Beyond appetite: Acylated ghrelin as a learning, memory and fear behavior-modulating hormone

Although often referred to as a hunger hormone, recent evidence highlights a neuroprotective function of acylated ghrelin (AG) and a substantial role in the regulation of declarative and aversive memories as well as fear behavior. As such


Introduction
As a major metabolic hormone, ghrelin is predominantly produced and released by gastric X/A-like cells (Date et al., 2000), although the peptide is also endogenously expressed in peripheral tissue and across the brain (Ferrini et al., 2009). In a series of catalytic steps, mature ghrelin is cleaved from its precursor preproghrelin. Furthermore, the peptide may be acetylated at Ser 3 by ghrelin O-acyltransferase, resulting in bioactive acylated ghrelin (AG) and its counterpart non-acylated ghrelin (DAG) (Yanagi et al., 2018). Once generated, AG and DAG are secreted into the blood stream in response to nutrient deficiency, fasting and energy shortage to encourage appetite and food-seeking behaviour (Cummings et al., 2001;Liu et al., 2008;Yanagi et al., 2018). Additionally, both variants of ghrelin readily cross the blood brain barrier (BBB), diffusing into areas such as the hypothalamus or hippocampus (Banks et al., 2002;Diano et al., 2006;Yanagi et al., 2018). Importantly, only AG, but not DAG, is capable of stimulating growth hormone secretagogue receptor type 1α (GHS-R1α) (Yanagi et al., 2018).
Although often referred to as a 'hunger hormone', AG and its ubiquitously distributed receptor GHS-R1α execute many important functions in the physiological system. For instance, GHS-R1α is expressed in the stomach, adrenal and pituitary gland, pancreas, ovaries, testicles, intestines and liver to regulate the hormonal and metabolic balance, in multiple hypothalamic nuclei and the dorsal vagal complex, including the afferent vagus nerve in the periphery, to induce feeding responses, and in the substantia nigra (SN), ventral tegmental area (VTA), dorsal and median raphe nuclei, cortex and the dorsal hippocampus, including Abbreviations: AG, acylated ghrelin; 5-HT, serotonin; 5-HTR, serotonin receptor; GH, growth hormone; γ-aminobutyric acidPTSD, (GABA), growth hormone secretagogue receptor type 1α (GHS-R1α)post-traumatic stress disorder; LA, lateral nucleus of the amygdala; basolateral amygdalaCTA, (BLA), dorsal raphe nucleus (DRN)conditioned taste aversion. the cornu ammonis (CA)1, CA2, CA3 and the dentate gyrus (DG), to navigate long-term memory, food/reward behaviour, motivation, spontaneity, anxiety and other cognitive or emotional aspects ( Fig. 1) (Abdalla, 2015;Date, 2012;Gnanapavan et al., 2002;Guan, Yu, Palyha, McKee, Feighner, Sirinathsinghji, Smith, VanderPloeg et al., 1997;Jiang et al., 2006;Yanagi et al., 2018). AG is thought to be the main mediator of the neuroprotective effects of caloric restriction (Bayliss et al., 2016), hence we have recently evaluated the potential of AG as a neuroprotective agent in Alzheimer's disease (AD) and Parkinson's disease (PD) (Reich and Holscher, 2020). Additionally, besides regulating feeding and food-related learning Serrenho et al., 2019), we highlight a substantial role of AG in the formation of various forms of memory and associated behaviors.

Synaptic plasticity and memory formation
Long-term potentiation (LTP) represents an interneuronal process that results in the permanent strengthening of synapses, thought to be the basis of memory formation. LTP is initiated by electrical impulses (action potentials) that arise due to mechanical high-frequency stimulation, the exposure to an environmental stimulus, or training in novel tasks. The latter lead to the opening of voltage-gated Ca 2+ channels (Ca V s), Ca 2+ influx, and the associated activation of various Ca 2+ -sensitive modulators that mediate the presynaptic liberation of glutamate into the synaptic cleft. Glutamate subsequently diffuses towards and interacts with excitatory α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptors (AMPARs), N-methyl-D-aspartate receptors (NMDARs) and kainate receptors on postsynaptic neurons, resulting in depolarization. Moreover, NMDAR-associated ion channel opening induces the postsynaptic influx of Ca 2+ , leading to the initiation of second messenger cascades that comprise kinases such as Ca 2+ / calmodulin-dependent protein kinase (CaMK), CaMKII/IV, PKC, tyrosine kinase or extracellular signal-regulated kinase (ERK). In conjunction, the latter kinases mediate the phosphorylation and incorporation of AMPARs into postsynaptic terminals and navigate the remodeling of dendritic spines. Moreover, ERK-and cyclic adenosine monophosphate (cAMP)/protein kinase A (PKA)-signaling induce the key transcription factor cyclic adenosine monophosphate (cAMP)/cAMP response element-binding protein (CREB) and the transcription of immediate early genes (IEGs), for instance c-Fos, early growth response protein 1 (Egr-1) or activity-regulated cytoskeleton-associated protein (Arc), to enhance the synaptic strength (Byth, 2014;Chaaya et al., 2018;Deak and Sonntag, 2012;Warburton, 2014). While the expression of IEGs may indicate plasticity, synaptic strength is electrophysiologically measured in form of total current flow, termed excitatory postsynaptic potential (EPSP), across postsynaptic membranes. Alternatively, γ-aminobutyric acid (GABA)-induced inhibitory postsynaptic potentials (iPSPs) may be assessed. Various events may facilitate EPSPs, and thus, drive LTP, such as the increased release of excitatory and decreased release of inhibitory neurotransmitters (GABA) at presynaptic terminals, the incorporation of further postsynaptic excitatory neurotransmitter receptors, the enhancement of neurotransmitter receptor affinities towards their substrates, or prolonged Ca 2+ channel opening times. Indeed, various excitatory and inhibitory neurotransmitters, including glutamate, GABA, dopamine, 5-HT, acetylcholine and noradrenaline (NA), their individual receptors as well as several ion channels cooperate in an intricated time-, contextand region-specific manner to regulate LTP and its counterpart longterm depression (LTD), which represents the selective weakening of synaptic strength (Chaaya et al., 2018;Deak and Sonntag, 2012;Palacios-Filardo and Mellor, 2019;Warburton, 2014).
On a cellular level, memory is generated by the synchronous firing of interconnected neurons during learning, leading to the stabilization of this neuronal network. While LTP and LTD in different brain regions drive distinct types of memory, see (Amin and Malik, 2013), the hippocampus imparts the formation of declarative long-term memories, including contextual/associative, semantic, episodic and spatial memories (Bird and Burgess, 2008). In a series of events, sensory information is transmitted from the (parahippocampal) perforant path and entorhinal cortex to DG granule cells, across the CA3 and towards CA1-located pyramidal neurons. The CA1 poses the final output station of the dorsal hippocampus, which projects to cortical fields to integrate memories or to initiate decision-making, while looping back to the perforant path to consolidate hippocampal input (Deak and Sonntag, 2012).
Importantly, hippocampal plasticity is thought to be a direct reflection of the ability to acquire and consolidate hippocampus-associated memories. Within the first 24 h, the hippocampus temporally retains the formed declarative information, yet, after approximately a week has passed, these memories eventually become hippocampus-independent. Indeed, other brain regions, such as the prefrontal cortex, mediate Fig. 1. Expression sites of GHS-R1α in the human brain. GHS-R1α is found at high densities in multiple hypothalamic nuclei (in particular the arcuate nucleus) to induce appetite, the hippocampal formation (CA1, CA2, CA3 and DG) to modulate declarative memory, the pituitary gland to affect hormone release and various midbrain sites (ventral tegmental area, dorsal raphe nuclei and more) to, amongst others, regulate dopamine and serotonin transmission (Airapetov et al., 2021). The shown relative mRNA expression levels (transcripts per kilobase million) were derived from the Human Protein Atlas (Uhlen et al., 2015). memory expression and store consolidated memories in the long-term (Beck and Pourie, 2013;Deak and Sonntag, 2012).
Besides the hippocampus, the presence of emotional stimuli engages the amygdala to drive the formation of emotional memories. Traditionally, research on emotional memories falls into two main categories, including classical cue-based Pavlovian conditioning, commonly with rodents, or emotional episodic memories (Dunsmoor and Kroes, 2019). As relevant for the scope of this review, we will expand on fear memory formation through Pavlovian and contextual fear conditioning in Section 5.1.
Furthermore, the process of creating a memory involves several steps. In the acquisition phase (i), as occurring during the training of rodents in a behavioral paradigm, the induction of cellular plasticity (LTP) in selected brain regions creates an initial, yet labile memory. Subsequently, when training ceases and the animal is returned to its cage, the memory trace is consolidated (ii) through CREB-mediated protein synthesis into a stable long-term memory. Finally, retrieval (iii) describes the recollection of a previously stored memory (Abel and Lattal, 2001;de Quervain et al., 2017).
As a major objective this review, we will critically appraise the currently available animal and human data to illustrate the modulatory impact of AG on memory formation, the underlying brain regions, biochemistry and behavioral changes, with a focus on the hippocampus and amygdala. Hereafter, we will summarize the in vivo and, as available, clinical data to show a supportive role of AG in the formation of hippocampus-dependent memory under healthy and demented conditions (Section 3). This will be followed by the respective neurotransmitter-, plasticity-and neurogenesis-modulating effects of AG in the hippocampus (Section 4). Finally, we will explore the impact of AG on various forms of fear memory, conditioned taste aversion (CTA) as well as the recently established clinical link to post-traumatic stress disorder (PTSD). Indeed, the context-dependent effects of AG on fear memory strength and PTSD involve a complex interplay between the ghrelin system, various amygdaloid nuclei, the hippocampus and serotoninsignaling (Section 5).

Acylated ghrelin enhances hippocampal memory
As presented below, animal studies advocate a learning and consolidation-enhancing effect of exogenously administered AG, synthetic ghrelin analogues as well as the ghrelin receptor GHS-R1α on hippocampus-dependent memory in wild-type and AD rodent models.

Ghrelin enhances spatial memory in wild-type animals
Various studies in healthy animals show that AG promotes the formation of hippocampus-dependent spatial memories. While these effects seem to be dose-dependent (inverse U-shaped), there is evidence for an acquisition and consolidation-improving outcome of AG treatment, GHS-R1α-modulating drugs and GHS-R1α itself on object recognition (Carlini et al., 2007(Carlini et al., , 2008Atcha et al., 2009;Diano et al., 2006;Ribeiro et al., 2020) and spatial memory (Atcha et al., 2009;Chen et al., 2011;Toth et al., 2010;Davis et al., 2011;Tian et al., 2019).
The administration of intra-hippocampal AG post-training enhanced the consolidation of short-term (1 h) and long-term (24 h) object recognition memory memories (Carlini et al., 2007(Carlini et al., , 2008, while the pre-training administration of ghrelin analogues enhanced the novel object exploration time and index (preferential investigation of the novel over familiar objects), suggesting an acquisition-boosting effect of AG (Atcha et al., 2009). Similarly, the central injection of AG following training ameliorated undernutrition-induced deficits in 1 h and 24 h object recognition memory (Carlini et al., 2008). In turn, ghrelin -/mice exhibited impaired object recognition memory that could be restored with the initial daily infusion of AG for 14 days (Diano et al., 2006). Notably, a preliminary report announced that the pre-training injection of a GHS-R1α inverse agonist weakened the recognition of novel and displaced objects (Luís F. Ribeiro et al., 2020). Therefore, both the tonic activity (~50% Gα q/11 -coupling in the absence of ligands (Damian et al., 2012)) and stimulation of GHS-R1α improve the consolidation and, possibly, acquisition of spatial recognition memories.
In experimental paradigms that assess hippocampus-associated spatial learning and memory, the 4 day-long daily and pre-training peripheral injection of a ghrelin mimetic improved the training performance of rodents in the Atlantis Water Maze (Atcha et al., 2009), while dorsohippocampal AG strengthened spatial memory acquisition and retainment in the Morris Water Maze (MWM) (Chen et al., 2011). However, the intra-CA1 infusion of low doses of AG (8 ng per hemisphere) prior to every other training day impaired spatial acquisition and memory retainment (Zhao et al., 2014;Zhu et al., 2013). Notably, while the hormone augmented hippocampus-dependent passive avoidance memories when given at 100 ng or 150 ng, central injections of < 50 ng AG resulted in a trend for poorer performance. 200 ng AG, in fact, significantly decreased the inhibitory memory of healthy mice in the T-Maze foot shock avoidance paradigm. This implies that an inverse U-shaped relationship between the hippocampal concentrations of AG and memory exists (Diano et al., 2006). Interestingly, the microinjection of AG into the amygdala following the first MWM training day, in a GHS-R1α-dependent manner, led to enhanced performance on the second day. This proposes that the putative plasticity-suppressing effects of AG in the amygdala (Section 5.2.2) may facilitate the hippocampal consolidation of spatial memories (Toth et al., 2010).
On the other hand, reports concerning GHS-R1α-deficient mice are mixed. Two studies showed that the deletion of GHS-R1α impeded spatial acquisition across the test days and attenuated reference memories during the probe trial in the MWM (J. F. Davis et al., 2011;Tian et al., 2019). However, a separate study by  found that GHS-R1α -/mice did not show any differences in spatial learning compared to wild-type littermates, but displayed heightened spatial memory in the MWM when tested 7 days, but not 1 day, following training. First, to provide an explanation for the latter results, it must be considered that GHS-R1α has neuroprotective effects (Section 3.2) and that both receptor deletion or the age-associated complex formation of accumulating Aβ 1-42 with GHS-R1α, which impairs the function of the ghrelin receptor, generate an AD-like phenotype that is more explicit with greater age (Tian et al., 2019). Indeed, (Tian et al., 2019) observed early deficits in spatial (MWM) memory formation and reductions in the hippocampal synapse densities in 4 month old GHS-R1α -/mice, while 9 month old GHS-R1α -/rodents further exhibited impaired hippocampal LTP-induction and postsynaptic CaM-KII phosphorylation. What improved the spatial memory retainment in 6 month old GHS-R1α -/mice 7 days post training is a mystery, however. To add to the puzzle,  also observed reduced contextual fear memory strength, another form of hippocampus-dependent memory, 30 days after conditioning. As we explain in Section 5.2.7, long-term increases in the plasma AG levels beyond 8 days, as achieved with caloric restriction, prolong the hippocampal (contextual fear) memory recall by stimulating neurogenesis (Hornsby et al., 2016;Gu et al., 2012;Pan et al., 2012).
In conclusion, the evidence emphasizes a memory formationenhancing role of AG and GHS-R1α in the hippocampus. How AG and GHS-R1α affect spatial memory retainment and recall over a multi-week period following training, however, requires further investigations.

Ghrelin rescues cognitive decline in Alzheimer's disease in vivo models
Besides healthy animals, AG rescued cognition in multiple animal models of AD. In the latter, AG or ghrelin mimetics enhanced the performance of rodents in passive avoidance, novel object recognition, MWM, Barnes Maze and Y-Maze paradigms, signifying improvements in the acquisition of inhibitory avoidance, reference and spatial memory (Dhurandhar et al., 2013;Diano et al., 2006;Eslami et al., 2018;Kang et al., 2015;Kunath et al., 2015;Madhavadas et al., 2014;Moon et al., 2011;Santos et al., 2017).
Importantly, AG improved the cognitive function of the AD animals by ameliorating their cerebral pathology, including Aß and Tau pathology, neuroinflammation, hyperglycemia, insulin resistance and the activation of GSK-3ß (see (Reich and Holscher, 2020) for details). This, in turn, resulted in enhanced EPSPs across CA3-CA1 synapses (even in control mice), rescued from the Aß-induced weakening of high-frequency stimulated LTP (Eslami et al., 2018;Santos et al., 2017), normalized brain acetylcholinesterase levels (Madhavadas et al., 2014) and prevented a decline in plasticity-associated p-CREB, CREB binding protein and p-300 levels plus the reduction in Egr-1 expression in the CA1 and DG regions (Bartolotti et al., 2016;Jeong et al., 2018). Likewise, AG-treatment upregulated CREB, p-CREB and brain-derived neurotrophic factor (BDNF) levels in streptozotocin-injected rats, improving cognition , while the intra-hippocampal co-administration of AG preserved MWM spatial memory in response to a seizure-inducing GABA A receptor antagonist (Babri et al., 2013). On the other hand, ghrelin -/mice showed attenuated spatial (recognition) memory, olfactory deficits and elevated micro-and astroglial immunoreactivity in the hippocampus (V. V. Santos et al., 2017).

Association studies of ghrelin and cognition are misleading in humans
While there are currently no human studies that assess the outcome of administered AG or its analogues on memory or cognition, it was reported that the plasma ghrelin levels were inversely associated with neuropsychological test scores in non-demented elderly individuals, including auditory working memory, verbal recall and confrontation naming (Spitznagel et al., 2010). The investigation had a few major weaknesses, however, such as limited sample size (35 participants), the use of overweight patients (BMI 28.35 ± 4.60) with a clinical history of cardiovascular risk factors that accelerate cognitive decline, such as hypertension (42.9 %) and elevated cholesterol levels (37.1 %), and the lack of a healthy control group. Also, plasma ghrelin levels are chronically reduced during obesity (Rigamonti et al., 2002;Shiiya et al., 2002;Tschop et al., 2001), which presumably resulted in skewed findings (Spitznagel et al., 2010).

The ghrelin system modulates hippocampal plasticity
As described in Section 3, there is robust evidence that AG enhances hippocampal learning and memory consolidation in animals. Therefore, Section 4 aims to condense the underlying memory-facilitating mechanisms of AG. Table 1 provides an overview of studies that have investigated the plasticity-altering and biochemical effects of AG and its constitutively active receptor, GHS-R1α, in the hippocampus. Subsection 4.1 will explore these studies and elucidate the established, but also still enigmatic, role of the ghrelin system in the hippocampal regulation of glutamatergic, GABAergic and dopaminergic neurotransmission. We will also highlight play a considerable role of AG and GHS-R1α in the mesolimbic dopamine pathway, with implications on feeding behavior, locomotion and novelty learning.

Acylated ghrelin promotes glutamatergic neurotransmission in the hippocampus
As illustrated in Fig. 2, both GHS-R1α and its ligand AG enhance glutamatergic neurotransmission in the dorsal hippocampus. Spatially, it was shown that GHS-R1α distributed across the cell body and dendrites of isolated hippocampal neurons and in the CA1, located in close proximity to hippocampal excitatory synapses (Ribeiro et al., 2014;Berrout and Isokawa, 2018). Using in vivo monitoring under anesthesia, the hippocampal infusion of AG induced slowly raising and long-lasting (>4 h) postsynaptic plasticity in a phosphoinositide 3-kinase (PI3K)/Akt-dependent and NMDAR-independent manner in the rat DG, resulting in improved spatial memory. Of note, even though AG did not affect the expression of high frequency-stimulated LTP, the AG-driven and delayed activation of ERK 1/2 (2 h) prevented a decline in these electrically evoked LTP (Chen et al., 2011). Similar to the latter investigation, AG did not elevate the magnitude of stimulus-induced (NMDAR-dependent) LTP in another ex vivo study. However, the application of AG lessened the LTP generation threshold in the DG, similarly resulting in GHS-R1α-dependent improvements in hippocampal passive avoidance memory consolidation (Ghersi et al., 2015). Besides the DG, there is evidence that AG or the GHS-R1α agonist MK-0677 dose-dependently elevated NMDAR-driven EPSPs across CA3-CA1 synapses (Ribeiro et al., 2014;Muniz and Isokawa, 2015), predominantly by increasing the synaptic AMPA/NMDA current ratio (Ribeiro et al., 2014) and facilitating LTP generation in the CA1 (Diano et al., 2006). Furthermore, AG dose-dependently stimulated the (4-aminopyridine-evoked) presynaptic glutamate release by hippocampal synaptosomes and in the DG in vivo, while enhancing the excitability of DG granule cells in the rat brain (Chen et al., 2011;Ghersi et al., 2015). This suggests that modifies the pre-and postsynaptic space, which, as explained below, involves the modulation of AMPARs.
Mechanistically, AG regulates the synaptic incorporation of AMPARs in a temporal manner, which can be categorized into short-term and long-term effects (see also Fig. 2). In the short-term, the use of a ghrelin mimetic rapidly elevated PI3K/Akt activity in hippocampal slices (~30 min) and evoked the PI3K-and PKA-co-dependent phosphorylation of GluA1-AMPAR subunits at Ser 845 , which promoted AMPAR surface exposition. Furthermore, simultaneous NMDAR channel activity was necessary to incorporate these exocytosed GluA1-AMPARs into synapses (Ribeiro et al., 2014). Notably, the Ser 845 -phosphorylation of GluA1-AMPARs was shown to be necessary for LTP execution in the CA1 (J.Y. H.K. Lee et al., 2010) and spatial memory retainment in the MWM (Lee et al., 2003). In this context, hippocampal GHS-R1α receptor interactions provide an explanation for the non-canonical induction of cAMP/PKA-signaling and the PKA-mediated AMPAR at Ser 845 -phosphorylation by AG. Generally, the canonical activation of GHS-R1α involves Gα q/11 -coupling, phospholipase C (PLC) activation and the simultaneous, PLC-mediated turnover of phosphatidylinositol 4,5-bisphosphate (PIP2) into inositol triphosphate (IP3) as well as the generation of diacylglycerol (DAGly). IP3 further triggers the liberation of calcium (Ca 2+ ) from intracellular ER stores. Then, DAGly and the IP3-driven accumulation of Ca 2+ induce protein kinase C (PKC). The induction of cAMP/PKA has been reported for GHS-R1α, yet is still under debate and conditional (Reich and Holscher, 2020;Yin et al., 2014). In this context, it was shown that GHS-R1α forms heteromers with D 1 R in neurons across the dorsal hippocampus (Jiang et al., 2006;Kern et al., 2015;Tian et al., 2019). AG and dopamine co-treatment triggered the heterodimerization of GHS-R1α and D 1 R, engaged Gα i/o (instead of Gα q/11 by GHS-R1α) and induced a multi-fold synergistic augmentation of cAMP accumulation, as compared to dopamine/D 1 R-signaling alone (Jiang et al., 2006). Notably, AG also induces the mesolimbic transmission of dopamine from the VTA to the hippocampus (Section 4.1.3). Likewise, the association of GHS-R1α with its inactive splicing variant, GHS-R1β, was shown to induce Gα i/o -coupling (Navarro et al., 2016). In this case, as observed in primary hippocampal and striatal neurons, the presence of dopamine or AG, but not both, initiated the GHS-R1β-necessitating GHS-R1β: GHS-R1α:D 1 R heteromer formation, a switch from Gα i/o to Gα s/olf and cAMP accumulation. There were no synergistic effects of AG and dopamine on cAMP accumulation anymore, however (Navarro et al., 2016). As such, AG may trigger the formation of GHS-R1α:GHS-R1β:D 1 R oligomers in the CA1, (CA3) or DG to drive non-canonical Gα s/olf -evoked cAMP/PKA-signaling in the hippocampus.
In the long-term, prolonged ghrelin agonist treatment (5-20 h), in a GHS-R1α-dependent fashion, resulted in PKC activity, the activation of stargazin (which is phosphorylated by PKC, but also CaMKII (Tomita Table 1 Hippocampal effects of AG, GHS-R1α agonists, antagonists and inverse agonists, receptor deletion or GHS-R1α receptor interactions on plasticity as well as glutamatergic (AMPARs and NMDARs), GABAergic and dopaminergic neurotransmission.

In vivo
Dorsohippocampal infusion into rats. LTP recording under anesthesia.
In agreement with the role of AG in AMPAR regulation, the i.c.v. microinjection of a GHS-R1α antagonist downregulated the transcriptional levels of GluA1-AMPARs in the rat hippocampus, worsening passive avoidance memory consolidation (Beheshti et al., 2020). Furthermore, the constitutive activity of GHS-R1α is involved. A pre-print paper of a study of a combination of hippocampal primary neurons and slice culture, reported that the use of GHS-R1α inverse agonists encouraged a more diffuse surface distribution of GluA1/GluA2-AMPARs, lessened their basal and glycine (NMDAR)-induced synaptic recruitment, weakened GluA1 phosphorylation at Ser 845 , decreased AMPA/NMDA current ratios at CA3 -CA1 synapses and worsened the murine acquisition of object memories following the i.p. injection of the BBB-penetrant GHS-R1α inverse agonist AZ12861903 (Ribeiro et al., 2020). Synoptically, in the short-term, the AG-induced oligomerization of hippocampal GHS-R1β:GHS-R1α:D 1 R leads to the Gα s/olf /cAMP/PKAdriven GluA1-AMPAR-phosphorylation at Ser 845 , priming them for the NMDAR activity-induced recruitment to hippocampal synapses. In the long term, the delayed activation of the GHS-R1α/PKC/stargazin axis, the stargazin-induced phosphorylation of GluA1-AMPARs at Ser 831 and the PKC-conveyed induction of ERK 1/2 maintain the synaptic insertion of AMPARs. The latter effects of AG on AMPAR recruitment, in combination with enhancing stimulus-evoked presynaptic glutamate release (Chen et al., 2011;Ghersi et al., 2015), likely explain the reduction in the LTP generation threshold in the CA1 and DG (Diano et al., 2006;Ghersi et al., 2015), the in vivo increase in postsynaptic excitability in DG granule cells and the sustainment of long-lasting LTP in the DG (Chen et al., 2011) following AG treatment.
Besides AMPARs, AG modulates NR1-and NR2-NMDAR subunits (Fig. 2). The latter may be composed of seven subunits, including the obligatory NR1 and glutamate-interacting NR2A-D or, in some cases, NR3A/B. Typically, NMDARs are found in the form of NR1/NR2A/NR2B triheterodimers or NR1/NR2(A/B) heterodimers in the forebrain. Interestingly, GHS-R1α is not only located closely to dendritic NMDARs in pyramidal neurons, but the bath application of AG enhanced the phosphorylation of NR1-NMDARs at Ser 896/897 by 20% and potentiated NMDAR-mediated EPSPs in the CA1. The latter plasticity enhancements were abolished following treatment with a GHS-R1α antagonist, inverse agonist or in GHS-R1α -/mice, while EPSPs were reduced in GHS-R1α -/+ heterozygous mice (Muniz and Isokawa, 2015). Mechanistically, AG induced hippocampal PKC activity in the CA1 following 5 h+ exposure times ex vivo (Ribeiro et al., 2014), with PKC known to drive the exocytosis-promoting phosphorylation of NR1-NMDAR subunits at Ser 896 (Scott et al., 2001;Tingley et al., 1997). Moreover, AG triggered cAMP/PKA-signaling and GHS-R1α-dependent NR1-NMDAR phosphorylation in the CA1 ex vivo (Cuellar and Isokawa, 2011). The interaction of AG with GHS-R1β:GHS-R1α:D 1 R oligomers and Gα s/olf -driven cAMP accumulation likely lead to the PKA-mediated NR1-NMDAR phosphorylation at Ser 897 (Fig. 2) (Tingley et al., 1997). Besides NR1, the 24 h (caption on next page) N. Reich and C. Hölscher bath application of AG modestly (~10%) heightened the number of NR2B-NMDAR-positive neurons in the CA1 and DG in organotypic slices. Moreover, whilst NR2B antagonism raised the LTP generation threshold in the DG and impaired step-down memory consolidation, the administration of AG, which lowered the DG LTP generation threshold even in the absence of the NR2B antagonist, prevented these detrimental outcomes (Ghersi et al., 2015). Similarly, intermittent fasting, a condition that is linked to the systemic release of AG, led to the hippocampal upregulation of NR2B-NMDAR subunits, enhanced plasticity and improved spatial memory in vivo (Fontan-Lozano et al., 2007). Another study showed that exogenous AG, in a GHS-R1α-dependent manner, activated Fyn kinase, leading to the Fyn-dependent phosphorylation of NR2B at Tyr 1336 in the CA1 (Berrout and Isokawa, 2018). Notably, Fyn-mediated NR2B phosphorylation at various Tyr residues prevents the endocytosis of synaptic NMDARs, thus sustaining LTP (W. Lu et al., 2015;Prybylowski et al., 2005). Moreover, Fyn is typically activated in response to the Gα s/olf /cAMP/PKA pathway in the CA1, for instance following the activation of D 1 Rs by dopamine (Trepanier et al., 2012;Yang et al., 2014) As such, AG-binding to Gα s/olf -coupling GHS-R1β: GHS-R1α:D 1 R heteromers probably accounts for the induction of Fyn (Fig. 2). In summary, the evidence suggests that AG/GHS-R1α-signaling mildly stimulates the expression of the NMDAR subunits in both the CA1 and DG, whilst inducing the phosphorylation of NR1 via PKC at Ser 896 and PKA at Ser 897 , as well as NR2B at Tyr 1336 by Fyn, in the CA1. This local enhancement of NR1 and NR2B phosphorylation in the CA1, which promotes the surface presentation and synaptic recruitment of NMDARs (Scott et al., 2001;Lu et al., 2015;Prybylowski et al., 2005), might partly be responsible for the augmentation of classic NMDAR-induced EPSPs that were observed in response to AG or MK-0677 treatment in CA3 -CA1 synapses (Ribeiro et al., 2014;Muniz and Isokawa, 2015). Intriguingly, the evidence suggests that AG and its receptor drive plasticity in a distinct manner in area CA1 (NMDAR activity-dependent) versus the DG (NMDAR-independent). As explained earlier, AG boosted classic NMDAR-dependent LTP in the CA1 and required NMDAR activity for the synaptic recruitment of GluA1-AMPARs (Ribeiro et al., 2014;Muniz and Isokawa, 2015). However, the GHS-R1α agonist MK-0677 and tonic activity of GHS-R1α did not increase NMDA currents, but elevated AMPA conductivity, in CA1 synapses (Ribeiro et al., 2014;Ribeiro et al., 2020). This suggests that NMDAR activity is required as a signal to trigger the postsynaptic AMPAR incorporation, whereby AG/GHS-R1α-signaling facilitates this process by stimulating AMPAR surface presentation and locking synapse-recruited AMPARs in place. In stark contrast to area CA1, NMDAR antagonism did not affect the LTP-potentiating, LTP-extending nor LTP generation threshold-lowering effects of AG in the DG (Chen et al., 2011;Ghersi et al., 2015). Chen et al. (2011) denoted this phenomenon as 'a new form of synaptic plasticity' that is dependent on the atypical activation of PI3K-signaling following AG treatment, yet does not necessitate high frequency stimulation or NMDAR induction. PI3K plays a crucial role in the DG, with PI3K inhibitors blocking the local presynaptic glutamate release and execution of LTPs (Kelly and Lynch, 2000). In particular, a training-induced increase in BDNF/TrkB/PI3K/Akt-signaling, as dependent on the BNDF-driven downstream activation of PI3K and ERK 1/2 (Gottschalk et al., 1999), mediate hippocampal plasticity and spatial memory formation (Mizuno et al., 2003). Indeed, AG was shown to activate the PI3K/Akt pathway in DG-derived neural stem cells (Chung et al., 2013;Johansson et al., 2008). AG might also indirectly induce Akt/PI3K-signaling via IGF-1, given that a GHS-R1α agonist was shown to elevate IGF-1 synthesis in the hippocampus of healthy rodents (Frago et al., 2002;Dyer et al., 2016). Further knockdown evidence suggests that GHS-R1α located on afferent fibers of the vagus nerve, as activated by circulatory AG (Date, 2012), influences the expression of the PI3K/Akt-inducing BDNF in, at least, area CA3, also affecting object recognition memory (Davis et al., 2020). Finally, the greater density of GHS-R1α:D1R-co-expressing neurons in the DG and CA3 compared to the area CA1 (Kern et al., 2015) might lead to subregional differences in AG-signaling that, possibly, alter the dependency on NMDAR activity in the CA1 and DG areas (Fig. 2).

Acylated ghrelin restricts neuroinhibitory GABA release in the hippocampus
AG and its cognate receptor further regulate GABAergic Fig. 2. Modulation of hippocampal synapses and plasticity-associated events by AG. In the hippocampal formation, AG potentiates NMDAR-induced EPSPs in the CA1, whilst lowering the LTP generation threshold and prolonging LTP expression in an NMDAR-independent manner in the DG. The latter involve the enhancement of the presynaptic glutamate release, the priming of AMPARs for their NMDAR activity-induced recruitment to synapses and the synaptic anchoring of AMPARs, but also NMDARs in the CA1, by AG-signaling. Additionally, both constitutively active GHS-R1α and GHS-R1α/AG-signaling block the presynaptic release of GABA by evoking the endocytosis of Ca V 2.1/Ca V 2.2 in the DG. By driving the latter changes, AG improves the formation of declarative memories. In the short term (~30 min), AG treatment non-canonically elicits the cAMP/PKA and PI3K/Akt pathway. Gα s/olf /cAMP/PKA-signaling is induced by the interaction of either AG or dopamine with GHS-R1α:GHS-R1β:D 1 R heteromers, leading to the exocytosis-promoting GluA1 phosphorylation at Ser 845 , which primes AMPARs for their activity-induced synaptic recruitment. PKA further phosphorylates NR1-NMDARs at Ser 896 and activates Fyn, which evokes NR2B-NMDAR phosphorylation at Tyr 1336 , resulting in the enhanced surface exposition and synaptic anchoring of NMDARs. It is still enigmatic how AG triggers GluA1-AMPAR (Ser 845 ) co-phosphorylating PI3K/Akt-signaling, possibly occurring indirectly through increasing the hippocampal and plasma IGF-1 levels (further discussed in the main text). In the long term (20 h), the binding of AG to GHS-R1α results in a receptor switch from Gα i/o to Gα q/11 and canonical PLC/PIP2/IP3/DAGly/PKC-signaling, which stimulates stargazin to enhance GluA1-AMPAR conductivity (Ser 831 phosphorylation) and synaptic AMPAR trapping, supports the synaptic localization of NMDARs (PKA-mediated NR1-NMDAR phosphorylation at Ser 897 ) and induces ERK/CREB to sustain long-term plasticity. Notably, the LTP threshold-reducing effects of AG have partially been associated with enhancing NO/nNOS activity in the DG. Further studies are necessary to investigate this phenomenon, since AG does not alter the intracellular Ca 2+ levels and drives plasticity by elevating AMPAR, but not NMDAR, currents in the hippocampus. Rising hippocampal AG levels, for example in response to caloric restriction, or AG treatment further transiently augment the presynaptic vesicle density, spinogenesis and synapse formation, promote neurogenesis (and, thus, memory resolution and pattern separation in the DG; see Section 4.2.2) and elevate Rag-1, IGF-1 and BDNF levels in the hippocampus; all events that are associated with improved plasticity and the formation of declarative memories. Lastly, AG evokes the mesolimbic transmission of dopamine to the hippocampus, while inhibiting the plasticitysuppressing release of 5-HT at synaptic terminals. Finally, in postsynaptic neurons and the absence of GHS-R1β, the co-binding of AG and dopamine induces the oligomerization of GHS-R1α with D 1 R. Once assembled, AG-binding to the receptor oligomers augments dopamine-induced cAMP accumulation across D 1 Rs, while the interaction of dopamine with GHS-R1α:D 1 Rs leads to non-canonical Gα q/11 -signaling, ultimately enhancing memory consolidation. Notably, GHS-R1α further dimerizes with 5-HT 2c R (not shown) in the hippocampus, whereby the interaction of 5-HT with GHS-R1α:5-HT 2c R complexes abolishes hippocampal AG-signaling, reduces plasticity and interferes with the consolidation of hippocampus-dependent (i.e. spatial or passive avoidance) memories. Dashed lines indicate incompletely understood signaling pathways. Abbreviations: acylated ghrelin (AG); growth hormone secretagogue receptor type 1α/β (GHS-R1α/GHS-R1β), dopamine receptor subtype 1 (D 1 R), serotonin (5-HT), γ-aminobutyric acid (GABA); γ-aminobutyric acid receptor (GABAR); serotonin receptor 2c (5-HT 2c R), voltage-gated Ca 2+ channel (CaV); α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor (AMPAR); N-methyl-D-aspartate receptors (NMDAR); excitatory postsynaptic potential (EPSP); inhibitory postsynaptic potential (iPSPs); cyclic adenosine monophosphate (cAMP); cAMP response element-binding protein (CREB); protein kinase A (PKA); phosphoinositide 3-kinase (PI3K); phospholipase C (PLC); phosphatidylinositol 4,5-bisphosphate (PIP2); inositol triphosphate (IP3); diacylglycerol (DAGly) protein kinase C (PKC); neuronal nitric oxide synthase (nNOS); mitogen-activated protein kinase (MEK), extracellular signal-regulated kinases (ERK); early growth response protein 1 (Egr-1); insulin-like growth factor 1 (IGF-1); insulin-like growth factor 1 receptor (IGF-1R); brain derived neurotrophic factor (BDNF); recombination activating gene 1 (Rag-1); dentate gyrus (DG); ventral tegmental area (VTA); dorsal raphe nucleus (DRN).
neurotransmission. Interestingly, it was revealed that the tonic activity of GHS-R1α dampens electric currents in Ca V 2.1 and Ca V 2.2. Indeed, the overexpression or deletion of GHS-R1α in primary hippocampal neurons confirmed that, predominantly by weakening Ca V 2.2 conductivity, GHS-R1α attenuated neuroinhibitory GABA release in vitro, while decreasing iPSPs in the granule cell layer of the DG ex vivo (Damonte et al., 2018) (see also Fig. 2). In area CA1, however, (Ribeiro et al., 2014) saw no differences in NMDA/GABA currents after 20 h GHS-R1α agonist treatment in organotypic slices, suggesting that AG might only suppress the release of GABA acutely or, selectively, in the DG. Likewise, in a Gα i/o -driven fashion, tonically active GHS-R1α suppressed N-and P/Q-type Ca 2+ currents by engaging Ca V β to trap Ca V α 1 subunits in the ER to reduce the surface exhibition of Ca V 2.1/Ca V 2.2, but also Ca V 1 and Ca V 3, in primary hypothalamic neurons (Lopez Soto et al., 2015;Mustafa et al., 2017). Moreover, Ca V -channel inhibition correlated with the expression levels of GHS-R1α (Lopez Soto et al., 2015). AG-binding to GHS-R1α evoked a switch from Gα i/o to Gα q/11 and G βγ , equally resulting in the quenching of Ca V 2.1 and (preferentially) Ca V 2.2 currents in a Ca V β-dependent manner. Indeed, Ca V 2 blockade by constitutively active GHS-R1α, synergistically augmented by receptor interaction with AG, suppressed stimulated, but not spontaneous, iPSPs in hypothalamic neurons (Lopez Soto et al., 2015). Given that GHS-R1α is expressed throughout the dorsal hippocampal formation (Guan et al., 1997;Zigman et al., 2006), GHS-R1α/AG-induced Ca V 2 internalization may attenuate GABA release not only in the DG (Damonte et al., 2018), but throughout the hippocampus.

The ghrelin system potentiates dopamine transmission and novelty learning
AG poses a major component of the food/reward machinery that is navigated by the mesolimbic transmission of dopamine. Although the locus coeruleus is the primary source of dopamine for the dorsal hippocampus, the VTA also partially supplies the hippocampus with dopamine (McNamara and Dupret, 2017;Serrenho et al., 2019).
Generally, caloric restriction and the associated plasma release of AG, in a GHS-R1α-dependent fashion, stimulate orexin-expressing neurons in the lateral hypothalamus which, subsequently, induce VTA dopaminergic neurons. The latter was necessary to induce postprandial dopaminergic spikes in vivo (Cone et al., 2014). Furthermore, the expression of GHS-R1α has been confirmed for VTA neurons (Abizaid et al., 2006). Receptor activation by its ligand AG leads to the mesolimbic projection of dopamine from the VTA to the nucleus accumbens, stimulating food-seeking behavior, meal intake and, if given centrally, locomotion in a GHS-R1α-dependent manner (Abizaid et al., 2006;Cornejo et al., 2018;Jerlhag et al., 2007Jerlhag et al., , 2012Quarta et al., 2009;Skibicka et al., 2011Skibicka et al., , 2012. Moreover, GHS-R1α co-localized with cholinergic neurons in the laterodorsal tegmental area that projects towards the VTA , whereby the AG-driven stimulation of locomotion, food intake and accumbal dopamine outflow necessitated nicotinic acetylcholine receptors (nAChRs) in the VTA Jerlhag et al., 2006Jerlhag et al., , 2008. On the other hand, the pre-test and systemic utility of GHS-R1α antagonists prevented reward-induced dopamine release, locomotor induction and psychostimulant/alcohol-encouraged conditioned place preference (Engel et al., 2015;Jerlhag et al., 2010Jerlhag et al., , 2009Skibicka et al., 2011Skibicka et al., , 2012Sustkova-Fiserova et al., 2014).
Intriguingly, the AG-driven stimulation of the mesolimbic VTA pathway boosts the formation of hippocampus-dependent memories. A study showed that the microinjection of AG in close proximity to the hippocampus, in a D 1 R-dependent manner, enhanced memory acquisition in the novel object task (Jacoby and Currie, 2011). Indeed, photostimulation revealed that VTA dopaminergic neurons project towards the CA1 to stabilize food reward/location memories during training and improve spatial recall in a crossword-like maze (McNamara et al., 2014). Furthermore, the intra-VTA injection of a GHS-R1α antagonist impeded the hippocampus-mediated consolidation of passive avoidance memories (Beheshti and Aslani, 2018). Another study found that the joint administration of ineffective doses of AG (intra-CA1) and nicotine (subcutaneous) could, synergistically, prevent morphine-driven impairments in the consolidation of passive avoidance memories (Nazari-Serenjeh et al., 2019). Thus, AG engages the dopaminergic-cholinergic reward axis in the VTA to potentiate hippocampus-dependent novelty (spatial) learning and the consolidation of passive avoidance memories.
Indeed, GHS-R1α augments hippocampal plasticity in cooperation with D 1 R (see Fig. 2). In the absence of ghrelin, GHS-R1α heterodimerized with D 1 R in neurons across multiple brain regions, including the hippocampal CA1-3 and DG, but also the cortex, midbrain, substantia nigra and the VTA (Jiang et al., 2006;Kern et al., 2015;Tian et al., 2019). Typically, the interaction of dopamine with D 1 R leads to recruitment of Gα s , intracellular cAMP amassment and PKA activation; all processes that have been associated with hippocampal long-term memory (Abel et al., 1997;Huang and Kandel, 1995). Following GHS-R1α:D 1 R heterodimer formation, however, the downstream signaling of dopamine is altered, leading the coupling of Gα q/11 to D 1 R (as opposed to Gα s ), PLC and IP3 activation, intracellular Ca 2+ release, the induction of CaMKII, the CaMKII-orchestrated phosphorylation of AMPAR subunits at sites associated with synaptic plasticity, synaptic rearrangements, the exocytosis of glutamate receptors and the expression of early hippocampal plasticity markers (Kern et al., 2015). Notably, the intra-DG infusion of a D 1 R agonist post extinction training enhanced the consolidation of extinction memories, while the pre-test application of these drugs heightened food location working memory in the T-Maze, suggesting that hippocampal dopamine-signaling improved spatial memory retrieval. However, the latter memory improvements by the dopamine agonist were abolished by the co-administration of a GHS-R1α antagonist or in response to GHS-R1α -/knockout, likely by interfering with GHS-R1α:D 1 R heterodimers (Kern et al., 2015). Besides dopamine, AG also acts as a ligand for hippocampal GHS-R1α:D 1 R heterodimers, which, at least in the absence of GHS-R1β, amplifies dopaminergic signaling. In vitro, the co-presence of dopamine and AG induced GHS-R1α:D 1 R heterodimerization, while the binding of AG to GHS-R1α:D 1 R complexes recruited Gα i/o (instead of the canonical coupling of Gα q/11 to GHS-R1α) to synergistically potentiate dopamine/D 1 R-driven cAMP accumulation (Jiang et al., 2006) (Fig. 2). Notably, recent post-mortem investigations discovered that Aβ directly interacts with amino acid residues 42-116 of GHS-R1α in the hippocampus of AD patients, which both desensitized GHS-R1α to AG and impaired the association of GHS-R1α with D 1 R. In turn, the 30 day co-injection of AG and dopamine analogues opposed the loss of GHS-R1α:D 1 R heterodimers, prevented Aβ 1-42 :GHS-R1α complex formation, lessened synapse loss, restored stimulation-induced LTP and raised neurogenesis in the CA1, resulting in enhanced spatial memory acquisition and reference memory in 5xFAD animals in the MWM (Tian et al., 2019). As such, being impaired in AD, the hippocampal interaction of GHS-R1α with D 1 R facilitates plasticity and declarative memory.

Other plasticity-and memory-related processes
Besides modulating neurotransmission, AG exerts other plasticityfacilitating effects in the hippocampus that will be presented below (Section 4.2.1). Furthermore, the neurogenesis-enhancing properties of AG strengthen hippocampal memory through an LTP-independent process, known as hippocampal pattern separation (Section 4.2.2).

Ghrelin induces transient structural changes in the hippocampus
Indeed, studies in healthy rodents demonstrated that peripherally administered AG traversed the BBB, diffused into the dorsohippocampal CA1, CA3 and DG, interacted with GHS-R1α, upregulated CA1 spine density and strengthened CA1 EPSPs. In contrast, ghrelin-null mice displayed lower spine synapse numbers in the CA1 (Diano et al., 2006). Strikingly, the cerebral deficits observed in GHS-R1α knockout mice, including decreased synapse density in the CA1, weakened stimulus-induced LTP and EPSPs across CA3-CA1 synapses and even Aβ burden, paralleled those in the 5xFAD animal model (Tian et al., 2019). This implies that both GHS-R1α and ligand-receptor interactions modulate the hippocampal architecture to facilitate plasticity.
In the DG, the intra-hippocampal infusion of AG dose-dependently increased the generation of nitric oxide (NO) by neuronal NO synthase (nNOS), lessened the hippocampal threshold to trigger LTP and improved passive avoidance memory consolidation. Furthermore, the hippocampal LTP threshold across all rodent groups, as lowered by AG, was inversely correlated to memory retainment. However, while a NOS inhibitor prevented the memory enhancement bestowed by AG, NOS blockade could only partially reverse the AG-evoked decrease in the LTP threshold . This suggests that AG facilitates LTP induction in the DG by NOS/NO-independent means as well, likely by augmenting glutamatergic and weakening GABAergic neurotransmission (see Fig. 2).
Generally, the activity of the Ca 2+ -sensitive nNOS is coordinated by the intracellular Ca 2+ levels, and its induction is linked to the enhancement of presynaptic vesicle recycling, glutamate release and LTPs in excitatory neurons of the hippocampus and cortex (Hardingham et al., 2013). As we argue in Section 4.1.1, AG-signaling does not appear to alter the intracellular Ca 2+ levels in hippocampal neurons, however. Furthermore, AG-driven LTP potentiation was NMDAR-dependent in the CA1 (Ribeiro et al., 2014;Muniz and Isokawa, 2015), but not DG (Chen et al., 2011;Ghersi et al., 2015). Even in the CA1, AG and GHS-R1α boosted AMPA, as opposed to NMDA, currents (Ribeiro et al., 2014;Ribeiro et al., 2020). Thus, it is unlikely, albeit not impossible, that AG induces NMDAR channel opening to elevate the intracellular Ca 2+ levels and nNOS activity in DG neurons. Indeed, the mechanisms underlying AG and GHS-R1α-induced plasticity in the DG, such as non-canonical PI3K/Akt activation, receptor interactions or, possibly, the engagement of IGF-1 and BDNF, require further studies.
AG was further shown to modulate synapses. In a GHS-R1α-mediated manner, AG increased the density of polymerized actin in rat hippocampal slices (Berrout and Isokawa, 2012). This indicates that AG rearranged dendritic spines to generate additional synapses, as observed in the hippocampus in vivo (Diano et al., 2006). In contrast, both the withdrawal of AG and GHS-R1α antagonism showed that these dendritic and synaptic changes are transient (Berrout and Isokawa, 2012). Similarly, in the cortex, in a GHS-R1αand dose-dependent fashion, AG boosted synapse densities, network outgrowth and activity (Stoyanova and le Feber, 2014;Stoyanova et al., 2013).
Interestingly, besides boosting synaptic vesicle and postsynaptic membrane deposit formation, AG elicited the expression of recombination activating gene 1 (Rag-1) in the CA3 region of wild-type mice (Wang et al., 2013). Although Rag-1 is primarily expressed by T-and B-lymphocytes, there is evidence that Rag-1, at least in part, contributes to the formation of spatial memory (Cushman et al., 2003;Fang et al., 2013;Marin and Kipnis, 2013).
In summary, when systemically released upon fasting, circulatory AG diffuses across the BBB into the hippocampus to induce transient dendritic actin polymerization, spine and synapse formation in area CA1, leading to greater stimulus-elicited LTP. In the DG, a yet poorly understood activation of nNOS by AG seems to lower the local LTP generation threshold. Finally, the AG-evoked expression of Rag1 might play a role in hippocampal memory.

Ghrelin facilitates neurogenesis and hippocampal pattern separation
Another mechanism that is associated with improved declarative memory formation is the enhancement of neurogenesis, and thus, hippocampal pattern separation by AG. Generally, in response to hippocampal activity, neuronal progenitor cells are born in the granular layer of the DG. These immature neurons subsequently mature, differentiate and are integrated as granule neurons into the DG. It has been proposed that plasticity mainly occurs in newborn DG neurons, allowing the preservation of already learned information that is stored in mature granule neurons. Although newborn neurons encode less detailed and more redundant information compared to mature DG granule neurons, their presence enhances memory resolution, hence facilitating the distinction of homogenous information (Aimone et al., 2011). Importantly, these immature DG neurons, whose total quantities are dependent on the rate of neurogenesis, contribute to hippocampal pattern separation. The latter describes the ability to discriminate and retain highly similar spatial, temporal or contextual information as well as temporally adjacent events as distinct memories (Aimone et al., , 2011Deng et al., 2010). For instance, the targeted elimination of immature DG cells, even though LTP in the CA1 area was not affected, resulted in learning deficits in the hippocampus-dependent trace conditioning task, which necessitates the association of a temporally separated CS and US (Shors et al., 2001). Similarly, the loss of neurogenesis led to impairments in the distinction of closely, but not more widely separated, stimuli (Clelland et al., 2009). Lack of neurogenesis also impaired long-term object recognition memory (48 h, but not 1 h or 24 h) (Pan et al., 2012) and, due to impeded allocentric discrimination, spatial relational memory formation and long-term memory retainment in the MWM (Dupret et al., 2008;Jessberger et al., 2009;Pan et al., 2012). However, the ablation of memory resolution-enhancing immature DG neurons did not affect more simple forms of spatial navigation, such as the habituation to a new environment or contextual fear conditioning (Dupret et al., 2008). Notably, ~4 week-old and highly plastic adult-born DG granule cells were shown to facilitate hippocampal recall following training, implying that elevating the pools of these immature DG neurons with neurogenesis supports memory conservation (Gu et al., 2012).
Both cell culture and in vivo studies support that AG boosts neurogenesis in the DG. In vitro, the application of AG augmented the proliferation, but not differentiation, of primary adult rat hippocampal neural stem cells via the GHS-R1α-dependent induction of the PI3K/Akt/mTor/ p70 S6K pathway, ERK 1/2 and STAT3 activation as well as the inactivation of GSK-3ß. The treatment with AG also heightened the transcriptional levels of GHS-R1α by these cells, suggesting the presence of a positive feedback loop (Chung et al., 2013). On the other hand, a synthetic ghrelin analogue rescued both apoptosis, necrosis and caspase-3 induction by stimulating PI3K/Akt-and ERK 1/2 -signaling in isolated and growth-factor deprived adult rat hippocampal progenitor neurons (Johansson et al., 2008). Thus, AG induces the non-canonical, Gα-associated activation of the PI3K/Akt and ERK 1/2 pathways in DG neural stem cells, leading to enhanced proliferation (PI3K/Akt and Jak2/STAT3) and survival (PI3K/Akt and ERK 1/2 ).
In vivo, both mice and dwarf rats were shown to display GHS-R1αexpressing immature neuroblasts in the granule cell layer of the DG (Hornsby et al., 2016;Li et al., 2013;Moon et al., 2009). Furthermore, the application of AG enhanced neurogenesis in healthy rodents (Hornsby et al., 2016;Kent et al., 2015;Zhao et al., 2014), dwarf rats , an AD (Moon et al., 2014) as well as a PD animal model (Elabi et al., 2018). Indeed, while the daily intraperitoneal treatment with AG for 8 days promoted the numbers of proliferating (BrdU) and total (DCX) DG neuroblasts in adult mice, the antibody-mediated depletion of plasma ghrelin for the same time period resulted in the inverse outcome (Moon et al., 2009). Indeed, the neurogenesis-heightening effects of extended caloric restriction, which elevates the plasma and hippocampal AG levels, were dependent on AG . Similarly, the daily peripheral AG injections for 2 weeks, to mimic food restriction-associated physiological levels of AG, boosted neurogenesis in the DG of healthy rats (Kent et al., 2015). On the other hand, the knockout of GHS-R1α did not alter the DG morphology or basal rate of neurogenesis, suggesting compensatory effects (Hornsby et al., 2016). However, in contrast to their wild-type littermates, GHS-R1α -/mice exhibited slower proliferation and exacerbated progenitor neuronal loss in the ventral DG in response to chronic social defeat stress (Walker et al., 2015). As discussed earlier, these pro-proliferative and neuroblast-preserving effects of AG and its receptor are supported by the activation of the neuroprotective PI3K/Akt pathway (and others) in cultured primary hippocampal progenitor neurons (Johansson et al., 2008;Chung et al., 2013). Interestingly, the acute injection of AG or overnight fasting further increased the transcription of neurogenic transcription factor (Egr-1) in the DG, while prolonged caloric restriction encouraged the differentiation of progenitor into mature DG neurons in wild-type, but not GHS-R1α knockout, mice (Hornsby et al., 2016). Egr-1 regulates the survival and differentiation of newborn neurons, their integration into the hippocampal learning circuit, dendrite and spine outgrowth as well as the expression of GluA1-AMPARs and several chloride ion channels, enabling immature DG neurons to receive glutamatergic and GABAergic input by other neurons (Veyrac et al., 2013). In opposition to the latter study, (Kent et al., 2015) observed no improvements in DG neuroblast differentiation following the 14 day i.p. administration of AG. Since the animals were culled 14 days after the final AG injection, any differentiation-boosting effects might have faded by the time the rodents were biochemically assessed. Besides Egr-1, despite the fact that AG drove neurogenesis in IGF-1-deficient dwarf rats , AG is a well-known, potent stimulator of the GH/IGF-1 axis, leading to the plasma release of the neurogenesis-stimulating and BBB-penetrant IGF-1 (Khatib et al., 2014;Nass et al., 2011;Nieto-Estevez et al., 2016;Pan and Kastin, 2000). Moreover, controversially, even though GHS-R1α-exhibiting immature DG neurons have been identified in vivo (Hornsby et al., 2016;Li et al., 2013;Moon et al., 2009), it was postulated that GHS-R1α is confined to adult DG granule cells, stimulating these to release the neurogenesis-promoting BDNF into the neurogenic niche (Buntwal et al., 2019). Peripheral AG might also stimulate the hippocampal production of BDNF across the vagus verge (Davis et al., 2020). Caloric limitation enhanced the hippocampal production of BDNF in both control and ghrelin -/mice, however, implying that BDNF release is ghrelin-independent . Synoptically, prolonged AG/GHS-R1α-signaling for 2 weeks, either achieved through regular AG injections or caloric restriction, stimulate the birth, proliferation, survival and possibly differentiation of immature DG neurons in vivo. Besides direct effects, AG likely supports neurogenesis by upregulating Egr-1 and IGF-1 levels in the DG, whilst the impact of AG on BDNF is questionable.
Importantly, AG-induced neurogenesis has been linked to the enhancement of hippocampal pattern separation and memory. Daily and 14-day-long peripheral injections of AG, to maintain plasma levels that would be anticipated in response to fasting, increased the numbers of DCX-positive young granule cells and augmented allocentric pattern separation in the spontaneous location recognition test (Kent et al., 2015). Similarly, the pre-training infusion of AG into dwarf rats, for a period of 28 days, encouraged neurogenesis, alternation behavior in the Y-maze as well as novel object exploration times, suggesting neurogenesis-and hippocampal pattern separation-associated improvements in spatial memory resolution and the distinction of (close) objects . As such, AG promotes the birth and proliferation of DCX-positive immature neurons in the DG that enhance hippocampal memory resolution and pattern separation, shown to facilitate allosteric spatial memory and object discrimination in rodents. These neurogenesis-imparted memory improvements are independent of the plasticity-enhancing effects of AG in the hippocampus.

Contextual and auditory fear conditioning
Besides the hippocampus-mediated formation of spatial memories and novelty learning, ample evidence suggests that AG and GHS-R1α modulate aversive and fear memory formation as well as fear behavior in vivo, with a recently discovered link to PTSD in humans. As a foundation, this section provides an overview of the different forms of fear conditioning and the engaged brain networks in rodents.
Fear conditioning is coordinated by a brain circuit that involves the prefrontal cortex (PFC), amygdala and hippocampus, as portrayed in Fig. 3. Of these brain areas, the hippocampus retains the context of the fear memory, whereas the amygdala encodes incoming fear stimuli, mediates their long-term consolidation and induces defensive behaviors. Furthermore, once formed, fear memories appear to be stored in the amygdala across the entire lifespan. Besides the amygdala and hippocampus, efferent inputs from the PFC regulate the amygdaloid activity (Gale et al., 2004;Homberg, 2012;Isaacs, 2015;Phelps, 2004).
To generate fear memories in animals, various paradigms may be employed. During classical (cued) fear conditioning, an amygdaladependent process, a (CS), such as a tone or light cue, is paired with an unconditioned stimulus (US), typically a foot shock. Following training, the rodents are re-exposed to the CS and the initiation of fear, expressed as freezing, is scored. In the contextual fear paradigm, the animal undergoes inescapable foot shocks in a dedicated conditioning chamber. Therefore, when re-tested the next day, the animal may memorize the contextual (hippocampal) association that the chamber triggers a foot shock (US). However, recent studies indicate that contextual fear conditioning also involves the separate acquisition of hippocampus-independent foot shock (US) memories. Lastly, hippocampus-dependent passive avoidance training implements both contextual fear conditioning and instrumental learning. In this case, the rodents have to choose between a bright and a foot shock-prompting dark compartment (step-through) or avoid descending onto a foot shock-triggering platform (step-down) (Homberg, 2012;Huff et al., 2016;Ö gren and Stiedl, 2013;Qi et al., 2018). The simplified serial model (depicted in Fig. 3) stipulates that the CS (such as a certain tone) and US (foot shock) are generated in the auditory cortex (thalamus) and the somatosensory cortex, respectively, and transmitted to the lateral nucleus of the amygdala (LA), a subnucleus of the basolateral amygdala (BLA). Additionally, LA neuronal activity is comodulated by sensory input from the thalamus. In the LA, an association between the CS and US is formed and the resulting fear memory is consolidated. Subsequently, LA neurons stimulate the central nucleus of the amygdala (CA) that, by consulting the periaqueductal grey, hypothalamus and brain stem, triggers the expression of fearful behavior. In opposition to auditory fear, contextual fear memories are created by relaying the hippocampal context that surrounds a fear memory to the LA. As recently postulated, initial plasticity in the BLA may generate a contextual fear memory trace (the 'foot shock memory') that is projected across the entorhinal cortex and ventral hippocampus to the dorsal hippocampus. Finally, following enhanced plasticity in the CA1 and CA3 regions, the contextual fear memory undergoes consolidation (Chaaya et al., 2018). Lastly, the ventromedial PFC (vmPFC) functions as a gatekeeper, enabling the BLA/CA-mediated expression of fear, once the respective CS or the context of a fear memory is presented to the animal (de Quervain et al., 2017;Ehrlich et al., 2009;Homberg, 2012;Parsons et al., 2006).
It was shown that NMDAR-signaling in the LA was mandatory for the learning of both auditory and contextual fear (Rodrigues et al., 2001). Likewise, in a foot shock-and context-separating passive avoidance paradigm, the post-training administration of a muscarinic cholinergic agonist into the BLA augmented the consolidation of both foot shock and contextual fear memories (Malin and McGaugh, 2006). As such, amygdaloid plasticity is implicated in the formation of both cued and contextual fear memories.
On the other hand, while hippocampal inactivation does not affect auditory fear memories, the hippocampus is necessary for the acquisition (CA1), consolidation (CA3) and retainment of contextual fear memories for a period of up to 2 weeks (Anagnostaras et al., 1999;Daumas et al., 2005;Kim and Fanselow, 1992;Maren et al., 1997;Phillips and LeDoux, 1992). In the long-term, however, the contextual fear memory is transferred and permanently stored in the BLA (Gale et al., 2004). Indeed, intra-hippocampal treatment with a muscarinic cholinergic receptor agonist promoted the consolidation of passive avoidance, but not foot shock, memories (Malin and McGaugh, 2006), while the hippocampal infusion of an NMDAR antagonist impeded contextual, but not auditory, fear acquisition (Bast et al., 2003). Likewise, mouse strains with superior NMDAR activity in the dorsal hippocampus acquired context-associated passive avoidance memories more easily (Baarendse et al., 2008). Indeed, contextual fear conditioning upregulated memory-facilitating ryanodine receptors and ER Ca 2+ channels 5-29 h post training, strengthening EPSCs in CA1 pyramidal neurons (More et al., 2018;Trifilieff et al., 2006). Contextual fear training also induced a biphasic stimulation of ERK 1/2 /CREB-signaling 0 -1 h and 9-12 h thereafter (More et al., 2018). This implies that contextual fear memory formation co-requires hippocampal and amygdaloid plasticity.

Ghrelin modulates amygdala-dependent forms of aversive memory in vivo
The literature emphasizes a role of AG in regulating anxiety as well as the stress-associated formation, but also extinction, of various amygdala-dependent forms of aversive memory in animals, including CTA, passive avoidance, auditory fear and contextual fear memory.
Beneath, the available evidence is presented and linked to the local effects of AG in the amygdaloid subnuclei.

Ghrelin's anxiolytic and anxiogenic effects are unrelated to fear memory
Dependent on the context, AG may exert anxiolytic or anxiogenic responses. As summarized elsewhere, systemic or central administrations of AG, in most cases, trigger anxiety in unstressed rodents (Fritz et al., 2020;Morris et al., 2018). Importantly, the association between anxiety, fear and fear memories has to be clarified. Anxiety may be defined as an unspecific and long-lasting expression of fear due to a previously learned fear context. In other words, to be placed into a foot shock-inducing conditioning chamber (context) may render the animal anxious. In contrast, fear is the acute display of fear in response to a fear cue (CS) (Homberg, 2012). It has been emphasized that anxiety, per se, is not necessarily linked to the increased formation of fear memories (Ögren and Stiedl, 2013). For instance, an anxious mouse strain (DBA/2J) acquired fear memories worse than their less anxious counterparts (C57BL/6J). Instead, the learning of contextual fear (passive Fig. 3. Brain networks involved in fear expression and the formation of auditory (cue-based) fear, contextual fear and passive avoidance memories. Auditory fear memories are amygdala-dependent and formed by pairing an incoming CS (tone) to a US (foot shock) in the LA. In contrast, contextual fear memories are predominantly hippocampusdependent. It has been hypothesized that a US creates an initial fear memory trace in the amygdala that is projected across hippocampal formations to the dorsal hippocampus. In the latter brain area, the US/fear memory trace (designated as 'foot shock memory' in some publications) is associated with the fear context and consolidated. Similar to contextual fear, the hippocampus consolidates passive avoidance memories. Notably, fear extinction requires the acquisition of a novel extinction memory. In this process, the US (foot shock) is paired with a non-fearful context in the hippocampus and consolidated in this brain area. Subsequently, whenever the animal is presented to the US, the hippocampus stimulates the prefrontal cortex to inhibit fear expression. As such, the newly acquired extinction memory is preferentially recalled over the original auditory or contextual fear memory, resulting in fear extinction. Adapted from (Homberg et al., 2012). avoidance) memories was dependent on LTP-induction in the dorsal hippocampus (Baarendse et al., 2008;Ö gren and Stiedl, 2013). This implies that, at least, the formation of contextual fear memories and the display of anxiety are separate phenomena.

Ghrelin exerts diverse direct effects on various amygdaloid subnuclei
Generally, the amygdala is anatomically separated into multiple subnuclei that are composed of excitatory pyramidal/principal neurons and inhibitory GABAergic interneurons. As one of these amygdaloid nuclei, the BLA consists of the LA, the basal nucleus of the amygdala (BA) and the accessory basal nucleus. While the LA projects to the basal and accessory basal nuclei, it is also connected to the medial nucleus of the amygdala (MA) and the CA (see (Tsvetkov et al., 2015) for the amygdaloid classifications and connections) (Bocchio et al., 2016). Studies revealed that GHS-R1α is present in the ventrolateral and medial divisions of the LA, in the posteroventral division of the MA (Alvarez-Crespo et al., 2012) and the CA (Cruz et al., 2013).
In the LA, AG appears to be neuroinhibitory. It was shown that the bath application of AG diminished the frequencies of EPSCs in LAlocated pyramidal neurons (Alvarez-Crespo et al., 2012). However, AG dose-dependently potentiated the firing rate of (unidentified) neuronal populations in the LA ex vivo, while intra-LA infused AG enhanced LA neuron spike frequency in a GHS-R1α-mediated manner in vivo. Unfortunately, it was neither clarified whether excitatory principal neurons or GABAergic interneurons express GHS-R1α, nor which type of neuron showed increased activity (Song et al., 2013). However, both the microinjection of AMPAR or NMDAR antagonists as well as AG into the LA blocked the acquisition of CTA (Song et al., 2013). Therefore, it is more likely that AG suppressed LA plasticity by inducing GABAergic LA interneurons.
In the CA, the superfusion of rat slices with AG synergized with ethanol to augment iPSPs in the CA, whereas a GHS-R1α antagonist attenuated these. Therefore, constitutive and ligand-induced GHS-R1αsignaling enhance GABAergic transmission in the CA (Cruz et al., 2013). Notably, the CA exclusively consists of GABAergic interneurons. By projecting to areas such as the hypothalamus and brainstem, the CA stimulates anxiety-like and ethanol withdrawal behavior (Jie et al., 2018), possibly contributing to the anxiogenic effects of AG (Fritz et al., 2020;Morris et al., 2018).

Ghrelin-signaling in the amygdala restricts conditioned taste aversion
Interestingly, AG-signaling in the LA, a subnucleus of the BLA, modulates CTA. Generally, CTA serves to memorize what types of flavor or smell are safe to ingest or potentially hazardous (nauseating). CTA mainly involves alterations in the firing rate and gene activity of a neuronal network across the nucleus tractus solitarius (NTS), parabrachial nucleus of the pons, the amygdala and several other brain structures, while the hippocampus is of lesser importance (see (Welzl et al., 2001)).
The pre-training infusion of AG into the LA attenuated the acquisition, but also the extinction, of CTA by rats, as evident 24 h later (Song et al., , 2013Zhu et al., 2013). However, the post-training or pre-test administration of AG did not interfere with the consolidation or recall of CTA (Song et al., 2013). Since CTA acquisition and recall were dependent on NMDAR and AMPAR activation in the LA (Song et al., 2013), AG presumably suppresses plasticity in LA neurons, as discussed in Section 5.2.2. On the other hand, the intra-LA application of a GHS-R1α inverse agonist, to ablate the constitutive activity of the receptor, enhanced the acquisition of CTA , whilst GHS-R1α antagonism prevented the blockade of CTA extinction by AG . Resembling CTA, when neonatal chicks had to learn choosing between methylantranilate-coated or neutral beads, the intracerebroventricular administration of AG following training impeded the development of methylantranilate-loathing memories. This suggests that AG might also impede the consolidation of CTA (Carvajal et al., 2009).
Given the latter findings and considering the orexigenic function of AG, the acylated hormone may restrict the adoption of food adversities. However, AG does not interrupt CTA recall nor extinction to guarantee that the hungry animal avoids the consumption of food that was previously identified as nauseating.

The cerebral microinjection of ghrelin enhances passive avoidance memory
Indeed, AG-signaling in various brain regions enhances contextretainment in passive avoidance paradigms. The post training central or targeted microinjection of AG into the hippocampus or CA1, amygdala or BLA and dorsal raphe nucleus (DRN) enhanced the consolidation of passive avoidance memories in the step-down (1 h and 24 h later) (Carlini et al., 2002(Carlini et al., , 2004(Carlini et al., , 2007Carlini, Ghersi et al., 2010;Carlini, Perez et al., 2010;Diano et al., 2006;Ghersi et al., 2011;Toth et al., 2009), step-through (24 h, 48 h and 72 h later) (Goshadrou et al., 2013) and T-Maze foot shock avoidance paradigms (1 week later) (Diano et al., 2006). Interestingly, the administration timing is crucial, since intra-hippocampal AG only augmented the consolidation of passive avoidance memories when given immediately after training, but not with a delay of 15 or 60 min (Ghersi et al., 2015). In turn, the central application of a GHS-R1α antagonist dose-dependently attenuated memory consolidation, while partially blocking acquisition, in the step-through trial (Beheshti and Shahrokhi, 2015).
Amongst the hippocampus, amygdala and DRN, the hippocampal administration of AG enhanced passive avoidance memory most potently (Carlini et al., 2004), in agreement with the hippocampal dependence of this form of learning (Ögren and Stiedl, 2013) and AG's plasticity-enhancing effects in the hippocampus (see Section 4 and Fig. 3). Indeed, the hippocampal threshold to induce LTP, as lowered by the hippocampal post training administration of AG, was inversely correlated to the latency times in the step-down test (Carlini, Perez et al., 2010). Moreover, the CA1 microinjection of AG ameliorated the morphine-induced deficits in inhibitory avoidance memories (Nazari-Serenjeh et al., 2019), presumably by restoring the morphine-driven impairments in the release of glutamate (Guo et al., 2005). Notably, AG did not enhance the acquisition or retrieval of passive inhibitory memories (Carlini, Ghersi et al., 2010). Instead, although intra-hippocampal injections of AG enhanced 1 h short-term passive avoidance memories in earlier studies (Carlini et al., 2007;Carlini, Perez et al., 2010), it was specified that AG selectively drives the consolidation of hippocampal long-term memories (Carlini, Ghersi et al., 2010). Besides the hippocampus, intra-BLA administered AG, in a GHS-R1α and BLA-dependent manner, enhanced the consolidation of passive avoidance memories (Goshadrou and Ronaghi, 2012;Toth et al., 2009), whereas the central, BLA, DG or VTA injection of a ghrelin antagonist following training attenuated the performance of rats in the passive avoidance paradigm 24 h later (Beheshti and Aslani, 2018;Beheshti et al., 2020).
Even though GHS-R1α-signaling in the BLA improved passive avoidance memory (Beheshti and Aslani, 2018;Goshadrou and Ronaghi, 2012;Toth et al., 2009), the application of the lesion-inducer lidocaine showed that the BLA is expendable for the consolidation of passive inhibitory memories (Goshadrou and Ronaghi, 2012;Tomaz et al., 1992). Instead, the contextual consolidation of passive avoidance memories, a form of declarative memory, is conveyed by the hippocampus (Bird and Burgess, 2008). Notably, the US (foot shock)-induced stimulation of the amygdala, as occurring during fear conditioning, transiently suppresses c-Fos activity and LTPs in the CA1, interfering with contextual fear learning (Dai et al., 2008;Waider et al., 2019). As such, it is possible that the microinjection of AG into the BLA restricts the local neuronal activity, hence favoring hippocampal plasticity and the formation of inhibitory avoidance memories. This concept is supported by the fact that systemic AG diminished c-Fos activity in the BLA (Hornsby et al., 2016), while intra-amygdaloid AG boosted the consolidation of hippocampus-dependent spatial memories in the MWM (Toth et al., 2010).
Taken together, the hippocampal administration of AG or microinjection into the BLA, which might improve hippocampal activity, facilitates the context consolidation of passive avoidance memories by enhancing local glutamatergic signaling (Section 4.1.1 and Fig. 2), transiently inducing synaptogenesis (Section 4.2.1) and, possibly, enhancing dopamine transmission by the VTA (Section 4.1.3). The timing is critical, however, and AG must be given immediately post training, in combination with the US-context pairing during conditioning, to observe these consolidation-enhancing effects. The dopaminergic and transient synaptic effects of AG might explain why AG further enhanced 1 h short-term passive avoidance memory in some studies.

Systemic ghrelin suppresses the consolidation of auditory fear memories
Although AG promotes the consolidation of the hippocampusdependent fear context in the passive avoidance trial (Section 5.2.4), the opposite effect is seen during amygdala-dependent auditory fear memory. A study showed that the systemic or intra-BLA application of a ghrelin analogue post conditioning blocked the retainment (Harmatz et al., 2017), or showed a trend towards impaired (Meyer et al., 2014), auditory fear memories in non-stressed rodents, whereas the peripheral injection of a GHS-R1α antagonist achieved the opposite result (Harmatz et al., 2017). Furthermore, the baseline circulatory ghrelin levels were inversely correlated to long-term (48 h) auditory fear memory strength. Of note, the ghrelin analogue did not affect the acquisition or retrieval of auditory fear, nor modulate contextual fear, suggesting that AG interferes with the amygdala-imparted consolidation of auditory fear (Harmatz et al., 2017). Indeed, a study showed that daily and 14 day-long intraperitoneal injections of AG, to maintain plasma levels that would be anticipated following caloric restriction, attenuated c-Fos induction in the BLA (Hornsby et al., 2016), the hub for auditory fear formation (Homberg, 2012).
When released upon fasting, one of AG's many functions entails the stimulation of appetite and foraging behavior (Yanagi et al., 2018). Therefore, it was proposed that AG prevents the formation of auditory (cue-based) fear memories in the absence of stress or life-threatening situations, because fear would interfere with the seeking of food (Harmatz et al., 2017).

Chronic stress-driven increases in plasma ghrelin potentiate auditory fear
It must be emphasized that the stress state of the animal influences auditory fear consolidation by AG. It is generally accepted that chronic, but not necessarily acute stress, augments the plasma secretion of AG (Fritz et al., 2020;Morris et al., 2018). Indeed, immobilization stress for at least 5 days heightened the plasma levels of ghrelin in rodents, even more in adrenalectomized littermates. Comparable to the 5 day intra-BLA or systemic administration of AG, these chronically immobilized PTSD mice displayed an increase in long-term auditory fear memory. Although the PTSD mice showed elevated plasma corticosterone levels, only the injection of a GHS-R1α antagonist, but not adrenalectomy, weakened the retainment of auditory fear memories (Meyer et al., 2014). Interestingly, another study discovered that a single injection of a ghrelin analogue only impaired the consolidation of auditory fear memories in non-, yet not chronically, stressed rodents (Harmatz et al., 2017).
In contrast to the auditory fear memory-suppressing outcome in the absence of or upon acute stress, this implies that the chronic stressdriven plasma secretion of AG, independent of the stress-associated secretion of HPA hormones, exacerbates the amygdaloid consolidation of cued (auditory) fear memories. For an in-depth explanation, please see Sections 5.3.5 and 5.3.6.

Ghrelin prolongs the recall of contextual fear memories
Regrettably, in vivo studies that investigate the effects of AGtreatment in the contextual fear paradigm are lacking. Nonetheless, when contextual fear conditioning was preceded by two weeks of caloric restriction (to enhance the plasma AG levels), neurogenesis as well as contextual fear retainment in these starved wild-type mice were elevated in a GHS-R1α-dependent manner 12 days after training, as compared to ad libitum-fed mice. However, contextual fear memory was unaffected during the acquisition phase or following 1 or 8 days in calorically restricted rodents. Given that AG acts as an appetitestimulating hormone, it was speculated that caloric restriction, raising the blood AG levels, simultaneously promotes fear context recall to heighten the likelihood of survival of the hungry animal (Hornsby et al., 2016). In agreement with the latter study, the genetic knockdown of GHS-R1α did not affect contextual fear acquisition and consolidation (24 h), yet it reduced long-term (30 day) contextual fear retainment . In this context of these findings, the selective post training optogenetic inhibition of highly plastic, ~4 week-old adult-born DG granule cells, but not immature DG neurons of other ages, diminished hippocampus-associated spatial (MWM) and contextual fear memory retrieval (Gu et al., 2012). The genetic attenuation of neurogenesis also deteriorated remote passive avoidance memory 21 days following conditioning (Pan et al., 2012). Therefore, the limited evidence proposes that long-term increases in plasma AG and the associated enhancement of neurogenesis in the DG (see Section 4.2.2) enlarge the pool of ~4 week-old, memory retrieval-facilitating DG immature granule neurons, thus extending the retainment (recall period) of remote contextual fear memories.

Acylated ghrelin augments the extinction of fear memories
Generally, memories are not erased per se. Instead, extinction necessitates the acquisition of a novel extinction memory that is preferentially recollected instead of a previous fear memory. It has been postulated that novel extinction memories are produced by the joint engagement of the hippocampus, vmPFC and BLA. Once the new extinction memory is formed, the hippocampus-stimulated vmPFC inhibits the LA, preventing the expression of fear when the CS is presented. Therefore, the incoming CS (tone) in the amygdala is paired with a nonfearful context in the hippocampus, replacing the original CS (tone)-US (foot shock) association that was formed by the auditory/somatosensory cortex and amygdala (Fig. 3). Notably, besides the auditory cortex, the auditory thalamus relays tone-based stimuli (CS) to the amygdala (de Quervain et al., 2017;Herry et al., 2010;Homberg, 2012;Parsons et al., 2006). Importantly, the acquisition of extinction memories necessitated amygdaloid plasticity, including NMDAR-and ERK-signaling, in the BLA (Falls et al., 1992;Herry et al., 2006;Lin et al., 2003;Lu et al., 2001) and LA (Sotres-Bayon et al., 2007).
Interestingly, AG enhanced the extinction of fear in an unexpected manner. When mice underwent auditory fear conditioning, two sessions of extinction training and a final fear memory trial, food-deprived animals acquired and retained extinction memories more effectively. However, both extinction learning and retainment could be prevented by the intra-LA infusion of a GHS-R1α antagonist (Huang et al., 2016). Thus, the plasma secretion of AG during caloric restriction, confirmed to be independent of glucocorticoids, mediated fear extinction by acting in the LA, possibly through inhibiting plasticity in this amygdaloid subnuclei (see 5.2.2). In line with the extinction-promoting effects of high plasma AG, (Hornsby et al., 2016) reported that, contrary to wild-type mice, GHS-R1α -/rodents showed poor contextual fear extinction over a period of 12 weeks. Notably, in the context of extinction, some studies suggest that classical Pavlovian conditioning triggers and sustains fear by transiently augmenting glutamatergic neurotransmission and auditory (CS) input across thalamus-LA synapses, whereas fear extinction might involve the targeted weakening of these synapses (Clem and Huganir, 2010;Kim et al., 2007). Strangely, food deprivation and intra-LA administered AG impaired LTD at thalamus-LA synapses in a GHS-R1α-dependent fashion (Huang et al., 2016). Thus, AG does not facilitate fear extinction by blocking auditory or sensory (CS) input into the LA. Rather, by inhibiting LTD, AG supports the thalamic CS transmission to the LA. Nevertheless, this mechanism may ease extinction by selectively potentiating the relay of the CS (such as a tone), but not the US (i.e. foot shock), to the LA. This, possibly, facilitates the pairing of the CS with a new, non-fearful context in the hippocampus. Indeed, (De la Casa, 2013) demonstrated that caloric restriction, which raises the circulatory AG levels, increased the attention of animals towards the CS (tone), preventing latent inhibition during auditory fear conditioning.
Mechanistically, AG improves the formation of extinction memories in cooperation with dopamine-signaling and via indirect effects in the hippocampus. Post training injections of a D 1 R agonist into the CA1 encouraged the extinction of both contextual fear and passive avoidance memories, whereas the use of a D 1 R antagonist weakened contextual fear extinction (Fiorenza et al., 2012). Strikingly, an intra-DG administered dopamine agonist promoted the consolidation of extinction memories, while the application of a GHS-R1α antagonist, by interfering with hippocampal GHS-R1α:D 1 R heterodimers, cancelled these effects (Jiang et al., 2006;Kern et al., 2015). Finally, as related to the pro-neurogenic effects of AG, mice with genetic deficits in neurogenesis displayed deteriorated contextual fear extinction (Pan et al., 2012).
As such, AG-driven mesolimbic dopamine-signaling (Section 4.1.3), likely in combination with improvements in hippocampal plasticity (Section 4), as well as the AG-induced potentiation of neurogenesis and memory resolution in the DG (Section 4.2.2) facilitate the formation of extinction memories to abolish fear. How LA-administered AG, which seemingly decreases local plasticity, boosts extinction memory requires further studies, however. In conjunction, similar to chronic selective serotonin reuptake inhibitor (SSRI) treatment (Deschaux et al., 2011), the AG-induced stimulation of 5-HT neurons in the DRN and the ensuing release of 5-HT in the amygdala (Hansson et al., 2014;Ogaya et al., 2011) may further suppress fear return after extinction training (Deschaux et al., 2011;Homberg, 2012).

Summary: the role of acylated ghrelin and GHS-R1α in the formation of aversive memories
• While conditional, an increase in the systemic or cerebral AG levels typically induces anxiety. These effects are unrelated to aversive memory formation, however. • GHS-R1α is expressed in the LA, MA and CA. AG-signaling likely suppresses LA plasticity by stimulating GABAergic interneurons in the LA, while augmenting (GABAergic) activity in the CA to drive anxiety. Despite the fact that GHS-R1α is not present in the BLA, the injection of AG into this amygdaloid nuclei inhibited auditory fear, while enhancing passive avoidance memories. This suggests that locally administered AG suppresses BLA activity, which appears to favor hippocampal activity. • The micro-infusion of AG into the LA inhibits the acquisition, extinction and, potentially, consolidation of CTA. • When administered into the hippocampus or amygdala (BLA) immediately after training, AG improves 1 h short-term working memory and the consolidation of hippocampus-dependent passive avoidance memories, as mediated by an enhancement in hippocampal plasticity (Section 4). • The systemic or BLA administration of AG following training prevents auditory fear memory consolidation. Moreover, the plasma AG levels are inversely correlated to auditory fear memory strength.
However, under conditions of chronic stress, which raises the circulatory AG levels, AG promotes auditory fear memory consolidation and retainment (further explored in Section 5.3).
• Elevated systemic AG levels, such as during caloric restriction, prolong contextual fear retainment by stimulating neurogenesis and, thus, facilitating hippocampal recall. However, in contrast to passive avoidance memory, the effects of AG injections on contextual fear memory formation are understudied. • AG boosts auditory and contextual fear memory extinction by enhancing the acquisition and consolidation of extinction memories. This involves AG-imparted improvements in hippocampal plasticity (Section 4), mesolimbic dopamine-signaling (Section 4.2.1) and neurogenesis (Section 4.2.2). Since AG, unintuitively, enhances the CS relay towards the LA (Fig. 3), this might facilitate the pairing of the CS with a non-fearful context in the hippocampus.

Ghrelin and its memory-modulating interaction with the serotonin system
The highly complex serotonin system is not only implicated in hippocampal memory formation, but it also regulates the expression of fear as well as fear memory formation. As such, this section illustrates the hippocampal and amygdaloid interaction of AG-and serotonin-signaling in memory and fear, as physiologically influenced by feeding. Furthermore, the biological mechanisms that allow AG to function as a fear memory-suppressor under acute stress, but fear memory-potentiator during chronic stress and PTSD, will be elucidated.

Serotonin controls the expression of cued and contextual fear
According to a popular theory, the transmission of 5-HT suppresses aversive thinking, mitigating the outcome of punishing events (Homberg, 2012). However, 5-HT also drives the acute display of cue-based fear. Generally, during classical Pavlovian conditioning, the US (foot shock) activates neuronal activity across the CA and DRN, whereas fear-potentiated startle selectively induces c-Fos in neurons of the dorsal region of the DRN and the medial subdivision of the CA (Spannuth et al., 2011). Regarding 5-HT, in vivo studies have revealed that the extracellular 5-HT levels in the amygdala, but also the vmPFC and nucleus accumbens, rose within 30-40 min following the US (foot shock) during conditioning or CS (tone) exposure on the test day (Inoue et al., 1993;Yokoyama et al., 2005). Moreover, immobilization stress for 20 min stimulated the expression of the HPA effector corticotropin-releasing factor (CRF) by the CA, leading to the CRF receptor-evoked transmission of 5-HT from the DRN to the CA to induce freezing (Forster et al., 2006;Merali et al., 1998;Merlo Pich et al., 1995;Mo et al., 2008). On the other hand, the gradual release of 5-HT in the vmPFC terminated stress/CRF/US-triggered cue-based fear display (Forster et al., 2006;Inoue et al., 1993;Kawahara et al., 1993).
Interestingly, the use of 5-HT-elevating drugs confirmed that 5-HT navigates auditory and contextual fear display differently. While the acute systemic treatment with SSRIs prior to the test session potentiated auditory freezing in response to the CS (tone) (Burghardt et al., 2007), it attenuated the expression of fear when it was learned in a contextual paradigm (Hashimoto et al., 1996(Hashimoto et al., , 2009Montezinho et al., 2010;Muraki et al., 2008;Nishikawa et al., 2007;Santos et al., 2006). In the latter, the selective pre-test microinjection of a SSRI into the amygdala or hippocampus, respectively, were both sufficient to block contextual fear expression (Inoue et al., 2004;Montezinho et al., 2010).

Ghrelin stimulates serotonin transmission by the dorsal raphe nuclei
5-HT originates from nine distinct, 5-HT-producing raphe nuclei in the brain stem. To varying degrees, these raphe nuclei innervate other brain regions throughout the CNS (Burke and Heisler, 2015;Wang and Aghajanian, 1977). Immunohistochemical investigations showed that GHS-R1α is prevalent in the DRN and median raphe nuclei (Guan et al., 1997). Besides upregulating a multitude of 5-HTRs in the DRN (Hansson et al., 2014), AG depolarized ~75% of DRN-resident 5-HT neurons in rat slice preparations (Ogaya et al., 2011), resulting in the amygdaloid accumulation of 5-HT in vivo (Hansson et al., 2014). The DRN almost exclusively innervates the amygdala , explaining the AG-evoked release of 5-HT in this brain area (Hansson et al., 2014).
Nonetheless, AG restricts the projection of 5-HT to selected brain areas. Indeed, AG blocked the release of 5-HT by hypothalamic synaptosomes (Brunetti et al., 2002), while intra-CA1-infused AG completely abolished the liberation of 5-HT in the rodent hippocampus (Ghersi et al., 2011). Similarly, prolonged caloric reduction, a condition that favors the plasma secretion of ghrelin, for 6 days or 4-5 weeks diminished 5-HT pools in the hypothalamus and hippocampus in vivo (Haider and Haleem, 2000;Haleem, 2009;Jahng et al., 2007). Given that 5-HT impedes hippocampus-dependent passive avoidance, contextual fear and spatial memory formation, while antagonizing AG-induced plasticity in the hippocampus (Section 5.3.3), AG favors hippocampal plasticity (Section 4) and declarative memory (Section 3) by restricting 5-HT neurotransmission towards the hippocampus.
Generally, the memory-and plasticity-associated effects of 5-HT in the dorsal hippocampus are region-specific. In the DG, acutely applied SSRIs weakened Arc expression, implying that 5-HT impedes DG plasticity (Ravinder et al., 2013). A study showed that a 5-HT 4 R agonist blocked mossy fiber-CA3-relayed LTPs and LTDs, but also LTDs at perforant path-DG synapses, proposing that 5-HT may restrict glutamatergic input to the CA3 region (Twarkowski et al., 2016). Interestingly, the optogenetic stimulation of 5-HT release in the hippocampal CA1 region, in a 5-HT 4 R-dependent manner, was shown to potentiate CA3-CA1 synaptic transmission and the retrieval of spatial memory (Kemp and Manahan-Vaughan, 2005;Teixeira et al., 2018). In the CA1, a neuroinhibitory role of 5-HT is apparent. 5-HT treatment hyperpolarized principal neurons in a 5-HT 1a R and stimulated the spontaneous activity of GABAergic interneurons in a 5-HT 3 R-conveyed manner in the CA1, thus blocking stimulation-driven LTPs and NMDA receptor activation in this region (Corradetti et al., 1992;Staubli and Otaky, 1994). Likewise, the utility of SSRIs showed that elevated extracellular levels of 5-HT impair LTP expression in CA1 pyramidal neurons (Igelstrom and Heyward, 2012;Mnie-Filali et al., 2006). Moreover, DRN-mediated 5-HT transmission induced hippocampal CA1/CA3 GABAergic interneurons, hyperpolarized a small subset of CA1 pyramidal neurons and had no effect on CA3 excitatory neurons (Varga et al., 2009). 5-HT also induced 5-HT 1b R receptors on CA1 pyramidal neurons, leading to the postsynaptic enhancement of AMPAR-mediated LTP at temporoammonic (TA)-CA1 synapses. Importantly, the latter stimulation of TA-CA1 synapses via 5-HT 1b R/5-HT seems to interfere with the consolidation of spatial memory, since a 5-HT 1b R antagonist augmented long-term spatial memory (Cai et al., 2013). On the other hand, the genetic lack of 5-HT augmented c-Fos neuronal activity in the CA1 and DG regions, supporting a plasticity-suppressing function of 5-HT. Moreover, US exposure (foot shocks) impeded LTPs, elevated LTDs and strengthened c-Fos induction in GABAergic CA1 interneurons only in wild-type, but not 5-HT-deficient, rodents. Thus, 5-HT-depleted rodents exhibited the enhanced acquisition, recall and 10 day-retainment of contextual fear (Dai et al., 2008;Waider et al., 2019), although displaying worsened spatial memory recollection (Dai et al., 2008). Synoptically, 5-HT-signaling impedes hippocampal plasticity in the CA1 and DG, but not CA3, interfering with the learning and consolidation of hippocampus-dependent spatial, passive avoidance and contextual fear memories (Cai et al., 2013;Carlini et al., 2007;Dai et al., 2008;Ghersi et al., 2011;Waider et al., 2019). 5-HT, however, enhances CA3-CA1-mediated spatial memory retrieval (Dai et al., 2008;Teixeira et al., 2018). Importantly, 5-HT further counteracts AG-signaling, interfering with both appetite and hippocampus-associated spatial and contextual fear memory. As reversible with 5-HT 2c R antagonists, it was unraveled that GHS-R1α formed heterodimers with 5-HT 2c Rs in primary rat hippocampal and hypothalamic neurons, which resulted in the blockade of GHS-R1α-mediated Ca 2+ -signaling. Moreover, the dimerization of GHS-R1α and 5-HT 2c Rs navigated appetite, since 5-HT 2c R blockers augmented food intake, whereas 5-HT 2c R agonists dampened feeding (Schellekens et al., 2015). Notably, even though AG stimulates 5-HT transmission by the DRN (Hansson et al., 2014;Ogaya et al., 2011), AG hampers the projection of 5-HT towards the hypothalamus and hippocampus (Brunetti et al., 2002;Ghersi et al., 2011) (see also Fig. 2). As confirmed in vivo, the intra-CA1 infusion of AG not only lowered the hippocampal 5-HT levels, but also enhanced the consolidation of passive avoidance memories. Furthermore, the hippocampal 5-HT pools were inversely correlated to the escape latencies, suggesting that 5-HT abolishes AG's plasticity and consolidation-boosting benefits in the hippocampus (Ghersi et al., 2011). Indeed, the intra-hippocampal administration of a SSRI blunted the enhancement of passive avoidance and object recognition memory by AG (Carlini et al., 2007).
These findings support the concept that locally available AG inhibits the synaptic release of 5-HT in the CA1 and DG, thus facilitating plasticity and the formation of hippocampus-dependent forms of memory, such as spatial or passive avoidance memories (see Sections 3.1 and 5.2.4).

Ghrelin mitigates auditory fear memory formation upon acute stress
Interestingly, greater circulatory levels of AG were inversely correlated to long-term auditory fear memory strength, while the systemic post training injection of a ghrelin analogue suppressed the consolidation of auditory fear memories (Harmatz et al., 2017). These results propose that AG functions as an auditory fear memory-mitigating factor when facing acute stress.
These fear memory-weakening effects seem to be based on three aspects. First, it has been implied that the binding of AG to GHS-R1αexpressing LA neurons suppresses principal neuron activity and plasticity (Section 5.2.2) (Alvarez-Crespo et al., 2012;Song et al., 2013). Similar to CTA, principal LA neurons mediate the learning and consolidation of cue-based (auditory) fear (Goosens and Maren, 2004;Monsey et al., 2011;Tipps et al., 2018). And indeed, AG suppressed the acquisition of CTA, which was dependent on an increase in LA plasticity (see Section 5.2.3) (Song et al., 2013).
Second, AG evokes 5-HT transmission from the DRN to the amygdala (Section 5.3.2). Generally, the acute systemic treatment with SSRIs, which increase extracellular 5-HT levels, enhances auditory fear acquisition and expression, whereas their chronic administration impairs auditory fear learning by rodents (as compiled by ). In fact, acutely given SSRIs inhibit 5-HT neurotransmission and the terminal release of 5-HT by the raphe nuclei by activating autoinhibitory 5-HT 1A Rs. In contrast, the prolonged exposure to SSRIs desensitizes 5-HT 1A Rs, which gradually enhances anti-depressive 5-HT output by raphe projections (Blier and Bergeron, 1995;Burghardt and Bauer, 2013;Gardier et al., 1996;Gray et al., 2013). Furthermore, DRN-induced 5-HT release inhibited the amygdaloid neuronal activity ex vivo (Wang and Aghajanian, 1977), while chronically administered SSRIs reduced amygdaloid activity in patients (Arce et al., 2008) and NR2B-NMDAR levels in the rodent BLA . Notably, 5-HT was hypothesized to function as a high-pass filter in the BLA, only allowing the transmission of sufficiently strong fear stimuli (Bocchio et al., 2016;Yamamoto et al., 2012). Since higher AG plasma levels were correlated to reduced auditory fear memory strength (Harmatz et al., 2017), long-term increases in AG may elevate the basal rate of 5-HT release in the amygdala to, similar to SSRIs, interfere with plasticity and fear memory formation. Indeed, the 14 day-long treatment with AG reduced c-Fos activity in the BLA (Hornsby et al., 2016).

Chronic stress inverts ghrelin's fear memory-attenuating effects
Curiously, animal studies indicate that chronic stress reverses AG's suppressive effects on auditory fear memory formation. Indeed, chronic immobilization stress prevented the attenuated consolidation of auditory fear by the systemically given ghrelin analogue MK-677 (Harmatz et al., 2017). Furthermore, the daily systemic injection of MK-677 for 5 days prior to conditioning, to reproduce the plasma levels of AG during chronic stress, did not affect fear acquisition, yet augmented the long-term retainment of auditory fear memories in a HPA axis (CRF and glucocorticoid)-independent manner (Meyer et al., 2014). In this context, AG treatment can selectively potentiate the HPA axis-mediated plasma release of glucocorticoids or cortisol (a human variant of glucocorticoids) by a yet to be fully characterized mechanism (Fritz et al., 2020;Morris et al., 2018), whereas glucocorticoids, in turn, are known to boost fear memory consolidation (see (de Quervain et al., 2017)). Importantly, chronic, but not acute, stress provokes multi-fold increases in the plasma AG levels, lasting up to years (Malik et al., 2020;Meyer et al., 2014;Yousufzai et al., 2018). Intriguingly, when GHS-R1α was systemically blocked during chronic immobilization stress and until fear conditioning took place, the PTSD mice did not show enhanced long-term auditory fear memories anymore (Meyer et al., 2014;Yousufzai et al., 2018). Therefore, chronic stress evokes long-term increases in the blood AG levels that, in a GHS-R1α-mediated and glucocorticoid-independent manner, encourage fear memory formation during PTSD in vivo. There is also evidence for a role of AG in augmenting PTSD in humans. Indeed, the plasma levels of AG were correlated to PTSD in a recent clinical study, accounting for 76.3 % of the PTSD severity. Although circulatory cortisol was co-elevated in PTSD patients, it had a Fig. 4. Brain region-specific effects of acylated ghrelin on memory formation in the presence or absence of chronic stress. The intricated effects of AG (grey arrows), 5-HT (green arrows) and chronic stress/PTSD (red arrows) in individual brain areas are highlighted. Dotted lines indicate hypothesized outcomes. Generally, AGinjections or fasting/caloric restriction induce 5-HT transmission from the DRN towards amygdaloid nuclei, such as the LA and BLA, while locally available AG in the hippocampus prevents the synaptic release of 5-HT in this brain area. Concertedly, the regional and 5-HT-associated effects of AG lead to the weakened consolidation of auditory fear memories, the reduced acquisition of CTA, enhanced fear memory extinction as well as the elevated consolidation of passive avoidance and declarative (spatial and object recognition) memories in unstressed rodents. Investigations that used caloric restriction and GHS-R1α knockout suggest that the ghrelin system extends the retainment of contextual fear memories by enhancing their recall, yet studies administering AG or synthetic analogues are warranted. In turn, chronic stress/PTSD exacerbate the plasma release of AG and alter AG-signaling in the brain, as indicated with red arrows, red text and a lightning symbol. These alterations seem to reverse the, at least auditory, fear memory-suppressing effects of AG in unstressed or acutely stressed rodents, rendering AG a biomarker for PTSD. For more details, please see the respective sections in the text. neglectable impact on PTSD severity (Malik et al., 2020), which is in agreement with the HPA axis-independent aggravation of auditory fear memory by AG in a PTSD animal model (Meyer et al., 2014).
As depicted in Fig. 4, the evidence from mostly animal investigations suggests that chronic stress alters AG-signaling at multiple levels, including the amygdala, thus altering the net effects of AG on fear memories. First, the expression of growth hormone (GH) in the rodent BLA was potentiated following chronic stress, while AG stimulated GH expression in cultured BLA neurons. Moreover, virus-mediated GH antagonism in the BLA prevented the enhanced retainment of auditory fear memories in response to the 10 day-long pre-training infusion of MK-677 (Meyer et al., 2014). Thus, the chronic enhancement of the AG plasma levels and, hence, GH expression in the BLA might facilitate the consolidation of auditory fear memories in PTSD rodent models (Meyer et al., 2014). To our knowledge, the expression of GHS-R1α has not been confirmed in the BLA in vivo, however, implying that AG might act on BLA-projecting neurons in neighboring amygdaloid regions or other brain structures. Second, 2 weeks of immobilization stress led to diminished AG-binding in the amygdala of rodents, symbolizing ghrelin resistance (Harmatz et al., 2017). In this context, the activation of GHS-R1α in the LA appears to inhibit the activity of principal LA neurons, presumably by increasing the firing of GABAergic LA interneurons, resulting in weakened CTA learning (see Section 5.2.2). This proposes that the downregulation of GHS-R1α in selected amygdaloid regions, such as the LA, might disinhibit the amygdala during PTSD. Third, given that chronic SSRI treatment blocks anxiety/panic, auditory and contextual fear by elevating the central 5-HT levels (Homberg, 2012), it is evident that the loss of amygdaloid 5-HT binding, as observed in patients (Murrough et al., 2011), worsens PTSD. Since AG as shown to elevate the amygdaloid synthesis of multiple 5-HTRs in rodents, especially neuroinhibitory 5-HT 1a Rs (Bobker and Williams, 1989;Cheng et al., 1998;Homberg, 2012), ghrelin resistance in the amygdala may exacerbate 5-HT desensitization (Murrough et al., 2011) and BLA stimulation (Holmes, 2008) during PTSD.

Synopsis of the interplay between the ghrelin and serotonin system under acute and chronic stress
• GHS-R1α is expressed in 5-HT neurons in the DRN, which project 5-HT to the amygdala in response to AG. However, local AG inhibits the synaptic release of 5-HT in the hypothalamus and hippocampus. • AG is released upon food restriction to stimulate appetite in the hypothalamus, whilst the local 5-HT release is heightened upon feeding (likely as a consequence of lowered AG levels, which inhibit the hypothalamic 5-HT release) to induce anorexigenic hypothalamic neurons. A similar antagonistic function is seen in the hippocampus, where AG-signaling blocks 5-HT transmission to facilitate plasticity and memory formation in the CA1 and DG, whereas the liberation of 5-HT in these brain regions and the inhibitory heterodimerization of GHS-R1α with 5-HT 2c Rs attenuate learning and memory consolidation. • Greater AG plasma levels suppress auditory fear memory consolidation and retainment. In contrast, chronic stress, which provokes massive increases in circulatory AG, enhances auditory fear memory retainment in rodents and PTSD symptoms in humans, independent of the stimulatory effects of AG on glucocorticoid/cortisol release by the HPA axis. • Acute cerebral 5-HT release enhances auditory (cued) fear expression and amygdaloid/hippocampal 5-HT projection blocks contextual fear display, whereas chronically elevated 5-HT levels in the CNS, such as following SSRI treatment, suppress both auditory and contextual fear. Given that AG induces amygdaloid 5-HT transmission, targeted studies are necessary to investigate if this affects fear memory expression. • Chronic stress might reverse the suppressive effects of AG on fear memory by stimulating GH expression in the BLA and inducing amygdaloid ghrelin resistance. The latter seemingly disinhibits the LA and leads to the loss of AG-evoked 5-HTR expression in the amygdala, especially neuroinhibitory 5-HT 1a Rs, possibly resulting in amygdaloid 5-HT desensitization and exacerbated fear display.

Conclusion
The presented studies support a fundamental role of AG in the formation of declarative and aversive memories, whilst influencing feeding-associated, but also inquisitive and fear behavior. An increase in plasma AG levels, as occurring during fasting, caloric restriction or direct hormone administration, result in the diffusion of BBB-penetrant AG into the brain. The constitutively active ghrelin receptor, GHS-R1α, as well as receptor stimulation by AG boost hippocampal synaptic plasticity (Fig. 2) and the birth of memory resolution-enhancing progenitor granule neurons in the DG, leading to the improved formation of spatial and object recognition memories as well as the consolidation of passive avoidance memories. By stimulating mesolimbic dopamine transmission in the VTA, which involves the dimerization of GHS-R1α with D 1 R in the hippocampus, AG not only prompts foraging behavior and locomotion, but also hippocampal reward and novelty learning. Presumably, the latter memory-enhancing effects of AG are mechanisms to allow the hungry mammal to remember its environment more vividly, including food-rich, fruitless or dangerous spots. Moreover, the plasma secretion of AG is responsible for the neuroprotective effects of caloric restriction (Bayliss et al., 2016), thus ameliorating age-associated neurodegeneration and cognitive deficits during AD (Reich and Holscher, 2020).
On the other hand, the intricate effects of AG in the amygdala, which involve the modulation of the 5-HT system, suppress the acquisition, but not extinction, of CTA, likely to avoid that the hungry animal starves to death due to adopting food adversities, yet does not consume nauseating food. Furthermore, even though typically anxiogenic, high AG levels suppress the consolidation and retainment of auditory fear memories, whilst extending the recall period of contextual fear memories under non-and acutely stressed conditions. These appear to be means to support food-seeking in a low risk manner. Chronic stress, however, alters AG-signaling in the amygdala, induces amygdaloid ghrelin resistance and reverses the fear memory-mitigating effects of AG, turning the hormone into a biomarker for PTSD (see Fig. 4). As such, beyond appetite, we highlight the underappreciated role of AG in modulating memory and, especially, fear behavior.

Declaration of interests
The authors have no financial or commercial conflicts of interest.