Cellular effects and clinical implications of SLC2A3 copy number variation

SLC2A3 encodes the predominantly neuronal glucose transporter 3 (GLUT3), which facilitates diffusion of glucose across plasma membranes. The human brain depends on a steady glucose supply for ATP generation, which consequently fuels critical biochemical processes, such as axonal transport and neurotransmitter release. Besides its role in the central nervous system, GLUT3 is also expressed in nonneural organs, such as the heart and white blood cells, where it is equally involved in energy metabolism. In cancer cells, GLUT3 overexpression contributes to the Warburg effect by answering the cell's increased glycolytic demands. The SLC2A3 gene locus at chromosome 12p13.31 is unstable and prone to non‐allelic homologous recombination events, generating multiple copy number variants (CNVs) of SLC2A3 which account for alterations in SLC2A3 expression. Recent associations of SLC2A3 CNVs with different clinical phenotypes warrant investigation of the potential influence of these structural variants on pathomechanisms of neuropsychiatric, cardiovascular, and immune diseases. In this review, we accumulate and discuss the evidence how SLC2A3 gene dosage may exert diverse protective or detrimental effects depending on the pathological condition. Cellular states which lead to increased energetic demand, such as organ development, proliferation, and cellular degeneration, appear particularly susceptible to alterations in SLC2A3 copy number. We conclude that better understanding of the impact of SLC2A3 variation on disease etiology may potentially provide novel therapeutic approaches specifically targeting this GLUT.

Carbohydrates are the major source of energy in humans. Carbohydrate intake, digestion, absorption and eventual distribution requires a well-balanced interplay between glycolytic and gluconeogenic hormones. For healthy individuals, this results in a relatively narrow range of plasma glucose levels, in order to protect cells from the toxic effects of hyperglycemia and life-threatening hypoglycemia. Before glucose can be used as an electron donor in glycolysis, it must be carried across the plasma membrane. The highly hydrophilic glucose molecule cannot enter cells passively, and therefore specific transmembrane glucose transporters (GLUTs) have evolved in all metazoan species, with a high grade of genetic conservation (Jia, Yuan, Lan, Xuan, & Jeon, 2019).
Sodium glucose transporters (SGLTs) exhibit a secondary active transport mechanism. However, transport via GLUTs utilizes the electrochemical gradient of glucose to facilitate diffusion in an energyindependent manner. Whereas SGLTs are expressed for the sole purpose of glucose absorption and reabsorption, particularly in intestinal cells and renal proximal tubular cells, GLUTs are ubiquitously expressed throughout the body, with single members of the GLUT-family showing differential tissue expression patterns. GLUTs are essential for glucose homeostasis as they are involved in the control of cellular glucose uptake and consumption, glucose storage, and hormonal regulation. Besides their role as cellular gatekeepers, GLUTs have also been suggested to play a role in glucose sensing, with implications for fasting glucose levels and diabetes risk (Dupuis et al., 2010).

| Towards cellular and clinical implications of SLC2A3 CNVs
Haploinsufficiency of GLUT1 leads to developmental delay, microcephaly, and neurological symptoms, such as epileptic seizures, spasticity, hypotonia, and complex motor symptoms (De Giorgis & Veggiotti, 2013). Rare loss-of-function mutations in GLUT2, GLUT9, and GLUT10 are associated with Fanconi-Bickel syndrome, renal hypouricemia, and arterial tortuosity syndrome, respectively. These diseases all follow an autosomal recessive inheritance pattern, and are due to protein truncation and/or impaired membrane trafficking of the affected GLUT (Coucke et al., 2006;Matsuo et al., 2008;Santer et al., 1997). Mutations in the other GLUTs have not been observed to cause disorders by a Mendelian inheritance mode.
However, in recent studies copy number variants (CNVs) of the SLC2A3 (GLUT3) gene were associated with the risk for various clinical phenotypes, such as attention-deficit/hyperactivity disorder (ADHD), rheumatoid arthritis (RA), chorea huntington, and congenital heart defects (Lesch et al., 2011;Mlynarski et al., 2015;Monteiro et al., 2017;Prakash et al., 2016;Veal et al., 2014;Vittori et al., 2014). This raises the question of whether there exists a common disease mechanism in different tissue types, that is dependent on SLC2A3 gene dosage.
In this context, we will first illustrate the unique characteristics of GLUT3, with emphasis on expression data and experimental manipulation of SLC2A3 gene dosage. Second, we will highlight the role of GLUT3 in carcinogenesis, and discuss whether research findings from metabolic cancer research can be translated into other clinical fields. We will then portray the potential impact of structural SLC2A3 variation on clinical phenotypes, focusing on neuropsychiatric, cardiovascular, and immune diseases. Finally, we will suggest future research to determine the effects of structural SLC2A3 variation, and propose hypotheses regarding protective and detrimental consequences of SLC2A3 duplication and deletion events.

| THE GLUCOSE TRANSPORTER GLUT3
The SLC2A family of GLUTs comprises 14 members, designated SLC2A1-SLC2A12, the myoinositol transporter HMIT (SLC2A13), and SLC2A14. GLUT3 is encoded by the SLC2A3 gene which is located at chromosome 12p13.31. Cloning of SLC2A3 was achieved by Kayano et al. (1988) using a complementary DNA library derived from fetal skeletal muscle. Consistent with other GLUTs, GLUT3 contains 12 transmembrane domains with both the amino-and the carboxyterminus located intracellularly, and carries a glycosylation site at the first extracellular loop (Augustin, 2010). The transmembrane protein comprises 496 amino acids accounting for a molecular weight of 54 kDa. GLUT3 has a high affinity (K m = 1.4) for its main substrate D-glucose (Colville, Seatter, Jess, Gould, & Thomas, 1993), but also has the capacity to transport galactose, mannose and ascorbic acid in vitro (Gould, Thomas, Jess, & Bell, 1991;Rumsey et al., 1997).
A general overview of GLUT3 structure, kinetics and expression patterns is given by Simpson et al. (2008).

| SLC2A3 copy number variation
It is thought that several duplication events took place during the evolution of the SLC2A family, creating novel GLUT genes with distinct properties regarding tissue and substrate specificity; however, several functional motifs have been highly conserved (Wilson-O'Brien, Patron, & Rogers, 2010). The detection of copy number variants among the GLUT genes (including SLC2A3) suggests that these recombination events are still occurring, accounting for both gene duplications and deletions. The detected structural variants around the SLC2A3 locus span 122-142 kb (range: chr.12: 7,987,761-8,129,708, hg19, own data) and comprise the entire SLC2A3 gene, the pseudogene NANOGP1, and the first two exons of SLC2A14 according to the gene structure published by Wu and Freeze (2002). As previously reported (Merker et al., 2017), our comparative genomic hybridization array data showed that chromosomal breakpoints were located in segmental duplications with high sequence similarity (first duplicon at chr12:7995630-7998390, second duplicon at chr12:8124315-8128199; fraction matching 9022 | 0.92). It therefore appears that non-allelic homologous recombination (NAHR) is the most likely cause of SLC2A3 CNVs.
However, a recent BLAST search has led us to conclude that, it is actually the other way around; SLC2A3 and NANOGP1 evolved by a NAHR event that led to duplication of SLC2A14 and NANOG approximately 29 million years ago, in the common ancestor Old World monkeys (Figure 1). The newly evolved SLC2A3 gene then has assumed the role of a high-affinity transporter for tissues with elevated glucose demand, whereas the phylogenetically older SLC2A14 gene appears to play a specialized role in the testis (Wu & Freeze, 2002). This view is supported by extensive sequence similarities between the NANOG gene and its numerous pseudogenes, amongst which NANOGP1 is one of the youngest (Fairbanks & Maughan, 2006). Under the assumption of NAHR with subsequent tandem duplications, only NANOG/SLC2A14 or SLC2A3/NANOGP1 can represent the true ancestral gene combination ( Figure 1). It is therefore highly unlikely that SLC2A14 and NANOGP1 originated from the same duplication event. As SLC2A14 expression is limited to a few specific tissues (Shaghaghi, Murphy, & Eck, 2016), and NANOGP1 is classified as a pseudogene, the SLC2A3 gene is of main clinical interest.
A meta-analysis containing 37,661 control subjects from different populations observed a SLC2A3 duplication frequency of 3.44% and deletion frequency of 0.79% (Table 1), suggesting that the SLC2A3 duplication is a common structural variant (>1%) whereas the SLC2A3 deletion is a rare variant (<1%). The different frequencies between duplication and deletion variants is likely due to a lower survival rate of embryos carrying a heterozygous deletion, as seen in rodents (Ganguly et al., 2007). This supports the hypothesis that duplications are generally under less selection pressure than deletions (Zarrei, MacDonald, Merico, & Scherer, 2015). To date, the highest SLC2A3 gene copy number ourselves and others have F I G U R E 1 Evolution of SLC2A3 and SLC2A3 CNVs. (a) Genealogical tree of SLC2A3 evolution in human ancestors. (b) Evolution of SLC2A3 by NAHR. (c) Emergence of SLC2A3 CNVs. In contrast to the common hypothesis that SLC2A14 evolved from SLC2A3 by a duplication event, and is the phylogenetically youngest member of the GLUT family, SLC2A3 together with NANOGP1 can be considered as the more recent sequences. A NAHR duplication event in the common ancestor (Catarrhini) of old world monkeys created a copy of SLC2A14 approximately 29 million years ago (blue star). This copy became expressed in brain and other structures and received the name SLC2A3. Due to the segment order, the new sequence appears more prone to duplication events as can be seen in the gibbon genome, where the duplication of SLC2A3 was integrated permanently (lower gray star), or in humans with the CNV (red star). Independent of the old world lineage, new world monkeys developed a duplication of the original SLC2A3 circa 18 million years ago (upper gray star). Some species also developed further duplications. CNV, copy number variant; GLUT, glucose transporter; NAHR, non-allelic homologous recombination detected is four (Vittori et al., 2014). It is likely that heterozygous loss-of-function mutations in SLC2A3 are generally tolerable, as approximately 1 in 130 people carry only one copy of the gene. This is supported by statistical evidence demonstrating a low probability of being loss-of-function intolerant (probability of being loss-of-function intolerant [pLI] score = 0.01, gnomad.broadinstitute.org). Considering the embryonal abortion of Glut3 null mice (Schmidt et al., 2009), and that no humans with zero SLC2A3 copies have been identified so far, we can infer that homozygous LOF mutations of SLC2A3, though, are incompatible with life.

| Physiological regulation of GLUT3 expression and function
SLC2A3 is most highly expressed in the brain, leading to its unofficial title of "neuronal glucose transporter". However, SLC2A3 is also expressed in other human tissues, such as cardiomyocytes, testis, spermatozoa, placenta, and white blood cells (Haber, Weinstein, O'Boyle, & Morgello, 1993;Kipmen-Korgun, Bilmen-Sarikcioglu, Altunbas, Demir, & Korgun, 2009;Shepherd et al., 1992). In the human brain, GLUT3 seems to be mainly localized to axonal and dendritic processes both in the neocortex and in deeper cortical structures (Mantych, James, Chung, & Devaskar, 1992), suggesting an important role for GLUT3 in ATP-dependent axonal transport processes and synaptic plasticity. In ultrastructural studies of rodent brain tissue, subcellular localization of GLUT3 to cellular processes of the neuropil but also cell bodies has been observed (Fields, Rinaman, & Devaskar, 1999;Leino, Gerhart, van Bueren, McCall, & Drewes, 1997). Both in the mouse and rat brain, highest GLUT3 expression can be found in hippocampal regions, Purkinje cells of the cerebellum, nuclear regions of the brain stem, and the neocortex (Bondy, Lee, & Zhou, 1992;Nagamatsu et al., 1992). SLC2A3 expression is not only tissue-specific, but also inducible. Metabolic, hormonal, medicinal and toxic factors can trigger an induction and/or suppression of SLC2A3 expression, which has been particularly observed under hypoglycemic and hypoxic conditions (Korgun et al., 2002;Wood, Wang, Lorente-Cebrian, & Trayhurn, 2007). Such regulation can enable potent upregulation or downregulation of gene expression, depending on the specific conditions and consequent energy demand. For example, the transcription factor hypoxia-inducible factor 1α (HIF1α) can upregulate SLC2A3 gene expression during hypoxia to facilitate sufficient increased glucose supply (Baumann, Zamudio, & Illsley, 2007;Thamotharan, Raychaudhuri, Tomi, Shin, & Devaskar, 2013). Additionally, HIF1α can be activated by phosphoinositide 3-kinase (PI3K)/Akt signaling upon binding of epidermal growth factor (EGF) and insulin-like growth factor 1 (IGF-1) at their respective receptors ( Figure 2; Kamei et al., 1999;Yu et al., 2012). The K m of GLUT3 is relatively low, leading to rapid saturation of its transport capacity in normoglycemic conditions, yet preserving transporter function in hypoglycemic conditions with a half-maximum reaction rate at 25 mg/dl blood glucose (Colville et al., 1993).
Energy-dependent glucose uptake can be fine-tuned by translocation of GLUT3 protein from intracellular vesicles to the plasma membrane. Although GLUT3 is generally considered an insulin-independent GLUT, multiple studies have suggested that insulin can trigger GLUT3 translocation in human leukocytes and rat neurons (Piatkiewicz, Czech, Taton, & Gorski, 2010;Uemura & Greenlee, 2006). GLUT3 protein has also been identified in the mitochondrial membrane of neuronal cell lines and PC12-cells (Leino et al., 1997;Thoidis et al., 1999), as well as in platelet α-granules, which migrate to the plasma membrane following thrombin stimulation (Heijnen, Oorschot, Sixma, Slot, & James, 1997). A possible translocatory mechanism for GLUT3 in cortical and hippocampal neurons has been proposed; after NMDA receptor binding calcium influx leads to activation of the NO-synthase pathway, with consecutive cGMP production followed by fusion of GLUT3 containing T A B L E 1 Meta-analysis of SLC2A3 duplication (gain) and deletion (loss) frequency  Figure 3).
In rodents, changes in GLUT3 expression occur concurrently with crucial steps of organ maturation in the central nervous system (CNS; Vannucci et al., 1998) and the heart (Grover- McKay, Walsh, & Thompson, 1999). These observations support the hypothesis that adaptive mechanisms of GLUT3 expression may be advantageous during pre-and postnatal developmental stages, when organ growth coincides with increased energy demand.

| Regulation of GLUT3 expression in cancer cells
It is well-known that malign dedifferentiation is accompanied by changes in cellular respiration, as originially demonstrated by Warburg who performed pioneering measurements of oxygen consumption in cancer cells (Warburg, 1956). Healthy cells produce the bulk of ATP by oxidative phosphorylation (OXPHOS), gaining 36 mol of ATP from only 1 mol of glucose. However, cancer cells preferentially bypass OXPHOS and use glycolysis only, even when enough oxygen is available as a terminal electron acceptor in the respiratory chain (aerobic glycolysis; Warburg, 1956). It is still unknown why cancer cells utilize this less effective method of energy production, which generates only 2 mol of ATP from 1 mol of glucose (Hsu & Sabatini, 2008). Warburg postulated that the metabolic shift from OXPHOS to aerobic glycolysis in cells was the main mechanism of cancer development (Warburg, 1956).
Currently, glycolytic cancer metabolism is considered an adaptation to intracellular alterations, including mitochondrial dysfunction due to oncogene activation or tumor suppressor loss (King, Selak, & Gottlieb, 2006). Moreover, rapid proliferation can exceed the tissue's vascular capacity, meaning that cancer cells often have to cope with a hypoxic tumor microenvironment in which glycolysis remains the only feasible option for energy production (Hsu & Sabatini, 2008). Due to inefficient ATP synthesis from glycolysis, the cancer cell's glucose demand is considerably elevated.
In support of this, GLUT1 and GLUT3 are overexpressed in many cancer cells, including tumors of the digestive system (Yamamoto et al., 1990), extracranial head and neck tumors (Mellanen, Minn, Grenman, & Harkonen, 1994), and gliomas (Boado, Black, & Pardridge, 1994). GLUT3 overexpression has additionally been associated with carcinoma grade and prognosis in non-small cell lung cancer (Younes, Brown, Stephenson, Gondo, & Cagle, 1997), laryngeal carcinoma (Baer, Casaubon, Schwartz, Marcogliese, & Younes, 2002), and oral squamous cell carcinoma (Ayala et al., 2010). F I G U R E 2 Signaling pathways for SLC2A3 and glycolytic regulation. PI3K/Akt signaling plays a key role for the regulation of both SLC2A3 expression levels and glycolytic activity by disinhibition of mTOR via TSC1/2. Consecutively, mTOR activates HIF1α, which binds hypoxia-responsive elements (HRE) in the SLC2A3 promotor region. The PI3K/Akt pathway can be activated by both extrinsic growth hormones (such as EGF and IGF-1), and by intrinsic factors (such as the oncogene Caveolin-1; Cav-1) which interacts with growth hormone receptors. Additionally, Cav-1 can increase glycolytic activity via Wnt pathway targeting of the oncogene c-MYC, and enhance SLC2A3 transcription by direct nuclear interaction with the transcription factor HMGA1. Loss of the tumor suppressor gene p53 during oncogenic transformation leads to increased glycolytic activity and enhanced SLC2A3 expression levels via NFκB signaling. This figure was designed in part using Servier Medical ART (https:// smart.servier.com). EGF, epidermal growth factor; HIF1α, hypoxiainducible factor 1α; IGF-1, insulin-like growth factor; IKK, IκB kinase; mTOR, mammalian target of rapamycin; NFκB, nuclear factor κB; OXPHOS, oxidative phosphorylation; PI3K, phosphoinositide 3kinase; TCA, tricarboxic acid cycle The proposed gene-dose effect of SLC2A3 CNVs on cellular glucose uptake, and the increased mutation rate in cancer cells (Jackson, 1998), suggest that SLC2A3 duplications could be enriched in tumor tissue thereby helping to cover elevated glucose demands.
The observation of increased SLC2A3 duplications in human embryonic stem cell lines (9 out of 69 cell lines; Laurent et al., 2011) additionally infers an increase of underlying NAHR events in highly proliferative tissue. Conversely, SLC2A3 deletions may disrupt the cancer cell's need for glucose, and thereby decelerate tumor growth.
However, to the authors' knowledge, there are currently no studies which have investigated an association between SLC2A3 copy number (intratumoral or inherited) and specific cancers or cancer prognosis. There are also no positive genome-wide association study (GWAS) findings in this regard.
Cancer research has provided a multitude of data on the intracellular signaling cascades which control both SLC2A3 expression levels and activity of glycolytic genes. Converging evidence points to a direct interaction of oncogenes and tumor suppressor genes with glucometabolic pathways. The oncogene Caveolin-1 (Cav-1) activates the canonical Wnt-pathway, which in turn targets the oncogene c-Myc with a consecutive increase in glycolytic activity (Tahir et al., 2013). Nuclear Cav-1 additionally leads to enhanced SLC2A3 expression by facilitating HMGA1 binding in the SLC2A3 promotor region (Ha & Chi, 2012). Furthermore, Cav-1 directly interacts with IGF-1 and insulin receptors, with consecutive activation of the PI3K/Akt pathway which boosts the expression of SLC2A3 and glycolytic genes via mammalian target of rapamycin (mTOR; Tahir et al., 2013). Loss of the tumor suppressor p53 also leads to heightened glycolytic activity and F I G U R E 3 The role of GLUT3 for excitatory neurotransmission. Glucose crosses the blood-brain barrier via GLUT1. Lactate, generated by anaerobic glycolysis in astrocytes, enters neurons with the help of the monocarboxylate transporter MCT2. Neurons can use both lactate and glucose (which enters the cell by facilitated diffusion via GLUT3) as substrates for oxidative phosphorylation for mitochondrial ATP generation. ATP is necessary for energy-dependent assembly of synaptic vesicles in the presynaptic terminal and is additionally used as cotransmitter. Upon arrival of action potentials at the terminal, Ca 2+ influx leads to fusion of synaptic vesicles with the presynaptic membrane, and neurotransmitter release into the synaptic cleft. Glutamate binding at NMDA receptors triggers Ca 2+ influx into the postsynaptic neuron. Ca 2+ and Akt lead to phosphorylation of NOS1, evoking translocation of GLUT3-containing vesicles to the plasma membrane, induced by cyclic GMP-dependent protein kinases (cGK). AMP-dependent kinases (AMPK), which are activated by increased AMP:ATP ratio, can additionally provoke GLUT3 translocation to the plasma membrane. GLUT3-mediated glucose supply is crucial for the generation of ATP, which is needed for restoration of resting membrane potential after depolarization during action potential conduction. This figure was designed in part using Servier Medical ART (https://smart.servier.com). EPSP, excitatory postsynaptic potential; GLUT3, glucose transporter 3 Even though there is presently no evidence for an association between SLC2A3 CNVs and cancer or cancer prognosis, the investigation of glucose metabolism and GLUTs in cancer research has led to valuable knowledge about the expressional regulation of GLUT3, which can inform further research on the cellular consequences of SLC2A3 CNVs. Cellular signaling upstream of GLUT3 transcriptional regulation has been well-characterized for the above reviewed pathways. However, it is currently unclear whether SLC2A3 CNVs can have gene dosage-dependent consequences for downstream effectors beyond the established glucometabolic processes of glycolysis and tricarboxic acid cycle (TCA). Therefore, the following sections are devoted to the consequences of SLC2A3 CNV in cellular models, and disease associations are discussed.

| Functional consequences of SLC2A3 expression changes on the cellular level
Studies from mouse and cell culture models have suggested that SLC2A3 CNVs may have an impact on the cellular level. Heterozygous SLC2A3 knockout (KO) mice showed a decrease in Slc2A3 messenger RNA (mRNA) and protein expression in whole brain lysates compared with wild-type (WT), yet only a small decrease in glucose uptake, which coincided with increased GLUT1 and MCT2 protein (Zhao et al., 2010). Two further studies using heterozygous KO models did not find any differences in brain glucose uptake between WT and Slc2a3 +/− mice, despite a 50% decrease in protein expression (Schmidt et al., 2008;Stuart et al., 2011). One possible explanation could be compensatory changes in other GLUT expression, as observed above, but the discrepancy may also potentially be explained by posttranslational translocation of GLUT3. In the case of reduced whole cell GLUT3, vesicular stores may be depleted first, leaving functional GLUT3 located at the plasma membrane mostly unaffected. However, in the studies described only whole brain glucose uptake and glucose uptake in astrocytes were measured, which are both highly dependent on GLUT1 as the major GLUT in astrocytes and the blood-brain barrier (Morgello, Uson, Schwartz, & Haber, 1995). When measuring glucose uptake in cell models of SLC2A3 CNV in future studies, inhibition of other GLUTs should be considered, with an inhibitor for GLUT1 already available (Siebeneicher et al., 2016).
Previous reports have suggested that SLC2A3 duplication increases SLC2A3 mRNA expression by more than 50%; Yang et al.  (Merker et al., 2017). This discrepancy between mRNA and protein expression/transport capacity in SLC2A3 duplication could be due to posttranscriptional or posttranslational modifications that may be necessary for the organism to flexibly adapt to changes in glucose supply and demand. An increase in GLUT3 protein half-life was described as a posttranslational mechanism in episodes of prolonged energy demand (Khayat, McCall, & Klip, 1998). Further studies, preferably in neuronal cell lines, should aim to elucidate whether SLC2A3 gene duplication can also influence protein expression.
There are not many studies which have investigated whether

| Functional consequences of SLC2A3 expression changes in animal models
Slc2a3 KO mice models have been created by several working groups.
Heterozygosity leads to early pregnancy loss in approximately 25% of affected mouse embryos (Ganguly et al., 2007), whereas a bi-allelic deletion of Slc2a3 is deleterious, leading to abortion at Day 12 p.c. It has therefore been suggested that GLUT3-mediated glucose uptake is pivotal for organogenesis (Schmidt et al., 2009). Slc2a3 +/− mice do not differ from WT animals in feeding behavior, body weight, blood-glucose, or insulin concentrations. However, electrophysiology has demonstrated that Slc2a3 +/− mice exhibit overall increased cerebrocortical activity, and an increased startle response to acoustic stimuli (Schmidt et al., 2008). Recently, a conditional tissue-specific postnatal Slc2a3 −/− KO mouse was designed, permitting normal embryonal development (Shin et al., 2018). In the neural-specific KO, the animals died between Day 15 and 31 after genetic manipulation. They showed diminished brain weight, reduced cortical thickness, and reduction of dendritic spines, as well as fewer ultrasonic vocalizations. Electrophysiological recordings also suggested cortical hyperexcitability.
When KO was restricted to the adult male limbic system normal survival was observed, but coincided with reduced locomotor activ-

| Neuropsychiatric disorders
In humans, the brain has the highest relative energy demand of any organ, consuming up to 20% of the body's oxygen and glucose intake, yet only accounting for 2% of the total bodily mass (Rolfe & Brown, 1997). The majority of the brain's energy is needed for neurotransmission; specifically, to restore the resting membrane potential, which collapses when neurons fire (Harris, Jolivet, & Attwell, 2012). Fatty acids cannot pass the blood-brain barrier, and therefore fatty acid oxidation cannot be used for ATP production in the brain. However, during prolonged fasting periods, ketone bodies can be used as alternative electron donors (Morris, 2005;Owen, 2005). Additionally, a lactate-shuttle from astrocytes to neurons has been extensively discussed as an additional potential source of neuronal energy supply. This mechanism enables transport of lactate from anaerobic glycolysis of astrocytes to neurons via monocarboxylate transporters MCT1/2/4 ( Figure 3). Consequently, neurons use this lactate for pyruvate synthesis to fuel TCA in aerobic respiration (Mason, 2017;Pellerin, Pellegri, Bittar, Martin, & Magistretti, 1998). Despite these other possible energy sources, it is still widely accepted that under non-fasting conditions the brain's energy demand is primarily provided by glucose (Clarke & Sokoloff, 1999;Lundgaard et al., 2015). Therefore alterations in GLUT3-mediated neuronal glucose uptake may have a large impact on the brain, and warrant investigation as a potential pathomechanism in neuropsychiatric disorders.
We first identified the SLC2A3 duplication in a genome-wide copy number variation analysis in an extended pedigree of patients with ADHD as an inherited variant (Lesch et al., 2011). In a consecutive case-control study we found a significant overrepresentation of the duplication in our German child and adult ADHD sample, but not in the Spanish and Dutch replication samples.
In a meta-analysis of all three samples, a significant additive effect of SLC2A3 copy number and a single nucleotide polymorphism (SNP) in the SLC2A3 3′-untranslated region was identified, suggesting a synthetic association of the common SNP variant with the more rare CNVs. We therefore hypothesized that SLC2A3 duplication-associated ADHD risk is dependent on genetic background, in a population-specific manner. On the endophenotypic level, duplication carriers underestimated the amount of calories in high-caloric food.
Furthermore, we found an influence of the duplication variant on EEG-measured event-related potentials in neurocognitive tasks.
These tasks comprised a continuous performance test (Go-NoGo paradigm) and an n-back task, which represent measures of cognitive response control as a frontal capacity and working memory performance, respectively. Our results suggest that SLC2A3 duplication variants may alter neural processing in tasks requesting cognitive efforts (Merker et al., 2017). The potential mechanisms behind this finding are currently unclear, but hypothetically could be related to impaired emotional decision-making as a prefrontal function of altered reward-related circuits, in concordance with the frontostriatal network hypothesis of ADHD (for review, see Cubillo, Halari, Smith, Taylor, & Rubia, 2012). Considering the neurodevelopmental nature of ADHD, fluctuating SLC2A3 expression that correlates with axonal sprouting and pruning (Khan, Rajakumar, McKnight, Devaskar, & Devaskar, 1999;Vannucci, 1994), and the severe stunting observed in homozygous KO mice (Shin et al., 2018), it is a plausible option that alterations in SLC2A3 gene dosage interfere with neurodevelopment.
This may not only apply to ADHD, but also to other neuropsychiatric disorders with a significant genetic contribution, such as schizophrenic psychosis and bipolar disorder. For the former, altered glucose metabolism was found in the striatum with reduced TCA and consequently enhanced anaerobic glycolytic activity (Dean, Thomas, Scarr, & Udawela, 2016). Supporting the latter, in an Amish pedigree family a diagnosis of bipolar disorder was observed in approximately 50% of SLC2A3 gene duplication carriers (Yang et al., 2009).
Besides the potential impact of GLUT3-mediated glucose uptake for neurodevelopmental disorders, alterations in GLUT3 expression have also been associated with neurodegenerative disorders, such as Alzheimer's(AD) and Huntington's disease (HD). Decreased GLUT3 expression levels in post-mortem brain tissue was observed for both diseases (Covarrubias-Pinto et al., 2015;Gamberino & Brennan, 1994;Liu, Liu, Iqbal, Grundke-Iqbal, & Gong, 2008). Decreased GLUT3 and GLUT1 protein levels have been associated with Tau hyperphosphorylation and neurofibrillary tangle load, implying a possible link between impaired brain glucose metabolism and AD Simpson, Chundu, Davies-Hill, Honer, & Davies, 1994).
Reduced GLUT expression has been associated with a decrease in HIF1α in the brain of patients with AD . Conversely, mounting evidence suggests a crucial role of the PI3K/Akt pathway with mTOR hyperactivation for the formation of senile plaques and 9028 | neurofibrillary tangles (Tramutola et al., 2015). However, increased mTOR signaling would result in enhanced glycolytic activity and GLUT3 expression via HIF1α (Figure 2). This apparent contradiction might be explained by an as yet unknown mechanism of SLC2A3 downregulation despite HIF1α activation, or a general deficit in protein synthesis caused by oxidative stress in late-stage neurodegenerative processes (Dasuri, Zhang, & Keller, 2013). Overexpression of p53 was also been repeatedly found in various neurodegenerative disorders (Morrison & Kinoshita, 2000), and could possibly be responsible for decreased GLUT3 levels, if p53 can evade negative regulation via the PI3K/Akt pathway (Figure 2).
In HD the SLC2A3 duplication has been associated with a later age of onset (Vittori et al., 2014). Glucose hypometabolism precedes cell decline and is already detectable in premanifest patients with HD (Ciarmiello et al., 2006), supporting the view that alterations in neuronal glucose metabolism are not only a result of advanced neurodegenerative cell damage but rather contribute to the pathophysiology in some way. Hence, a reduction in GLUT3-mediated neuronal glucose uptake seems to be a common feature of neurodegenerative diseases, and initial evidence suggests a protective effect of the SLC2A3 duplication variant for HD (Vittori et al., 2014). This is possibly due to a larger glucose transport capacity, confering metabolic resistance against oxidative stress. However, the potential role of glucometabolic changes in neurodegeneration should not be overestimated, as these changes may just be a side-effect of causative pathophysiological changes. Current CNV GWAS have so far not associated SLC2A3 CNVs with AD (Swaminathan et al., 2012) or age of onset of AD (Szigeti et al., 2014). However, the data has indirectly suggested a genetic association between SLC2A3 and AD; an intronic SNP rs10845990 in the neighboring SLC2A14 gene was among the 12 top hits (though not genome-wide significant), with the G-allele conferring higher risk for AD (Shulman et al., 2011). This finding was replicated in a subsequent case-control study within a different population (Wang et al., 2012). A database search on expression quantitative trait loci derived from whole blood (www.eqtlgen.org) shows that the rs10845990 G-allele is not only associated with lower SLC2A14 expression, but also with lower SLC2A3 and higher C3AR expression levels (Võsa et al., 2018). C3AR encodes a complement receptor whose brain expression correlates with cognitive decline in humans, and which has been associated with Tau pathology in a mouse model (Litvinchuk et al., 2018). Therefore, the functionality of the SLC2A14 Alzheimer risk SNP might be due to cis-regulation of SLC2A3 and C3AR, thereby connecting possible glucometabolic dysregulation and complement-mediated neuroinflammation in neuronal cell damage.
Taken together, current findings indicate the relevance of GLUT3 for neurodevelopmental and neurodegenerative processes.
Future research should focus on consequences of SLC2A3 overexpression and deletion for neuronal glucose uptake, and determine whether there are downstream cellular signaling targets which depend on SLC2A3 gene dosage. It should be emphasized that no human studies have so far associated SLC2A3 deletion variants with neuropsychiatric disorders, and our own control cohorts include several adult patients without any obvious symptoms of a neuropsychiatric disorder that are also carriers of heterozygous SLC2A3 deletion.

| Cardiovascular diseases
The human heart is versatile in its ability to oxidize different energy substrates, including glucose, fatty acids, ketone bodies, lactate and amino acids. It is therefore less dependent on glucose metabolism than the brain. Due to its vital hemodynamic role and scarce ATP storage capacity, which only provides enough ATP for several beats, the heart depends on a steady supply of substrate fuels (Grynberg, 2005). This energy supply is mainly provided by fatty acid oxidation, accounting for up to 70% total energy under normal physiological conditions in healthy adults (Opie, 1991). Although GLUT4 and GLUT1 are thought to be the primary cardiac GLUTs in humans and rodents (Grover-McKay et al., 1999;Kraegen et al., 1993), GLUT3 is also expressed in the human heart: In fetal cardiomyocytes, GLUT3 expression increases from Week 10-15, and then subsequently decreases until Week 21 gestational age (Grover-McKay et al., 1999). This suggests an important role for GLUT3 in cardiogenesis. Supporting this, a recent study found that SLC2A3 mRNA expression in fetal myocardium was 15.6-fold higher than in newborns (Kong, Liu, & Lu, 2008). These changes in expression may reflect the switch from prenatal anaerobic glycolysis to postnatal fatty acid oxidation as a preferential source of ATP generation (Ascuitto & RossAscuitto, 1996).
SLC2A3 duplication has been repeatedly associated with congenital syndromic heart defects, including in Turner syndrome (Prakash et al., 2016) and 22q11.2 deletion syndrome (Mlynarski et al., 2015). In the latter, it was suggested that increased gene dosage might interact with genes in the 22q11.2 region, increasing risk in a "two hit model" (Mlynarski et al., 2015). These findings are supported by the increased maternal serum glucose levels observed in tetralogy of Fallot (Priest, Yang, Reaven, Knowles, & Shaw, 2015).
These data, combined with the knowledge that mature heart formation requires strict regulation of energy for normal cell guidance and proliferation (Kloesel, DiNardo, & Body, 2016), suggest a path from impaired glucose metabolism to cardiac misdevelopment. An alternative hypothesis for the markedly increased frequency of SLC2A3 duplication in congenital heart defects would be a gain-offunction effect due to improved energetic supply, perhaps promoting survival in the face of otherwise lethal cardiac development impairments.
There are currently no studies demonstrating an association of SLC2A3 CNVs with coronary heart disease, postischemic outcomes, or cardiac remodeling resulting in ventricular hypertrophy. However, overexpression of GLUTs has been demonstrated to have cardioprotective effects post-ischemia in mice (Luptak et al., 2007).
Furthermore, chronic hypertrophic stress has been observed to cause a switch from fatty acid oxidation back to glycolysis as the preferential source of ATP supply in cardiomycytes, similar to in the fetus (Barger & Kelly, 1999). A beneficial effect of SLC2A3 duplication ZIEGLER ET AL.
| 9029 in cardiodegeneration therefore seems feasible, but is currently lacking evidential support. Lastly, GLUT3 is also essential for the secretory and aggregatory activity of platelets, and therefore may potentially influence the development of atherosclerotic lesions . In a mouse conditional double KO of GLUT3 and GLUT1, reduced platelet activation and thrombogenic potential was observed, suggesting a role of GLUT3 in thrombus formation .
Taken together, the data suggests that SLC2A3 CNVs may have an impact on cardiac development in syndromal disorders but systematic efforts to disentangle the role of glucometabolic alterations in cardiovascular diseases are lacking. Therefore, current data warrants further investigation of the potential role of GLUT3 in cardiovascular diseases, particularly in cardiogenesis and ischemic heart failure. As the heart relies on glycolytic energy exploitation only under certain conditions, such as during cardiogenesis and hypertrophic stress, this might be a promising starting point for researching the role of GLUT3-mediated glucose supply in developmental and degenerative heart disease.

| Immunodisorders
Activation of both the innate and adaptive immune systems is associated with increased proliferation, increased protein synthesis (for antibody and chemokine production), cytoskeletal rearrangements (for targeted cell migration), and enhanced secretory activity. These cellular changes cause a rapid increase in energy demand, as demonstrated by elevated glucose utilization in activated immunecompetent cells (Calder, Dimitriadis, & Newsholme, 2007;Newsholme, Costa Rosa, Newsholme, & Curi, 1996). In response to this increased energy demand, immune cells can upregulate their GLUT expression (including GLUT3) by up to 6-fold (Fu, Maianu, Melbert, & Garvey, 2004;Maratou et al., 2007). This has also been demonstrated in CD4 + T-cells, which showed increased GLUT1 and GLUT3 expression followed by a concomitant elevation of glycolytic rate, promoting T-cell activation, survival and expansion of T-effector cells (Macintyre et al., 2014). Increased GLUT expression may represent a general adaptive mechanism to high energetic demands, indirectly reflecting disease activity under certain conditions. For example, in autoimmune hepatitis disease activity appears to correlate with enhanced glucose metabolism and strong GLUT3 expression in inflammatory IgG4 + cells (Araki et al., 2018). However, chronic demyelinated lesions in multiple sclerosis conversely exhibit a reduction in axonal GLUT3 (Nijland et al., 2014), although this may represent adaptation to a decreased energy demand following neurodegenerative damage, rather than a direct response to a pathogenic catalyst. It is currently unclear whether inflammatory activity could potentially be decreased by reducing GLUT3-mediated glucose uptake in leukocytes. Further studies should aim to elucidate this possibility, as it could potentially represent a novel therapeutic approach for the treatment for immune system disease by GLUT inhibition.
SLC2A3 CNV, specifically deletion, has been suggested to play a role in autoimmune diseases. Frequency of SLC2A3 deletion was found significantly lower in patients with RA than healthy controls (odds ratio = 0.5) in a large study of a Swedish cohort, a finding which could be replicated in two other samples, bringing the total cohort number to 6124 cases (deletion frequency 0.78%) and 9511 controls (deletion frequency 1.13%; Veal et al., 2014). These data suggest that SLC2A3 may provide a protective effect against RA. However, in a higher-powered independent replication study, this association was not observed (Simpfendorfer et al., 2019). Although initial results may have been a false positive, it is also possible that there is a population-specific effect of SLC2A3 CNVs on different disease phenotypes (Merker et al., 2017 sue-specificity and individual pathological predisposition, reflecting a modifying rather than a causative role of SLC2A3 CNVs for disease activity. This is supported by the data which suggests that the SLC2A3 gene is not particularly sensitive to loss-of-function mutations ((pLI score = 0.01, gnomad.broadinstitute.org). There is therefore no general "ideal" SLC2A3 copy number. We suggest that dynamic cellular processes, which rely on a high ATP supply, are especially susceptible to alterations in SLC2A3 copy number. This susceptibility may become apparent when energy demands alter due to normal aging or pathological activity.
Additional SLC2A3 copies may convey advantageous effects in disease conditions where energy metabolism becomes insufficient, such as cellular degradation and hypertrophy, by increasing overall glucose uptake capacity. In neurodevelopmental disorders, such as ADHD or autism however, SLC2A3 CNVs could disturb highly energy-dependent axonal organization and pruning during the formation of neural networks. Similarly, SLC2A3 deletion may improve disease states that are characterized by uncontrolled cell proliferation (such as autoimmune diseases and neoplasms), yet potentially intensify neurodegeneration. Embryonal differentiation, organogenesis, maturation, cancer formation, and degenerative processes in a broad range of organ systems, such as the CNS, heart, and immune-competent cells are therefore important targets for future investigations.
To date, the majority of data is only correlational, with no definitive cause-and-effect relationship between SLC2A3 gene dosage and clinical phenotypes reported. Future studies should aim to follow a multidimensional approach in an attempt to elucidate this potential relationship. Such an approach requires the integration of data from basic molecular biology, electrophysiology, neuropsychology and imaging, to provide a clearer picture of the possible impacts of structural SLC2A3 variation. With development of an SLC2A3 overexpression mouse and the advent of induced pluripotent stem cells (Jansch et al., 2018) there will be two new models available to study the influence of SLC2A3 CNVs on both the molecular and behavioral level.
Finally, the designation of GLUT3 as a "neuronal glucose transporter" may have led to its role in peripheral tissues being overlooked. However, newly identified associations between SLC2A3 CNVs and various clinical disorders demonstrate that GLUT3-mediated bioenergetic imbalances may be important in a range of tissue types, reflecting a broad spectrum of diseases. This research avenue is of high relevance not only due to general interest in glucometabolic pathomechanisms, but also due to the potential of GLUTs as promising novel therapeutic targets.

ACKNOWLEDGMENTS
Sincere thanks go to Johanna Zöller for assisting with the design of

CONFLICT OF INTERESTS
None of the authors declare any financial conflicts of interest.

AUTHOR CONTRIBUTION
GCZ wrote the first draft of the manuscript. PA performed BLAST searches and found that SLC2A3 descends from SLC2A14. All authors participated in literature research, conception, design, and discussion of the manuscript, and approved its final version.

DATA AVAILABILITY STATEMENT
Data sharing is not applicable to this article as no new data were created or analyzed in this study.