Melanoma antigen-specific T cells deficient for Sphk1 (pMel-SphK1−/<sup>−</sup> T cells) were shown to maintain a central memory phenotype and have a reduced propensity to differentiate into Treg as compared to wild-type T cells (pMel T cells). Tumour growth was significantly slower upon adoptive transfer of pMel-SphK1−/<sup>−</sup> T cells, as compared to mice injected with wild-type pMel T cells [164].

Recent findings also show that S1P secretion, via the S1P transporter Spinster Homologue 2 (Spns2), reduced CD8<sup>+</sup> T cell function and therefore promoted lung metastasis. Indeed, the deletion of Spns2, either globally or in a lymphatic endothelial cell-specific manner, was associated with an increased ratio of effector T-cells to immunosuppressive Tregs, in the lungs of Spns2-deficient mice intravenously injected with B16F10 or HCmel12 murine melanoma cells. This resulted in a reduced pulmonary metastatic burden as compared to what was observed in wild-type animals [165].

#### *5.3. Ceramide and Its Derivatives in the Immune Response*

Several studies have shown that ceramide metabolism could regulate the immune response in different melanoma models. Firstly, A-SMase-deficient B16-F1 melanoma cells engrafted in mice display an inflammatory TME and are infiltrated by high levels MDSCs and Tregs and low levels of DCs. A-SMase overexpression in these cells restores CD8<sup>+</sup> and CD4<sup>+</sup> T cells and DCs infiltration while reducing levels of infiltrating MDSCs and Tregs, thereby reducing tumour growth [166].

Secondly, KRN7000, a synthetic alpha-galactosylceramide [167], showed promising results in enhancing NK, NKT, CD8<sup>+</sup> T cells and M1 infiltration in the syngeneic murine B16 metastatic melanoma model [168] but further investigation needs to be conducted.

Thirdly, it was recently reported that liposomes enriched in C2-ceramide were shown to reprogram the immune TME in a PKCζ-dependent manner in B16-F10-bearing mice. Under these conditions, TAMs shifted towards an M1 phenotype and CD8<sup>+</sup> and Th1 cells infiltration was enhanced while intra-tumour MDSCs and Tregs levels were reduced [169].

Finally, gangliosides also represent attractive targets for immunotherapies as they are abundant in melanoma cells [24] and recognised by NKT cells [170]. GM2 [171] and N-glycolyl GM3 (NGcGM3), in particular, have been the main gangliosides used as targets for the development of anti-melanoma antibodies and vaccine [172,173]. In addition, gangliosides are also known to be shed by melanoma cells [174,175] and exert a pro-apoptotic effect on DCs [176,177]. 3F8, a monoclonal anti-GD2 mAb, demonstrated anti-proliferative and pro-apoptotic activity in human melanoma cell lines [178] but clinical studies focused on neuroblastoma [179] and medulloblastoma [180] patients. More recently, GD2 has been considered as a promising target for the treatment of melanoma patients using either CAR-T cell therapy [181] or the immunocytokine hu14.18-IL2, an anti-GD2 humanised mAb linked to two molecules of IL-2 and administered to patients with recurrent resectable stage III or IV melanoma [182].

#### *5.4. Melanoma-Derived Exosomes Are Vectors of Immunosuppression*

As mentioned above, Trajkovic et al. showed that the production of ceramide by nSMase 2 was part of the mechanisms involved in exosome budding [134]. In addition to favouring progression and metastasis, melanoma-derived small extracellular vesicles, often defined as exosomes, also carry immunomodulatory molecules that impair anti-tumour immune responses. In vitro, exosomes released from B16F0 murine melanoma cells inhibited the proliferation of T cells by delivering PTPN11(SHP-2) mRNA and protein [183]. B16F10-derived exosomes can also activate the mitochondrial apoptotic pathway of CD4<sup>+</sup> T cells in vitro and in vivo, thereby increasing tumour growth and reducing T cell infiltration [184]. The authors proposed that the miRNA cargo of exosomes (e.g., miR-690) inhibited the expression of anti-apoptotic proteins in CD4<sup>+</sup> T cells. In addition, small EVs produced by A375 human melanoma cells were shown to be able to reduce MHC class I molecules to the cell surface of primary human monocytes and THP-1 cells and downregulate the expression of endogenous MHC class I and II molecules in DCs [185]. Exosomes from metastatic melanoma-derived cell lines inhibited TCR signalling and cytokines secretion in CD8<sup>+</sup> T cells by transferring an array of miRNA cargo [186]. Moreover, tumour-derived exosomes harvested from melanoma patients' plasma were shown to induce the apoptosis, inhibit proliferation and decrease the activation of CD8<sup>+</sup> T cells. They were also able to downregulate NKG2D expression on NK cells [187]. In metastatic melanoma patients, circulating exosomal PD-L1 suppressed CD8<sup>+</sup> T cell activity. The authors reported that the pre-treatment level of circulating exosomal PD-L1 was a better predictor of clinical response to anti-PD-1 therapy than total circulating PD-L1 [188]. Importantly, Poggio et al. showed that Pdl1 knockout or exosome depletion by knocking out Smpd3, the gene encoding nSMase2, was sufficient to restore the anti-tumour immune response and to induce an efficient anti-tumour immune-memory response in the murine TRAMP-C2 prostate cancer model [189].

The relationship between SL metabolism and exosome-mediated immunosuppression in melanoma is not well understood, yet it could be a major mechanism of resistance to immunotherapy and thus deserves further investigation.

#### **6. Potential Therapeutic Strategies for Melanoma Patients**

Historically, when tumour resection was not possible or failed, chemotherapy was used to treat melanoma. dacarbazine (DTIC) has been approved as first-line treatment for advanced-stage melanoma and has remained for more than 30 years the standard chemotherapy despite no clear overall survival benefits [190–192].

The identification of melanoma driven mutations such as BRAF V600E allowed for a real breakthrough in the treatment of patients with metastatic melanoma. The emergence of BRAF targeted agents such as vemurafenib [54] and dabrafenib [55] allowed tremendous progresses in the field of personalised medicine and demonstrated survival benefits in metastatic melanoma patients as compared to dacarbazine-treated patients. Subsequently, the MEK inhibitor was also approved as treatment for this pathology as it showed survival benefits for patients displaying the BRAF V600E mutation [193]. Combination therapy using BRAF and MEK inhibitors such as cobimetinib [56] is nowadays one of the first line treatment for patients with BRAF V600E metastatic melanoma. This treatment results in higher rates as well as extended duration of response and decreases the cutaneous toxicities observed with the BRAF inhibitor monotherapy. Unfortunately, such therapeutic approaches remain constrained by the inevitable emergence of resistance to single-pathway blockade [194].

Immune checkpoint blockade (ICB) was the first therapeutic strategy to provide sustained responses and survival for advanced melanoma patients, even after treatment discontinuation [195–197]. Administering monoclonal antibodies targeting the immune checkpoint PD-1, alone or in combination with anti-CTLA-4 blocking antibodies is, to date, the standard of care for advanced melanoma patients. Independently of BRAF mutation status, patients treated with the anti-PD-1 and anti-CTLA-4 combo achieve at five years a progression-free survival and an overall survival of 36% and 52%, respectively [198]. Unfortunately, half of the patients do not respond or develop early resistance to ICB

and exhibit severe immune-related adverse events (IRAE) [199]. Although treatment discontinuation due to adverse events seems not to affect the outcome for patients treated with ICB combination therapy [198], IRAEs tend to be associated with a better outcome for patients treated with anti-PD1 monotherapy [200].

Targeted therapies and ICB have deeply changed therapeutic management of patients with metastatic melanoma but all these therapeutic approaches still need improvement. Understanding the mechanisms that underlie resistance to these treatments is of utmost importance to improve the outcome of melanoma patients.

Here, we review some studies, which identified alterations in the SL metabolism as a cause of melanoma resistance to treatment (Table 2) and other studies using SL-related molecules as monotherapy or combined therapy to fight melanoma (Table 3).


**Table 2.** SL-metabolising enzymes regulate the response of melanoma to therapy.


**Table 2.** *Cont*.



#### **7. Conclusions**

As developed in this review, changes in SL metabolism that contribute to melanomagenesis, tumour progression and therapeutic resistance are multiple (Figure 4).

The action of sphingolipids (via the enzymes that control their metabolism, and transporters) is likely mediated by modifications in key regulatory processes including the phenotypic switch and EV-mediated cell-cell communication. Interestingly, these metabolic alterations could be envisioned as potential biomarkers and be exploited to better characterise tumour progression in melanoma patients. As a matter of fact, we identified that a reduced expression of SMS1 was significantly associated with a worse prognosis in metastatic melanoma [29]. AC was also identified as a potential biomarker for the prognosis of melanoma [211]. Moreover, we recently demonstrated that human invasive melanoma cells had lower AC levels and activity than proliferative melanoma cells [19]. In accordance, high AC expression was observed in node-negative stage II melanomas [18].

**Figure 4.** Role of sphingolipid metabolism in melanoma progression and immune response. SL metabolites and SL-metabolising enzymes whose levels and expression are increased, decreased or implicated are marked in red, blue or black, respectively. AC, acid ceramidase; CD8+, CD8<sup>+</sup> T cells; Cer, ceramide; CerS, ceramide synthase; GC, glucosylceramidase; GCS, glucosylceramide synthase; M1, M1 macrophages; M2, M2 macrophages; MDSC, myeloid-derived suppressor cells; NK, natural killer cells; NKT, natural killer T cells; S1P, sphingosine 1-phosphate; S1P1/2/3, S1P receptor type 1/2/3; SMases, sphingomyelinases; SMS, sphingomyelin synthase; SPC, sphingosylphosphorylcholine; SphK, sphingosine kinase; SPL, S1P lyase; Th1, Th1 CD4<sup>+</sup> T cells; Treg, tegulatory T cells.

It is also interesting to note that a strong association between increased serum levels of gangliosides and high Breslow index or high Clark level as well as the presence of ulceration has been reported in melanoma patients, suggesting that circulating gangliosides may serve as potential markers for melanoma staging [212].

Monitoring the expression of SL-metabolising enzymes as well as SL levels could also be used to track the response to therapy in melanoma. Indeed, we previously showed that AC expression was associated to the response of melanoma cells to dacarbazine. Whereas overexpression of AC conferred resistance to dacarbazine, AC downregulation sensitised tumour cells to the drug [66]. DTIC triggered AC degradation and this effect was accompanied with an increased ceramide/S1P ratio. Our recent results also reveal that a distinct SL profile, i.e., a tendency for increased very long-chain ceramide species, was observed in the plasma of patients with melanoma who achieve a response to a BRAF-targeted therapy as compared with patients with progressive disease [64]. Finally, we recently discovered that melanoma patients with low SphK1 expression had significantly longer progression-free survival and overall survival than those with high SphK1 expression and patients with high SphK1 expression mostly failed to respond to anti-PD-1 therapy. These findings support the

hypothesis that SphK1 expression represents a potential biomarker to predict tumour progression and resistance to anti-PD-1 in metastatic melanoma patients [22].

It would now be of great interest to evaluate the possible association between these SL metabolic alterations and the mutation status of oncogenes such as BRAF or NRAS as well as immune responses in metastatic melanoma patients. This will be performed in patients treated with anti-PD-1 in combination or not with anti-CTLA-4 in a prospective clinical trial (IMMUSPHINX: NCT03627026) we are currently conducting in our institute.

**Funding:** This research was funded by INSERM, Paul Sabatier University, Fondation Association pour la Recherche sur le Cancer [B. Ségui (R19179BB) and N. Andrieu-Abadie (R18167BB)], Société Française de Dermatologie [N. Andrieu-Abadie (R18126BB)], Fondation Toulouse Cancer Santé [B. Ségui (R19225BB)] and Institut National du Cancer [N. Andrieu-Abadie (R19243BP)]. The work also received funding from the Transcan-2 Research Program, which is a transnational R&D program jointly funded by national funding organisations within the framework of the ERA-NET Transcan-2 [N. Andrieu-Abadie (TRANS201601250)]. The APC was funded by Fondation Association pour la Recherche sur le Cancer [N. Andrieu-Abadie (R18167BB)]. L.C. is a recipient of a fellowship from Fondation pour la Recherche Médicale.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Review* **Transcriptional Regulation of Sphingosine Kinase 1**

#### **Joseph Bonica 1, Cungui Mao 2,3, Lina M. Obeid 1,2,3,**† **and Yusuf A. Hannun 1,2,3,\***


Received: 5 October 2020; Accepted: 5 November 2020; Published: 8 November 2020

**Abstract:** Once thought to be primarily structural in nature, sphingolipids have become increasingly appreciated as second messengers in a wide array of signaling pathways. Sphingosine kinase 1, or SK1, is one of two sphingosine kinases that phosphorylate sphingosine into sphingosine-1-phosphate (S1P). S1P is generally pro-inflammatory, pro-angiogenic, immunomodulatory, and pro-survival; therefore, high SK1 expression and activity have been associated with certain inflammatory diseases and cancer. It is thus important to develop an understanding of the regulation of SK1 expression and activity. In this review, we explore the current literature on SK1 transcriptional regulation, illustrating a complex system of transcription factors, cytokines, and even micro-RNAs (miRNAs) on the post transcriptional level.

**Keywords:** sphingosine kinase 1; SK1; microRNA; transcription factor; hypoxia; long non-coding RNA

#### **1. Introduction**

Sphingolipids are a class of cellular lipids that are involved in both maintaining cell structure and mediating cellular signaling processes. Sphingomyelins are important in cell membrane structure, while ceramides, sphingosine, and sphingosine-1-phosphate (S1P) are involved in cell signaling and are thus referred to as bioactive sphingolipids [1,2]. Ceramides can be produced by one of three pathways. The de novo pathway begins with the activity of serine-palmitoyl transferase to eventually generate ceramide. The hydrolytic pathway generates ceramide (and sphingosine) from the hydrolysis of complex sphingolipids. The salvage pathway involves re-generating ceramide and sphingolipids via salvage of sphingosine generated in the lysosome and then re-incorporated into ceramide. All three bioactive sphingolipids are known to have important signaling consequences in both healthy and diseased cell states. Ceramide and sphingosine are known to induce cell death, senescence, and/or differentiation and are generally antiproliferative [1,2]. S1P, on the other hand, is best known to increase cell survival, proliferation, angiogenesis, migration and invasion, and immune cell egress [2–4]. S1P typically signals through one of five G-protein coupled receptors, named S1PR1-5, although there is evidence it also signals intracellularly [5]. Due to the delicate and sometimes opposing nature of the signaling processes utilizing sphingolipids, the absolute and relative concentration of these lipids are tightly controlled [6]. As such, the enzymes involved in the metabolism of sphingolipids are also closely regulated, with their expression and activities modulated by the concentrations of lipids in the cell.

Sphingosine kinases catalyze the phosphorylation of sphingosine into S1P. There exist two known major isoforms of sphingosine kinase, products of the two distinct genes, sphingosine kinase 1 (SK1) and sphingosine kinase 2 (SK2). SK1 is typically localized to and more active at the plasma

membrane [7,8], while SK2 is localized to the nucleus upon activation [9,10]. SK1 is the more closely studied of the two isozymes, and understanding its regulation and activity is important due to the pro-proliferative and pro-survival activities of its product S1P; this is especially true in cancer. SK1 is known to be highly upregulated in many cancers, such as breast cancer, colon cancer, head and neck cancer, and glioblastoma [11–14], and its upregulation is associated strongly with poor prognosis and increased cancer metastasis [15–19]. SK1 has also been shown to contribute to chemoresistance in cancer. In CML, for instance, higher expression of SK1 led to disruption in the ratio between C18 ceramide and S1P, contributing to imatinib resistance [20]. In melanoma, high SK1 expression is shown to contribute to resistance to immune checkpoint inhibitors [21]. Metabolically, SK1 drives the formation of S1P while also serving to clear sphingosine, and thus providing an exit from the sphingolipid metabolic network. As such, SK1 may also regulate the levels of ceramide and possibly other upstream sphingolipids. SK1 is thus considered a particularly key regulator of the levels of bioactive sphingolipids. Therefore, it is important to understand the mechanisms of SK1 regulation, to generate further study in cancer, and to identify potential drug targets.

SK1 regulation is well studied and has been discussed in several reviews [7,22,23]. However, it should be noted that work looking specifically at transcriptional regulation of the enzyme is somewhat limited. The gene for sphingosine kinase 1 is located on chromosome 17q25.2 [24]. Work on the rat sphk1 gene showed 6 exons, although 6 alternative first exons were also detected [25]. The human SPHK1 gene is 7 exons in length and 11,276 bp long.

The current body of work reveals that SK1 expression is regulated by several different transcription factors, indicating the enzyme's importance in different signaling pathways. These pathways encompass such important conditions as neuronal growth, hypoxia, ischemia, and cancer. It is known that three primary isoforms of SK1 exist, namely, SK1a, SK1b, and SK1c [25–27]. The expression and distribution of these isoforms is partly governed by the methylation of CpG islands in the SK1 promoter [25]. At 384 amino acids, SK1a is the smallest isoform and the most highly expressed; SK1b is 398 amino acids long, and SK1c is 470 [27]. While not much is known about differences in activity or biological effects of these isoforms, there is evidence that SK1a and SK1b interact with different proteins in breast cancer [28]. Additionally, SK1b appears to be resistant to the SK1 inhibitor Ski in prostate cancer cells [29]. Otherwise, the three isoforms differ only in the size of their N-terminal regions [24]. There is also increasing evidence that SK1 regulation is partially governed by microRNAs, or miRNAs, and that dysregulation of miRNAs in cancer is responsible for higher expression of SK1. In this review, we discuss what is known about SK1 transcriptional and post-transcriptional regulation, what signaling pathways effect and are affected by SK1 regulation, and what further work needs to be done to fully understand regulation of this important enzyme.

#### **2. Transcriptional Regulation-Sp1**

Much work on SK1 transcriptional regulation implicates specificity protein 1, or Sp1, as an important transcription factor. In rat PC-12 cells, it was shown that neuronal growth factor (NGF) led to increased SK1 expression by increasing the expression of Sp1 [30]. In these cells, Sp1 was shown to interact specifically with exon 1d of the SK1 gene. Sp1 has also been shown to regulate SK1 in humans; for instance, Sp1 has been demonstrated to be packaged and transported in exosomes to upregulate SK1 in nearby cells, which protects from ischemia/perfusion injury [31]. Here, Sp1 is shown to interact with the region 0.5–0.6 kb upstream of the SK1 promoter (Figure 1). In a hepatocellular cancer model, blocking of Sp1 activity and expression with the known Sp1 inhibitor peretinoin decreased SK1 levels both in vitro and in vivo [32]. Interestingly, Sp1 is overexpressed in several cancers, such as breast, pancreatic, lung, glioma, and thyroid [33]. Many of these cancers also show upregulated SK1. Much like SK1, higher levels of Sp1 in these cancers are correlated with increased severity, stage, angiogenesis, and metastasis. This indicates an important relationship between SK1 and Sp1 and implies that increased Sk1 activity and expression can be used as a readout for conditions that increase Sp1 activity and/or expression.

**Figure 1.** Location of transcription factor binding on the SK1 Gene. Approximate locations of *SPHK1* transcription factors in relation to the *SPHK1* promoter region. Transcription factors have been shown to bind both upstream and downstream of the promoter. E2F is known to be important in regulating SK1 in cancer, while Sp1 is associated with upregulating SK1 in cancer and in response to neuronal growth factors. In hypoxic conditions, HIF2a and Lmo2 were shown to upregulate SK1, especially in cancer models. AP-1 upregulates SK1 in response to the cytokine IL-1b. MicroRNAs (miRNAs) have been shown to bind to certain sequences in the 3'UTR of the *SPHK1* transcript.

#### **3. Hypoxia and Ischemia**

Due to S1P's importance in angiogenesis, SK1 has long been studied as an important element in response to hypoxia. As oxygen levels decline, the body needs more blood vessels to move blood and oxygen quickly. This need for further angiogenesis serves to stimulate the upregulation of SK1 in hypoxic conditions. In endothelial cells, hypoxia is curiously shown to regulate SK1 but not SK2 [34], highlighting the former's importance in angiogenesis. SK1 has a hypoxia response element (HRE) in its promoter (Figure 1), and as such, it responds to hypoxia sensitive transcription factors [34]. The regulation of SK1 in response to hypoxia has been demonstrated several times in living systems. For instance, SK1 has been shown to be upregulated in arteries after short periods of hypoxia [35]. Acute and chronic hypoxia have also been shown to upregulate SK1 in human pulmonary smooth muscle cells [36]. SK1 is also implicated in the regulation of two main hypoxia-induced transcription factors (HIFs), HIF1a and HIF2a [37].

Ischemia, which often leads to hypoxia, is known to regulate SK1, leading to both protective and deleterious effects on the involved system. For instance, SK1 was demonstrated to be substantially upregulated in a mouse stroke model, and its upregulation increased inflammatory response and poorer outcome [38]. A large increase in SK1 expression was similarly shown in the area of stroke lesion [39]. SK1 has also been shown to be upregulated in activated microglia, and to play a key role in the inflammatory response to cerebral ischemia-reperfusion (IR) [39–42]. Cerebral IR was shown in brain tissues to upregulate SK1, which in turn increased IL-17A expression in primary microglia [40]. The neuronal injury following cerebral IR in this model was reduced via administration of the SK1 inhibitor PF-543, cementing SK1's role in driving inflammatory injury in this system. Further work demonstrated that SK1 affects IL-17 expression via upstream effects on TRAF2 expression and NFκB activation [41]. Crosstalk between SK1 and TLR2 has also been implicated in the inflammatory response to cerebral IR, as both were shown to be upregulated in microglia after cerebral IR [42]. In addition to PF-543, treatment of mice with fingolimod (a pro-drug functional antagonist of S1PR1) reduced hemorrhagic transformation and stroke injury [39]. These results suggest that targeting SK1 represents a potential treatment option for stroke and post-stroke injury, which have limited treatment options. Interestingly, SK1 upregulation induced by the anesthetic isoflurane is actually protective against intestinal injury in a renal ischemia model known to result in intestinal injury [43]. A separate renal IR injury model in mice showed that the increase in SK1 expression mediated the severity of the injury [44]. Upregulation of SK1 via exposure to conditioned media from mesenchymal stem cells in endothelial colony-forming cells potentiated the revascularization of endothelial colony-forming cells, considered an exciting means of treating infarct damage [45]. However, SK1 is also shown to be upregulated after myocardial infarction, and it contributes to dysfunctional cardiac remodeling and heart failure [46]. The specific transcription factors and regulatory pathways upstream of SK1 upregulation in ischemia do not appear to be widely studied; however, and future research in this area can improve our understanding of SK1's various apparent roles in the ischemic response.

There is evidence that SK2 expression, on the other hand, offers protection against ischemic injury in a potentially compensatory manner. For instance, SK2 has been shown to be important in hypoxic preconditioning, and its activity and expression were required for the protective effects of such conditioning [47]. In cardiomyocytes, hypoxic preconditioning was responsible for increased expression of SK2, which was required to prevent apoptosis in ischemic conditions [48].

Hypoxia is an important response in cancer, due to often low oxygen conditions in the tumor; as such, hypoxic pathways are considered potential therapeutic targets in cancer [49]. SK1 is known to have several roles in cancer-induced hypoxia. Interestingly, most evidence points to HIF2alpha and not HIF1alpha as the primary regulator of SK1 in hypoxia [50,51]. In the U87MG glioma cell line, knockdown of HIF2a but not HIF1a led to downregulation of SK1 [50]. Transfection of cells with the SK1 promoter, however, led to upregulation of SK1 in response to CoCl2 treatment to simulate hypoxia, and several HRE elements were identified in the SK1 promoter [50]. HIF2a has also been associated with SK1 upregulation in clear cell renal carcinoma (ccRC) cells, where again knockdown of HIF2a led to reduced SK1 expression [51].There also exists evidence of a feedback loop between expression of HIF2a and SK1 in certain systems; a different model of ccRC from the one above seems to implicate SK1 in HIF2a regulation [52]. Another transcription factor important in angiogenesis, Lim domain only 2 transcription factor (Lmo2), has been shown to directly upregulate SK1 by binding to a sequence +3986 bases from the promoter [53]. This demonstrates the importance of SK1 in the hypoxic response, and especially the hypoxic cancer response.

#### **4. Cytokines**

SK1 and S1P are known to be involved in several inflammatory signaling pathways [23], and SK1 expression and activity are themselves regulated by several cytokines. In glioblastoma, it was found that SK1 is upregulated via the cytokine IL-1 through a JNK/c-Jun dependent pathway [26]. Since additional data show that high expression of SK1 correlates negatively with glioblastoma prognosis [17], understanding the mechanism of SK1 upregulation in response to cytokines is vital to understanding the disease severity. Interestingly, IL-1B only seems to upregulate the SK1a and SK1c isoforms in this system, with little effect on SK1b. Here, SK1 was shown to be upregulated by the binding of transcription factor AP-1 to the first intron at +587 to +593 [26].

In leukemia macrophage THP-1 cells, LPS initiated toll-like receptor 4 signaling stimulated SK1 expression, which led to the accumulation of IL-6 [54]. While evidence exists for SK1's involvement in inflammatory pathways, evidence for cytokines themselves, leading to SK1 upregulation continues to be somewhat limited. It should be noted though, that cytokines have been also shown to activate SK1 through non-transcriptional mechanisms TNFα, for instance, is known to increase SK1 activity by inducing the phosphorylation of Ser 225, necessary for the trafficking of SK1 to the plasma membrane [55]. SK1 has also been shown to be activated by tumor necrosis factor receptor-associated factor 2 (TRAF2) via direct binding to the protein [56].

Transforming growth factor-B (TGF-B) signaling is known to upregulate and activate SK1, and SK1 activity has been shown to be important in many TGF-B dependent pathways. Treatment of fibroblasts with TGF-B led to increased SK1 expression, which was required for TIMP-1 upregulation [57,58]. TGF-B induced upregulation of SK1 has been shown to be important in coronary artery disease and liver fibrosis [58,59]. In breast cancer, TGF-B driven SK1 upregulation has been implicated in bone metastasis [60]. Despite the apparent link between the well-studied TGF-B signaling pathway and SK1 upregulation, the exact transcription factors and regulatory elements governing SK1's response to TGF-B are currently not well established.

#### **5. E2F and Long Non-Coding RNAs**

In certain cancers, there is evidence of SK1 expression being under the control of transcription factors known as E2F transcription factors. This family of eight transcription factors is known to regulate proliferation, apoptosis, and stress responses [61]. Increased expression of E2Fs result in oncogenic activity in several cancers [62]. In head and neck squamous cell carcinoma, SK1 expression was shown to be downstream of E2F7, which is a direct transcription factor of SK1 [63]. In this cancer, E2F7-driven upregulation of SK1 is associated with anthracycline resistance; inhibition of E2F7 activity reduced SK1 levels and sensitized tumor cells to anthracycline chemotherapies. However, not all E2F family transcription factors regulate SK1 expression through direct actions on its promoter. For instance, the transcription factor E2F1 has been shown in liver carcinoma to regulate SK1 via the long non-coding RNA called HULC, or highly upregulated in liver cancer [64]. The connection between E2F transcription factors and SK1 regulation, especially in cancer, is one that merits further investigation based on these intriguing data.

Long non-coding RNAs, or lncRNAs, are RNA molecules of at least 200 nucleotides in length that do not code for any proteins [65]. While this type of RNA is incompletely understood, it is believed that they are largely regulatory in nature and known to be involved in epigenetic regulation [65,66]. While lncRNA's role in disease is not completely understood, there exists compelling evidence that they are dysregulated in certain cancers [67]. In the case discussed above, the lncRNA HULC sequesters micro-RNA 107, a miRNA targeting E2F1; this preserves E2F1 and allows it to bind to the promoter [68]. As shown in Figure 1, E2F1 binds to the SK1 promoter at −147 to −140, stimulating expression. HULC's activity in maintaining high SK1 levels is not limited to the liver, however, it has also been shown to upregulate SK1 in non-small cell lung carcinoma [69]. In that case, higher levels of SK1 prevent apoptosis and stimulate proliferation in the cancer cells.

Occasionally, lncRNAs are not transcribed from intergenic regions but from the antisense strand of a gene itself [66–68]. The SK1 gene, *SPHK1*, has its own antisense lncRNA called Khps1, of which several different subtypes exist [69]. Khps1 has been shown in rats to govern CpG island methylation of the *Sphk1* gene, and thus governing tissue differential expression of SK1 [69]. Khps1 has also been found to regulate SK1 by directly associating with histones and altering their structure [70]. However, work on the disease-specific effects of Khps1 on SK1 expression remains limited.

#### **6. MicroRNAs**

A growing body of work implicates microRNAs, or miRNAs, in SK1 regulation in both healthy and diseased states. MiRNAs are short oligonucleotides (typically ~22 nt in length), which degrade mRNAs by binding to their 3'untranslated region (3'UTR), targeting them for processing by the RISC [71]. Much of this regulation is relevant to cancer, where miRNAs are found to be frequently dysregulated. MiRNA dysregulation has, in fact, been linked to several diverse types of cancer [72]. There is also evidence of separate dysregulation of miRNA in the tumor microenvironment [73].

Several different miRNAs have been implicated in SK1 regulation, especially in cancer (Table 1). MiR-124, for instance, has been shown to regulate SK1 expression in a variety of cancers, with implications in invasion and metastasis, proliferation, and tumor formation. The 3- -UTR binding site of miR-124 in SK1 has been defined. Typically, miR-124 is downregulated in cancer, which leads to increased expression of SK1 and associated biology. In ovarian cancer, for instance, miR-124 overexpression in cell lines was shown to lead to SK1 degradation, which leads to reduced tumor invasion and migration [74]. This loss in invasive potential was restored with overexpression of SK1 in the cells. It has also been demonstrated that ovarian cancer-associated fibroblasts revert to normal fibroblasts when exposed to exosomes containing miR-124, and this is mediated via the degradation of SK1 in the CAFs [75]. MiR-124 also regulates SK1 in gastric cancer, leading to reduced proliferation and tumorigenicity [76]. MiR-124 has been shown to directly target SK1 in an osteosarcoma model, affecting proliferation, invasion, and matrix metalloprotease expression [77]. Outside of cancer, miR-124 promotes cell death in myocardial infarction by downregulating SK1 [78].


**Table 1.** microRNAs Known to Downregulate SK1.

MiR-506 has also been established as an SK1 regulator in several different systems. In hepatocellular carcinoma, for instance, MiR-506 downregulated SK1 and inhibited angiogenesis [79]. This work also established a negative correlation between MiR-506 expression and SK1 expression in hepatocellular carcinoma tissues. MiR-506 is also downregulated in pancreatic cancer, and restoring its expression disrupts the SK1/AKT/NFkB axis [80]. This change both enhanced cancer cell death and increased sensitivity to gemcitabine, a chemotherapeutic commonly used in the treatment of pancreatic cancer. In osteosarcoma, a close relative of miR-506 called miR-506-3p inhibits SK1, leading to reduced invasion. Interestingly, the expression of this miRNA even causes the cells to lose metastatic potential via mesenchymal-to-epithelial transition [81].

While much work has focused on those two miRNAs in regulating SK1 expression, evidence has linked other miRNAs to SK1 regulation as well. MiR125b, for instance, has been shown to regulate SK1 expression in bladder cancer [82] and Alzheimer's disease [68]. In the former case, overexpression of miR-125b reduced SK1 levels and proliferation [82]; in the latter case, expression of miR-125b was correlated to more severe disease, as SK1 was downregulated and cell death increased. In bladder cancer, the miRNA-613 is downregulated, leading to upregulation of SK1 and increased proliferation, migration, and EMT [83]. In colon cancer, miR-659-3p reduced SK1 and sensitized them to cisplatin [84], while in K562 leukemia cells, miR-659-3p reduced proliferation by targeting SK1 [68]. Hypoxic conditions downregulated expression of miR-1-3p in pulmonary smooth muscle cells, which led to upregulation of SK1 [85].

#### **7. Discussion**

Sphingosine kinase 1, or SK1, is a key enzyme in the sphingolipid pathway, as it converts the pro-death and pro-senescence lipid sphingosine into the pro-survival S1P. The dysregulation of SK1 has been associated with severity in several diseases, especially cancer. Higher expression of SK1 in several types of cancer is associated with poor survival and increased disease severity. High expression of SK1 increases cancer severity by driving proliferation, angiogenesis, metastasis, and chemoresistance through increased production of S1P.

While regulation of SK1 via post-translational modifications such as phosphorylation and proteolysis is well known, the means for transcriptional regulation of SK1 are somewhat less established. It is known that the transcription factor Sp1 upregulates SK1 under certain conditions, which is important in neuronal growth and possibly cancer. Hypoxia is also known to be vital to SK1 upregulation, with HIF2a being the primary transcriptional factor of SK1 in this case. In ischemia, upregulation of SK1 seems to have complex effects on recovery after injury, with the enzyme correlating with better or worse prognosis depending on the system. Despite these varying effects in ischemia, the transcription

factors regulating SK1 expression in response to ischemia are not well defined. Further study of the transcriptional regulation of SK1 in ischemia can help broaden understanding of the development of ischemic injury and help establish SK1 as a drug target or possible upstream drug targets in this system. Cytokine signaling has also been shown to regulate SK1 expression, although again, the transcription factors governing this are not well established.

In cancer, the E2F family of transcription factors has recently been demonstrated to affect SK1 expression, which in turn was shown to improve chemoresistance and angiogenesis in tumors. Further work elucidating the connection between E2F transcription factors and SK1 in cancer would go a long way towards understanding the mechanism of these transcription factors in regulating disease. E2Fs have been shown to sometimes regulate SK1 expression via microRNAs, or miRNAs. Indeed, several different miRNAs have been found to regulate SK1 expression, and many of these miRNAs are downregulated in cancer.

Transcriptional upregulation of SK1 in several different diseases makes it an attractive therapeutic target. Some work has been done exploring the effect of small molecule inhibitors of known SK1 transcription factors on certain diseases. For instance, the pan-E2F inhibitor HLM006474 was shown to induce cell death in models of metastatic melanoma and lung cancer [86,87]. However, no work has been done to assess how disruption of SK1 expression may be related to these effects. The acyclic retinoid peretinoin has been shown to prevent carcinogenesis in liver fibroblasts by downregulating SK1 via Sp1 inhibition [32]. However, since transcription factors typically regulate the expression of several genes and since the effects of SK1 in several systems are well established, targeting of SK1 and the S1P pathway is more likely to be an effective treatment option. Experimentally, treatment with both the SK1 inhibitor PF-543 and the S1PR1antagonist fingolimod seemed to alleviate neuronal injury. Several in vivo studies have been conducted using various SK1 inhibitors, probing their effects on diseases such as asthma, sickle-cell anemia, multiple sclerosis, myocardial infarction, arthritis, and several cancers [88].

However, few clinical trials of SK1 inhibitors have been conducted. One such trial, looking at the putative SK1 inhibitor safingol in conjunction with cisplatin in solid tumors, showed that the inhibitor was well tolerated in patients [89]. Despite these safety data, little subsequent work has been done to measure the efficacy of SK1 inhibitors in human patients. In fact, a search of clinical trials occurring currently in the United States on clinicaltrials.gov reveals only 8 clinical trials, and they are mostly probing inhibitors of SK2 rather than SK1. Such work would be especially welcome in cancer, where SK1's role is well established and where there remains a substantial need for targeted therapies. Further investigation should also be done on the role of SK1 in ischemia injury, as the initial results appear promising, and there is an enormous lack of pharmacological options for treatment.

While considerable efforts have been applied to understanding SK1 regulation post-translationally, many elements of SK1 transcriptional regulation remain poorly understood. Some direct regulation via transcription factors has already been established, and we should examine SK1 levels in conditions where it is known that one of these transcription factors is more active. Post-transcriptional regulation also continues to be studied, and research into SK1 regulation via miRNAs is growing particularly rapidly. As we further elucidate just how miRNAs are regulated and effect certain disease states, so should we further study how miRNA downregulation of SK1 contributes to the disease state.

**Author Contributions:** Writing—original draft preparation: J.B.; writing—reading and editing: J.B., C.M., and Y.A.H.; relevant research and literature study of SK1: J.B. and L.M.O. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was partly supported by NIH grant R01 GM130878.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Dissecting Gq**/**11-Mediated Plasma Membrane Translocation of Sphingosine Kinase-1**

**Kira Vanessa Blankenbach 1,**†**, Ralf Frederik Claas 1,**†**, Natalie Judith Aster 1,**†**, Anna Katharina Spohner 1, Sandra Trautmann 2, Nerea Ferreirós 2, Justin L. Black 3, John J. G. Tesmer 4, Stefan O**ff**ermanns 5, Thomas Wieland <sup>6</sup> and Dagmar Meyer zu Heringdorf 1,\***


Received: 25 August 2020; Accepted: 27 September 2020; Published: 29 September 2020

**Abstract:** Diverse extracellular signals induce plasma membrane translocation of sphingosine kinase-1 (SphK1), thereby enabling inside-out signaling of sphingosine-1-phosphate. We have shown before that Gq-coupled receptors and constitutively active Gαq/<sup>11</sup> specifically induced a rapid and long-lasting SphK1 translocation, independently of canonical Gq/phospholipase C (PLC) signaling. Here, we further characterized Gq/<sup>11</sup> regulation of SphK1. SphK1 translocation by the M3 receptor in HEK-293 cells was delayed by expression of catalytically inactive G-protein-coupled receptor kinase-2, p63Rho guanine nucleotide exchange factor (p63RhoGEF), and catalytically inactive PLCβ3, but accelerated by wild-type PLCβ<sup>3</sup> and the PLCδ PH domain. Both wild-type SphK1 and catalytically inactive SphK1-G82D reduced M3 receptor-stimulated inositol phosphate production, suggesting competition at Gαq. Embryonic fibroblasts from Gαq/<sup>11</sup> double-deficient mice were used to show that amino acids W263 and T257 of Gαq, which interact directly with PLCβ<sup>3</sup> and p63RhoGEF, were important for bradykinin B2 receptor-induced SphK1 translocation. Finally, an AIXXPL motif was identified in vertebrate SphK1 (positions 100–105 in human SphK1a), which resembles the Gα<sup>q</sup> binding motif, ALXXPI, in PLCβ and p63RhoGEF. After M3 receptor stimulation, SphK1-A100E-I101E and SphK1-P104A-L105A translocated in only 25% and 56% of cells, respectively, and translocation efficiency was significantly reduced. The data suggest that both the AIXXPL motif and currently unknown consequences of PLCβ/PLCδ(PH) expression are important for regulation of SphK1 by Gq/11.

**Keywords:** sphingosine kinase; sphingosine-1-phosphate; G-protein-coupled receptors; Gαq/<sup>11</sup>

#### **1. Introduction**

Sphingosine-1-phosphate (S1P) is a multifunctional lipid mediator involved in organismal development and homeostasis of the immune, cardiovascular, nervous, and metabolic systems [1]. The metabolism of S1P is evolutionarily highly conserved, comprising sphingosine kinases (SphK), lipid phosphate phosphatases, S1P phosphatases, and S1P lyase [2]. In vertebrates, S1P activates five G-protein-coupled receptors (S1P-GPCRs), S1P1–5 [3]. These receptors differentially couple to Gi, Gq/11, and G12/<sup>13</sup> proteins, and thereby regulate cell proliferation, survival, migration, adhesion, and Ca2<sup>+</sup>-dependent functions [1]. S1P-GPCRs are ubiquitously expressed and implicated for example in angiogenesis, maintenance of vascular tone and permeability, and immune cell trafficking. Accordingly, S1P-GPCRs play a role in autoimmunity, inflammation, fibrosis, and cancer [1]. Beyond these well-established effects of extracellular S1P, several roles and targets have been described for intracellular S1P. Examples include endocytic membrane trafficking [4,5], Ca2<sup>+</sup> mobilization, regulation of histone deacetylases, and mitochondrial respiration (reviewed in [6]). Of note, all of these studies show highly localized signaling by SphK, supporting the early conclusion that SphK localization is a key to function [7]. There are two SphK isoforms, which are derived from different genes and differ in tissue expression, structure, subcellular localization, regulation, and function. SphK2 has been observed in the cytosol, ER, mitochondria, and nucleus, whereas SphK1 is mainly found in the cytosol and may translocate to the plasma membrane upon stimulation [7–10]. Thus, SphK1 seems to be poised to generate S1P for cellular export, and thereby trigger S1P-GPCR cross-activation, which is known as inside-out signaling. A prominent example for inside-out signaling by S1P is observed in fibroblasts, in which SphK1 activation by platelet-derived growth factor (PDGF) leads to cross-activation of the S1P1 receptor, which then mediates PDGF-induced cell migration [11]. Numerous other studies have shown the importance of the SphK/S1P axis for auto- and paracrine S1P signaling, and diverse transporters mediating S1P export have been identified (reviewed in [1,12]).

SphK1 is regulated transcriptionally, translationally, and post-translationally by many different pathways [8–10]. Acute activation of SphK1, with or without membrane translocation, can be induced by growth factors (for example PDGF, epidermal growth factor, nerve growth factor), cytokines (for example tumor necrosis factor-α, interleukin-1β, transforming growth factor-β), immunoglobulin receptors, or GPCRs (for review, see [8–10]). Several mechanisms have been described for plasma membrane translocation of SphK1. Phorbol-12-myristate-13-acetate (PMA)-induced translocation [13] involved phosphorylation of SphK1 at S225 by extracellular signal-regulated kinases (ERK)-1/2 [14]. In contrast, SphK1 translocation by oncogenic K-Ras was dependent on ERK but independent of S225 phosphorylation [15]. SphK1 translocation by both PMA and oncogenic Ras required calcium-and-integrin-binding-protein-1 (CIB1) [16]. Another pathway for acute activation and translocation of SphK1 is phosphatidic acid production [17]. Membrane binding of SphK1 has been attributed to a hydrophobic patch involving L194, F197, and L198 [4]. Furthermore, a highly positively charged site composed of K27, K29, and R186 was shown to form a single contiguous interface with the hydrophobic patch, mediating electrostatic interactions of SphK1 with membranes [18]. Finally, based on SphK1 crystal structures [19,20], Adams et al. have suggested that a dimeric quaternary structure may play a role in curvature-dependent targeting of SphK1 to the plasma membrane, and suggested how phosphorylation at S225 and protein binding to the C-terminus may potentially unmask membrane association determinants in SphK1 [21].

Our own studies have focused on regulation of SphK by GPCRs. Whereas overall SphK activity can be stimulated via Gi as well as via Gq pathways (reviewed in [22]), we have shown that specifically Gq-coupled receptors induce a rapid and long-lasting translocation of SphK1 to the plasma membrane [23,24]. SphK1 translocation was further induced by overexpression of constitutively active Gα<sup>q</sup> and Gα11, but not Gαi, Gα12, or Gα<sup>13</sup> [23]. Importantly, Gq-mediated SphK1 translocation was independent of phosphorylation at S225, because SphK1-S225A translocated after stimulation of the M3 receptor in HEK-293 cells or the B2 receptor in C2C12 myoblasts, similarly to the wild-type enzyme [23,25]. Classical Gq/11/phospholipase C (PLC) signaling pathways were not involved in SphK1 targeting. Thus, neither cell-permeable diacylglycerol analogues or PMA, which induce activation of protein kinase C, nor thapsigargin or ionomycin, which induce increases in [Ca2<sup>+</sup>]i, were able to induce SphK1 translocation to the extent that it was induced by M3 receptor stimulation. Even a combined pretreatment with PMA plus ionomycin for about 8 min, which caused a minor SphK1 translocation by itself, did not prevent a subsequent marked SphK1 translocation stimulated by the M3 receptor. Furthermore, the involvement of Ca2+/calmodulin, phospholipase D, tyrosine kinases, Rho kinase, and mitogen-activated protein kinase kinase was ruled out by specific inhibitors [23].

We, therefore, studied the regulation of SphK1 by Gq/<sup>11</sup> signaling pathways in more detail. We identified and characterized a motif conserved in vertebrate SphK1, with similarities to Gαq/<sup>11</sup> binding motifs in direct Gq effectors, which is required for Gq-mediated SphK1 translocation. We also showed that PLCβ, beyond its canonical downstream effectors, is important in inducing the most rapid membrane SphK1 translocation upon GPCR stimulation, and that this is mimicked by the pleckstrin homology (PH) domain of PLCδ1.

#### **2. Materials and Methods**

#### *2.1. Materials*

Carbachol, bradykinin, and fatty-acid-free bovine serum albumin (BSA) were purchased from Sigma-Aldrich (Sigma-Aldrich Chemie GmbH, Taufkirchen, Germany). S1P was from Biomol GmbH (Hamburg, Germany). All other materials were from previously described sources [24,26].

#### *2.2. Plasmids*

The 3xHA-S1P1 in pcDNA3.1 was obtained from the Missouri S&T cDNA Resource Center (Rolla, MO, USA). Gαqi5-G66D was kindly provided by Dr. Evi Kostenis (University of Bonn, Bonn, Germany) [27]. The plasmid for expression of Gαq-YFP was a kind gift from Dr. Catherine Berlot (Weis Center for Research, Danville, PA, USA) [28]. Plasmids for expression of Gαi2, Gα<sup>q</sup> wild-type, Gαq-Q209L-EE, Gαq-T257E, Gαq-Y261N, Gαq-W263D, Gαq-D321A, Gαq-Y356K, Gα15-Q212L-EE, G-protein-coupled-receptor-kinase-2 (GRK2)-K220R, and the bradykinin B2 receptor have been described previously [23,29–33].

Plasmids for expression of murine YFP-SphK1 (YFP-mSphK1), human GFP-SphK1 (GFP-hSphK1), human SphK1-G82D (hSphK1-G82D), and human mCherry-SphK1 (mCherry-hSphK1) have been described before [23,24]. Human SphK1-cerulean (hSphK1-cerulean) is a synthetic sequence with optimized codons (Mr. Gene, Regensburg, Germany) deduced from human SphK1 (GenBank accession number AF200328.1) and cerulean fluorescent protein (GenBank accession number ACO48272.1), which was cloned into the pcDNA3.1 vector using HindIII and XhoI. SphK1-F197A-L198Q-GFP and SphK1-L194Q-GFP were kindly provided by Dr. Pietro De Camilli (Yale University School of Medicine, New Haven, CT, USA) [4]. For experiments with the SphK1 mutants, mCherry-hSphK1-A100E-I101E and mCherry-hSphK1-P104A-L105A, these mutants and a second construct of mCherry-hSphK1 wild-type were designed according to the human SphK1 sequence described in GenBank accession number NM\_001142601.2. All three constructs were in pmCherry-C1 vector (Clontech/Takara Bio Europe, Saint-Germain-en-Laye, France) and obtained from Proteogenix (Schiltigheim, France). Full-length p63Rho guanine nucleotide exchange factor (p63RhoGEF) in pmCherry-C1 vector was a kind gift from Dr. Dorus Gadella (University of Amsterdam, Amsterdam, The Netherlands; Addgene plasmid #67896; http://n2t.net/addgene:67896; RRID:Addgene\_67896) [34]. PLCβ<sup>3</sup> (GenBank accession number NM\_000932), C-terminally tagged with TurboGFP in pCMV6-AC-GFP vector, was obtained from OriGene Technologies (product #RG224268; Rockville, MD 20850, USA). PLCβ3-H332A-GFP in pCMV6-AC-GFP vector was obtained from Proteogenix (Schiltigheim, France). PLCδ1(PH)-CFP was a kind gift from Dr. Michael Schäfer (University of Leipzig, Leipzig, Germany) [35].

#### *2.3. Cell Culture and Transfection*

HEK-293 cells stably expressing the M3 muscarinic acetylcholine receptor were cultured in Dulbecco's modified Eagle's medium (DMEM/F12) supplemented with 10% fetal calf serum, 100 U/mL penicillin G, and 0.1 mg/mL streptomycin as described [23]. Stock cultures of HEK-293 cells were grown in the presence of 0.5 mg/mL G418. Mouse embryonic fibroblasts (MEFs) from CIB1-deficient mice were made by J.L. Black in the laboratory of Dr. Leslie V. Parise (University of North Carolina at Chapel Hill, NC, USA) [36,37]. These MEFs, along with MEFs from Gαq/<sup>11</sup> double-deficient mice [38], were cultured in DMEM/F12 medium with 10% fetal calf serum, 100 U/mL penicillin G, and 0.1 mg/mL streptomycin. Transfection of HEK-293 cells was performed with Lipofectamine 2000 (Invitrogen GmbH, Karlsruhe, Germany), while MEFs were transfected with Turbofect (Fermentas, St. Leon-Rot, Germany) according to the manufacturer's instructions. For microscopy, the cells were seeded onto poly-l-lysine-coated 8-well slides (μ-slide; ibidi GmbH, Martinsried, Germany). Before experiments, the cells were kept in serum-free medium overnight.

#### *2.4. Measurement of SphK1 Translocation*

SphK1 translocation was analyzed using fluorescently labelled SphK1 constructs and confocal laser scanning microscopy as described recently [24]. Cells grown on poly-l-lysine-treated 8-well slides (μ-slide; ibidi GmbH, Martinsried, Germany) were incubated in Hank´s balanced salt solution (HBSS) containing 118 mM NaCl, 5 mM KCl, 1 mM CaCl2, 1 mM MgCl2, 5 mM glucose, and 15 mM HEPES, pH 7.4. Fluorescence microscopy was performed with a Zeiss LSM510 Meta inverted confocal laser scanning microscope equipped with a Plan-Apochromat 63×/1.4 oil immersion objective (Carl Zeiss MicroImaging GmbH, Göttingen, Germany). The following excitation (ex) laser lines and emission (em) filter sets were used: CFP and cerulean: ex 458 nm, em band-pass 465–510 nm; GFP: ex 488 nm, em long-pass 505 nm; GFP in combination with mCherry: ex 488 nm, em band-pass 505–530; YFP: ex 514, em band pass 525–600; mCherry: ex 561 nm, em band-pass 575–630 nm. Translocation half-times were determined by measuring the fluorescence intensity within defined cytosolic regions and fitting exponential functions to the translocation-induced decay in cytosolic fluorescence intensity. For estimation of translocation efficiency, the translocated fraction was calculated from these exponential curves as % decay in cytosolic fluorescence. While the average translocation half-time was ~5 s throughout all experiments, the translocated fraction values were comparatively variable between experiments because their calculation was influenced to a certain degree by cell shape changes. For quantification of SphK1 plasma membrane localization in cells co-transfected with Gαq-Q209L-EE (Figure 6G,H), we measured the fluorescence profiles of individual cells using the ZEN software (Carl Zeiss MicroImaging GmbH, Göttingen, Germany), and calculated the ratios of plasma membrane and cytosolic fluorescence intensities.

#### *2.5. Inositol Phosphate Production*

Inositol phosphate production was measured as described recently [24]. Briefly, HEK-293 cells labelled with 1 μCi/mL myo-2-[3H]-inositol (23.75 Ci/mmol; Perkin Elmer Life and Analytical Sciences, Rodgau-Jügesheim, Germany) were stimulated with 100 μM carbachol in HBSS containing LiCl for 20 min at 37 ◦C. The reaction was stopped by addition of 2 mL ice-cold methanol. The cells were scraped from the dishes, 1 mL H2O and 2 mL chloroform were added, and the aqueous phase was transferred to Poly-Prep AG 1-X8 columns (Bio-Rad, Hercules, CA, USA). After washing with H2O and 50 mM ammonium formate, inositol phosphates were eluted with 5 mL of 1 M ammonium formate and 0.1 M formic acid. The radioactivity was measured by liquid scintillation counting.

#### *2.6. Western Blotting*

Cells grown to near confluence on 6 cm-dishes were lysed, the proteins were separated by SDS gel electrophoresis and blotted onto polyvinylidene difluoride membranes. The SphK1 antibody, directed against the C-terminus of human SphK1, was a kind gift from Drs. Andrea Huwiler (University of Bern, Bern, Switzerland) and Josef Pfeilschifter (Goethe-University Frankfurt, Frankfurt, Germany) [39]. Anti-mCherry antibody (ab125096) was from Abcam (Cambridge, UK). Anti-β-actin (A5441) was from Sigma-Aldrich Chemie GmbH (Taufkirchen, Germany), HRP-conjugated secondary antibodies were from GE Healthcare (Freiburg, Germany), and the enhanced chemiluminescence system was from Millipore Corporation (Billerica, MA, USA).

#### *2.7. High-Performance Liquid Chromatography Tandem Mass Spectrometry*

S1P (d18:1) concentrations were determined by high-performance liquid chromatography tandem mass spectrometry as described recently [24].

#### *2.8. Data Analysis and Presentation*

Fluorescence images were edited with the ZEN software (Carl Zeiss MicroImaging GmbH, Göttingen, Germany). Statistical tests, curve fitting, and calculations of translocation half-times were done with Prism-5 (GraphPad Software, San Diego, California, USA). Averaged data are expressed as means ± SD or means ± SEM from the indicated number (*n*) of cells, samples, or experiments, respectively.

#### **3. Results and Discussion**

Similar to our previous publications [23,25], we observed a rapid and long-lasting translocation of both murine and human SphK1 to the plasma membrane upon stimulation of the M3 muscarinic acetylcholine receptor in HEK-293 cells (Figures 1, 2, 5, and 6). To confirm the involvement of Gαq/<sup>11</sup> in this system, we analyzed the influence of a catalytically inactive mutant of GRK2, GRK2-K220R. GRK2-K220R is unable to phosphorylate G protein-coupled receptors but directly binds Gαq/<sup>11</sup> and Gβγ, and acts as a Gαq/11/Gβγ scavenger [29] (see also [40]). When co-expressed with YFP-mSphK1, GRK2-K220R strongly delayed, reduced, and in several cells even fully prevented translocation of the enzyme by the M3 receptor (Figure 1A–C). Together with our previous observation that constitutively active Gα<sup>q</sup> induced SphK1 translocation, this result indicated that M3 receptor-induced translocation was mediated by Gαq.

Another binding partner and direct effector of Gαq/<sup>11</sup> is p63RhoGEF [41,42]. It is known that p63RhoGEF activates RhoA but also competes with PLCβ for activated Gαq/11, and vice versa [41]. Therefore, we tested the influence of p63RhoGEF on SphK1 translocation. As shown in Figure 1D–F, overexpressed mCherry-p63RhoGEF strongly delayed and reduced M3 receptor-induced translocation of GFP-hSphK1. This result indicated that SphK1 is not activated by p63RhoGEF downstream signaling and matches our previous observation that a Rho kinase inhibitor did not prevent Gq//11–mediated SphK1 translocation [23]. The data obtained with GRK2-K220R and p63RhoGEF can be explained by competition of the different Gαq/<sup>11</sup> effectors for binding at the active Gα<sup>q</sup> or Gα<sup>11</sup> subunit. Thus, the data suggest two possibilities: (1) that SphK1 is activated downstream of PLCβ, and (2) that SphK1 directly interacts with and is translocated by active Gαq/11. To analyze these possibilities further, we next tested the influence of PLCβ on SphK1 translocation.

**Figure 1.** Influence of catalytically inactive GRK2 and p63RhoGEF on M3 receptor-induced SphK1 translocation. HEK-293 cells stably expressing the M3 muscarinic acetylcholine receptor were transfected as indicated and translocation of SphK1 was monitored by live cell imaging with a confocal laser scanning microscope. (**A**–**C**) Cells were transfected with YFP-mSphK1 and either catalytically inactive GRK2 (GRK2-K220R) or control vector. Time series were acquired with ~3 images/s and 100 μM carbachol was added after ~25 s. Images were taken from representative time series at 10 and 60 s, thus showing localization of YFP-mSphK1 before and ~35 s after stimulation with carbachol, respectively. The line graphs show the corresponding time courses of cytosolic fluorescence intensity, measured in the indicated cytosolic regions. (**C**) The SphK1 translocation half-times were measured by fitting exponential curves to the translocation-induced decay in cytosolic fluorescence intensity. Each dot represents a single cell. Translocation half-times >100 depict the number of cells which did not respond. Note: \*\*\* *p* < 0.0001 in *t*-tests with Welch's correction for unequal variances. (**D**–**F**) Cells were transfected with GFP-hSphK1 (green) and either mCherry or mCherry-p63RhoGEF (red). Representative images were taken at high spatial resolution immediately before and after the acquisition of time series, during which only GFP fluorescence was monitored with 1 image/s. The line graphs showing time courses of cytosolic fluorescence intensity correspond to the cells shown in the images. Carbachol was added at the indicated time points. (**F**) SphK1 translocation half-time was measured as described in (C). Note: \* *p* < 0.05, \*\*\* *p* < 0.0001 in *t*-tests with Welch's correction for unequal variances. Micrometer bars, 10 μm.

As shown in Figure 2A–D, overexpression of catalytically inactive PLCβ3-H332A-GFP caused a significant delay in M3 receptor-induced translocation of mCherry-hSphK1. In contrast, wild-type PLCβ3-GFP slightly but significantly accelerated translocation of mCherry-hSphK1. While the inhibitory effect of PLCβ3-H332A could be caused both by competition at Gα<sup>q</sup> and by its activity as a GTPase-activating protein (GAP), the acceleration by wild-type PLCβ<sup>3</sup> shows that its GAP activity was surmounted by its stimulatory effect. Interestingly, acceleration of M3 receptor-induced SphK1 translocation was also observed upon expression of the PH domain of PLCδ1, which serves as a sensor for phosphatidylinositol-4,5-bisphosphate (PI(4,5)P2) [43]. As described [35,43], PLCδ1(PH)-CFP was localized at the plasma membrane in unstimulated cells and rapidly translocated to the cytosol after stimulation of the M3 receptor (Figure 2E). The velocity of PLCδ1(PH)-CFP translocation was not altered by co-expression of YFP-mSphK1; however, translocation of YFP-mSphK1 was significantly accelerated by co-expression of PLCδ1(PH)-CFP (Figure 2F). Since the PH domain of PLCδ<sup>1</sup> does not induce DAG and IP3 production, its effect on SphK1 is rather due to PI(4,5)P2 binding or other cellular effects, for example competition with other PI(4,5)P2 binding proteins. In fact, PLCδ1(PH) has been

shown to reduce the amount of phosphatidylinositol-4-phosphate-5-kinase (PIP5K) at the plasma membrane, and thereby the cellular level of PI(4,5)P2 [44]. According to this report, expression of PLCδ1(PH) will rather decrease than increase PLCβ catalytic activity, but nevertheless accelerated SphK1 translocation. Importantly, RhoA, the downstream effector of p63RhoGEF, activates type I PIP5K [45]. Thus, it is possible that (over)expression of PLCβ<sup>3</sup> or PLCδ1(PH) accelerated SphK1 translocation while p63RhoGEF delayed SphK1 translocation by decreasing and enhancing PI(4,5)P2 levels, respectively. However, other tools manipulating PI(4,5)P2 levels such as the multiple pathway inhibitor genistein [46] did not alter SphK1 translocation velocity in our cells ([23] and data not shown). Consequently, the mechanisms by which (over)expression of PLCβ<sup>3</sup> and PLCδ1(PH) accelerate SphK1 translocation remain unclear.

**Figure 2.** Influence of PLCβ<sup>3</sup> and PLCδ(PH) on M3 receptor-induced SphK1 translocation. HEK-293 cells stably expressing the M3 receptor were transfected as indicated. (**A**–**D**) Cells were transfected with mCherry-hSphK1 (red) and GFP (**A**), catalytically inactive PLCβ<sup>3</sup> (PLCβ3-H332A-GFP) (**B**) or PLCβ<sup>3</sup> wild-type (PLCβ3-WT) (**C**). Representative images were taken at high spatial resolution immediately before and after the acquisition of time series during which only mCherry fluorescence was monitored at 1 image/s. The line graphs showing time courses of cytosolic fluorescence intensity correspond to the cells shown in the images. Carbachol was added at the indicated time points. (**D**) SphK1 translocation half-times were measured by fitting exponential curves to the decay in cytosolic fluorescence. The translocated fraction represents the decay in cytosolic fluorescence intensity in % of initial fluorescence. Each dot represents a single cell. Translocation half-times >100 depict the number of cells which did not respond. Note: n.s., not significant; \* *p* <0.05, \*\*\* *p* < 0.0001 in *t*-tests with Welch's correction for unequal variances. (**E**,**F**) Cells were transfected with PLCδ(PH)-CFP (cyan), YFP-mSphK1 (yellow), or both. Carbachol-induced translocation of the two proteins was studied in time series with ~3–4 images/s. (**E**) Images were taken from a representative time series with double-transfected cells before and ~45 s after addition of carbachol. The line graph shows the time course of cytosolic fluorescence intensity for both CFP and YFP. (**F**) Translocation half-times for PLCδ(PH)-CFP and YFP-mSphK1, both alone and in combination. Note: n.s., not significant; \*\*\* *p* < 0.0001 in *t*-tests with Welch's correction for unequal variances. Micrometer bars, 10 μm.

Interestingly, not all members of the Gα<sup>q</sup> subfamily (Gαq, Gα11, Gα14, and Gα15/16, with Gα<sup>15</sup> being the murine homologue of human Gα16) interact with the established targets in the same manner. For example, Gα15/<sup>16</sup> does not bind to GRK2, and binds to but does not activate p63RhoGEF (reviewed in [40]). With the aim of possibly separating PLCβ activation and SphK1 translocation, we expressed constitutively active Gα15-Q212L-EE and studied its influence on SphK1 localization. As shown in Figure 3A, mCherry-SphK1 was strongly localized at the plasma membrane in Gα15-Q212L-EE-transfected cells. Thus, PLCβ activation and SphK1 translocation could not be separately targeted by using Gα15/16.

**Figure 3.** (**A**) SphK1 translocation by constitutively active Gα15. HEK-293 cells stably expressing the M3 receptor were transfected with mCherry-hSphK1 and either control vector or Gα15-Q212L-EE. Shown are two representative images each. Bars, 10 μm. (**B**,**C**) Influence of SphK1 overexpression on M3 receptor-induced inositol phosphate production. HEK-293 cells were transfected with GFP or GFP-hSphK1 (**B**), or with the pcDNA3.1 vector or hSphK1-G82D (**C**). The formation of [3H]inositol phosphates ([3H]IPx) was measured in [3H]inositol-labelled cells stimulated with 100 μM carbachol (Carb.) for 20 min in the presence of LiCl. The expression of GFP-hSphK1 and hSphK1-G82D was confirmed with an anti-SphK1 antibody. The data are means ± SEM from *n* = 9 (B) or *n* = 8 (C) independent experiments, each performed in triplicate. Note: n.s., not significant; \* *p* < 0.05, \*\* *p* = 0.01 in paired *t*-test.

To further analyze the mutual interactions of SphK1 and PLCβ, we next studied the influence of overexpressed SphK1 on M3 receptor-induced accumulation of [3H]inositol phosphates in [ 3H]inositol-labelled cells. As shown in Figure 3B, GFP-SphK1 had no influence on basal inositol phosphate production, but significantly reduced M3 receptor-stimulated inositol phosphate production. Moreover, SphK1-G82D, which is a catalytically inactive mutant [47], had the same effect (Figure 3C), indicating that it was due to protein–protein interactions and independent of S1P signaling. This result indeed suggests that SphK1 competed with PLCβ for Gαq/<sup>11</sup> in the context of a living cell, although it remains possible that SphK1 binds to PLCβ, thereby reducing its activity.

Next, we used embryonic fibroblasts from mice deficient in both Gα<sup>q</sup> and the related Gα<sup>11</sup> [38] to study structural requirements of Gαq/<sup>11</sup> for inducing SphK1 translocation. In Gαq/<sup>11</sup> double-deficient MEFs, both hSphK1-cerulean and YFP-mSphK1 were localized in the cytosol, and their localization did not change upon stimulation of the B2 bradykinin receptor unless Gαq-YFP or Gα<sup>q</sup> wild-type were co-transfected (Figure 4). Of note, in cells expressing Gα<sup>q</sup> wild-type or Gαq-YFP, SphK1 was localized to a small part at the plasma membrane, even under control conditions (Figure 4B,E). This was not the case in cells lacking both Gα<sup>q</sup> and Gα<sup>11</sup> (Figure 4A,D). Using these cells, we confirmed that stimulation of the Gi-coupled S1P1 receptor did not induce SphK1 translocation, even when S1P1, GFP-hSphK1, and Gαi2 were co-transfected (Figure 4C). However, expression of the chimeric G-protein, Gαqi5-G66D, which links Gi-coupled receptors to Gq signaling pathways [27], enabled S1P1 to induce SphK1 translocation (Figure 4C). In cells co-expressing the B2 receptor, YFP-mSphK1, and Gα<sup>q</sup> wild-type, 10 μM bradykinin induced translocation of SphK1 with an average half-time of 5.8 ± 0.6 s (mean ± SEM, *n* = 31 cells; Figure 4E,G). Several Gα<sup>q</sup> mutants were able to fully restore B2 receptor-induced SphK1 translocation in Gαq/<sup>11</sup> double-deficient MEFs, with translocation half-times that did not significantly differ from that of Gα<sup>q</sup> wild-type. These were Gαq-Y261N (t1/<sup>2</sup> = 7.9 ± 1.4 s, *n* = 17), Gαq-D321A (t1/<sup>2</sup> = 4.1 ± 0.5 s, *n* = 14), and Gαq-Y356K (t1/<sup>2</sup> = 4.5 ± 0.5 s, *n* = 17) (all means ± SEM; Figure 4G). In cells expressing Gαq-W263D, B2 receptor-stimulated SphK1 translocation was significantly delayed, with a half-time of 10.8 ± 1.1 s (mean ± SEM, *n* = 18 cells; Figure 4G). Importantly, SphK1 translocation was very slow in cells expressing Gαq-T257E and typically visible only after 2–3 min, for which reason the average translocation half-time was not determined (Figure 4F). Taken together, the amino acids T257 and W263 of Gα<sup>q</sup> are important for targeting of SphK1. Interestingly, Gαq-T257, Gαq-Y261, and Gαq-W263 are implicated in Gαq/GRK2 interaction, as mutation of these residues abolished Gα<sup>q</sup> binding to GRK2 [31,48]. Furthermore, PLCβ activation was completely inhibited by mutation of Gαq-R256/T257 to alanines [49]. Finally, mutants Gαq-Y261N and Gαq-W263D had reduced binding to p63RhoGEF, while Gαq-T257E neither bound nor activated p63RhoGEF [33]. Thus, amino acid T257 of Gα<sup>q</sup> plays a major role in binding or activation of PLCβ, p63RhoGEF, and GRK2, and likewise is important for SphK1 translocation. This result is in agreement with both hypotheses: (1) that SphK1 is activated downstream of PLCβ, and (2) that SphK1 competes with PLCβ, p63RhoGEF, and GRK2 for the same Gα<sup>q</sup> binding site.

Next, we wondered which structural elements in SphK1 were required for Gαq-mediated translocation of the enzyme. As described above, SphK1 membrane translocation by PMA and oncogenic Ras involved CIB1, which binds to the calmodulin binding site of SphK1 [16,50]. Amino acids L194, F197, and L198 of hSphK1 were important for CIB1 binding, and the double mutant hSphK1-F197A-L198Q had reduced CIB1 binding and did not translocate to the plasma membrane in response to PMA, while its catalytic activity remained nearly intact [16,50]. Another study localized L194, F197, and L198 within a hydrophobic patch on the surface of SphK1 and demonstrated that hSphK1-L194Q and hSphK1-F197A-L198Q did not bind to acidic liposomes in vitro and were not recruited to tubular membrane invaginations induced by cholesterol extraction in living cells [4]. Hence, this hydrophobic patch is regarded as essential for curvature-sensitive membrane binding of SphK1 [9,21]. We show here that the two hSphK1 mutants, hSphK1-L194Q-GFP and hSphK1-F197A-L198Q-GFP, did not visibly translocate to the plasma membrane in response to M3 receptor activation in HEK-293 cells (Figure 5A–C). Interestingly, while usually there were only low levels of wild-type SphK1 in the nuclei of transfected HEK-293 cells, there was significant fluorescence in the nuclei of cells transfected with hSphK1-L194Q-GFP (Figure 5B). Furthermore, the mutant, hSphK1-F197A-L198Q-GFP, was strongly localized to the nuclei, with some cells expressing even more fluorescence in the nucleus than in the cytoplasm (Figure 5C). This observation suggests that mutations in this region might possibly disrupt a nuclear export sequence, although the two known nuclear export sequences in hSphK1 comprise amino acids 147–155 and 161–169 [51]. Because of the mentioned involvement of hSphK1-L194, -F197, and -L198 in CIB1 binding [16], we furthermore studied Gq/11-dependent SphK1 translocation in CIB1-deficient MEFs. As shown in Figure 5D, the B2 receptor was clearly able to induce GFP-hSphK1 translocation in cells lacking CIB1. In addition, the translocation half-time was not altered (data not shown). Taken together, we demonstrate that this region comprising L194, F197, and L198 in hSphK1 is important for Gq/11-dependent SphK1 translocation, very likely because this hydrophobic patch is required for membrane binding. CIB1, however, does not play a role in Gq-mediated SphK1 translocation.

**Figure 4.** Identification of Gα<sup>q</sup> residues required for Gq/11-mediated SphK1 translocation. (**A**–**G**) SphK1 translocation was analyzed in Gαq/11-double-deficient MEFs. (**A**,**B**) The cells were co-transfected with the bradykinin B2 receptor and hSphK1-cerulean without (**A**) and with (**B**) Gαq-YFP. Images were taken before and after addition of 10 μM bradykinin as indicated. (**C**) The cells were co-transfected with the S1P1 receptor, GFP-hSphK1, and either Gαi2 or Gαqi5-G66D as indicated. Images were taken before and ~3 min after addition of 1 μM S1P. (**D**–**G**) The cells were co-transfected with the B2 receptor, YFP-mSphK1, and Gα<sup>q</sup> wild-type (Gαq-WT) or various Gα<sup>q</sup> mutants as indicated. Images were taken at a high resolution before and ~5 min after addition of 10 μM bradykinin. The time course of SphK1 translocation was measured by taking series of images at lower spatial resolution at ~1 image/400 ms. Cytosolic fluorescence intensity was measured in selected regions as indicated in the inserts in (**E**,**F**), and translocation half-times were calculated by fitting exponential curves to the decay in the cytosolic fluorescence intensity. (**G**) Translocation half-times obtained with the various Gα<sup>q</sup> mutants. Data are means ± SEM; *n* = 31 (Gαq-WT); *n* = 14–18 (Gα<sup>q</sup> mutants). Note: n.s., not significant; \*\*\* *p* < 0.0001 in one-way ANOVA followed by Bonferroni´s post-test. The micrometer bars represent 10 μm.

PLCβ and p63RhoGEF bind to the effector binding site of Gα<sup>q</sup> primarily via their conserved ALXXPI motifs (X represents any amino acid) [40]. Although both enzymes have additional domains that contribute to the interaction with active Gαq, mutation of the conserved leucine in this motif (L859 in human PLCβ3, L475 in human p63RhoGEF) is sufficient to eliminate Gα<sup>q</sup> binding [40]. Similarly, although GRK2 lacks the ALXXPI motif, it contains a structurally equivalent leucine (L118) that is essential for Gα<sup>q</sup> binding [40]. Interestingly, there is a similar motif, AIXXPL, in vertebrate SphK1, with isoleucine instead of leucine in position 2 and leucine instead of isoleucine in position 6 of this motif (Figure 6A). Comparison of different SphK1 homologues shows the conservation of this motif among vertebrates, with small variations concerning the leucine/isoleucine substitutions, such as ALXXPL in *Gallus gallus* and AIXXPI in *Xenopus laevis* (Figure 6A). We did not find such a motif in non-vertebrate SphK, such as *Drosophila melanogaster* or *Caenorhabditis elegans* SphK, in agreement with the current view that S1P-GPCRs, which are ultimately targeted by SphK1 plasma membrane translocation, have evolved with the vertebrates (see [1]). To study the functional importance of the AIXXPL motif, we generated the mutants hSphK1-A100E-I101E and

hSphK1-P104A-L105A as fusion proteins with *N*-terminal mCherry. When expressed in HEK-293 cells, both mCherry-hSphK1-A100E-I101E and mCherry-hSphK1-P104A-L105A were detected by an anti-mCherry antibody and by an antibody directed against the C-terminus of human SphK1, and had the same molecular weight as mCherry-hSphK1 wild-type (Figure 6B). The double bands seen in Figure 6 were also seen with mCherry alone, and thus caused by the fluorescent tag (Figure 6B). Furthermore, expression of all the SphK1 mutants elevated intracellular S1P concentrations, as measured by high-performance liquid chromatography tandem mass spectrometry. In cells transfected with mCherry, the concentration of S1P was 1050 ± 150 pg/mg protein (*n* = 8), while expression of mCherry-hSphK1 wild-type increased S1P to 1500±200 pg/mg protein (*n*=6; *p*<0.001). Cells expressing mCherry-hSphK1-A100E-I101E had S1P concentrations of 1300 ± 160 pg/mg protein (*n* = 9; *p* < 0.05), and cells expressing mCherry-hSphK1-P104A-L105A had 1500 ± 190 pg/mg protein (*n* = 9; *p* < 0.001) (all values represent means ± SD, with significance tested in one-way ANOVA). These results indicated that all of the mutants were catalytically active.

**Figure 5.** Role of the hydrophobic patch in SphK1 for Gq/11-mediated SphK1 translocation. (**A**–**C**) HEK-293 cells stably expressing the M3 receptor were transfected with GFP-hSphK1, hSphK1-L194Q-GFP, or hSphK1-F197A-L198Q-GFP. Translocation of SphK1 mutants was studied upon stimulation of the cells with 100 μM carbachol. Shown are images before and after stimulation, and time courses of cytosolic fluorescence from representative experiments. In (**C**), the cytosolic fluorescence of the cell in the upper left could not be evaluated because of the strong change in cell shape. The micrometer bars represent 10 μm. (**D**) Role of CIB1 for Gq-mediated SphK1 translocation. MEFs from CIB1-deficient mice were transfected with GFP-hSphK1 and the B2 receptor. Shown are images and time courses of cytosolic fluorescence from a representative time series during which 10 μM bradykinin was added after 35 s. The two images, thus, show localization of GFP-hSphK1 before and 85 s after addition of bradykinin. Micrometer bars, 20 μm.

**Figure 6.** Identification of structural elements in SphK1 required for Gq/11-mediated translocation. (**A**) Sequence alignment of diverse SphK1 homologues between amino acids 81–120 of human SphK1. (**B**) Western blot analysis of mCherry-hSphK1-A100E-I101E and mCherry-hSphK1-P104A-L105A expressed in HEK-293 cells. (**C**–**E**) Analysis of localization and translocation of mCherry-hSphK1-A100E-I101E and mCherry-hSphK1-P104A-L105A in HEK-293 cells stably expressing the M3 receptor. The cells were stimulated with 100 μM carbachol as indicated. Shown are images before and after stimulation, and time courses of cytosolic fluorescence from representative experiments. Since translocation of mCherry-hSphK1-P104A-L105A was variable, two examples are shown here for this mutant. (**F**) Summary of translocation half-times and translocated fractions from (C–E). Each dot represents a single cell. Translocation half-times >100 depict the number of cells which did not respond. Note: \*\* *p* < 0.01 in unpaired *t*-test. (**G**,**H**) Influence of constitutively active Gα<sup>q</sup> on subcellular localization of hSphK1 mutants. HEK-293 cells were transfected with mCherry-hSphK1 wild-type, mCherry-hSphK1-A100E-I101E, or mCherry-hSphK1-P104A-L105A, plus either control vector or Gαq-Q209L-EE as indicated. (**G**) Representative images. All micrometer bars, 10 μm. (**H**) Quantification of SphK1 plasma membrane localization was performed by measuring the fluorescence profiles of individual cells and calculating the plasma membrane/cytosol fluorescence ratios. Each dot represents a single cell. Note: \* *p* < 0.05, \*\*\* *p* < 0.0001 in one-way ANOVA followed by Dunnett's multiple comparisons test.

Next, we studied plasma membrane translocation of the two mutants in response to M3 receptor activation in HEK-293 cells. In unstimulated cells, both mCherry-hSphK1-A100E-I101E and mCherry-hSphK1-P104A-L105A were localized in the cytosol of the cells and only a small part was sometimes seen in the nucleus, similar to mCherry-SphK1 wild-type (Figure 6C–E). Interestingly, carbachol-induced translocation of mCherry-hSphK1-A100E-I101E occurred in only 6 of 24 cells (25%), while mCherry-hSphK1 wild-type translocated in 25 of 26 cells (96%) in this set of experiments (Figure 6F). The translocation half-times of mCherry-hSphK1-A100E-I101E in the 6 cells with translocation were not significantly different from those of mCherry-hSphK1 wild-type (Figure 6F). However, translocation efficiency, measured as the % decrease in cytosolic fluorescence, was significantly lower with mCherry-hSphK1-A100E-I101E than with the wild-type enzyme (14.0 ± 2.5%, *n* = 25 versus 23.6 ± 1.4%, *n* = 6; mean ± SEM; Figure 6D,F). The other mutant, mCherry-hSphK1-P104A-L105A, translocated in 19 of 34 cells stimulated with carbachol (56%), while mCherry-hSphK1 wild-type translocated in 100% of cells in this set of experiments (Figure 6E,F). Again, in the cells which had a response, translocation efficiency of this mutant was significantly reduced (16.5 ± 1.6%, *n* = 19 versus 23.4 ± 1.5%, *n* = 16; mean ± SEM), while translocation half-times were not significantly different when compared to the wild-type enzyme (Figure 6F). In cells co-transfected with constitutively active Gαq-Q209L-EE, mCherry-hSphK1-A100E-I101E remained cytosolic in the majority of cells, while mCherry-hSphK1-P104A-L105A was localized at the plasma membrane to a variable extent (Figure 6G). Quantification of the plasma membrane/cytosol fluorescence ratios revealed that Gαq-Q209L-EE-induced membrane attachment of both mutants was significantly reduced compared to the wild-type enzyme, and that the A100E-I101E mutant was again more affected (Figure 6H). Thus, the results for the two mutants support a role for the AIXXPL motif in Gq targeting of SphK1. Gq-mediated translocation was much more strongly affected by mutation of A100 and I101 to glutamate than by mutation of P104 and L105 to alanine. This might be explained by the stronger disruption of the domain by the negative charges of the two glutamates, or by a higher relevance of AI compared to PL. Intriguingly, within the AIXXPL motif, I101 corresponds to L859 in human PLCβ<sup>3</sup> and L475 in human p63RhoGEF, which are most important for Gq binding [40]. We think that the mutations did not disrupt the general structure of hSphK1, as (1) the molecular weight and subcellular localization in unstimulated cells were normal, (2) both mutants were able to translocate at least in some cells, and (3) S1P concentrations were elevated in cells expressing SphK1-A100E-I101E and SphK1-P104A-L104A indicating catalytic activity. The remaining responses of the mutants might be due to incomplete disruption of the domain, contribution of other parts of the enzyme (see below), or to a second parallel pathway that might involve phosphorylation.

Taken together, we present functional data showing that SphK1 translocation by Gq/11-coupled receptors is prevented by (over)expression of diverse Gαq/<sup>11</sup> effectors or binding partners, suggesting that SphK1 is targeted either via PLCβ or directly by activated Gαq/11. The fact that expression of catalytically inactive hSphK1-G82D reduced receptor-stimulated inositol phosphate production argues in favor of the latter possibility, although the role of PLCβ in this scenario remains unclear. Furthermore, we show that SphK1- s conserved AIXXPL motif is involved in translocation of the enzyme by Gαq. Further studies are required to examine whether this motif indeed mediates direct interaction of SphK1 with Gαq/11. We do not exclude that there are other structural elements in SphK1 that directly or indirectly interact with Gαq. In fact, we had shown before that both the *N*-terminus and C-terminus of hSphK1 (except for the TRAF2 binding site) were required for M3 receptor-induced translocation [23]. The fragment with *N*-terminal deletion of the first 110 amino acids (hSphK1111−384), thus without AIXXPL motif, did not translocate. The fragment with C-terminal deletion of 27 amino acids in addition to the TRAF2 binding site (hSphK11−350) also did not translocate, suggesting that there are other unknown elements in this region required for interaction with Gαq/11, the membrane, or other regulatory proteins (see also discussion in [21]).

There are numerous examples of the importance of SphK1 in Gq/<sup>11</sup> signaling and functional responses. Gq/11-coupled receptors engaging SphK1 include not only muscarinic receptors [52] and bradykinin receptors [25], but also protease-activated receptors [53,54], angiotensin receptors [55], or histamine receptors [56,57], just to name a few. These examples show that SphK1 was involved in Gq/11-dependent regulation of the vascular endothelium and smooth muscle, Gq/11-mediated myogenic differentiation of skeletal muscle, or Gq/11-regulated inflammatory responses. From its eminent role in vascular regulation, it was hypothesized recently that SphK1 might be a therapeutic target in pulmonary hypertension [58], in addition to its roles in inflammation, fibrosis, and cancer [10]. Constitutively active Gq/<sup>11</sup> proteins act as oncogenes [59], and thus given the role of SphK1 in cancer, it will be interesting to unravel a possible interconnection. Another important theme is the emerging role of SphK1 in epithelial–mesenchymal transition [60,61]; however, it should be kept in mind that besides Gq/<sup>11</sup> proteins, there are many other pathways regulating SphK1 expression and activity, and which may be important in this context. Nevertheless, we strongly believe that unravelling the mechanism(s) by which Gq/<sup>11</sup> regulates SphK1 will further help to understand the functional roles of this enzyme and facilitate its targeting by potential therapeutics.

**Author Contributions:** Conceptualization, D.M.z.H., T.W.; acquisition of the data, K.V.B., R.F.C., N.J.A., A.K.S., D.M.z.H., S.T., N.F.; resources, J.L.B., J.J.G.T., S.O., T.W., D.M.z.H.; evaluation and discussion of the data, K.V.B., R.F.C., N.J.A., A.K.S., J.J.G.T., S.O., T.W., D.M.z.H.; writing, review and editing, K.V.B., R.F.C., J.J.G.T., T.W., D.M.z.H. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was funded by the Deutsche Forschungsgemeinschaft (FOG-784; SFB1039 TP A04, B07, and Z01), the LOEWE Lipid Signaling Forschungszentrum Frankfurt (LiFF), and National Institutes of Health grants HL071818 and CA221289 to J.J.G.T.

**Acknowledgments:** We thank Andrea Huwiler (University of Bern, Switzerland) and Josef Pfeilschifter (Goethe-University Frankfurt, Germany) for the SphK1 antibody, and Catherine Berlot (Weis Center for Research, Danville, PA, USA) for the Gαq-YFP expression plasmid. The expert technical assistance of Luise Reinsberg, Agnes Rudowski, and Nicole Kämpfer-Kolb is gratefully acknowledged.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
