Two types of corticosteroid receptors exist, the type I high affinity mineralocorticoid receptors (MR) and type II lower affinity GRs (Reul and de Kloet, 1985). Although sharing almost identical DNA-binding domains, MR and GR can exert distinct cellular functions as the genes they bind to show little overlap (Datson et al., 2001). GRs are widely distributed throughout the brain in most neurons and glia. GRs have a lower affinity for physiological GCs and only become substantially activated as hormone levels rise following stress. MR is expressed in neurons only and has a more restricted distribution with high expression confined particularly to the hippocampus (Figure 1), and septum. MRs having a 10-fold higher affinity for physiological corticosteroids (GCs and the mineralocorticoid aldosterone) are extensively occupied under basal conditions when hormone levels are low (Reul and de Kloet, 1985; McEwen et al., 1986).

In addition to the delayed genomic effects via intracellular MR and GR, it has become evident from recent work that GCs also affect brain function through rapid non-genomic membrane-associated mechanisms (Groeneweg et al., 2011) (Figure 2). The latter mode of action explains the rapid (minutes) effects of GCs on the excitability and activation of neurons in several brain regions (e.g., hypothalamus, hippocampus, amygdala, and prefrontal cortex) and provides a physiological basis for rapid effects on behavior (de Kloet et al., 2008). A membrane-localized form of MR appears to mediate the rapid GC signaling in the hippocampus (Karst et al., 2005). However, not all rapid GC effects occur via MRs with some [e.g., increase in spine density of hippocampal neurons (Komatsuzaki et al., 2005)] depending on membrane-located GRs rather than MRs and others [e.g., long-term potentiation (LTP) induction (Wiegert et al., 2006), NMDA-dependent neurotoxicity (Xiao et al., 2010)] occurring independent of MR or GR but mediated possibly through as yet unidentified membrane-localized receptors.

Short term increases in GC levels are normally adaptive and beneficial but prolonged exposure to elevated GC levels such as during chronic stress or as a consequence of failure or impairment of negative feedback control of GC secretion (e.g., Cushing's syndrome, major depression) will lead to excessive GC responses and to pathology in the periphery (e.g., diabetes, hypertension, osteoporosis, central obesity) and CNS (e.g., depression, impaired learning, and memory). Cushing's syndrome patients with hyper-secretion of cortisol show reduced hippocampal volume and impaired performance on hippocampal learning tasks (Starkman et al., 1992). Notably, successful treatment to correct the excessive secretion of GCs reverses the pathology including hippocampal structural recovery and the restoration, to some extent, of mood, learning, and memory (Starkman et al., 1999). Conversely, severely reduced GC levels, as in Addison's disease, also result in pathology and cognitive impairments (Tytherleigh et al., 2004) and are treated with lifelong corticosteroid replacement. Thus both too little and too much GCs can have detrimental effects on memory, emphasizing the crucial need to maintain optimal levels of GCs for health and survival.

Whether the consequence of the receptor activation is positive or negative for memory depend largely on the context of the situation/event to be remembered, the timing and the magnitude of the increased GC levels (de Kloet et al., 1999; Joels et al., 2006). Acute stress or increased GC levels occurring around the time of learning and within the context of the event to be remembered enhances memory consolidation. In contrast, they impair memory if occurring either before or a considerable time after the learning task (Quervain et al., 1998; Cazakoff et al., 2010). Evidence, mainly from animal studies, suggests that GCs preferentially enhances memory consolidation of emotionally arousing experiences (Roozendaal et al., 2006). Animal learning tasks, including the Morris watermaze and radial arm maze for spatial memory training, are designed to be generally affectively arousing because they require motivation to elicit changes in behavior. Even learning tasks that include no rewarding or aversive stimulation, such as the object recognition task, induces modest novelty-induced stress or arousal during training (Okuda et al., 2004). GCs released during learning appear essential for establishing enduring memories (De Kloet et al., 1998). Thus, small increases in GCs enhance hippocampus-mediated learning and memory while larger, prolonged elevations impair memory (Lupien and McEwen, 1997; Kim and Diamond, 2002). This follows the inverted U-shaped dose-response relationship between GC levels and effects on hippocampal LTP, an electrophysiological phenomenon associated with synaptic strengthening which is one of the major cellular mechanisms underlying learning and memory. Low moderate GC levels occupy predominantly high affinity MR which increases LTP and memory, while high GC levels occupy the lower affinity GR (in addition to MR) and impair LTP and memory (Pavlides et al., 1995, 1996; Kim and Diamond, 2002; Kim et al., 2007). One recent study supports the inverted-U-shaped relationship between intrinsic stress intensity (i.e., increased endogenous GCs induced by factors associated with the learning task, in this case water temperature) and spatial memory in the radial arm watermaze. Thus, rats trained at 19°C made fewer errors than rats trained at either more (16°C) or less (25°C) stressful conditions (Salehi et al., 2010).

The hippocampus not only plays a central role in the processing of spatial and contextual information (Morris et al., 1982; Moser et al., 1995) but also exerts an inhibitory influence over HPA function (Jacobson and Sapolsky, 1991). With its high density of MRs and GRs, the hippocampus is also particularly sensitive to the deleterious actions of chronic GC excess, potentiating neurotoxicity, dendritic atrophy, and perhaps even neuronal loss (Sapolsky, 1987). The idea that excess GCs could promote aging of the hippocampus was first established over 30 years ago following a study that showed a positive correlation between hippocampal aging (astrocyte reactivity as a marker of neuronal damage) and plasma levels of corticosterone in aging rats (Landfield et al., 1978). While only a few studies have shown high GC levels or stress actually cause hippocampal neuron loss (Uno et al., 1989; Sousa et al., 1998), much evidence support chronic stress (or high GCs) causing hippocampal atrophy (Watanabe et al., 1992; Magarinos and McEwen, 1995).

In humans, including those with Cushing's disease, Alzheimer's disease, depression, and normal aging, higher cortisol levels have been associated with poorer memory and hippocampal shrinkage/neuronal loss (De Leon et al., 1988; Wolkowitz et al., 1990; Newcomer et al., 1994; Mitchell and Dening, 1996; Lupien et al., 1998; Karlamangla et al., 2005; MacLullich et al., 2005). Moreover, increased HPA activity, as a consequence of impaired HPA axis negative feedback control, has been hypothesized to contribute to the decline in cognitive function, including deficits on spatial tasks with aging (Ohta, 1981; Lupien et al., 1998; McEwen et al., 1999). In support, the extent of age-related cognitive impairments in rodents and humans correlates with increased HPA activity (Issa et al., 1990; Meaney et al., 1995; Yau et al., 1995; Lupien et al., 1996, 1998). Aging rodents also show difficulty with hippocampus-dependent tasks that require a spatial mapping strategy; acquisition deficits have been reported in spatial information processing tests including Barnes hole-board task, radial arm mazes, and Morris watermaze (Ingram et al., 1981; Barnes and McNaughton, 1985; Gage et al., 1989; Issa et al., 1990; Gallagher et al., 1993; Yau et al., 1994, 1995).

Although GCs act on the hippocampus to modulate the formation of new memories, it also acts at other sites, notably the amygdala and prefrontal cortex, which functionally interact. Thus, GC effects on hippocampal LTP and memory can be blocked by lesions to the BLA (Roozendaal and McGaugh, 1997; Kim et al., 2001) and much evidence supports a role of the BLA via emotional arousal-induced noradrenergic activation and GR actions in the modulation of memory consolidation and working memory (Roozendaal and McGaugh, 2011). Given that amygdala function is affected by aging (Iidaka et al., 2002; Charles et al., 2003), it might be expected that fear conditioning would also be disrupted with aging and hence influence hippocampal memory deficits. However, age-related impairments in spatial memory appear not to be influenced by emotional and contextual memories, which tend to be preserved with aging (Comblain et al., 2004; Gould and Feiro, 2005; May et al., 2005; Bergado et al., 2011; Broster et al., 2012).

As a group, aged rats show impaired spatial learning. However, there are substantial inter-individual differences in performance (Yau et al., 1995). This increased variation in aged rats allows subdivision into categories (cognitively-unimpaired, cognitively-impaired) in the watermaze according to their latency to find the submerged platform on the last days of training relative to the mean latency of young controls (Issa et al., 1990; Tombaugh et al., 2002) or to a learning index score computed from probe trials (during retraction of the platform to the bottom of the pool during 30 s of 90 s trials) (Gallagher et al., 1993; Wilson et al., 2004; Robitsek et al., 2008). Aged rats have also been subdivided into impaired and unimpaired categories according to their spatial memory performance in a Y-maze two-trial spatial recognition task where the mean value of the percentage of visits in the novel arm of impaired rats were not different from chance values (Vallee et al., 1997). We typically find approximately ~20–25% of aged Lister hooded rat cohorts differ significantly (>2.5 SD) from young controls in the watermaze (escape latency and probe times) and another ~20–25% not significantly different (<0.5 SD) from young controls (Figure 3). It is the cognitively impaired groups that selectively show elevated plasma corticosterone levels (Issa et al., 1990; Yau et al., 1995), and reduced hippocampal corticosteroid receptor density (Issa et al., 1990; Yau et al., 1995) compared to both young and similarly aged cognitively unimpaired rats (Issa et al., 1990). While aged rats with watermaze performances not significantly different from young rats have been sub-classed as “cognitively unimpaired,” it is important to note that they are not cognitively the same as young rats. Indeed, we found that aged Lister hooded rats, categorized as unimpaired by their performance in the conventional spatial memory watermaze task, were impaired in a more cognitively demanding delayed matching-to-place paradigm spatial memory watermaze task (Steele and Morris, 1999) when compared to young controls (Yau and Seckl, unpublished).

Aged mice also show impaired spatial memory in the watermaze (Verbitsky et al., 2004; Pawlowski et al., 2009) and increased plasma CORT levels (Yau et al., 2001, 2007; Holmes et al., 2010) such that higher CORT levels correlates with impaired learning in the watermaze (Yau et al., 2001). However, in contrast to aged rats, subdividing aged mice cohorts into cognitively aged-impaired and aged-unimpaired groups is less clear-cut with large variations in performance found even in young controls (Pawlowski et al., 2009). Mice in general take longer to train in the watermaze than rats and show less consistent performance with some aged mice tending to float rather than swim. The Y-maze spatial recognition task, in contrast to the watermaze, is an ethologically relevant test based on the rodent's innate curiosity to explore novel areas and presents no negative or positive reinforcers and little stress for the rodents. When spatial memory was tested in the Y-maze using a 2 h inter-trial interval, young C57BL/6J mice can still remember the novel arm (exploring it more than the other two arms) while aged mice were overall impaired as a group (not distinguishing the novel arm between the three arms). As in the watermaze, inter-individual differences in Y-maze spatial memory performances exist in aged mice, but with a smaller percentage of aged mice “unimpaired” (<20%) while the majority were “impaired” (Yau and Seckl, unpublished).

11β-HSD1 is the predominant isoform in the adult rodent and human brain, where it is widely distributed with particularly high expression in the hippocampus, cerebellum, and cortex in both neurons and glia cells (Moisan et al., 1990; Sandeep et al., 2004). 11β-HSD2 whilst highly expressed in the fetal CNS until mid-gestation, in the adult brain is restricted to the nucleus of the solitary tract (NTS) in mice and this plus a few other scattered brain stem and diencephalic nuclei in rats. 11β-HSD2 acts as a dehydrogenase to inactivate GCs before they can bind to receptors. It is best noted for its role to exclude GCs from otherwise non-selective MRs in the distal nephron, thus allowing aldosterone selectivity. 11β-HSD1 in contrast functions as a 11β-reductase (regenerating active GCs from inert 11-DHC) in intact cells, thus locally “amplifying” GC levels within brain cells as well as in liver, adipose tissue, immune system cells etc (Figure 4). This direction of action, far from protecting neurons against GC excess, would be anticipated to increase local intraneuronal GC levels, potentiating their effects including toxicity. Consistent with this hypothesis, in vitro otherwise inert 11-DHC potentiates kainate neurotoxicity in hippocampal cells in culture, as a consequence of its conversion to active corticosterone by 11β-HSD1 expressed in the neurons, an effect lost in the presence of an 11β-HSD inhibitor (Rajan et al., 1996). The importance of such regeneration of GCs within cells in vivo was shown in mice homozygous for targeted disruption of the 11β-HSD1 gene (Kotelevtsev et al., 1997). 11β-HSD1 appears to be the only 11β-reductase, at least in mice, since knockout (KO) animals cannot convert 11-DHC to active corticosterone and showed evidence of reduced tissue GC actions (e.g., resist hyperglycaemia induced by obesity or stress) (Kotelevtsev et al., 1997). So are there brain effects of 11β-HSD1 deficiency?

Expression of 11β-HSD1 in brain sites (prefrontal cortex, hippocampus, hypothalamus) and pituitary that underpin negative feedback actions of GCs suggests that this enzyme may influence HPA axis activity. In order to maintain tissue GC levels in feedback sites that normally express 11β-HSD1, higher levels of plasma GCs would be predicted to result as a consequence of loss of local production of active GCs at these sites. Indeed mice lacking 11β-HSD1 on the 129/MF1 strain showed evidence of reduced HPA axis feedback sensitivity (elevated nadir levels of plasma corticosterone, enlarged adrenal glands, and exaggerated GC response to an acute stressor) (Harris et al., 2001). However, the effects of 11ß-HSD1 deficiency on HPA axis activity were lost when the mice were bred onto another genetic background strain. Thus, 11β-HSD1 KO mice congenic on a C57BL/6J background show normal nadir plasma corticosterone levels and efficient negative feedback regulation thought to be due to compensatory increased GR expression in the hippocampus and paraventricular nucleus of the hypothalamus (Carter et al., 2009). Activation of the HPA axis is therefore not an inevitable consequence of 11β-HSD1 deficiency or inhibition. The genetic background appears crucial in governing the HPA axis response to 11β-HSD1 deficiency or inhibition.

Different genetic background mouse strains can exhibit marked differences in learning and memory (Owen et al., 1997). The original 11β-HSD1 KO mice generated on the 129 Ola background learnt a cued version of a watermaze task (Yau et al., 2001), but had difficulty learning the classical task. Subsequently, aged 11β-HSD1 KO mice congenic on the C57BL/6J background, the strain of choice in many behavioral tests, also showed an improved cognitive phenotype resisting the spatial memory impairments observed in many aged control mice, this time observed in the standard reference memory watermaze and also Y-maze spatial recognition tasks (Yau et al., 2007). Even a 50% reduction in 11β-HSD1 in heterozygous 11β-HSD1 KOs is enough to prevent age-related spatial memory impairments (Sooy et al., 2010).

Aged C57BL/6J mice showed increased 11β-HSD1 expression in the cortex (layer V) and CA3 cells of the hippocampus compared to young controls. Moreover, hippocampal and cortical 11β-HSD1 mRNA levels correlate with impaired spatial learning in the watermaze (Holmes et al., 2010). The selective increase of 11β-HSD1 in the CA3 subregion of the memory-impaired hippocampus may be of functional significance since it is the CA3 cells that undergo dendritic atrophy following chronic restraint stress or corticosterone injections (Magarinos and McEwen, 1995), both of which impair spatial memory (Luine et al., 1994; Conrad et al., 1996; Wright and Conrad, 2005; Hoffman et al., 2011). CA3 appears crucial for memory acquisition and consolidation in the watermaze (Florian and Roullet, 2004). Furthermore, CA3 cells selectively fail to rapidly encode new spatial information in memory-impaired aged rats (Wilson et al., 2005). This suggests that the implied increase of GC levels driven by variable overexpression of 11β-HSD1 in CA3 of aged mice may be a major contributor to the spatial memory deficits with aging. The mechanisms underlying the upregulation of 11β-HSD1 in CA3 and cortex of aged mice are unknown. GCs and stress elevate hippocampal 11β-HSD1 in young animals affording a possible feed-forward system to amplify GC action (Low et al., 1994). Consistent with increased 11β-HSD1 levels in the hippocampus impairing memory with aging, transgenic mice overexpressing 11β-HSD1 in the forebrain under the CAMIIK promoter resulting in 50% increase in the hippocampus, developed premature memory impairments with deficits in hippocampus-dependent learning tasks (watermaze spatial reference memory and passive avoidance memory) at 18 months (Holmes et al., 2010).

In the absence of hippocampal neuron loss, age-related spatial learning deficits in rats may be the result of more subtle changes in synaptic structure or function (Rapp and Gallagher, 1996; Smith et al., 2000). Consistent with this idea, impaired hippocampal synaptic plasticity, specifically LTP in aged rats relates to individual differences in spatial learning ability (Deupree et al., 1991; Bach et al., 1999). Indeed, LTP induced in the CA1 region using theta-frequency stimulation (5 Hz) was selectively impaired in hippocampal slices from a subpopulation of aged rats that had previously shown poor spatial learning in the water maze (Tombaugh et al., 2002). Elevated corticosterone levels (presumably occupying both MR and GR) also impairs Primed Burst Potentiation (PBP), a low threshold form of LTP, in the hippocampal CA1 in the rat, while low levels of corticosterone (that would occupy mostly MRs) facilitates PBP (Diamond et al., 1992). Although the involvement of other hormones and/or brain regions may also play a role, high levels of corticosterone directly impairs hippocampal synaptic potentiation (Alfarez et al., 2002). In addition, induction of long-term depression (LTD), which in contrast to LTP weakens rather than strengthens synaptic contacts by repeated stimulation (Bear and Malenka, 1994), is enhanced by high corticosterone levels (Coussens et al., 1997) and during aging (Norris et al., 1996, 1998).

Primed burst stimulation induces lower potentiation in the hippocampus of young animals compared to the LTP induction protocol (Diamond et al., 1994, 1996; Alfarez et al., 2002), but few reports have shown this reliably in aged animals (Moore et al., 1993). We found very little PBP (a single 200 Hz stimulus followed 180 ms later by a burst of four stimuli at 200 Hz) in hippocampal slices from either 25 months old 11β-HSD1 KO or age matched controls. However, LTP induced with a single 100 Hz tetanus, was increased in hippocampal slices from aged 11β-HSD1 KO mice compared to aged-matched C57BL/6J controls (Yau et al., 2007). Thus, an increase in LTP in the hippocampus of aged 11β-HSD1 KO mice may, in part, underlie their retention of spatial memory with age.

Selective 11β-HSD1 inhibitors that can cross the blood brain barrier have recently been developed (Webster et al., 2007). Two weeks peripheral treatment with a CNS active selective 11β-HSD1 inhibitor (UE1961) in aged C57BL/6J mice improved spatial memory in the Y-maze compared to vehicle treated age-matched controls (Sooy et al., 2010). Moreover, intracerebroventricular administration of another selective 11β-HSD1 inhibitor (UE2316) for 2 weeks also reversed spatial memory impairments in aged C57BL/6J mice confirming mediation by brain 11β-HSD1 inhibition (Yau et al., unpublished). Thus, spatial memory impairments in aged mice are not always a consequence of irreversible brain structural changes and the effects of the inhibitor are most likely the consequence of reduced intracellular GC levels during spatial learning and recall. Previous studies have shown that it is the consequence of increased GCs as a result of hippocampal structural changes (e.g., chronic stress-induced dendritic atrophy) impairing HPA axis feedback that has the more important influence on spatial memory performance since such impairments can be prevented on the day of testing by blocking GC synthesis (Roozendaal et al., 2001; Wright et al., 2006).

Depending on the duration of treatment, selective 11β-HSD1 inhibitors may result in resistance of target tissues to GCs, including HPA axis regulatory centers of the brain and pituitary. This could lead to compensatory activation of the HPA axis in an attempt to compensate for the GC deficiency in negative feedback loci expressing 11β-HSD1 in a similar manner to that observed in 11β-HSD1 KO mice on the 129 genetic background (Harris et al., 2001). A hyperactive HPA axis would lead not only to excess plasma cortisol, but perhaps also to excess mineralocorticoid precursors (corticosterone and deoxycorticosterone) and androgens, all of which may manifest as unwanted side effects, such as hypertension and hirsutism/hyperandrogenization in women. The degree of HPA axis activation may depend on the genetic background of the individual, as in mice, as well as the levels of 11β-HSD1 in these tissues, which may differ under various environmental conditions (e.g., with stress, diet, aging). There is also the question of whether inhibition of 11β-HSD1 will affect other types of memory other than spatial memory in cognitively impaired aged individuals. In particular, emotional and fear associated memories, involving the basolateral amygdala, are known to be enhanced with increased GCs (Roozendaal and McGaugh, 2011). Hence, 11β-HSD1 inhibitors may compromise the strength of such memories but whether this is the case remains to be determined.

GCs play a role in the regulation of peripheral glucose mobilization and metabolism. Much less is known about the effects of GCs on glucose metabolism in brain tissues. GCs inhibit glucose utilization in the hippocampus (Sapolsky, 1986). Adrenalectomy decreases serum glucose levels and increases hippocampal glucose utilization (Kadekaro et al., 1988). While exogenously administered GCs adversely affect human memory consolidation and recall (Newcomer et al., 1994), glucose improves memory performance in normal elderly individuals (Hall et al., 1989) and in subjects with probable Alzheimer's disease (Craft et al., 1993). Studies in rodents suggest that hyperglycemia and high-fat diets adversely affect hippocampal function (Kamal et al., 2000; Molteni et al., 2002). In humans, hyperglycemia associate with decline in cognitive function with aging (Convit et al., 2003). Since 11β-HSD1 KO mice lack the enzyme in all tissues, it is possible that the cognitive protection in aged 11β-HSD1 KO mice may reflect, in part, their improved metabolic profiles (Kotelevtsev et al., 1997; Morton et al., 2001), rather than solely the direct effects of enzyme deficiency in the brain. However, aged 11β-HSD1 KO mice show no significant difference in glucose tolerance compared to age-matched controls (Yau et al., 2007). Moreover, the 11β-HSD inhibitor carbenoxolone improved cognitive function in aged humans without altering plasma glucose levels (Sandeep et al., 2004). Thus, the implication is that direct CNS effects of 11β-HSD1 deficiency or inhibition are more likely responsible for the observed improved cognitive function.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.