Molecular studies
The brains of three adult heterozygous ALDH1L1 – eGFP-L10a bacterial artificial chromosome (BAC) transgenic mice were processed for immunocytochemistry using antibodies against enhanced green fluorescent protein (eGFP), an astrocytic marker (anti-GFAP), an oligodendrocytic marker (CNP), and a neuronal marker (anti-NeuN). GFP immunofluorescence was restricted to non-neuronal, non-oligodendrocytic cells (Fig. 1), consistent with previous evidence [13]. Some GFP-labeled cells did not express GFAP, also in line with published data [13, 19, 20]. Thus, these results confirm that ALDH1L1 is a more inclusive astrocytic marker than GFAP.
Sleep/wake pattern and response to sleep deprivation are normal in ALDH1L1 – eGFP-L10a mice
We first tested whether ALDH1L1 – eGFP-L10a mice showed a normal sleep/wake cycle by making chronic electroencephalographic (EEG) recordings that lasted for several days, including during and after sleep deprivation. Mice (n = 5) showed robust entrainment to the light–dark cycle, sleeping mainly during the day and staying awake most of the night, as expected for nocturnal animals. During baseline, NREM sleep accounted for 39.2 ± 1.1 % of the 24-hr cycle, and 50.4 ± 2.3 % of the light period, while the corresponding values for rapid eye movement (REM) sleep were 7.4 ± 0.4 % and 10.4 ± 0.4 % (Fig. 2a). As expected, NREM sleep was predominant in the first part of the light phase relative to the second part (Fig. 2a). The mean duration of NREM and REM episodes, number of brief arousals, and several other sleep parameters were all within the range reported for most mouse lines (e.g. [21–24]).
In mammals, SWA – the EEG power between 0.5 and 4.5 Hz – reflects both the need for sleep and its depth [18]. Indeed, SWA levels increase as a function of prior wake duration and decline in the course of sleep. Moreover, arousal thresholds are higher when SWA is higher. Changes in SWA were studied after ALDH1L1 – eGFP-L10a mice were deprived of sleep during the first 4 hr of the light period. During sleep deprivation, they spent less than 2 % of their time in NREM sleep, and REM sleep was completely abolished. In the first 2 hr of recovery sleep, NREM sleep duration was significantly increased relative to baseline (Fig. 2b), while REM sleep and total sleep amounts did not change (not shown). SWA during the first hour of recovery sleep was increased relative to baseline levels in both the frontal (Fig. 2c) and parietal cortices (not shown). Thus, we conclude that sleep patterns and mechanisms of sleep regulation in ALDH1L1 – eGFP-L10a mice are normal.
TRAP enriches for astrocytic-specific genes
Three groups of mice were used in the TRAP study (n = 6 per group): awake mice (W) were collected during the dark phase (~3–5 am) at the end of a long period of wake (>1 hr, interrupted by periods of sleep of <5 min), and after spending at least 70 % of the previous 6–7 hr awake. Sleeping mice (S) were collected during the light period (~3–5 pm), at the end of a long period of sleep (>45 min, interrupted by periods of wake of <4 min), and after spending at least 75 % of the previous 6–7 hr asleep. Sleep-deprived mice (SD) were spontaneously awake during most of the dark phase and then kept awake during the first 4 hr of the light period by exposure to novel objects and by tapping on the cage when slow waves appeared on the EEG.
We checked whether TRAP accurately enriched for astrocytic-specific genes using two approaches. First, we randomly selected six astrocytic immunoprecipitated samples (IP) and six unbound samples (UB) from the S, W and SD groups (two mice per group) and performed quantitative polymerase chain reaction (qPCR) for the genes coding for the astrocytic marker GFAP, the neuronal marker SYT1 and the oligodendrocyte marker MBP. In all mice, GFAP expression was markedly higher in the IP sample relative to the UB sample, while the opposite was true for MBP and SYT1 (Fig. 3a). We then used the TRAP microarray data to calculate the IP/UB ratio for genes previously identified as enriched in astrocytes, neurons, mature oligodendrocytes (MOs), oligodendrocyte precursor cells (OPCs), microglia, endothelial cells or pericytes [25]. Astrocytic-specific genes were highly expressed in all IP samples, whereas the large majority of genes specific for neurons, MOs, OPCs, microglia, endothelial cells or pericytes were enriched in all UB samples (Fig. 3b). Thus, we can confirm that TRAP consistently enriched for astrocytic-specific genes in all experimental groups. In addition, in the IP samples we measured the expression of several genes involved in pathways known to be important for astrocytic functions [13, 19, 25, 26]. Expression levels of several of these genes were affected by behavioral state (Fig. 3c). This was the case for Gjb6, Ppp1r3c, and Adcy10, which were upregulated in both W and SD relative to S. Other genes, such as Slc1a3 and P2ry2, increased their expression only in SD, while P2ry4 showed higher expression in S than SD. To test whether changes observed at the transcription level were followed by changes at the protein level, we considered the list of all astrocytic-specific genes that were differentially expressed across behavioral states, and whose corresponding protein was reported to be exclusively expressed in astrocytes. From this list, we selected Gjb6 and Slc1a3, which code, respectively, for the gap-junction subunit connexin30 (Cx-30) and for the glutamate transporter GLAST, two well-characterized astrocytic proteins. Both genes showed increased mRNA levels in SD relative to S (>30 %, P < 0.01). Western blotting experiments using cerebral cortex homogenates of S (n = 4) and SD (n = 4) mice found that the expression of both Cx-30 and GLAST was also increased in SD relative to S (P = 0.02 for both proteins, Fig. 3c, right panel).
Astrocytic genes are affected by sleep and wake independent of time of day
We compared astrocytic IP samples (18 in total) from S, W and SD mice (n = 6 per group) to identify the specific effects of sleep and wake on astrocytic gene expression. To be included in the analysis, probesets needed to be called “present” in at least one sample, which was the case for ~74 % of all probesets (33,255 out of 45,101). All these 33,255 probesets were analyzed, whether or not they were enriched in IP samples relative to UB samples. This is because our primary goal was to identify sleep/wake-dependent genes expressed in astrocytes, independent of whether they were overrepresented in this cell type relative to neurons or oligodendrocytes.
First, we compared IP samples collected at 3 am (W mice) with IP samples collected at 3 pm (S + SD) and found that 152 probesets, representing 139 unique genes, were differentially expressed (P < 0.01, W vs S + SD). Thus, the expression levels of ~0.5 % of astrocytic genes changed due to time of day, independently of behavioral state (Additional file 1: Table S1 and Additional file 2: Table S2). Next, we compared IP samples collected after sleep with IP samples collected after spontaneous wake and sleep deprivation, and found that 498 probesets, representing 451 unique genes (~1.4 %), changed their expression because of behavioral state independently of time of day (P < 0.01, S vs W + SD, Additional file 3: Table S3 and Additional file 4: Table S4). As in previous work [27], we considered “sleep” genes to be only those that were upregulated in the sleep group (6–7 hr of sleep) relative to both the spontaneous wake group (6–7 hr at night), and the forced wake group (4 hr of sleep deprivation after spending most of the night awake), whereas “wake” genes were those with higher expression in both spontaneous wake and sleep deprivation groups than in the sleep group. Since spontaneous wake occurred at night and sleep deprivation during the day, sleep and wake genes as defined here reflect changes due to behavioral state rather than circadian factors or exposure to light.
The vast majority of genes (396, representing 1.3 % of all analyzed probesets) that changed their expression because of behavioral state were wake genes, while only 55 genes (0.1 % of all analyzed probesets) were sleep genes. A statistical approach using gene annotation enrichment analysis (DAVID) was carried out, together with an extensive analysis of the literature, to elucidate the biological processes, molecular functions and cellular components associated with sleep and wake in astrocytes (Fig. 4, Additional file 5: Table S5 and Additional file 6: Table S6). Despite the short list of sleep genes, we found enrichment of genes involved in cell development and proliferation (Pax3, Pax7, Rab38, Tsnaxip1, Spata4, Sphk2 and Pik3ca) and biosynthesis (Fbp1 and Pigs). Other sleep genes included Slc16a1, which codes for the MCT1 transporter involved in the astrocyte–neuron lactate shuttle [28, 29] and whose activity in the hippocampus is essential for long-term memory formation [30]. By contrast, the functional categories enriched during wake included cell metabolism (Adcy10, Foxf2, Cox8a, Cox19 and Ppp1r3c), nucleotide binding (Gem, Pak3, Rhoh, Trio, Kif15 and Ttl), anatomical structure development (Klf4, Cdh6, Ptp4a1, Uhrf1, Trp53, Eif2b5 and Hoxc6), endocytosis (Mrc2, Ehd2, Hip1 and Mertk), and ion binding (Kcnmb2, Kcnh5, Kcnj1 and Cacna1a). Overall, a large category of astrocytic genes modulated by behavioral state was related to the extracellular matrix and/or the cytoskeleton, including sleep genes Arap3 (involved in actin cytoskeleton remodeling), collagen type IV alpha 4 (Col4a4), Ceacam1, the actinin-associated LIM protein Pdlim3, and actin gamma (Actg1). Wake genes included Trio, Gem and Synj2, all of which have been involved in cytoskeleton modifications and the elongation of astrocytic processes, as well as genes coding for integrins (Itgal, Itgb6 and Itgb1) and other extracellular matrix proteins (Efemp2, Emilin3 and Emid2), the mannose receptor C2 that binds and internalizes collagen (Mrc2), the proteoglycan syndecan 4 (Sdc4), the endoglycosidase heparanase (Hpse), and others (Mitd1, Ptp4a1, Ehd2, Nrap and Smagp).
As discussed before, we applied a conservative criterion to define wake genes, considering only those upregulated in both spontaneous wake and forced wake. When the two wake conditions were compared to each other, we found 584 differentially expressed unique genes, including 200 upregulated in W and 384 upregulated in SD (Additional file 7: Table S7). The number of differentially expressed genes was two- to threefold higher when S was compared to either W or SD: 1014 genes in S vs W (204 upregulated in S, 810 in W) and 1439 in S vs SD (208 upregulated in S, 1231 in SD). This suggests that W and SD are more similar to each other than either wake condition to sleep. The biological processes most enriched in W and SD were cell adhesion and metabolism, respectively (Additional file 8: Figure S1), but the interpretation of these differences is not straightforward, because W and SD differ by many factors, including “quality” of wake (spontaneous vs forced), duration of wake (SD > W by several hours), presence or absence of light, and time of day (3 am vs 3 pm). Of note, most genes differentially expressed between W and SD also differed between S and SD (Additional file 7: Table S7), indicating that their expression reflects the duration of wake and/or the specific presence of forced wake. Only a minority (~30 %) of genes instead differed between W and SD but not in other comparisons (S vs SD, S vs W-SD, 3 am vs 3 pm), including 77 genes known to be upregulated in SD and 41 upregulated in W. These could be “light” genes, responding to the presence or absence of light, respectively (Additional file 7: Table S7).
Ultrastructural studies
Astrocytic coverage of cortical dendritic spines increases after extended wake
Astrocytes are plastic cells. They are capable of expanding their fine peripheral astrocytic processes (PAPs) in close proximity to synapses in response to increased synaptic activity [4, 31, 32]. In vitro and in vivo experiments have shown that these plastic events can take place in a few hours and have functional consequences for synapses [4, 33]. Our microarray results suggested that astrocytic pathways involved in cytoskeleton modification and elongation of PAPs were modulated by sleep and wake. To test this hypothesis directly, we used SBF-SEM to study PAP dynamics in relation to changes in behavioral state. Specifically, we focused on the interaction between PAPs and dendritic spines in layer II of the prefrontal cortex, because supragranular layers are highly plastic [17] and the SWA of NREM sleep, which reflects the need for sleep and its depth, is largest in frontal areas [18]. The great majority (>95 %) of cortical spines are associated with excitatory synapses [34, 35]. We focused on dendritic branches of similar size (diameter 0.5–1.3 μm, length 9–29.6 μm) and within each segmented portion of the dendrite all spines were analyzed (filopodia were not included in the analysis). The dataset included 317 spines from a sleep group (S, 6–8 hr of sleep, three mice, ~100 spines per mouse), 368 spines from an acute sleep deprivation group (SD, 6–8 hr of sleep deprivation, three mice, ~150 spines per mouse), and 310 spines from a chronic sleep restriction group (CSR, 4 days of chronic sleep restriction with ~70 % of sleep loss, three mice, ~100 spines per mouse). For each spine, the spine head, the axon–spine interface (ASI), i.e. the interfacing surface between the axonal bouton and the spine head, and the PAPs were manually segmented and reconstructed in 3D (Fig. 5a–d; Additional file 9: Table S8).
First, we classified spines depending on whether or not they were reached by PAPs. PAP positive (PAP+) spines were defined as those contacted by PAPs in any location of the spine head, independent of the presence of PAPs contacting the spine neck and/or the axonal bouton. Across all animals, ~81 % of spines were PAP+ and ~19 % were PAP negative (PAP−) (Fig. 5e), consistent with findings in layer IV of the mouse barrel cortex, where only ~10 % of spines lack astrocytic contact [31]. Compared to PAP+ spines, PAP− spines had smaller mean head volume (0.053 ± 0.05 μm3 vs 0.11 ± 0.12 μm3, Kolmogorov–Smirnov, P < 0.0001) and smaller mean ASI (0.11 ± 0.13 μm2 vs 0.24 ± 0.27 μm2, Kolmogorov–Smirnov, P < 0.0001; data not shown). Between-group analysis showed that the number of PAP+ and PAP– spines did not differ across groups (PAP+, P = 0.96; PAP−, P = 0.96; Fig. 5e).
Next, for PAP+ spines, we calculated the astrocytic coverage, i.e. the extent of the astrocytic surface apposed to the spine head. In all three groups, we found a positive correlation between spine head volume and apposed astrocytic surface (Fig. 5f). Thus, not only larger spines are more likely to be contacted by astrocytes, but the extent of this contact also increases with spine size. Crucially, we also found that the apposed astrocytic surface differed across groups (Kruskal–Wallis, P = 0.0009), being higher in CSR relative to both S (P < 0.0001) and SD (P < 0.0001; Fig. 5g). We also calculated the ratio between apposed astrocytic surface and the whole spine head surface, and found that this parameter differed across the three groups (Kruskal–Wallis, P < 0.0001), again with CSR having higher values than both S (P = 0.0005) and SD (P = 0.003; Fig. 5h).
We further assessed whether the increased astrocytic coverage observed in CSR was limited to spines with a specific size by subdividing spine heads into three subgroups: small (head volume between 0th and 50th percentiles), medium (50th to 75th percentiles) and large spines (>75th percentile). The increase in apposed astrocytic surface in CSR was significant in medium (SD vs CSR, P = 0.03; S vs CSR, P = 0.027) and large spines (SD vs CSR, P = 0.0003; S vs CSR, P = 0.0002; Fig. 5i). Finally, previous work has shown that large mature spines usually contain the spine apparatus, an organelle composed of stacked smooth endoplasmic reticulum [36]. We tested whether the increased coverage observed in CSR mice applied specifically to spines with the spine apparatus, and found that this was the case (SD vs CSR, P = 0.03; S vs CSR, P = 0.0003; Fig. 5j, k).
Wake brings astrocytic processes closer to the synaptic cleft
Astrocytes can contact spine heads in two different ways, depending on whether or not the contact includes the area around the synaptic cleft (ASI, Fig. 6a–d). Thus, we subdivided PAP+ spines in two subgroups, PAP+ASI– and PAP+ASI+, which accounted for ~25 % and ~75 % of all PAP+ spines, respectively, and tested whether their number was affected by sleep and wake. We found that this was the case (PAP+ASI–, Kruskal–Wallis, P = 0.0058; PAP+ASI+, Kruskal–Wallis, P = 0.012). Specifically, the proportion of PAP+ASI+ spines was higher in SD (P = 0.02) and CSR (P = 0.0033) than in S, while the opposite was true for PAP+ASI– (Fig. 6e). Of note, in all groups large spines were approached by PAPs at the ASI level more frequently than small spines (Kolmogorov–Smirnov, P < 0.0001; data not shown). To rule out that the observed PAP changes were unique to conditions of forced wake, such as SD and CSR, we also studied mice sacrificed after at least 6 hr of spontaneous wake (W, three mice, 279 spines). The proportion of PAP+ASI+ spines in W was significantly higher than S (68.23 ± 6.26 % vs 51.09 ± 7.88 %, P = 0.023; Additional file 10: Figure S2), and comparable to SD (MW, P = 0.59) and CSR (MW, P = 0.79), indicating that the movement of PAPs toward the synaptic cleft is not caused by the stress of sleep deprivation.
Next, we measured whether the extent of the astrocytic coverage at the level of the synaptic cleft also increased with wake. We calculated the portion of ASI perimeter that was contacted by PAP (astrocytic perimeter) and the ratio between the astrocytic perimeter and the whole ASI perimeter (Fig. 6d). Both parameters increased in CSR relative to both S (astrocytic perimeter, P < 0.0001; ratio, P = 0.0005) and SD (astrocytic perimeter, P = 0.002; ratio, P = 0.0001; Fig. 6f, g). Furthermore, since in all groups we found a positive correlation between astrocytic perimeter and ASI (S, r = 0.41, P < 0.0001; SD, r = 0.36, P < 0.0001; CSR, r = 0.51, P < 0.0001), and the size of the ASI is positively correlated with the spine head volume (P < 0.0001 in all groups), we subdivided the spines depending on head size or presence of a spine apparatus, as done before. The ratio between astrocytic perimeter and ASI perimeter increased significantly in CSR compared to S in spines of the largest size (MW, P = 0.0027), as well as in spines with a spine apparatus (MW, P = 0.003), although CSR showed a trend to be higher relative to both S and SD also in spines without spine apparatus (not shown). The ratio between astrocytic perimeter and ASI perimeter also increased in CSR relative to SD in small spines (MW, P = 0.002). Overall, the CSR changes at the synaptic interface are consistent with those seen at the level of the entire spine head, with the most significant increases in astrocytic coverage occurring in large spines with the spine apparatus.
Finally, we tested whether the increased extension of the astrocytic coverage in CSR may result from an overall enlargement of the astrocytic profiles in the neuropil. We measured the PAP volume in a portion of neuropil surrounding every spine considered in the previous analysis (Fig. 6h, i). The volume occupied by PAPs did not change across experimental groups, although an increasing trend in CSR was noticeable (S, 8.31 ± 6.55 %; SD, 8.61 ± 8.17 %; CSR, 9.19 ± 6.8 %; S vs CSR, MW, P = 0.1; S vs SD, MW, P = 0.97; SD vs CSR, MW, P = 0.07; not shown). However, the PAP surface-to-volume ratio was significantly higher in CSR than S (P < 0.0001) and SD (P < 0.0001; Fig. 6j). Since the PAP volume is slightly increased in CSR relative to S and SD, an increase of the PAP surface-to-volume ratio can be explained only by an increase of the astrocytic surface. Thus, extended periods of wake have little effect on PAP volume, but increase the PAP surface in the neuropil.