Astrocytic β₂-adrenergic receptors mediate hippocampal long-term memory consolidation

Significance

Experiences are remembered long-term when these memories are formed in a state of arousal and heightened emotion. The arousal-induced release of noradrenaline is critical for modulating consolidation, the process that establishes long-term memory. Although the effects of pharmacological manipulation of adrenergic signaling on memory stability are already being investigated in the clinical setting, how adrenergic receptors mediate long-term memory consolidation remains unclear. This study reports a previously unidentified mechanism with important translational implications: The noradrenergic receptors that in the hippocampus mediate memory consolidation are β₂-adrenergic receptors (β₂ARs) expressed in astrocytes. These receptors are necessary for the learning-evoked release of lactate from astrocytes, which then is required to support the neuronal molecular changes essential for long-term memory formation.

Abstract

Emotionally relevant experiences form strong and long-lasting memories by critically engaging the stress hormone/neurotransmitter noradrenaline, which mediates and modulates the consolidation of these memories. Noradrenaline acts through adrenergic receptors (ARs), of which β₂-adrenergic receptors (βARs) are of particular importance. The differential anatomical and cellular distribution of βAR subtypes in the brain suggests that they play distinct roles in memory processing, although much about their specific contributions and mechanisms of action remains to be understood. Here we show that astrocytic rather than neuronal β₂ARs in the hippocampus play a key role in the consolidation of a fear-based contextual memory. These hippocampal β₂ARs, but not β₁ARs, are coupled to the training-dependent release of lactate from astrocytes, which is necessary for long-term memory formation and for underlying molecular changes. This key metabolic role of astrocytic β₂ARs may represent a novel target mechanism for stress-related psychopathologies and neurodegeneration.

Emotional arousal and the stress hormone noradrenaline (NA) mediate and modulate memory consolidation, the process required to stabilize long-term memory (1). NA contributes to this effect through the activation of α- and β-adrenergic receptors (βARs) in brain regions such as the amygdala and the hippocampus, which are critical for encoding and consolidating memories (2). Of particular interest is the action of NA through βARs because blockers of these receptors, such as propranolol, have been reported to interfere with memory consolidation and strengthening (3–5) and have been suggested as potential therapeutics to treat anxiety disorders, including panic disorder (6) and posttraumatic stress disorder (7). All three subtypes of βARs—β₁, β₂, and β₃—are present in the central nervous system with distinct distributions (8): In the rat hippocampus, a region required for the consolidation of explicit/episodic memories, β₁ARs predominate, and their pattern of distribution is different from that of β₂ARs and β₃ARs (9–11), suggesting possible distinct roles for these receptor subtypes. Pharmacological investigations using agonists (e.g., NA itself) and antagonists (e.g., propranolol) that bind to both β₁ARs and β₂ARs have suggested βARs have a similar role in memory encoding, modulation, and retrieval in humans and rodents (3, 5, 12). Studies on animal models lacking β₁ARs or treated with selective β₁AR agonists or antagonists revealed the critical role of this adrenergic receptor subtype in both synaptic plasticity and memory formation, with particular emphasis on retrieval (12–14). Genetic deletion of β₂ARs results in impaired memory modulation by stress or corticosterone and impaired hippocampal plasticity, in agreement with a pharmacologically established role of β₂ARs in amygdala-dependent memory modulation and its effect on hippocampal function and in prefrontal cortex (2, 15, 16). However, little is known about the role of hippocampal β₂ARs in hippocampus-dependent long-term memory formation and processing.

In general, the functional contributions of βARs to memory processes have been thought to result mainly from their effect on neurons and therefore have been studied largely in neuronal/synaptic models (reviewed in ref. 17). However, in addition to being expressed in pre- and postsynaptic compartments of neurons, βARs are also found in other cell types, particularly in astrocytes (18–21). More specifically, it has been suggested that β₂ARs in the nervous system are expressed predominantly in glia (18, 21–23), whereas β₁ARs are found primarily in neurons, at synaptic junctions (18, 22, 23). This suggestion raises the question of which βAR subtype (β₁AR or β₂AR) and which βAR-expressing cells mediate hippocampal memory consolidation.

NA activates glycogenolysis and subsequent lactate release (24), a process known to occur mainly in astrocytes and not in neurons; in fact, glycogen is stored almost exclusively in astrocytes (25, 26). Glycogenolysis and its activation by NA are critical for memory consolidation in chicks (27). Furthermore, in the rat hippocampus, glycogenolysis, which results in lactate increase and astrocyte–neuron lactate transport, mediates memory consolidation (28, 29), in vivo hippocampal long-term potentiation, and the induction of learning-dependent molecular changes, including activity-regulated cytoskeletal protein (Arc) and phosphorylation of cAMP response element-binding protein at serine 133 (pCREB) and cofilin (p-cofilin) (29). Using the hippocampus-dependent task inhibitory avoidance (IA), here we asked which astrocytic and/or neuronal β₁ARs and/or β₂ARs in the hippocampus mediate memory consolidation.

Results

βARs Mediate Memory Consolidation via Astrocytic Lactate.

A bilateral injection of the β₁+β₂AR antagonist propranolol into the rat dorsal hippocampus (dHC) 15 min before training had no effect on short-term memory tested 1 h after training (Fig. 1A) but significantly impaired long-term memory tested 1 d and 7 d after training (Fig. 1B), suggesting that hippocampal β₁ARs and/or β₂ARs mediate IA memory consolidation. Given that NA promotes glycogenolysis and lactate release from astrocytes (30), which are necessary for memory consolidation (29), we tested whether the effect of propranolol on memory consolidation is linked to astrocytic lactate release. We found that coadministration of l-lactate with propranolol fully and persistently rescued memory impairment. In contrast, an equicaloric concentration of d-glucose had no effect. A threefold higher concentration of glucose rescued memory impairment at 1 d but not at 7 d after training (P < 0.0001; two-way repeated-measures ANOVA followed by Bonferroni’s post hoc tests) (Fig. 1B). Together, these results indicate that the critical role of β₁ARs and/or β₂ARs in memory consolidation is likely dependent on lactate and hence on astrocytic mechanisms.

Fig. 1.

l-lactate rescues the impairment of long-term IA memory and underlying molecular mechanisms produced by propranolol. Memory retention is expressed as mean latency ± SEM (in seconds). (A) Short-term memory tested 1 h after training following a bilateral dHC injection of propranolol given 15 min before training (n = 7). (B) Long-term memory tested 1 d after training (test 1) and 6 d later (test 2) following a bilateral dHC injection of vehicle or propranolol, in the presence or absence of l-lactate, or of an equicaloric concentration of d-glucose, or of a three times higher concentration (3×) of d-glucose, given 15 min before training (n = 7–13). (C–E) Densitometric Western blot analysis and representative images for Arc (C), pCREB (D), and pCaMKIIα (E) in dHC extracts from untrained (naive) rats injected with vehicle and rats injected 15 min before training with vehicle, l-lactate, propranolol, or propranolol + l-lactate and killed 30 min (for Arc) or 20 h (for pCREB and pCaMKIIα) after training. Data are expressed as mean protein percentage ± SEM of mean values (100%) in vehicle-injected naive rats. Each sample protein densitometric value was normalized to that of relative actin stained on the same blot (n = 5–10). *P < 0.05; **P < 0.01; ***P < 0.001. Numeric values and detailed statistical analyses are reported in Tables S1 and S2.

Table S1.

Mean latencies and statistical analyses related to Fig. 1

Treatment	n	Mean latency (s) ± SEM		Statistics
		Test 1	Test 2
Fig. 1A
Vehicle	7	250.9 ± 32.3	–	t12 = 0.62, P > 0.05
Propranolol	7	284.4 ± 43.0	–
Fig. 1B
Vehicle	11	363.6 ± 26.7	356.5 ± 48.8	Treatment: F4,43 = 12.91, P < 0.0001
Propranolol	13	103.0 ± 12.7	75.3 ± 9.0	Time: F1,43 = 10.37, P = 0.002
Propranolol + l-lactate	8	262.0 ± 49.3	275.3 ± 64.3	Interaction: F4,43 = 7.70, P < 0.0001
Propranolol + d-glucose	7	120.5 ± 34.3	120.5 ± 27.5
Propanolol + 3× d-glucose	9	288.8 ± 46.9	121.3 ± 15.8

Table S2.

Western blot relative densitometry and statistical analyses related to Fig. 1

Treatment	n	% of naive ± SEM	Statistics
Fig. 1C: Arc
Naive–vehicle	8	100 ± 11.3	F4,33 = 5.34, P = 0.002
Trained–vehicle	8	227.6 ± 27.3
Trained–l-lactate	6	207.7 ± 53.4
Trained–propranolol	9	87.3 ± 14.3
Trained–propranolol + l-lactate	7	221.5 ± 39.7
Fig. 1D: pCREB
Naive–vehicle	9	100 ± 11.1	F4,31 = 6.15, P = 0.001
Trained–vehicle	8	158.2 ± 15.0
Trained–l-lactate	5	153.1 ± 17.8
Trained-propranolol	9	108.4 ± 14.1
Trained–propranolol + l-lactate	5	182.7 ± 22.5
Fig. 1E: pCaMKIIα
Naive–vehicle	9	100 ± 13.6	F4,36 = 4.91, P = 0.003
Trained–vehicle	10	171.7 ± 16.0
Trained–l-lactate	6	200.5 ± 22.6
Trained–propranolol	10	108.8 ± 16.5
Trained–propranolol + l-lactate	6	185.1 ± 29.4

Propranolol injected bilaterally into the dHC 15 min before training also blocked the learning-dependent increase of molecular changes known to underlie synaptic plasticity and memory formation; these included the immediate early gene Arc 30 min after training (Fig. 1C), and phosphorylation of CREB (Fig. 1D) and of calcium calmodulin kinase II α (CaMKIIα) (Fig. 1E) 20 h after training, time points at which these molecular changes have been previously established as learning dependent (29, 31). Lactate coadministration rescued all these molecular impairments (P < 0.01; one-way ANOVA followed by Newman–Keuls post hoc test). The levels of total CREB and CaMKIIα remained unchanged (P > 0.05) (Fig. S1 and Table S3). Hence, βARs in the hippocampus engage astrocytic mechanisms to support memory formation.

Fig. S1.

No change in CREB and CaMKIIα concentration was seen with training in the presence or absence of propranolol and/or lactate treatments. (A and B) Densitometric analyses of Western blots and representative images for CREB (A) and CaMKIIα (B) obtained with dHC extracts from untrained (naive) rats injected with vehicle and rats injected 15 min before training with vehicle, l-lactate, propranolol, or propranolol + l-lactate and killed 20 h after training. Data are expressed as the percentage (mean ± SEM) of mean values (100%) in vehicle-injected naive rats (n = 5–10 per group). Numeric values and detailed statistical analyses are reported in Table S3.

Table S3.

Western blot relative densitometry and statistical analyses related to Fig. S1

Treatment	n	% of naive ± SEM	Statistics
Fig. S1A
Naive–vehicle	9	100 ± 7.4	F4,35 = 0.40, P > 0.05
Trained–vehicle	10	103.6 ± 6.6
Trained–l-lactate	6	103.7 ± 9.6
Trained–propranolol	9	95.3 ± 3.5
Trained–propranolol + l-lactate	6	93.2 ± 10.3
Fig. S1B
Naive–vehicle	9	100 ± 6.6	F4,30 = 0.24, P > 0.05
Trained–vehicle	7	98.6 ± 10.5
Trained–l-lactate	5	96.1 ± 10.1
Trained–propranolol	9	91.9 ± 7.3
Trained–propranolol + l-lactate	5	106.5 ± 22.1

β₂ARs, but Not β₁ARs, Mediate the Learning-Dependent Lactate Increase and Long-Term Memory Formation.

Next we sought to dissociate the roles of the βAR subtypes, β₁AR and β₂AR, in regulating training-induced astrocytic lactate release and memory consolidation. In agreement with our previous finding (29), we found that training leads to a significant increase in lactate levels in the dHC as measured by microdialysis; this increase lasted for more than 60 min and returned to baseline by ∼90 min. However, this lactate increase was completely abolished by a systemic (i.p.) injection of propranolol or of the β₂AR-selective antagonist ICI 118,551, but not the β₁AR-selective antagonist betaxolol, given 30 min before training (P < 0.0001; two-way ANOVA followed by Bonferroni post hoc tests) (Fig. 2A). Compared with vehicle-injected control rats, propranolol- and ICI 118,551-injected rats, but not betaxolol-injected rats, had significantly impaired long-term memory when tested 1 d after training (P < 0.0001; one-way ANOVA followed by Bonferroni post hoc tests) (Fig. 2A). The efficacy of betaxolol was confirmed in parallel by reproducing the previously established effect of this compound on retrieval of contextual fear conditioning (CFC) (Fig. S2 and Table S4) (12). In addition, as with systemic injections, bilateral injection of ICI 118,551 into the dHC 15 min before training impaired long-term memory when tested 1 d and 7 d after training, whereas injection of betaxolol had no effect (P < 0.001; two-way repeated-measures ANOVA followed by Bonferroni post hoc tests) (Fig. 2B). Following the 7-d test, rats received a reminder footshock (RS)—the unconditioned stimulus in a different context—and testing 1 d later showed that the memory deficit persisted, suggesting that the amnesia was likely caused by impaired consolidation rather than by inhibition of memory expression (P < 0.01; Student’s t test) (Fig. 2B). Hence, both the IA-induced increase in hippocampal lactate and long-term memory consolidation require β₂ARs but not β₁ARs. We therefore focused our subsequent investigations on β₂ARs.

Fig. 2.

Propranolol and ICI 118,551, but not betaxolol, block both training-evoked l-lactate increase in the dHC and long-term IA memory formation. (A, Left) The relative concentration of l-lactate in the dHC was measured by microdialysis in freely moving rats injected i.p. with vehicle, propranolol, betaxolol, or ICI 118,551 30 min before training. Untrained (naive) rats injected with vehicle were used as controls. Baseline data were collected for 20 min before training, and data collection continued for 90 min after training. Data are expressed as the percent of baseline l-lactate concentration ± SEM (mean of the first two samples set at 100%; n = 4–7). (Right) Long-term memory retention of the rats tested in microdialysis expressed as mean latency (in seconds) ± SEM (n = 6–7). (B) Long-term memory tested 1 d (test 1) and 7 d (test 2) after training and following an RS (test 3) in rats that received a bilateral dHC injection of vehicle, ICI 118,551, or betaxolol 15 min before training. Data are expressed as mean latency (in seconds) ± SEM (n = 6–14). **P < 0.01; ***P < 0.001. Numeric values and detailed statistical analyses are reported in Tables S5, S6, and S7.

Fig. S2.

Betaxolol impairs CFC retrieval. Memory retention is expressed as percentage of freezing (± SEM). Betaxolol was injected i.p. 30 min before CFC retrieval that occurred 1 d after training (n = 8 per group). *P < 0.05. Numeric values and detailed statistical analyses are reported in Table S4.

Table S4.

Percent freezing and statistical analyses related to Fig. S2

Treatment	n	% freezing ± SEM
Vehicle	8	57.7 ± 7.7
Betaxolol	8	32.1 ± 5.9

Statistics: t₁₄ = 2.53, P = 0.024.

Table S5.

Lactate concentration and statistical analyses related to Fig. 2

Time, min	Naive			Vehicle			Propranolol			Betaxolol			ICI 118,551
	Mean	SEM	n	Mean	SEM	n	Mean	SEM	n	Mean	SEM	n	Mean	SEM	n
−10	100.9	7.5	7	104.0	3.9	7	100.3	3.2	7	86.2	9.1	7	104.4	9.0	6
0	99.1	7.5	7	96.0	3.9	7	99.7	3.2	7	113.8	9.1	7	95.6	9.0	6
10	104.9	8.0	7	148.6	11.0	7	102.0	5.0	7	142.8	9.2	7	91.1	9.3	6
20	98.7	6.2	7	165.3	13.3	7	94.4	8.4	7	135.9	11.1	7	85.7	12.0	6
30	100.8	7.4	7	153.2	15.8	7	86.3	5.8	7	134.4	9.4	7	86.4	11.2	6
40	98.0	8.1	7	154.3	10.7	7	89.8	5.2	7	156.1	10.7	7	89.1	10.0	6
50	103.1	9.1	7	174.3	23.9	6	93.2	9.0	7	151.8	12.3	7	96.0	12.4	6
60	108.3	9.9	7	122.8	27.7	7	86.0	5.5	7	166.2	16.1	7	99.6	14.4	6
70	111.8	14.1	7	121.7	10.9	7	91.2	8.0	7	150.9	19.9	7	96.1	15.0	6
80	105.3	6.9	7	131.9	17.4	4	104.7	9.9	4	141.3	12.7	7	90.5	12.7	5
90	115.0	10.7	7	104.9	15.5	4	92.5	9.8	4	120.4	16.6	7	100.6	12.6	5

Statistics: Treatment: F_4,304 = 35.88, P < 0.0001. Time: F_10,304 = 2.08, P = 0.03. Interaction: F_40,304 = 1.92, P = 0.001.

Table S6.

Mean latencies and statistical analyses related to Fig. 2A

Treatment	n	Mean latency(s) ± SEM
Vehicle	7	297.1 ± 32.8
Propranolol	7	119.3 ± 21.9
Betaxolol	7	276.5 ± 21.6
ICI 118,551	6	86.2 ± 16.1

Statistics: F_3,23 = 19.97, P < 0.0001.

Table S7.

Mean latencies and statistical analyses related to Fig. 2B

Treatment	n	Mean latency(s) ± SEM		Statistics	n	Mean latency(s) ± SEM	Statistics
		Test 1	Test 2			Test 3
Vehicle	14	358.4 ± 45.4	313.9 ± 45.8	Treatment: F2,24 = 6.97, P = 0.0002	8	300.2 ± 57.9	t12 = 3.18, P = 0.008
Betaxolol	7	292.6 ± 67.7	291.6 ± 43.7	Time: F1,24 = 2.01, P > 0.05	–	–
ICI 118,551	6	121.9 ± 49.6	54.6 ± 10.0	Interaction: F2,24 = 0.46, P > 0.05	6	81.1 ± 16.1

Astrocytic, Not Neuronal, Hippocampal β₂ARs Mediate Long-Term Memory Formation.

The expression and distribution of β₂ARs on different cell types is not clear; although some studies suggest that β₂ARs are expressed mainly in glia (18, 21–23), others report that they are found on both astrocytes and neurons (20). To determine the selective functional contribution of astrocytic and/or neuronal β₂AR in long-term memory formation, we used cell-specific virus-mediated knockdown of the receptors. The astrocyte-specific promoter glial fibrillary acidic protein (short gfaABC1D) was packaged into adeno-associated virus 9 (AAV9), and the neuron-specific human synapsin (hSyn) promoter was packaged in AAVDJ; both were tagged with eGFP and injected into the dHC. Two weeks after injection, histochemical staining revealed that the expressed fluorescence of eGFP was distributed throughout the dHC (Fig. 3A). The AAV9-short gfaABC1D-eGFP expression was selectively astrocytic, as revealed by its colocalization with the endogenous astrocytic marker GFAP but not with the neuronal marker NeuN (Fig. 3 B and C) or with the microglial marker ionized calcium-binding adaptor molecule 1 (Iba1) (Fig. S3). Conversely, AAVDJ-hSyn-eGFP expression was found selectively in neurons and not astrocytes (Fig. 3 D–F).

Fig. 3.

AAV under the short gfaABC1D or the hSyn promoter selectively targets astrocytes or neurons, respectively. dHC infection of AAV9-short gfaABC1D-eGFP (A–C) or AAVDJ-hSyn-eGFP (D–F) 2 wk after bilateral injection (A, D). Shown are representative images of rostral-to-caudal brain sections [bregma −2.4 to −4.4 mm (45)]. All images of whole dHC are composite tile scans. In AAV9-short gfaABC1D-eGFP–infected cells, eGFP expression colocalizes with immunostaining of GFAP (B) but not NeuN (C) in subregions CA1 and CA3 and the dentate gyrus (DG). In AAVDJ-hSyn-eGFP–infected cells, eGFP expression colocalizes with NeuN (E) but not GFAP (F) in subregions CA1 and CA3 and the dentate gyrus.

Fig. S3.

AAV under short gfaABC1D does not infect microglia. dHC infection of AAV9-short gfaABC1D-eGFP 2 wk after bilateral injection. In cells infected with AAV9-short gfaABC1D-eGFP, eGFP expression does not colocalize with immunostaining of Iba1 in subregions CA1, CA3, and dentate gyrus (DG).

To knock down β₂ARs, we used two methods: shRNA miR (Sh) or antisense (AS) sequences against β₂ARs. shRNA were designed by Vector Biolabs (SI Materials and Methods). The Sh and AS sequences were engineered into the astrocyte-specific virus, resulting in AAV9-short gfaABC1D-eGFP-β₂AR-Sh and AAV9-short gfaABC1D-β₂AR-AS (henceforth referred to as “gfa-Sh” and “gfa-AS,” respectively) or into the neuron-specific virus, resulting in AAVDJ-hSyn-eGFP-β₂AR-Sh and AAVDJ-hSyn-β₂AR-AS (henceforth referred to as “hSyn-Sh” and “hSyn-AS,” respectively). Relative scrambled (Scr) or sense (S) sequences were used as controls.

First, we measured the ability of the Sh and AS viruses to affect the selective expression of the β₂AR protein. Using β₁AR/β₂AR-knockout mouse brains (32) we used Western blot analyses and/or immunohistochemistry to conduct specificity testing of several available antibodies indicated to bind β₂ARs specifically, but we failed to identify any specific reactivity (Table S8). Thus, to quantify β₂AR and β₁AR protein levels, we used receptor autoradiography binding assays. As previously established (10), this method is based on the quantification of iodo-(-)-cyanopindolol ([¹²⁵I] CYP) binding in the presence of serotonergic inhibitors (SB 224289 and WAY 100,635) and of the β₁AR inhibitor CGP-20712A to assess the expression of β₂AR or in the presence of the β₂AR inhibitor ICI 118,551 to assess the expression of β₁AR. Using this method, we quantified the distribution of β₁AR and β₂AR binding and confirmed the previously described differential distributions of these receptors in the hippocampus (Fig. 4A) (10). We observed that β₂AR binding occurs in the lacunosum moleculare and the stratum oriens subregions (Fig. 4B), which are enriched in astrocytes rather than neurons (Fig. 4C), whereas β₁AR binding was diffused throughout the hippocampus (Fig. 4B).

Table S8.

β₂AR antibodies tested in Western blot and/or immunohistochemistry

Source or reference	Catalog no.	Type	Target region	Application
Abcam	ab 36956	Polyclonal	Hippocampus	Western blot
				Immunohistochemistry
	ab 182136	Monoclonal	Hippocampus	Western blot
			Cerebellum
			Amygdala
			Liver
			Hippocampus	Immunohistochemistry
	ab 61778	Polyclonal	Hippocampus	Western blot
			Cerebellum
			Prefrontal cortex
			Liver
			Hippocampus	Immunohistochemistry
			Prefrontal cortex
	ab 176490	Polyclonal	Hippocampus	Immunohistochemistry
Santa Cruz	sc 569	Polyclonal	Hippocampus	Western blot
			Cerebellum
			Amygdala
			Liver
			Hippocampus	Immunohistochemistry
			Cerebellum
Cell Signaling Technologies	#8513	Monoclonal	Hippocampus	Western blot
			Cerebellum
			Prefrontal cortex
			Liver
			Hippocampus	Immunohistochemistry
Proteintech Group	13096-1-AP	Polyclonal	Hippocampus	Western blot
				Immunohistochemistry
Strader et al. (48, 49)	BAR 248	Polyclonal	Hippocampus	Western blot
	BAR 404

Western blot and immunohistochemistry images are available upon request.

Fig. 4.

β₂AR and β₁AR autoradiography. (A) Anatomical schematic representation of coronal brain sections at bregma −3.6 mm (45) relative to representative examples of receptor autoradiography: negative control and [¹²⁵I] CYP binding to β₂AR or β₁AR. (B) dHC autoradiography for β₂AR or β₁AR binding. (C) Composite tile scan images of immunostaining with anti-GFAP (astrocytes) or anti-NeuN (neurons) antibodies of a dHC section shown at the same bregma coordinates. DG, dentate gyrus; LMol, stratum lacunosum moleculare; MoDG, stratum moleculare of the dentate gyrus; Or, stratum oriens; Pyr, stratum pyramidale; Rad, stratum radiatum.

Compared with relative Scr controls, the infection of both gfa-Sh and hSyn-Sh resulted in a significant decrease in β₂AR binding without affecting the binding levels of β₁AR (P < 0.001 Student’s t test) (Fig. 5 A and B), providing evidence for the specific knockdown of β₂AR in astrocytes or neurons, respectively. Notably, the knockdown of the astrocyte-selective virus was more robust (about 50%) than that of the neuron-selective virus (about 20%), as is consistent with a more abundant β₂AR binding codistribution with astrocytes than with neurons and with previous literature suggesting higher expression of β₂AR in glia (18, 21–23).

Fig. 5.

ShRNA-mediated knockdown of β₂AR in astrocytes but not in neurons impairs IA long-term memory. This impairment is rescued by l-lactate. (A and B) β₂AR and β₁AR binding, expressed in disintegrations per minute (dpm), (n = 6) measured 2 wk after injection of gfa-Scr or gfa-Sh (A) or hSyn-Scr or hSyn-Sh (B). (C) Short-term memory tested 1 h after training and expressed as mean latency ± SEM (in seconds) in rats injected with either gfa-Scr or gfa-Sh 2 wk before training (n = 6). (D) Long-term memory tested 1 d (test 1) and 7 d (test 2) after training and after an RS (test 3), expressed as mean latency ± SEM (in seconds), in rats injected with either gfa-Scr or gfa-Sh 2 wk before training (n = 7–9) and with a bilateral dHC injection of vehicle or l-lactate administered 15 min before training. (E) Long-term memory tested 1 d (test 1) and 7 d (test 2) after training, expressed as mean latency ± SEM (in seconds), in rats injected with either hSyn-Scr or hSyn-Sh 2 wk before training (n = 12). *P < 0.05; **P < 0.01; ***P < 0.001. Numeric values and detailed statistical analyses are reported in Tables S10 and S11.

Table S10.

Receptor binding (dpm) and statistical analyses related to Fig. 5

Treatment	n	dpm ± SEM	Statistics
Fig. 5A: β2 AR
gfa-Scr	6	222.7 ± 14.46	t10 = 5.10, P = 0.0005
gfa-Sh	6	120.9 ± 13.77
Fig. 5A: β1AR
gfa-Scr	6	618.1 ± 25.29	t10 = 0.84, P > 0.05
gfa-Sh	6	645.0 ± 19.82
Fig. 5B: β2AR
hSyn-Scr	6	285.9 ± 9.190	t10 = 4.72, P = 0.0008
hSyn-Sh	6	238.6 ± 4.024
Fig. 5B: β1AR
hSyn-Scr	6	551.4 ± 49.40	t10 = 0.25, P > 0.05
hSyn-Sh	6	538.4 ± 13.38

Table S11.

Mean latencies and statistical analyses related to Fig. 5

Treatment	n	Mean latency(s) ± SEM				Statistics
		Test 1	Test 2	n	Test 3
Fig. 5C						t10 = 1.03, P > 0.05
gfa-Scr	6	239.2 ± 44.66	–	–	–
gfa-Sh	6	308.1 ± 49.60	–	–	–
Fig. 5D						Tests 1 and 2: Treatment: F3,28 = 11.54, P < 0.0001
gfa-Scr–vehicle	9	549.7 ± 114.5	522.4 ± 112.0	9	579.1 ± 100.2	Time: F1,28 = 5.86, P = 0.02
gfa-Scr–l-lactate	8	641.4 ± 109.0	457.5 ± 75.3	8	558.3 ± 103.3	Interaction: F3,28= 0.85, P > 0.05
gfa-Sh–vehicle	8	185.4 ± 37.9	135.6 ± 32.3	7	154.9 ± 41.6
gfa-Sh–l-lactate	7	900 ± 0	718.1 ± 117.7	7	692.9 ± 106.5	Test 3: F3,27 = 5.67, P = 0.004
Fig. 5E						Treatment: F1,22 = 0.07, P > 0.05
hSyn-Scr	12	583.3 ± 69.0	516.8 ± 73.6	–	–	Time: F1,22 = 2.03, P > 0.05
hSyn-Sh	12	600.2 ± 79.0	555.4 ± 88.7	–	–	Interaction: F1,22 = 0.08, P > 0.05

Next we determined the effect of virus-mediated knockdown of β₂ARs on memory retention. gfa-Sh bilaterally injected into the dHC 2 wk before training had no effect on short-term memory tested 1 h after training (Fig. 5C) but impaired long-term memory 1 d and 7 d later (P < 0.0001; two-way repeated-measures ANOVA followed by Bonferroni post hoc tests) and following an RS (P < 0.01; one-way ANOVA followed by Bonferroni post hoc tests) (Fig. 5D). Lactate injected bilaterally into the dHC 15 min before training fully and persistently rescued the memory impairment (Fig. 5D). In contrast, 2 wk after hippocampal injection, hSyn-Sh elicited no change in long-term memory retention compared with Scr controls (P > 0.05; two-way repeated-measures ANOVA) (Fig. 5E). To rule out the contribution of behavioral ceiling retentions and/or suboptimal time of infection to the lack of an effect of neuronal β₂AR knockdown on memory, we also tested the effect of a longer, 4-wk infection of hSyn-Sh and of a weaker memory evoked by milder-intensity training (0.6 mA footshock) given either 2 or 4 wk after infection. Memory tested 1 d and 7 d after training showed no differences in any of these conditions (P > 0.05; two-way repeated-measures ANOVA) (Fig. S4 and Table S9), indicating that neuronal β₂ARs are not critical for IA memory consolidation.

Fig. S4.

Prolonged hSyn-Sh hippocampal infection and/or footshock training (0.6-mA intensity) has no effect on memory retention. Memory retention is expressed as mean latency ± SEM (in seconds). Rats were trained with a 0.9-mA footshock 4 wk after viral infection (n = 4 per group) (A), 0.6-mA footshock 2 wk after viral infection (n = 6–8 per group) (B), or 0.6-mA footshock 4 wk after viral infection (n = 7 per group) (C). All animals were tested 1 d (test 1) and 7 d (test 2) after training. Numeric values and detailed statistical analyses are reported in Table S9.

Table S9.

Mean latencies and statistical analyses related to Fig. S4

Treatment	n	Mean latency (s) ± SEM		Statistics
		Test 1	Test 2
Fig. S4A				Treatment: F1,6 = 1.11, P > 0.05
hSyn-Scr	4	638.1 ± 150.5	432.2 ± 100.4	Time: F1,6 = 2.86, P > 0.05
hSyn-Sh	4	722.2 ± 105.6	654.3 ± 106.7	Interaction: F1,6 = 0.73, P > 0.05
Fig. S4B				Treatment: F1,12 = 0.02, P > 0.05
hSyn-Scr	8	344.9 ± 68.1	350.9 ± 86.8	Time: F1,12 = 0.48, P > 0.05
hSyn-Sh	6	366.7 ± 74.8	296.7 ± 81.6	Interaction: F1,12 = 0.68, P > 0.05
Fig. S4C				Treatment: F1,12 = 0.08, P > 0.05
hSyn-Scr	7	523.6 ± 104.1	371.1 ± 95.5	Time: F1,12 = 1.10, P > 0.05
hSyn-Sh	7	489.8 ± 148.4	495.0 ± 145.9	Interaction: F1,12 = 1.26, P > 0.05

Similar results were obtained using gfa-AS and hSyn-AS. Four weeks after infection, both gfa-AS and hSyn-AS significantly reduced β₂AR binding compared with S controls (P < 0.05; Student’s t test), whereas β₁AR binding remained unchanged (Fig. 6 A and B). Gfa-AS had no effect on short-term memory (Fig. 6C) but significantly impaired long-term memory tested at 1 d and 7 d after training (Fig. 6D). The memory impairment was rescued by a bilateral injection of lactate into the dHC 15 min before training (P < 0.0001; two-way repeated measures ANOVA followed by Bonferroni post hoc tests) (Fig. 6D). In contrast, hSyn-AS had no effect on long-term memory (P > 0.05; one-way ANOVA) (Fig. 6E).

Fig. 6.

Antisense-mediated knockdown of β₂AR in astrocytes but not in neurons impairs IA long-term memory. The impairment is rescued by l-lactate. (A and B) β₂AR and β₁AR binding expressed in disintegration per minute (n = 6) measured 4 wk after injection of gfa-S or gfa-AS (A) or hSyn-S or hSyn-AS (B). (C) Short-term memory tested 1 h after training and expressed as mean latency ± SEM (in seconds) of rats injected with either gfa-S or gfa-AS 4 wk before training (n = 5 or 6). (D) Long-term memory tested 1 d (test 1) and 7 d (test 2) after training, expressed as mean latency ± SEM (in seconds) of rats injected with either gfa-S or gfa-AS 4 wk before training (n = 10–13). A bilateral dHC injection of vehicle or l-lactate was administered 15 min before training. (E) Long-term memory tested 1 d (test 1) and 7 d (test 2) after training, expressed as mean latency ± SEM (in seconds) of rats injected with either hSyn-S or hSyn-AS 4 wk before training (n = 11 or 12). *P < 0.05; **P < 0.01; ***P < 0.001. Numeric values and detailed statistical analyses are reported in Tables S12 and S13.

Table S12.

Receptor binding (dpm) and statistical analyses related to Fig. 6

Treatment	n	dpm ± SEM	Statistics
Fig. 6A: β2AR
gfa-S	8	230.0 ± 17.98	t14 = 2.16, P = 0.049
gfa-AS	8	158.5 ± 27.80
Fig. 6A: β1AR
gfa-S	8	622.4 ± 13.93	t14 = 0.22, P > 0.05
gfa-AS	8	625.8 ± 6.519
Fig. 6B: β2AR
hSyn-S	6	238.8 ± 17.88	t10 = 2.67, P = 0.023
hSyn-AS	6	138.3 ± 33.07
Fig. 6B: β1AR
hSyn-S	6	509.4 ± 20.55	t10 = 0.48, P > 0.05
hSyn-AS	6	521.6 ± 14.90

Table S13.

Mean latencies and statistical analyses related to Fig. 6

Treatment	n	Mean latency(s) ± SEM		Statistics
		Test 1	Test 2
Fig. 6C				t9 = 0.31, P > 0.05
gfa-S	5	326.8 ± 54.5	–
gfa-AS	6	301.6 ± 58.0	–
Fig. 6D				Treatment: F3,43 = 11.52, P < 0.0001
gfa-S–vehicle	12	590.6 ± 82.8	419.9 ± 70.5	Time: F1,43 = 13.25, P = 0.0007
gfa-S–l-lactate	13	566.4 ± 76.8	385.8 ± 68.0	Interaction: F3,43 = 0.51, P > 0.05
gfa-AS–vehicle	10	150.2 ± 37.2	93.2 ± 19.9
gfa-AS–l-lactate	12	455.3 ± 60.8	269.93 ± 43.5
Fig. 6E				Treatment: F1,21 = 0.54, P > 0.05
hSyn-S	12	677.9 ± 72.7	514.8 ± 79.0	Time: F1,21 = 7.31, P = 0.01
hSyn-AS	11	548.1 ± 90.5	492.4 ± 72.0	Interaction: F1,21 = 1.76 P > 0.05

Together, these data show that β₂ARs expressed by hippocampal astrocytes rather than neurons are the critical effectors of the adrenergic-mediated effect on long-term memory formation. Specifically, astrocytic β₂ARs play a critical role in memory consolidation by activating lactate release that supports long-term memory formation and its underlying molecular changes.

SI Materials and Methods

Animals.

Adult male Long–Evans rats weighing between 200 and 250 g were used. Animals were individually housed and maintained on a 12-h light/dark cycle. Experiments were performed during the light cycle. All rats were allowed ad libitum access to food and water and were handled for 3 min/d for 5 d prior any procedure. For all experiments, rats were randomly assigned to different groups. All protocols complied with the National Institutes of Health Guide for the Care and Use of Laboratory Animals (44) and were approved by the Icahn School of Medicine at Mount Sinai Institutional Animal Care and Use Committee or by the New York University Animal Welfare Committee.

Cannula Implants.

Bilateral hippocampal surgeries were performed as previously described (31). Rats were anesthetized with ketamine (75 mg/kg) and xylazine (10 mg/kg), and stainless steel cannulas (22 gauge; Plastics One) were stereotaxically implanted (4 mm posterior to bregma, 2.6 mm lateral from midline, and 2 mm ventral). After surgery, rats recovered for 7 d before undergoing any procedure. After completion of an experiment, brains were fixed in 10% formalin, and 50-μm sections were cut on a Leica vibratome to verify cannula placements. Rats with incorrect placement were discarded from the study.

IA.

IA was carried out as described previously (31). The IA chamber (Med Associates) consisted of a rectangular Perspex box divided into a safe compartment and a shock compartment. The safe compartment was white and illuminated by a light fixture fastened to the compartment wall. The shock compartment was black and unilluminated. Foot shocks were delivered to the grid floor of this chamber via a constant current scrambler circuit. The two compartments were separated by an automatically operated sliding door. During training sessions, each rat was placed in the safe compartment with its head facing away from the door. After 10 s the door was automatically opened, allowing the rat access to the shock chamber. The door closed 1 s after the rat entered the shock chamber, and a brief foot shock (0.6 or 0.9 mA for 2 s) was administered. Latency to enter the shock compartment was taken as a measure of acquisition. The rat then was returned to its home cage. Retention tests were performed at the indicated times by placing the rat back into the safe compartment and measuring the latency to enter the shock compartment. Foot shock was not administered on the retention test, and testing was terminated at 540 or 900 s. The duration and intensity of the RS were identical to that given during training, but the RS was carried out in a novel chamber with transparent walls that was placed in a separate, well-lit room. Training and testing procedures were performed blind to treatment conditions.

CFC.

Rats were conditioned in a fear-conditioning chamber, which consisted of a rectangular Plexiglas box (30.5 × 24.1 × 21.0 cm) with a metal grid floor (Model ENV-008; Med Associates). Rats were placed in the chamber for 120 s and then received a 1.5-mA footshock for 2 s. One minute later, the rats were removed. Testing occurred 1 d later; rats were placed in the same chamber for 5 min. No footshock was delivered during testing. All experiments were video recorded, and freezing, defined as lack of movement except for breathing, was scored by an experimenter blind to the treatment conditions.

Drug Injections.

Propranolol (Sigma-Aldrich) was dissolved in PBS containing 10% DMSO. Sodium l-lactate, d-glucose, and ICI 118,551 (Sigma-Aldrich) and betaxolol (Tocris Bioscience) were dissolved in PBS (pH 7.4), or in 10% DMSO when delivered in the same experiment as propranolol. Systemic injections were given i.p. at 10 mg/kg (propranolol, betaxolol, and ICI 118,551) in a 0.5-mL volume at the indicated time points. Intrahippocampal injections (1 μL per hippocampal side) were performed with the infusion needles extending 1.5 mm beyond the cannula. At the indicated time points before training, the rats received bilateral hippocampal injections at a rate of 0.333 μL/min via an infusion pump. The injection needle was left in place for 3 min following the injection to allow complete dispersion of the solution. Intrahippocampal concentrations were propranolol: 5 μg/μL; betaxolol: 2.2 μg/μL; ICI 118,551: 5 μg/μL; l-lactate: 100 mM; d-glucose: 50 mM or 150 mM.

Western Blot Analysis.

Ten to fifty micrograms of total protein extract per lane were resolved using 7.5 or 10% SDS/PAGE and analyzed by Western blot as described previously (31). Primary antibodies were rabbit anti-pCREB (1:2,000; Cell Signaling), rabbit anti-CREB (1:1,000; Millipore), rabbit anti-Arc (1:1,000; Synaptic Systems), rabbit anti-pCaMKII (1:5,000; Cell Signaling), and mouse anti-CaMKII (1:4,000; Millipore). We also tested the following primary anti-β₂AR antibodies (Table S8) according to the manufacturer’s recommendation or as previously described: ab36956 and ab182136, ab61778 (46) and ab176490 (all from Abcam); sc-569 (47) (Santa Cruz Biotechnology); 8513 (Cell Signaling Technologies); and 13096-1-AP (Proteintech Group). BAR248 and BAR404 (48, 49) were a generous gift from Chiye Aoki, Center for Neural Science, New York University, New York. Actin (1:5,000; Santa Cruz Biotechnology) was used for loading normalization. HRP-coupled specific secondary antibodies (1:4,000; Santa Cruz Biotechnology) were incubated in Tris-buffered saline (TBS) containing 5% (wt/vol) skim milk, or LI-COR secondary antibodies, anti-rabbit IRDye800CW and anti-mouse IRDye680 (1:10,000), were incubated in TBS containing 0.1% Tween and 0.01% SDS, for 1 h at room temperature. Membranes were scanned on the LI-COR Odyssey imager under nonsaturating conditions. Data were quantified using pixel intensities with the Odyssey software according to the manufacturer’s protocols (LI-COR). Alternately, ECL detection (GE Healthcare Bio-Sciences) was used, and relative densitometric analysis was done using ImageJ software (NIH).

Immunohistochemistry.

Rats were killed with pentobarbital (100 mg/kg) or chloral hydrate (750 mg/kg) and then perfused transcardially with ice-cold 0.1-M PBS for 3 min followed by 4% (wt/vol) paraformaldehyde (PFA) for 15 min at a flow rate of 9 mL/min. Brains were postfixed overnight in 4% (wt/vol) PFA, followed by a 2- to 3-d cryoprotection in 30% (wt/vol) sucrose/PBS at 4 °C. Serial sets of 30-μm coronal sections from fixed brains were collected on a cryostat. Fixed sections were incubated in blocking solution [3% (vol/vol) serum, 0.3% Triton X-100], followed by incubation with primary antibody overnight in 3% serum, 0.3% Tween 20 [mouse anti-GFAP (1:400; Millipore), mouse anti-NeuN (1:200; Millipore), and rabbit anti-Iba1 (1:400; Wako)]. We also tested the following primary anti-β₂AR antibodies (Table S8) according to the manufacturer’s recommendation or as previously described: ab36956, ab182136, ab61778 (46), and ab176490 (all from Abcam); sc-569 (47) (Santa Cruz Biotechnology); 8513 (Cell Signaling Technologies); and 13096-1-AP (Proteintech Group). Sections were washed and subsequently incubated with Alexa-Fluor 488–conjugated secondary antibody (1:400; Fisher Scientific) for 2 h at room temperatures. Sections were mounted on gelatin-coated slides and coverslipped with Vectashield mounting medium containing DAPI (Vector Laboratories). Images were captured on a Leica SP5 or SP8 confocal microscope or an Olympus VS120 fluorescent microscope.

Microdialysis.

Microdialysis was carried out as previously described (29). Cannulae (CMA 12), probes (CMA 12 elite, 2 mm long), and fluorinated ethylene propylene (FEP) tubing were from CMA Microdialysis, Inc. Rats underwent surgery as described for cannula implants but were implanted with microdialysis cannulae unilaterally targeting the left dHC. Rats were handled for 10 min/d for 5 d before microdialysis. On the day of microdialysis, the probe was inserted into the guide cannula and perfused with artificial cerebrospinal fluid containing 119 mM NaCl, 26 mM NaHCO3, 2.5 mM KCl, 1 mM NaH₂PO₄, 1.25 mM Na₂HPO₄, 2 mM CaCl2, and 4 mM MgCl2 at a rate of 3 μL/min, collected every 10 min. The first two samples were discarded, and then, after 20 min of baseline collection, rats were trained in IA with a 0.9-mA footshock; sample collection was continued for 90 min after training. Samples were subsequently analyzed using the Abcam Lactate Fluorescence Assay Kit. Because baseline lactate concentrations showed interindividual differences, data from each animal were expressed as percentages of baseline, and the individual values were calculated accordingly for each animal.

Viral Constructs and Injections.

Sh sequences were identified by Vector Biolabs using second-generation prediction algorithms; the candidates were embedded in miR arms (sequence proprietary) and screened using A549 cells. Plasmids overexpressing rat β₂AR were cotransfected with the plasmids containing Sh or Scr sequences driven by a CMV promoter. A luciferase assay was used to select the sequence with the highest β₂AR knockdown; Sh (GCTG-TAAGGAGGATGTAAACTTCCT-GTTTTGGCCACTGACTGAC-AGGAAGTTCATCCTCCTTA) and Scr (CCGGCAACAAGATGAAGAGCACCAACTCGAGTTGGTGCTCTTCATCTTGTTGTTTTT) sequences were engineered into plasmids pAAV-short gfaABC1D-eGFP, a generous gift from Bernard Schneider, École Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland, and pAAV-hSyn-eGFP, a generous gift from Karl Deisseroth, Stanford University, Stanford, CA. Packaging of AAV9 and AAVDJ (50) viruses was carried out by Stanford’s Gene Vector and Virus Core. Mark Kay, Stanford University, Stanford, CA provided the pRC-DJ plasmid used to produce AAVDJ. AS and S sequences targeting the first 770 bp of the rat β₂AR coding sequence were generated using rat hippocampal cDNA. RNA was extracted with TRIzol (Invitrogen) and reverse transcribed using the Bio-Rad iScript first-strand synthesis kit (Bio-Rad). PCR (Bio-Rad) was performed using Pfu DNA polymerase (Invitrogen) using the following primers: AS forward: 5′-TCCGTGCCCGCTCCTC-3′ and reverse: 5′CCGGCCATGGAGCCAC-3′; and S forward: 5′-CCGGCCATGGAGCCAC-3′ and reverse: 5′-TCCGTGCCCGCTCCTC-3′. Resulting sequences were engineered into pAAV-short gfaABC1D-eGFP and pAAV-hSyn-eGFP in place of the eGFP sequence; they were packaged into AAV9 [generated at the EPFL; design, production, and titration of the AAV9 vector for antisense expression in astrocytes have been described previously (51)] and AAVDJ (generated at Stanford’s Gene Vector and Virus Core), respectively. All constructs were verified by sequencing. All viral genomic copies (GC) were determined by TaqMan (Applied Biosystems) quantitative PCR. Viruses were produced at titers of 10¹³ GC/μL. Two microliters of virus were injected per hippocampal side at a rate of 0.2μL/min. The injection needle was left in place for 10 min following injection to allow complete dispersion of the solution.

Receptor Autoradiography.

Receptor autoradiography was carried out as previously described (10). Brains were flash-frozen for autoradiography analysis to assess βAR binding. Sections were cut at 20 μm on a cryostat, mounted on Superfrost Plus Slides (Fisher Scientific), and stored at −80 °C. Slides were fixed for 2 min in 0.1% PFA and washed in Tris buffer (50 mM Tris⋅HCl, 120 mM NaCl, 5 mM KCl; pH 7.4). To determine β₁AR- and β₂AR-specific binding, slides were incubated for 150 min at room temperature in a 150-pM [¹²⁵I] CYP (Perkin-Elmer) solution containing 1 μM SB 224289 hydrochloride, 1 μM WAY 100635 maleate (Tocris), and 100 nM CGP 20712 dihydrochloride (Tocris) to assess β₂AR-specific binding or 50 nM ICI 118,551 hydrochloride (Sigma) to assess β₁AR-specific binding in Tris buffer with 10 mM MgCl₂ and 0.1% BSA. To assess nonspecific binding, a separate set of slides was incubated in 150-pM [¹²⁵I] CYP with all the above inhibitors. All slides then were washed in 4 °C Tris buffer with MgCl₂ and then in ice-cold water. Slides were dried and then exposed to Carestream Kodak Biomax MR film for 1 d. Films were developed with Kodak GBX Developer and Fixer. Regions of interest were quantified as previously described (52). Optical densities for βAR binding (throughout the hippocampus for β₁AR and in the lacunosum moleculare and stratum oriens for β₂AR) were measured using ImageJ, and [¹²⁵I] CYP standards were used to convert the uncalibrated optical densities to disintegrations per minute. Background was subtracted using binding measurements from the corpus callosum. Brain sections from β₁AR/β₂AR-knockout mice (32) in pilot experiments confirmed the specificity of the assay.

Statistical Analyses.

For sample-size estimation using power analyses, we used a power analysis calculator (G*Power3). For biochemical studies, power calculation of one-way ANOVA indicated that a minimum sample size of four or five rats per group was necessary to achieve power of 0.8 and an error probability of 0.05. For behavioral experiments, a similar power analysis calculated the requirement of a minimum sample size of six for two-way ANOVA to achieve power of 0.8 and an error probability of 0.05. Statistical analyses were designed using the assumption of normal distribution and similar variance among groups. Data were analyzed using Prism 6 (GraphPad Software, Inc.) using one-way or two-way ANOVA or two-way repeated-measures ANOVA followed by Bonferroni or Newman–Keuls post hoc tests. When two groups were compared, a Student’s t test was used. All tests were two-sided.

Discussion

Using two distinct, virus-mediated, cell-specific knockdown approaches, we showed that hippocampal β₂ARs, but not β₁ARs, and specifically β₂ARs expressed on astrocytes but not on neurons, are required for IA memory consolidation. The critical action of the hippocampal β₂ARs is coupled to the training-evoked lactate release.

These results together with the previously established roles for β₁ARs in synaptic plasticity and CFC memory retrieval of both rats and mice (12, 14) show that β₁ARs and β₂ARs contribute differentially to hippocampal memory function and processes. We speculate that in the hippocampus β₁ARs may activate neuronal responses during retrieval and that β₂ARs activate astrocytic metabolic coupling to neurons to mediate memory consolidation. This differential role of β₁ARs and β₂ARs in memory processes seems to be part of a broader picture of distinct, and in some cases opposite, roles of the β₁ARs and β₂ARs in memory processes such as consolidation and retrieval which involve different brain regions in different memory systems (13, 33). Our findings of a cell-specific role of β₂ARs in hippocampal astrocytes may contribute to explaining controversial data on systemic or genetic ablation vs. intraregional inhibition of receptor subtypes (12, 15). Understanding these differential roles and contributions of glial vs. neuronal βARs, perhaps in different neuronal populations (excitatory vs. inhibitory neurons) as well as in different brain regions, is key to unraveling the mechanisms of NA-mediated responses. This understanding also should help the clinical use of βAR antagonists or agonists, and the understanding of the relative behavioral outcomes.

Similar to the results described in chicks (34), our data showed that astrocytic β₂ARs in the hippocampus play a critical role in memory consolidation. Although the abundance of βAR subtypes does not correlate in chick and mammalian brains (34), it seems that β₂ARs expressed by astrocytes have conserved roles in memory formation by controlling glycogenolysis (27). In addition, lactate production from aerobic glycolysis coupled to β₂AR stimulation does not seem to occur exclusively in the brain; in fact, it has also been described as occurring in muscle during shock states associated with a reduced or maintained blood flow (35).

Our results are consistent with several previous findings concerning functional regulation, mechanisms of action, and anatomical targeting of NA in the mammalian brain, where NA, produced by the locus coeruleus (LC), projects diffusely to a variety of brain areas, including the neocortex (36) and the hippocampus (37, 38). First, because NA is released in the extracellular space from junctional varicosities along its fibers, it can act on extrasynaptic receptors, which are localized on astrocytes in particular (39). Second, in the hippocampus the highest noradrenergic innervation is found in the dentate gyrus and in the stratum lacunosum moleculare, where we found an enrichment of β₂ARs (Fig. 4B). Third, one cellular consequence of NA release during LC firing in particular behavioral states or evoked by glutamatergic inputs is glycogenolysis (30). Fourth, NA signaling is necessary for long-term potentiation (17), memory consolidation (2), and the induction of plasticity and memory genes such as Arc, pCREB, and pCaMKIIα in the hippocampus in vivo (Fig. 1).

Our findings that the training-dependent lactate increase in the hippocampus requires β₂ARs and not β₁ARs and that supplying lactate rescues the IA memory impairment caused by hippocampal astrocytic β₂AR knockdown strengthens the conclusion that astrocytically generated lactate is a key mediator of hippocampus-dependent memory formation under arousing conditions (29). Our results do not dissect whether this lactate originates solely from glycogenolysis or also from glutamate-mediated glycolysis, as proposed by the astrocytic–neuronal lactate shuttle hypothesis (40). As suggested by Hertz and Gibbs (41), the level of stress/arousal evoked by the task may dictate whether glycogenolysis and/or glycolysis are recruited to provide the lactate necessary for memory consolidation. It is possible that the relevance of the experience, and thus the experience-evoked stress level, determines whether astrocytic β₂ARs are recruited as essential mechanisms of memory consolidation.

Furthermore, similar to previous observations with the glycogenolysis inhibitor 1,4-dideoxy-1,4-imino-d-arabinitol hydrochloride (DAB) (29), a direct supply of glucose into the hippocampus was unable to replicate the effect of lactate, and only at higher concentrations were the memory impairments caused by β₂ARs disruption transiently rescued. Although further experiments are needed to dissect this issue, one possible explanation is that activity-dependent processes promote glucose entry into astrocytes rather than directly into neurons. Glucose then would be metabolized into lactate and finally transported from astrocytes to neurons (42). If this is the case, direct hippocampal delivery of lactate therefore might be more efficient.

Because the dysregulation of β₂ARs in astrocytes is associated with multiple sclerosis, Parkinson’s disease, and Alzheimer’s disease (43), we suggest that the disruption of the metabolic role of astrocytic β₂ARs in regions supporting cognitive functions may contribute to the pathological features of those diseases. Hence, targeting astrocytic β₂ARs mechanisms may help prevent or repair these disorders and/or their precipitation by stress.

Materials and Methods

All animal protocols complied with the National Institutes of Health Guide for the Care and Use of Laboratory Animals (44) and were approved by the Icahn School of Medicine at Mount Sinai Institutional Animal Care and Use Committee or by the New York University Animal Welfare Committee. Adult male Long–Evans rats were implanted with cannulae and/or injected with AAV bilaterally into the dHC (from bregma: anteroposterior, −4 mm; mediolateral, ±2.6 mm; and dorsoventral, −3.5 mm) as described in ref. 31. IA training (0.9 mA for 2 s) and testing were performed as described in ref. 31. Drug injections (i.p.) of propranolol (Sigma-Aldrich), betaxolol (Tocris Bioscience), or ICI 118,551 (Sigma) were done at 10 mg/kg. Intra-dHC injections (1 μL per side) of propranolol (5 μg/μL), betaxolol (2.2 μg/μL), ICI 118,551 (5 μg/μL), l-lactate (100 mM), or d-glucose (50 mM or 150 mM, as indicated) were performed as described previously (29, 31). For Western blot analysis, dHC protein extracts were done as described in ref. 31; primary antibodies were rabbit anti-pCREB (1:2,000; Cell Signaling), rabbit anti-CREB (1:1,000; Millipore), rabbit anti-Arc (1:1,000; Synaptic Systems), rabbit anti-pCaMKII (1:5,000; Cell Signaling), and mouse anti-CaMKII (1:4,000; Millipore). Immunohistochemistry was performed using mouse anti-GFAP (1:400; Millipore), mouse anti-NeuN (1:200; Millipore), or rabbit anti-Iba1 (1:400; Wako). Microdialysis samples, collected every 10 min, were analyzed using the Abcam Lactate Fluorescence Assay Kit as described in ref. 29. Receptor autoradiography was performed as described in ref. 10; [¹²⁵I] CYP (Perkin-Elmer) containing 1 μM SB 224289 (Tocris), 1 μM WAY 100635 (Tocris), and 100 nM CGP 20712 (Tocris) or 50 nM ICI 118,551 (Sigma) was used. Data were analyzed using one-way or two-way ANOVA or two-way repeated-measures ANOVA followed by Bonferroni or Newman–Keuls post hoc tests or Student’s t test.