Adenosine metabolism in the CNS has been extensively reviewed elsewhere (Cunha, 2001; Latini and Pedata, 2001; Fredholm et al., 2005a,b). Unlike classical neurotransmitters, which are stored in synaptic vesicles and released by exocytosis for exclusive actions on synapses (Fredholm et al., 2005a,b), adenosine distributes extra- and intra-cellularly throughout tissues in the CNS and maintains a basal level in the range of 50–200 nM (Latini and Pedata, 2001). The intra- and extra-cellular adenosine levels exist in a state of dynamic exchange due to the influence of both equilibrative and concentrative nucleoside transporters (Baldwin et al., 2004; Gray et al., 2004). Additionally, metabolic enzymes such as ADA, ADK, SAH-hydrolase, and NTs, continually cycle adenosine through a variety of pathways (Boison and Shen, 2010). The input sources of adenosine are also varied, though strictly dependent on the metabolic state of the cell. Extracellularly, adenosine is produced by the breakdown of adenine nucleotides (such as ATP) by a variety of ecto-NTs, which include ecto-nucleoside triphosphate diphosphohydrolase CD39 and the 5′-nucleotidase CD73. Intracellularly, adenosine can be generated via the dephosphorylation of AMP or the hydrolysis of SAH by cytosolic enzymes 5′-NT or SAH-hydrolase, respectively (Schubert et al., 1979; Broch and Ueland, 1980) (Figure 1).

In the CNS, adenosine levels are largely regulated by glial cells, and two main types of equilibrative adenosine transporters that exist in astrocytic membranes. Intracellular removal of adenosine is executed by degradation to inosine or phosphorylation to AMP by ADA and ADK, respectively (Pak et al., 1994; Boison et al., 2010). The release and uptake of adenosine are mediated by bi-directional nucleoside transporters, depending on the adenosine concentration gradient between the extracellular and intracellular spaces (Gu et al., 1995). It is also worth pointing out that of adenosine levels between extracellular and intracellular spaces are usually maintained in the same range due to highly efficient cellular equilibrative nucleoside transporters in most cells for facilitated diffusion (Cass et al., 1999).

As a key neurotransmitter, adenosine signaling plays a significant role in neuronal excitability and neuroprotection, as well as synaptic and non-synaptic neurotransmission in the CNS. The adenosinergic system also interacts with several other major neurotransmitters, such as dopamine, glutamate, acetylcholine noradrenaline, serotonin (5-hydroxytryptamine), and endocannabinoids (Sperlagh and Vizi, 2011; Shen et al., 2013); these interactions have, however, been reviewed and discussed extensively (Sebastiao and Ribeiro, 2000; Fredholm et al., 2005a,b; Chen et al., 2013). Of particular interest in this review is the ability of adenosine to modulate neurotransmitter release and neuronal excitability in the CNS via activation of four different subtypes of G-protein coupled adenosine receptors (ARs): A₁R, A_2AR, A_2BR, and A₃R (Fredholm et al., 2005a,b). Through ARs, adenosine influences a wide range of brain functions under physiological and pathophysiological conditions, including neuroprotection, neuroinflammation, sleep, psycholocomotion, anxiety, cognition and memory, and neuropsychological disorders (Dunwiddie and Masino, 2001; Fredholm et al., 2005a,b; Shen et al., 2008, 2012). The interactions among subtypes of ARs and between adenosinergic signaling and other neurotransmissions may involve molecular mechanisms at multiple levels, such as (i) direct cross-talk between receptors on the cell membrane, (ii) downstream intracellular second messenger systems among different receptors, (iii) trans-synaptic actions, and (iv) interactions between neurons and glial cells (Sebastiao and Ribeiro, 2000; Stevens et al., 2002; Dare et al., 2007; Shen and Chen, 2009; Boison et al., 2010).

Through AR-dependent actions, adenosine modulates the release of several neurotransmitters in the CNS, including dopamine, glutamate, GABA, serotonin, noradrenaline, acetylcholine, and cannabinoids (Fredholm and Dunwiddie, 1988; Cunha, 2001; Dunwiddie and Masino, 2001). Among the four subtypes of ARs, A_2AR and A₁R play the most significant roles in the interaction between the adenosinergic signaling pathway and other neurotransmitters, while A_2AR and A₃R have a limited effect (Fredholm et al., 2005a,b; Chen et al., 2013). Even so, the AR-mediated manipulations on neurotransmissions involve a high degree of complexity (Dixon et al., 1997; O’Kane and Stone, 1998; Lopes et al., 1999; Gorter et al., 2002). Firstly, this regulation is primarily mediated by the interplay of A₁R and A_2ARs, as these two subtypes have different partnering sets of G-proteins mediating opposing excitatory vs. inhibitory actions. Specifically, the activation of G_i/o-coupled A₁R (and A₃R) mediates the adenosine-dependent inhibition of neurotransmitter release at pre-synaptic neuronal terminals and represses post-synaptic neuronal (Snyder, 1985; Fredholm and Dunwiddie, 1988; Latini et al., 1996). Contrastingly, activation of G_s/olf-coupled A_2AR (and A_2BR) exerts excitatory activity of stimulating the release of glutamate, acetylcholine, and GABA in the striatum and hippocampus (Kirkpatrick and Richardson, 1993; Kirk and Richardson, 1994; Cunha et al., 1995a,b; Sebastiao and Ribeiro, 1996). Secondly, the actions from pre- and post-synaptic sites also contribute to the complexity of AR actions. Presynaptically, stimulation of A₁R suppresses the release of transmitters including dopamine, glutamate, 5-HT, acetylcholine, GABA, and noradrenaline; whereas activation of presynaptic A_2AR facilitates the release of acetylcholine, GABA, and glutamate (Sperlagh and Vizi, 2011). Post-synaptically, activation of A_2AR inhibits activity of dopamine D₂ receptor (D₂ R), while A₁R activation counteracts functions of dopamine D₁ receptor (D₁R) (Ferre et al., 1991; Fredholm et al., 2005a). A_2ARs serve a permissive role in activating cannabinoid receptor (type 1) (CB1R)-mediated inhibition of excitatory transmission, while A₁R prevents the CB1R-mediated reduction of glutamate release. In addition, adenosine signaling affects the activity of post-synaptic AMPAR and NMDAR by control of glutamate release (Sebastiao and Ribeiro, 2009; Sperlagh and Vizi, 2011). Thirdly, the formation of these AR heteromeric receptor complexes also allows for their regulation at the second messenger level. For example, ARs can form heteromers between ARs themselves or with other receptor types, such as glutamate receptors, dopamine D₁ and D₂ receptors, or cannabinoid receptors. While the A_2AR-D₂R heteromers are the first demonstrated epitope–epitope electrostatic interactions underlying receptor heteromerization (Ferre et al., 2004), the capability of A₁R and A_2AR to form homo- or hetero-oligomers with both other receptors and ARs alike has widened their biological roles in developmental, physiological, and pathological situations. In addition to AR actions, studies suggest that adenosine regulating enzymes may also affect epigenetic modifications (Kiese et al., 2016), including alteration of transmethylation and histone modifications (Borea et al., 2016). All of the above regulatory levels comprise a high degree of complexity of adenosinergic actions.

The proliferation and differentiation of OPCs, as well as their capability to myelinate axons, are partly controlled by neurotransmitters linked to adenosine modulation; the activation of these neurotransmitter receptors contributes to the damage of OLs and disrupted myelination in pathological scenarios (Karadottir and Attwell, 2007; Kolodziejczyk et al., 2009; Hamilton et al., 2017). For example, glutamate receptors are expressed in oligodendroglial cells and affect their development (Karadottir et al., 2005). Under normal physiological conditions, glutamate acts through the activation of NMDAR to increase OPC migration, whereas activation of the AMPAR or kainate receptors (but not NMDAR) decreases the proliferation of OPCs (Karadottir and Attwell, 2007) by altering both intracellular sodium concentration and potassium channel function. Blocking NMDA receptors with phencyclidine will result in the inability of OLs to mature (Lindahl et al., 2008).

Dopamine has been reported to modulate CNS myelination (Belachew et al., 1999) through underlying mechanisms that are not fully elucidated. It was found that the dopamine antagonist haloperidol increases OPC proliferation but inhibits their differentiation, an action likely to occur via D₃R (Bongarzone et al., 1998). To clarify the effect of the different subtypes of dopaminergic receptors involved in oligodendroglial development, genetic approaches may need to be utilized in the future. OPCs also express acetylcholine receptors, and the activation of muscarinic acetylcholine M3 receptors increases OPC proliferation through activation of the MAPK signaling pathway; conversely, anti-muscarinic treatment has been seen to accelerate functional myelin repair (Abiraman et al., 2015). Meanwhile, inactivation of nicotinic acetylcholine receptors (nAChR) can inhibit proliferation of OPCs, as evidenced by the ability of the nAChR antagonist mecamylamine to suppress OPC differentiation (Imamura et al., 2015).

Together, while the adenosinergic system plays crucial roles in glutamatergic and dopaminergic neurotransmissions at multiple molecular and cellular levels (Stevens et al., 2002; Barnea-Goraly et al., 2004; Matos et al., 2015), the studies in this section suggest that adenosine signaling can influence the regulation of oligodendroglial development and remyelination via neurotransmitters’ actions itself.

Clinical and experimental evidence indicates that A₁R plays a role in the modulation of neuroinflammation (Si et al., 1996). A decreased expression of A₁R in peripheral mononuclear cells and decreased adenosine levels in plasma were seen in patients with MS. In addition, expression of A₁Rs on CD45⁺ glial cells was decreased by 50% in MS patients and similar observations have been made in patients with other CNS diseases (Haselkorn et al., 2010). From animal studies it was shown that pharmacological activation of A₁R by N6-cyclohexyladenosine (CHA) can induce remyelination and protect existing myelin in a rat model of optic chiasm demyelination. Further study revealed that this protective effect on myelinating cells is mediated by A₁R by potentiating regeneration of endogenous neural progenitors (Asghari et al., 2013). Apparently, activation of A₁Rs is an important inhibitory mechanism against neuroinflammation in MS and has also been observed to attenuate neuroinflammation and demyelination in EAEs animal models (Si et al., 1996; Yao et al., 2012). In accordance with these findings, studies have revealed that inactivation of A₁R resulted in a progressive-relapsing form of EAE in A₁R KO mice, with worsened demyelination and axonal injury, increased microglial proliferation, and enhanced expression of matrix metalloproteinase-12 (MMP-12), iNOS, as well as proinflammatory gene interleukin-1beta (IL-1β) (Si et al., 1996). The modulation of microglial proliferation was also seen in pharmacological approaches, as both non-selective and selective A₁R agonists attenuated the proliferation of rat microglia, and activation of A₁R was shown as an endogenous inhibitor of the microglial response (Turner et al., 2002; Fogal et al., 2010). Therefore, the interplay between A₁R and neuroinflammation is a reciprocal regulatory mechanism; namely, microglial A₁R is downregulated during neuroinflammation in animal models of EAE and patients with MS (Hasko and Pacher, 2008), and the inactivation of A₁R, in turn, exacerbates EAE and MS progression. Thus, modulation of neuroinflammation by A₁R manipulation may offer interesting therapeutic opportunities for MS and other demyelinating diseases (Yao et al., 2012). In developmental stages, however, adenosine acts via A₁Rs and induced diffused WMI. Immature OLs express more A₁Rs compared with mature cells in vitro. Adenosine plays an important role in OL lineage progression by causing OL precursor cells to differentiate prematurely. A₁R stimulation induces white matter loss and mediates hypoxia-induced ventriculomegaly (Mayne et al., 2001). Aiming at modulating this process may provide a novel strategy for WMI treatment (Sitkovsky et al., 2004).

A_2AR also exerts significant anti-immune and anti-inflammatory actions on immune cells (Vincenzi et al., 2013; Ingwersen et al., 2016; Liu et al., 2018). In human studies, A_2ARs are up-regulated in MS patients and administration of the A_2AR agonist CGS-21680 can reduce lymphocyte proliferation of MS patients. This activation of A_2AR mediates a significant inhibition of tumor necrosis factor alpha (TNF-α), IL-6, IL-1β, IL-17, interferon gamma (IFN-γ), and represses cell proliferation, expression of very late antigen (VLA)-4, and activation of NF-kB (Genovese et al., 2009). Further research has been conducted in animal models to assess the pattern of A_2AR expression in neuropathologically damaged tissue. A_2AR is seen to be expressed in infiltrating immune cells, as well as the surrounding endothelium; it is also up-regulated on T cells and macrophages/microglia in EAE lesions (Mills et al., 2008). Likewise, endothelial A_2AR is detected in demyelinated areas in MS brain samples (Mills et al., 2008). Moreover, A_2AR agonist CGS-21680 can ameliorate EAE neurological deficiencies in mice. In sum, A_2AR has been shown to have an overall influence in mediating BBB function in CNS demyelinating diseases (Matos et al., 2013).

Activation of A_2AR in vitro inhibits the migration of CD4⁺ T cells, macrophages, and primary microglia, and suppresses macrophage/primary microglia-mediated phagocytosis of myelin. Along with this finding, A_2AR-specific agonists have also been seen to inhibit myelin-specific T cell proliferation ex vivo and ameliorate EAE (Mills et al., 2008). A_2AR agonists suppressed in vivo primary mechanical injury and secondary inflammatory tissue damage after spinal cord injury; they can also reduce leukocyte recruitment and the activation of the JNK signaling pathway in oligodendroglia, thus protecting the spinal cord from injury-induced demyelination (Paterniti et al., 2011). In contrast, A_2AR KO mice develop increased morbidity, exacerbated neurobehavioral deficits, and generally more severe EAE pathology than their WT littermates. A_2AR KO mice exhibit a severe phenotype of demyelination in EAE with enhanced infiltration of inflammatory cells in the spinal cord and cerebral cortex, accompanied by reduced expression of anti-inflammatory cytokines and increased expression of pro-inflammatory cytokines in the CNS and blood (Hasko and Pacher, 2008). However, studies also demonstrated that there is a dual role of A_2AR in autoimmune neuroinflammation; for instance, A_2AR-specific agonists inhibited the proliferation of myelin-specific T cell ex vivo and ameliorated EAE, whereas application of the same agonist after the onset of disease exacerbated progression of non-remitting EAE. This suggests that activation of A_2AR exerts complex effects on chronic autoimmune neurodegeneration (Wei et al., 2013). The paradoxical effects of A_2AR may be dependent on pathophysiological conditions; further studies are needed to clarify the manipulatory role of A_2AR on demyelinating disorders as well as the potential shifting of the predominant role of A₁R and A_2AR in the pathogenesis of demyelination.

In developmental models, adenosine signaling pathway acts via A_2AR to inhibit OPC differentiation (Coppi et al., 2013). Increased nitric oxide release by microglia, up-regulated neuronal glutamate release, and decreased glutamate uptake by astrocytes are all results of A_2AR activation, and leads to aggravated excitotoxicity (Cook, 1990). Blockade of A_2AR, on the other hand, has been revealed to have a negative effect on demyelination in the spinal cord injury model (Masino et al., 2011).

The A_2BR is a relatively low-affinity adenosine receptor compared to A₁R and A_2AR; in fact, the activation of A_2BR requires a pathologically enhanced level of adenosine (Fredholm et al., 2005a; Chen et al., 2013). A clinical study revealed upregulated A_2BR in the peripheral leukocytes of MS patients; this is in line with findings which showed increased A_2BRs in lymphoid tissues of EAE mice (Lazarus et al., 2011). Pharmacological blockade of A_2BR (via A_2BR antagonists MRS-1754 and CVT-6883) relieved the symptoms and pathological changes of EAE, accompanied by a suppression of IL-6 production and Th17 cell differentiation. A promising work from Gonzalez-Fernandez et al. (2014) demonstrated that activation of A₃R may contribute to oligodendroglial cell death in optic nerve and white matter ischemic damage. However, further investigations are required to fully understand the effects of A_2BR and A₃R in the myelination and demyelination processing.

A patient study from Masino et al. (2009) using a customized parent-based questionnaire indicated an observable relationship between activities that were predicted to increase adenosine levels in the brain and parental observations of relieved behavioral manifestations of ASD. Their results suggest, in general, that increased adenosine signaling activity can benefit sleep disorders, seizures, and social and behavioral dysfunction in ASD (Stubbs et al., 1982; Shen et al., 2008). Augmentation of adenosine signaling activity may serve as a novel therapeutic strategy for ASD with multiple potential beneficial effects in alleviating core symptoms and several comorbidities of ASD (Bottini et al., 2001). However, no direct evidence of dysregulated adenosine in the brain has been tested with respect to the manifestations of ASD (Hettinger et al., 2008). Adenosine metabolic enzymes were also found to have links to ASD. A reduced activity of ADA was reported in the sera of autistic children (Freitag et al., 2010), and a low-activity polymorphism (Asp8Asn) of ADA gene was found to be significantly associated with ASD patients (Ansari et al., 2017a; Amodeo et al., 2018). The above lines of research suggest that reduced ADA activity may be a risk factor for the development of ASD.

The A_2ARs are found expressed predominantly in the caudate nucleus, where they exert their influence in ASD pathology (Ahmad et al., 2017a). Genetic studies reveal that one of the polymorphisms of A_2AR gene, rs2236624-CC, has a nominal association with ASD and three others, rs3761422, rs5751876 and rs35320474, and influences phenotypic variability in ASD symptoms (Ahmad et al., 2017a). Interestingly, the A_2AR gene is located on chromosome 22q11.23, and large 22q11.2 duplications and deletions were observed in individuals with ASD; this correlation suggests a possible role of A_2AR variants in mediating phenotypic expression in ASD (Ahmad et al., 2017a).

Repetitive behaviors and restricted interests are commonly related to neurodevelopmental disorders which are diagnostic for ASD. As a means of exploring these phenotypes, BTBR T⁺ Itpr3tf/J (aka BTBR) mice can be utilized as an animal model of idiopathic autism due to their strong and consistent autism-relevant behaviors (Ahmad et al., 2017b; Ansari et al., 2017b). Studies showed that A_2AR agonism via the adenosine analog CGS-21680 caused a reduction of repetitive and inflexible behaviors in these BTBR mice, whereas A_2AR antagonist SCH-58261 treatment exacerbated repetitive behaviors. This antagonism was also seen to cause a decreased expression of IL-27 and IkB-α and an increased expression of NF-kB p65 and toll-like receptors, TLR2, TLR3, and TLR4 in the brain (Tanimura et al., 2010; Campbell et al., 2013; Lopez-Cruz et al., 2017). Correspondingly, other pharmacological studies have shown that activation of A₁R and A_2AR via co-administration of A_2AR agonist CGS-21680 and A₁R agonist CPA attenuates stereotypy (repetitive behaviors) in a second animal model of ASD in a dose-dependent manner (Zhu et al., 2011). A_2AR inactivation also affects social behaviors and anxiety in the autism spectrum; A_2AR knockout mice show an anxiety profile, higher levels of sociability, and a reduced sensitivity to social novelty (Matute et al., 2001). Pharmacological manipulation of A_2AR can conversely potentiate motivation to work. Together, these studies support the notion that augmentation of AR activities in the CNS could be a promising therapeutic strategy for ASD.

In addition, a third player, A₃R, was shown to have two rare coding variants (Leu90Val and Val171Ile) in a genetics study using four single-nucleotide polymorphisms (SNPs) representing common haplotypes of 958 families with autism. It was revealed that these variations were found at an increased rate in ASD cases (Alberdi et al., 2006). Further in vitro analysis revealed that these same coding variants, Val90-A₃R and Ile171-A₃R, link to an elevated 5-HT re-uptake activity which responds differently to agonism of the 5-HT transporter. This suggests that these hyperfunctional coding variants of A₃R may have an impact on ASD risk (Alberdi et al., 2006; Chez et al., 2007).