The level of brain glucose is a process finely regulated. Glucose transporter 1 (GLUT1) seems to be the primary transporter controlling brain glucose entry (Cardoso et al., 2010). The high affinity of this transporter for glucose justifies the level found in the brain which is about 30% of the blood level and close to 2 mM at baseline (~7–8 mM blood glucose). This is the case in the hypothalamus or other areas including the hippocampus and striatum for instance (McNay and Gold, 1999; McNay et al., 2001; de Vries et al., 2003; Dunn-Meynell et al., 2009; Langlet et al., 2013). Interestingly, hypothalamic glucose level does not seem to fluctuate significantly between meals (Dunn-Meynell et al., 2009) or even after fasting (Langlet et al., 2013). Langlet and colleagues even found that glucose level in the ARC is slightly increased after fasting due to an increased permeability of the blood-brain-barrier (BBB). Thus, only profound manipulations of blood glucose level would induce changes in hypothalamic levels including insulin-induced hypoglycemia or pathological hyperglycemia. Such conditions would make hypothalamic levels to fluctuate between 0.2 and 4.5 mM (see for review Routh et al., 2014). Nevertheless, the finding of specialized neurons able to detect changes above 2.5 or even 5 mM suggests that, in confined areas, glucose level could be increased closer to level found in the blood. This could be the case around fenestrated capillaries of the BBB present in the ventral part of the ARC (Ciofi, 2011; Langlet et al., 2013). In support of this hypothesis, HGE and HGI neurons have only been found in the ARC (Fioramonti et al., 2004) and not in the VMN (Song et al., 2001). In this later study, neurons activated by increased glucose level above 5 mM have been found in the VMN but due to pre-synaptic glutamate inputs and not though a direct detection (Song et al., 2001). Local changes of glucose level around fenestrated capillaries would not be detectable by the available techniques including the microdialysis which does not present sufficient spatial resolution to monitor changes in confined environments. Determining (1) whether HGE- and HGI-like neurons can be found in extra-hypothalamic circumventricular organs in which the BBB is fenestrated (e.g., area postrema of the hindbrain, the subfornical organ and the vascular organ of lamina terminalis); (2) the real concentration of glucose in these micro-environments, will help characterizing brain spots containing these specialized glucose-sensing neurons as well as their physiological functions.

Some studies using electrophysiology on brain slices suggested that POMC neurons could be either GE or HGE-type neurons as their activity is modulated by changes in extracellular glucose level (Ibrahim et al., 2003; Claret et al., 2007; Parton et al., 2007). Other studies including one from our lab did not find POMC neurons directly glucose-sensing (Wang et al., 2004; Fioramonti et al., 2007). Nevertheless, a study from Parton et al. showed that α-melanocyte-stimulating hormone (α-MSH) release on hypothalamic chunks is increased in response to increased glucose level (Parton et al., 2007). This shows that POMC neurons can be activated by increased glucose level. Nevertheless, to our knowledge, no study has shown that glucose directly modulates the activity of POMC neurons. One could hypothesize that the modulation of POMC activity by glucose is due to presynaptic inputs rather than a direct detection by POMC neurons themselves. The hypothesis that POMC neurons would not detect changes in glucose level directly is supported by a recent work showing that the frequency of excitatory postsynaptic currents onto POMC neurons is modulated by glucose (Hu et al., 2014). To get further insight into the direct glucose-sensing properties of POMC neurons, we studied their activity using calcium-imaging on freshly dissociated hypothalamic cells from POMC-GFP mice (for detailed method, see Chretien et al., 2017). In such preparation, dissociated cells are not in contact from each other (Figure 2). Thus, this strategy allows the study of direct glucose-sensing properties of hypothalamic neurons (Dunn-Meynell et al., 2002; Kohno et al., 2003; Kang et al., 2004; Vazirani et al., 2013; Chretien et al., 2017). As presented in Figure 2, we found that ~8% hypothalamic neurons tested were identified as HGE neurons as they harbor a transient increase in [Ca²⁺]_i in response to a raise in glucose level from 2.5 to 10 mM. Nevertheless, none of the HGE neurons identified were GFP-expressing neurons (n = 11; Figure 2). Altogether, these data plus those from the literature suggest that glucose-sensing neurons modulate the melanocortin system but that POMC neurons themselves do not directly detect changes in glucose level. Further studies are still needed to determine the identity of ARC GE and HGE neurons. Knowing better the identity of hypothalamic (and extra-hypothalamic) glucose-sensing neurons is necessary to decipher the physiological functions controlled by these neurons.

To determine mechanisms of brain glucose-sensing, studies have initially tried to determine similarities and differences from the known gold-standard glucose-sensor, the pancreatic β-cell (Yang et al., 1999). Thus, Michael Ashford was the first to show that ATP-dependent potassium (K_ATP) channels are involved in glucose-sensing of GE neurons (Ashford et al., 1990a,b). Nevertheless, it is not possible to determine whether GE and/or HGE-like neurons were studied as those pioneer studies used large glucose changes from 0 to 10 mM. Later, studies from Vanessa Routh's laboratory demonstrated that K_ATP channels are indeed involved in the detection of decreased glucose level by GE neurons (2.5–0.1 mM level window; Song et al., 2001; Wang et al., 2004). Nevertheless, they showed that the change in K_ATP conductance plateau above 2.5 mM (Wang et al., 2004), suggesting that K_ATP channels could only mediate detection below this level, and consequently, only in GE neurons. In addition, Guy Rutter's group suggested that K_ATP channels would not be involved in glucose detection above 2.5 mM in a study showing that intracellular ATP level is not increased in hypothalamic neurons in response to increased glucose level from 3 to 15 mM (Ainscow et al., 2002). We confirmed that K_ATP channels are not involved in HGE neurons. Indeed, we found that these channels are essentially closed at 5 mM. We also showed that HGE neurons can be found in K_ATP-deficient mice (Fioramonti et al., 2004) whereas GE neurons cannot be detected in these mice (Miki et al., 2001). Yang et al previously suggested that K_ATP-independent mechanisms would be involved in 5–20 mM glucose detection as the K_ATP-opener diazoxide failed to prevent increased glucose detection (Yang et al., 1999). Instead, we found that a non-selective cationic conductance was involved in the glucose response of HGE neurons (Fioramonti et al., 2004). More recently, we demonstrated that transient-receptor-potential canonical type 3 (TRPC3) channels are required for glucose detection by HGE neurons (Chretien et al., 2017). Involvement of non-selective cationic conductance has also been identified in the response to glucose of GnRH neurons even though the identity of the channel responsible for glucose-sensing in these neurons is yet to be determined (Roland and Moenter, 2011b). TRPC channels are involved in the response to insulin and leptin of ARC kisspeptin and POMC neurons (Qiu et al., 2010, 2011, 2014, 2017). Together, these studies open new line of research for hypothalamic glucose- and, more generally, nutrient-sensing mechanisms (Chretien et al., 2017).

The mitochondrial respiratory chain represents one the main sources of ROS. Increasing reduced equivalents (NADH,H⁺ and FADH₂) as a result of glucose oxidation leads to increased electron transfer chain activity, generating an elevated but physiological superoxide anions production. Our group showed that cerebral injection of glucose in rodents leads to a short-lived mROS signaling, in the form of H₂O₂ (see for review Leloup et al., 2011). We and others also showed that this increase in mROS level is necessary for the of glucose-sensing neurons to increased glucose (Chretien et al., 2017) and the regulation of food intake and insulin secretion (Leloup et al., 2006; Andrews et al., 2008; Colombani et al., 2009; Carneiro et al., 2012). Thus, increased mROS levels, rather than just the ATP/ADP ratio, constitute a signal that mediates the stimulatory effect of glucose on some hypothalamic neurons. Therefore, mROS signaling is consistent with the NADH mechanism suggested earlier for the glucose sensing mechanism (Yang et al., 1999; Ainscow and Rutter, 2002). In addition, the fact that TRPC3 channels form a redox-sensitive complex reinforces the role of mROS in the signaling involved in glucose detection.

Mitochondria are organized into a tubular network that continuously changes its shape and motility, mediated by fission and fusion mechanisms. Mitochondrial dynamics (fusion and fission) have been linked to the balance between energy demand and nutrient supply. Our group suggested that mitochondria dynamics (fission) play a significant role in mROS production in response to increased glucose level. In the MBH, we highlighted glucose-induced Drp1-dependent mitochondrial fission is an upstream regulator for mROS signaling, and consequently, a key mechanism in hypothalamic glucose sensing (Carneiro et al., 2012). More recently, Diano's and Claret's teams showed that fission and fusion mechanisms are involved in the detection of increased or decreased glucose level by hypothalamic POMC or SF1 neurons (Schneeberger et al., 2013; Toda et al., 2016; Ramirez et al., 2017; Santoro et al., 2017). Even though the involvement of mitochondria dynamics seems to be a critical mechanism involved in glucose detection and in the regulation of energy and glucose homeostasis, the signaling pathways leading to such changes in mitochondria morphology is unclear and needs to be studied further. The idea that such mechanisms are specific to glucose-sensing neurons also needs to be addressed.

Some studies highlighted that metabolism-independent mechanisms take part in neuronal glucose-sensing. Thus, it has been shown that the sodium-glucose cotransporters (SGLT)-inhibitor phloridzin, blocks the response to increased glucose level of some hypothalamic neurons (Yang et al., 1999; O'Malley et al., 2006). Furthermore, the non-metabolizable substrate of SGLTs, α-methylglucopyranoside, mimics the effect of increased glucose level on a majority of HGE-like neurons (O'Malley et al., 2006). SGLT1 or SGLT3 could be the transporters responsible for glucose detection as they are expressed in hypothalamic neurons. The hypothesis on SGLT3 is particularly interesting since this transporter is expressed in human cholinergic neurons where it acts as a glucose-activated sodium channel which does not transport glucose (Diez-Sampedro et al., 2003). Interestingly, SGLT3 has been shown to be the sensor mediating glucose detection of the portal vein (Delaere et al., 2012).

Another metabolism-independent mechanism has been suggested to mediate glucose detection in the hypothalamus through sweet taste receptors. These receptors present in papillae of the tongue mediate the sweet sensation of sugars and sugar substitute (Laffitte et al., 2014). These receptors are heterodimers composed of T1R2/T1R3 subunits which have also been identified in several extra-gustatory tissues and cells including the pancreatic β-cell and the brain (Laffitte et al., 2014). In the brain, the expression of T1R2 and T1R3 subunits is particularly enriched in the ARC (Ren et al., 2009; Herrera-Moro Chao et al., 2016). Recently, Yada's group showed that half of the HGE-like neurons in the ARC (activated by the increase in glucose level from 1 to 10 mM) are excited by the application of the sweetener sucralose. They also showed that the sweet taste receptor inhibitor gurmarin, blocks the response to glucose of 2/3 of HGE-like neurons (Kohno et al., 2016). Finally, in this study, they also show that sucralose activates neurons of the ARC which are not POMC neurons, reinforcing the idea that POMC neurons do not detect directly changes in glucose level. These data highlight a new mechanism involved in hypothalamic glucose-sensing. Further studies are however necessary to determine the physiological role of the sweet taste receptors expressed in hypothalamic glucose-sensing neurons in the control of energy homeostasis. It will also be interesting to determine the ionic channel mediating the effect of sweeteners or glucose downstream sweet taste receptor. The involvement of K_ATP or TRPC channels has always been linked to metabolic-dependent mechanisms. However, one could hypothesize that these channels mediate sweet taste receptor-dependent glucose-sensing. Twik-related acid-sensitive potassium-like (TASK) 1/3 channels could also be a relevant candidate as they take part in metabolism-independent mechanism of orexin GI neurons in the LH (Gonzalez et al., 2008, 2009).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The reviewer GTD and handling Editor declared their shared affiliation.