Abstract
Most animal species operate according to a 24-hour period set by the suprachiasmatic nucleus (SCN) of the hypothalamus. Despite the lack of direct synaptic connections to the hippocampus, the rhythmic activity of the SCN modulates memory-dependent processes in the hippocampus. What accounts for these effects has long remained unknown. Here, we show that there are cell-type specific structural and functional changes that take place in the hippocampus and that modulate the strength of excitatory synaptic transmission in area CA1 with circadian rhythmicity. Pyramidal neurons change the surface expression of NMDA receptors. Fine astrocytic processes glide through the neuropil, leading to changes in the time course of glutamate uptake and integration of temporally clustered excitatory synaptic inputs. These findings identify important mechanisms through which neurons and astrocytes modify the synaptic environment and contribute to the local storage of information in the hippocampus and the dynamics of learning and memory processing.
Introduction
The circadian rhythmicity with which mammals sleep, feed, regulate their body temperature and engage in reproductive behaviors is controlled by the main circadian master clock, the SCN of the hypothalamus, and by the coordinated activity of other semi-autonomous ancillary oscillators distributed in the brain and in peripheral tissues (Guilding and Piggins, 2007). Most neuronal projections from the SCN remain confined within the medial hypothalamus, and control the circadian release of corticosteroids into the bloodstream (i.e. gluococorticoids and mineralocorticoids) (Lehman et al., 1987; Meyer-Bernstein et al., 1999; Kalsbeek and Buijs, 2002). A few projections reach the inter-geniculate leaflet and paraventricular nucleus of the thalamus and are functionally important for the rhythmic production of melatonin by the pineal gland (Moore, 1996). Corticosteroids and melatonin have important regulatory effects, which take place through the activation of receptors expressed broadly in the brain and in peripheral organs (Reppert et al., 1996; Dubocovich et al., 2003; Dubocovich, 2007). The rhythmic control of gene and protein expression is fully established during embryonic development (Namihira et al., 1999; Reick et al., 2001; Wakamatsu et al., 2001; Sumova et al., 2012). The endogenous rhythmic activity of the SCN and of the pineal gland can be powerfully entrained by environmental cues, or zeitgebers, the most potent of which is light. By activating direct retino-hypothalamic glutamatergic connections, light evokes a fast-onset, sustained increase in the firing rate of SCN neurons, ultimately leading to reduced melatonin production and, in nocturnal animals, reduced glucocorticoid release (Moore and Lenn, 1972; Stephan and Zucker, 1972; Meijer et al., 1998; Ishida et al., 2005). Therefore, in mice, the production of glucocorticoids, mineralocorticoids and melatonin all peak during the dark phase of the circadian cycle (ZT12-24).
Interestingly, synaptically excitability in the hippocampus fluctuates during circadian rhythms (Barnes et al., 1977) and more than 10% of hippocampal genes and proteins show circadian fluctuations (Debski et al., 2017). Learning and memory formation, two functions that are conserved across species and that in mammals rely on the activity of the hippocampal formation, are also modulated by circadian rhythms (Ruby et al., 2008; Gerstner et al., 2009; Smarr et al., 2014; Shimizu et al., 2016; Snider et al., 2016; Rawashdeh et al., 2018). For example, behavioral studies indicate that mice trained to context fear conditioning protocols acquire the conditioning more slowly if trained during the dark (D) phase of the circadian cycle (ZT12-24) (Chaudhury and Colwell, 2002), a time when long-term plasticity is also reduced (Winson and Abzug, 1977, 1978). Interestingly, this circadian control of hippocampal function and plasticity is retained also in reduced slice preparations, as evidenced by the reduced magnitude of long-term potentiation (LTP) at hippocampal Schaffer collateral synapses in slices prepared during the D-phase (Harris and Teyler, 1983; Raghavan et al., 1999). The persistence of these effects in slices suggests that they are independent of the ongoing activity of extra-hippocampal inputs and might be due to the presence of one or more diffusible factors that are not easily washed out of slices, but whose identity has not been revealed. On one hand, increased glucocorticoid levels during the active phase of the circadian cycle are thought to facilitate memory formation and cognition (Hui et al., 2004; Yuen et al., 2011; Whitehead et al., 2013). On the other hand, melatonin has been proposed to be both sufficient and necessary for poor memory formation and plasticity during the D-phase (Ozcan et al., 2006; Rawashdeh et al., 2007).
The propensity of excitatory synapses to express LTP depends on their initial strength, which in turn is influenced by the density, composition and time course of activation of glutamate receptors (Migliore et al., 2015). Glutamate clearance relies on passive diffusion and active transport, mostly mediated by astrocytes’ glutamate transporters GLAST and GLT-1. Since astrocytes are competent circadian oscillators (Prolo et al., 2005; Brancaccio et al., 2017; Brancaccio et al., 2019), and impaired glutamate uptake degrades LTP (Katagiri et al., 2001; Scimemi et al., 2009), we hypothesized that these cells, in addition to neurons, might also contribute to circadian changes in hippocampal plasticity. Our data show that NMDA receptor activation is reduced during the D-phase, a time when astrocytes retract their fine processes from excitatory synapses and glutamate clearance is prolonged. These changes lead to decreased temporal summation of AMPA EPSPs and reduced LTP during the D-phase. Together, our data provide a new mechanistic understanding of the molecular and cellular mechanisms modulating daily changes in hippocampal plasticity.
Results
The hippocampus is a competent circadian oscillator
Central and peripheral circadian clocks are genetically equipped to generate rhythms during embryonic development, but continue their maturation during the first postnatal week. At P10, when synaptogenesis in the SCN and other peripheral oscillators is complete, rats show regular changes in their body temperature that are controlled by the rhythmic activity of circadian oscillators (Weinert, 2005; Sumova et al., 2008). We confirmed that this holds true also in juvenile mice aged P14–21, from which we measured the body temperature using micro-transponders implanted under the skin of the shoulder (Fig. 1A-B). Their mean body temperature (∼35.8°C) displayed consistent oscillations of ±0.6°C, with detectable peaks at the end of the light (L)-phase (ZT12) and troughs at the end of the dark (D)-phase (ZT0). We confirmed that the expression of clock genes like Arntl and Per2 in the mouse hippocampus is not limited to the SCN. Accordingly, using an RNAscope fluorescent in situ hybridization assay (Fig. 1C-E), we showed that Per2 expression increased significantly in acute slices prepared at ZT15.5 (D-phase) compared to those prepared at ZT3.5 (L-phase) in the SCN (SCNL: 2.9e-3±0.6e-3 (n=28), SCND: 14.8e-3±1.8e-3 (n=33), ***p=3.5e-7), the pyramidal cell layer of hippocampal area CA1 (CA1-PCL: 30.1e-3±4.6e-3 (n=31), CA1-PCD: 44.1e-3±3.9e-3 (n=36), *p=0.023) and stratum radiatum (CA1-SRL: 1.1±0.2 (n=33), CA1-SRD: 10.9±1.1 (n=32) ***p=2.4e-10). By contrast, the expression of Arntl did not change significantly between these two time points in any of the brain regions mentioned above (SCNL: 4.2e-3±0.5e-3 puncta/µm2 (n=28), SCND: 4.9e-3±1.1e-3 puncta/µm2 (n=33), p=0.54; CA1-PCL: 14.6e-3±1.6e-3 puncta/µm2 (n=31), CA1-PCD: 11.0e-3±1.9e-3 puncta/µm2 (n=36), p=0.16; CA1-SRL: 2.1e-3±0.5e-3 puncta/µm2 (n=33), CA1-SRD: 1.7±0.5 puncta/µm2 (n=32), p=0.56). These findings are important because they support the notion that hippocampus, not just the SCN, is a competent circadian oscillator.
Circadian rhythmicity shapes long-term plasticity at Schaffer collateral synapses
Consistent with previous work, we also detected changes in the magnitude of theta burst stimulation-induced (TBS)-LTP at Schaffer collateral synapses, using a stimulation protocol that aims to mimic the endogenous electroencephalographic activity measured in the rodent hippocampus during exploratory learning tasks (Buzsaki, 2002). The TBS protocol induced LTP in slices prepared in the L-phase but not in those prepared in the D-phase (LTPL: 1.17±0.03 ms-1 (n=14), ***p=3.9e-4; LTPD: 0.97±0.06 ms-1 (n=7), p=0.64; Fig. 1F-G). This was not a photic effect, but rather a circadian one, because LTP could still be induced in slices prepared at ZT3.5 from mice kept under constant darkness conditions (LTPDD: 1.14±0.03 ms-1 (n=14), **p=1.1e-3). The effect recorded from slices prepared in the D-phase was significantly different than the one observed in slices prepared in the L- and subjective L-phases (LTPD vs LTPL: *p=0.014; LTPD vs LTPDD: *p=0.029; Fig. 1G).
If the detected changes in LTP were due to circadian differences in the activation of corticosteroid and melatonin receptors, blocking them pharmacologically should not affect plasticity during the L-phase but should rescue it during the D-phase. The two main types of corticosteroids produced by the adrenal gland are glucocorticoids and mineralocorticoids. Consistent with our hypothesis, blocking NR3C1 mineralocorticoid receptors with spironolactone (10 µM; LTPL: 1.19±0.03 ms-1 (n=7), **p=1.4e-3) and blocking NR3C2 glucocorticoid receptors with mifepristone (1 µM; LTPL: 1.17±0.02 ms-1 (n=8), ***p=1.9e-4) did not occlude LTP in slices prepared during the L-phase (Fig. 2A,B). C57BL/6 mice carry two mutations in biosynthetic enzymes for melatonin (Ebihara et al., 1986; Kasahara et al., 2010; Peirson et al., 2018) and MT1/2 melatonin receptor gene expression is low in the hippocampus (Saunders et al., 2018). Consistent with these findings, blocking MT1/2 melatonin receptors with luzindole (50 µM) in slices prepared during the L-phase did not occlude LTP (LTPL: 1.16±0.04 ms-1 (n=11), **p=1.4e-3; Fig. 2C). No LTP was detected in the D-phase in the presence of luzindole (LTPD: 1.05±0.04 ms-1 (n=7), p=0.22; Fig. 2F). By contrast, blocking NR3C1 (LTPD: 1.13±0.02 ms-1 (n=8), ***p=5.9e-4) or NR3C2 receptors (1.23±0.04 ms-1 (n=8), ***p=9.1e-4) in slices prepared during the D-phase rescued LTP (Fig. 2D,E). These data not only confirm previous works showing that LTP expression is modulated during the circadian cycle, but also identify corticosteroids as key molecules responsible for this effect.
The activation of NMDA receptors on CA1 pyramidal cells varies with circadian rhythmicity
We performed experiments to determine whether the effects described above could be accounted by changes in pre- or post-synaptic function (Fig. 3). We alternated single and paired electrical stimuli (10-20 V, 50 µs) to evoke glutamate release from Schaffer collaterals. We recorded AMPA EPSCs using a Cs-based internal solution while keeping CA1 pyramidal cells at a holding potential of −65 mV, in the presence of the GABAA receptor antagonist picrotoxin (100 µM). After recording a stable baseline of AMPA EPSCs, we blocked them by bath applying the AMPA receptor antagonist NBQX (10 µM) and switched the holding potential to +40 mV to record NMDA EPSCs. Our measures of AMPA and NMDA EPSC paired-pulse ratio (PPR) remained similar during the L/D-phases, suggesting that there was no change in pre-synaptic release probability (AMPA PPRL: 2.7±0.1 (n=19), AMPA PPRD: 2.7±0.1 (n=13), p=0.98; NMDA PPRL: 2.3±0.2, NMDA PPRD: 2.6±0.3, p=0.62; Fig. 3A,B). In these experiments, the stimulus strength was set to evoke AMPA EPSCs of similar amplitude during the L/D-phases (AMPA EPSC ampL: 72±7 pA, AMPA EPSC ampD: 93±12 pA, p=0.15). Their rise time (AMPA EPSC riseL: 1.9±0.1 ms, AMPA EPSC riseD: 2.2±0.2 ms, p=0.15) and half decay time (AMPA EPSC t50L: 10.0±0.4 ms, AMPA EPSC t50D: 10.7±0.8 ms, p=0.47) were also similar, suggesting no change in the time course of AMPA receptor activation when deliverin single stimuli to Schaffer collaterals (Fig. 3C,D). The amplitude of the NMDA EPSCs, however, was smaller in the D-phase (NMDA EPSC ampL: 94±15 pA, NMDA EPSC ampD: 47±7 pA, **p=8.2e-3), with no change in their rise and half decay time (NMDA EPSC riseL: 5.6±0.2 ms, NMDA EPSC riseD: 5.8±0.4 ms, p=0.75; NMDA EPSC t50L: 99±7 ms, NMDA EPSC t50D: 101±7 ms, p=0.85; Fig. 3E,F). The reduced NMDA EPSC amplitude led to reduced NMDA/AMPA ratio in the D-phase, which had not been previously reported (NMDA/AMPAL: 2.2±0.3, NMDA/AMPAD: 1.0±0.2, **p=3.7e-3; Fig. 10G,H). We confirmed these results in separate experiments in which RuBi-glutamate (100 µM) was superfused through the recording chamber and uncaged using a 5 ms duration light pulse generated by a SOLA-SE light engine (Lumencor, Beaverton, OR) and delivered to the sample by connecting the light source to the epi-fluorescence port of the microscope using a light guide (2.7 mW at the specimen plane; Fig. 3I-L). We recorded electrically-evoked and flash-evoked AMPA and NMDA EPSCs in the same cells. Once again, only the amplitude of electrically-evoked NMDA EPSCs was smaller during the D-phase (AMPA EPSC ampL: 110±17 pA (n=5), AMPA EPSC ampD: 73±19 pA (n=5), p=0.18; NMDA EPSC ampL: 40±7 pA, NMDA EPSC ampD: 13±5 pA, *p=0.017; Fig. 3I,J) as was that of flash-evoked NMDA EPSCs, not AMPA EPSCs (AMPA Flash-EPSC ampL: 80±16 pA, AMPA Flash-EPSC ampD: 112±25 pA, p=0.31; NMDA Flash-EPSC ampL: 254±42 pA, NMDA Flash-EPSC ampD: 69±28 pA, ***p=7.8e-3; Fig. 3K,L).
These effects were not associated with gross changes in the density of excitatory synaptic contacts onto CA1 pyramidal cells, as indicated by the presence of a similar spine density in dendrites from biocytin-filled neurons in slices prepared during the L/D-phases (L: 0.64±0.03 (n=85), D: 0.65±0.02 (n=105), p=0.90; Fig. 4A-B).
Circadian changes in astrocyte glutamate clearance affect the temporal summation of AMPA EPSCs
The results presented so far point to the existence of regulatory mechanisms that alter NMDA receptor activation and LTP expression at Schaffer collateral synapses onto CA1 pyramidal cells. A circadian reduction in the weight of NMDA currents is not expected to change the summation of AMPA or NMDA EPSCs evoked by temporally clustered stimuli, like those that compose each burst of the LTP-inducing TBS stimulation. Accordingly, the peak NMDA EPSC recorded when delivering trains of five pulses at 100 Hz to CA1-PCs was similar during the L/D-phases (Fig. 5A). Surprisingly, however, the summation of AMPA EPSCs was reduced in the D-phase (Fig. 5A). Using a kinetic model of AMPA and NMDA receptors we showed that these effects could be recapitulated if, in addition to the reduced activation of NMDA receptors in the D-phase, there is also a prolongation in the time course of the evoked extracellular glutamate concentration profile in the extracellular space. This would promote AMPA receptor desensitization, reducing the temporal summation of AMPA EPSCs. The effect on NMDA EPSCs would be marginal, because the desensitization rate of NMDA receptors is significantly lower compared to that of AMPA receptors (Fig. 5B). This hypothesis was supported by a multi-compartmental NEURON model that takes into account the complex morphology of CA1 pyramidal cells in the mouse hippocampus and the random spatial distribution of excitatory inputs through the apical dendritic tree in stratum radiatum (Fig. 5C,D). The model confirmed that the smaller AMPA EPSC summation in the D-phase can be accounted for by a 3-fold increase in the recovery rate of the peak conductance of these receptors (from 10 to 30 ms), consistent with an increased desensitization of AMPA receptors during the D-phase. The small reduction in NMDA EPSC summation during the D-phase was accounted for by a 15% reduction in the peak conductance of NMDA receptors (Fig. 5C,D). This is consistent with experimental data showing reduced NMDA receptor activation during the D-phase (Fig. 3). Together, these findings identify two important phenomena contributing to circadian changes in excitatory synaptic transmission in hippocampal area CA1.
A prolongation in the lifetime of synaptically-released glutamate in the extracellular space can lead to changes in the time course of synaptically-activated currents (STCs), which can be recorded from astrocytes using experimental and analytical tools that we and others have extensively characterized (Diamond, 2005; Scimemi et al., 2009; Scimemi and Diamond, 2013; Sweeney et al., 2017). Stimulation of excitatory Schaffer collaterals evokes glutamate release in stratum radiatum, and causes the onset of a complex current waveform in astrocytes (Fig. 6A). Because of the low resistance of the astrocytes’ cell membrane, the first part of this current is a transient outward current corresponding to the extracellular fiber volley, a readout of action potentials travelling along Schaffer collaterals. This fiber volley is not followed by a field current generated by glutamate receptor activation, because all recordings were performed in the presence of AMPA and GABAA receptor blockers (NBQX 10 µM, picrotoxin 100 µM). What follows the fiber volley is an inward current representing the overlay of a sustained potassium current, through which astrocytes remove potassium ions accumulated in the extracellular space from action potential firing in neurons, and a glutamate transporter-mediated current (i.e. the STC) due to the coordinated movement of glutamate with Na+, K+ and H+ across the membrane (Zerangue and Kavanaugh, 1996). As expected, small changes in the stimulation strength across cells led to proportionate changes in the amplitude of the fiber volley and of the potassium current (Fig. 6B). The range of stimulus intensity used in our experiments evoked STCs of similar amplitude and rise time, across the L/D-phases (STC ampL: 25.0±3.1 pA (n=7), STC ampD: 23.8±3.1 pA (n=7), p=0.79; STC riseL: 1.5±0.2 ms (n=7), STC riseD: 1.7±0.2 ms (n=7), p=0.34; Fig. 6C). However, STCs decayed more slowly in the D-phase, as evidenced by measuring the half decay time (t50L: 8.0±0.3 ms (n=7), t50D: 11.9±0.8 (n=7), **p=2.2e-3) and the centroid (<t>L: 8.5±0.4 ms (n=7), <t>D: 14.3±1.3 ms (n=7), **p=3.6e-3; Fig. 6C). The kinetics of the STC are similar to those of the facilitated portion of the STC (fSTC), obtained by subtracting STCs evoked by single and paired stimulation (100 ms inter-pulse interval) (Diamond, 2005; Scimemi et al., 2009; Scimemi and Diamond, 2013; Scimemi et al., 2013; Sweeney et al., 2017). A low concentration of the broad spectrum glutamate transporter antagonist TFB-TBOA (1 µM) reduced the fSTC amplitude (norm fSTC ampL: 0.43±0.05 (n=7) ***p=1.9e-5, norm fSTC ampD: 0.49±0.06 (n=7), ***p=9.5e-5, L vs D: p=0.47) and prolonged its kinetics similarly during the L/D-phases (norm fSTC riseL: 1.3±0.1 **p=2.3e-3, norm fSTC riseD: 1.1±0.1, ***p=8.7e-4, L vs D: p=0.066; norm fSTC t50L: 1.2±0.1, **p=7.8e-3, norm fSTC t50D: 1.1±0.1, *p=0.017, L vs D: p=0.070; norm fSTC <t>L: 1.2±0.1, *p=0.02, norm fSTC <t>D: 1.1±0.1, *p=0.029, L vs D: p=0.097; Fig. 6D). The fSTCs are useful because they can be used to derive the time course of glutamate clearance from astrocytes (Diamond, 2005; Scimemi and Diamond, 2013; Sweeney et al., 2017). This analysis indicated that the time of glutamate clearance from astrocytic membranes is slower during the D-phase (<t>clearanceL: 7.4±0.4 ms (n=7), <t>clearanceD: 9.0±0.5 ms (n=7), L vs D: *p=0.035; Fig. 6E,F), suggesting that the glutamate uptake capacity of these cells varies physiologically at different times of the day.
These effects were not confounded by changes in the passive membrane properties of astrocytes across the L/D-phases (Fig. 7). The I/V plots, obtained by measuring the steady-state current response to 100 ms long voltage steps (Fig. 7A), showed a similar profile during the L/D-phases (Fig. 7B). The membrane resistance (RmL: 25.0±1.5 MOhm (n=45), RmD: 27.0±1.1 MOhm (n=40), p=0.28; Fig. 7C-D) and capacitance were also similar during the L/D-phases (CmL: 12.0±1.1 pF (n=44), CmD: 15.3±1.7 pF (39), p=0.11; Fig. 7C-D). These values were significantly lower than the membrane resistance and capacitance of CA1 pyramidal cells (RmL: 255.6±32.2 MOhm (n=13), ***p=1.1e-5, RmD: 201.0±17.8 MOhm (n=10), ***p=4.3e-6; CmL: 56.0±5.5 pF (n=13), ***p=2.8e-6, CmD: 50.9±7.1 pF (n=10), ***p=6.5e-4), which also did not differ significantly during the L/D-phases (RmL vs RmD p=0.15; CmL vs CmD p=0.57; Fig. 7E). The astrocytes’ resting membrane potentials remained similar during the L/D-phases (VrL: −82±4 mV (n=9), VrD: −81±4 (n=6), p=0.88; Fig. 7F), as did their level of dye coupling, measured via confocal imaging of biocytin-filled cells (couplingL: 12±2 (n=11), couplingD: 12±1 (n=7), p=0.98; Fig. 7F).
Slower glutamate clearance in the D-phase was not due to reduced expression of the astrocyte glutamate transporters GLAST and GLT-1, which we measured in protein extracts from mouse hippocampi dissected at ZT3.5 or ZT15.5 using Western blotting (Fig. 8). Glutamate transporters assemble as trimers that can only be partially broken into monomers and dimers during SDS-PAGE. Thus, the total band intensity of GLAST monomers and dimers, normalized to the band intensity of our loading control β-actin, was 1.49±0.13 (n=20) during the L-phase and 1.33±0.17 (n=20) during the D-phase (p=0.47; Fig. 8A,B), representing a non-significant 11±8% reduction of GLAST expression during the D-phase (p=0.30; Fig. 8C). Similarly, the total normalized band intensity of GLT-1 was 0.77±0.08 (n=16) during the L-phase and 0.78±0.06 (n=16) during the D-phase (p=0.90; Fig. 8D-E), leading to a non-significant 3±6% change of expression during the D-phase (p=0.78; Fig. 8F).
The gross structure of the hippocampal neuropil does not change during the circadian cycle
Other potential mechanisms that could account for the slower glutamate clearance during the D-phase are widespread changes in the structure of the neuropil (i.e. cell swelling) or more subtle, local rearrangements of the synaptic environment, which we previously documented (Sweeney et al., 2017). We analyzed the overall cell membrane and cytoplasm composition of hippocampal slices prepared at ZT3.5 and ZT15.5 using a marker-free holo-tomographic approach, which allowed us to distinguish plasma membranes from cytoplasmic compartments based on differences in their refractive-index gradient (Cotte et al., 2013). We confirmed that this technique can be used to resolve widespread changes in the plasma membrane/cytoplasm (M/C) ratio in hippocampal area CA1 (Fig. 9). To do this, we prepared acute hippocampal slices and stored them for one hour in saline solutions of different osmolality that are known to induce cell swelling or shrinkage (Fig. 9A-D). We then fixed the slices, mounted them on slides and collected z-stacks using the 3D Cell Explorer digital holo-tomographic microscope (Nanolive, Ecublens, Switzerland). Plasma membranes and cytoplasmic compartments were detected and digitally stained using the refractive index values and gradients shown in Fig. 9B. The M/C ratio, calculated as the ratio between the number of voxels that labelled plasma membranes (yellow) and cytoplasm (magenta), provides a readout of the surface area-to-volume ratio of cell processes in hippocampal area CA1. As expected, the M/C ratio increased with increasing osmolality and induced cell shrinkage (Fig. 9C,D). There was a tight correlation between changes in the M/C ratio and changes in osmolality in both the pyramidal cell layer (Fig. 9C) and stratum radiatum, where the M/C ratio was larger due to the absence of cell bodies (Fig. 9D). Having confirmed the sensitivity of this imaging technique, we used it to measure the M/C ratio in slices prepared during the L/D-phases. The parameters used to label the plasma membrane and cytoplasm during the L/D-phases were similar among each other and with respect to those used in the osmolality experiments (Fig 9B,F). We used them to measure the M/C ratio in stratum radiatum, the domain of area CA1 that contains most astrocytes and excitatory synapses (Fig. 9E-G). The lack of difference between the measures obtained during the L/D-phases suggests that no major cell swelling/shrinking occurs in stratum radiatum during the L/D-phases. (M/C ratioL: 0.13±0.02 (n=12), M/C ratioD: 0.10±0.02 (n=14), p=0.27).
Though valuable, this analysis cannot identify cell-specific structural changes that might occur only in astrocytes. To test whether changes in the overall structure of astrocytes occur during the L/D-phases, we acquired confocal images of biocytin-filled astrocytes and analyzed their maximal intensity projections before and after protein-retention expansion microscopy (proExM), a technique originally developed by Tillberg et al., 2016. Using proExM, we can isotropically expand biocytin-filled astrocytes and overcome a caveat of confocal microscopy applied to astrocyte imaging: the spatial resolution of confocal microscopy can be larger than the size of astrocytes’ distal processes (resx,y=λ/2·NA=488/2·1.4=174 nm). We subjected hippocampal slices collected at Zt3.5 and ZT15.5 to three, 15 min rounds of expansion (Fig. 10A), which lead to a roughly 3.5-fold increase in slice area (norm areaL: 3.4±0.3 (n=6), norm areaD: 3.5±0.1 (n=6), p=0.84; Fig. 10B-D) and a 2-fold increase in slice perimeter (norm perimeterL: 1.8±0.1 (n=6), norm perimeterD: 1.9±0.1 (n=6), p=0.57; Fig. 10E-G). Before expansion, the biocytin-filled astrocytes imaged by confocal microscopy showed similar surface area during the L/D-phases (areaL: 3,891±478 μm2 (n=15), areaD: 3,961±224 μm2 (n=17), p=0.90; Fig. 11A,C,E,G). We imaged astrocytes processed using proExM using two-photon laser scanning microscopy. Even in this case, we found no difference in the average surface area of astrocytes during the L/D-phases (areaL: 26,400±3,680 μm2 (n=9), areaD: 25,200±3,888 μm2 (n=6), p=0.83; (Fig. 11B,D,F,H)), suggesting that the gross morphology of these cells does not change during the circadian cycle.
Astrocytes reduce the number of fine processes during the dark phase of the circadian cycle
The data outlined above do not rule out the occurrence of more subtle changes in astrocyte morphology, for example in the density of its smallest branches and/or in their proximity to excitatory synapses, which can be resolved using 3D axial STEM tomography (Sousa et al., 2011; Sweeney et al., 2017). We identified astrocytic processes in 5 µm wide, 1-1.5 µm thick tissue samples because of their lack of spines and clear cytoplasm, which distinguishes them from neurons. There were fewer astrocytic processes during the D-phase (Fig. 12A,B). Their volume and surface area did not change during the L/D-phases (volL: 0.019±0.001 µm3 (n=73), volD: 0.023±0.002 µm3 (n=37), p=0.11; areaL: 0.78±0.04 µm2 (n=73), areaD: 0.89±0.05 µm2 (n=37) p=0.080; Fig. 12C,D). Because of this decrease in the number of astrocytic processes, the nearest neighbor distance between each post-synaptic density and their closest astrocyte process increased more than 2-fold from 155±18 nm (n=59) to 342±63 nm (n=28) during the L- and D-phases, respectively (**p=7.6e-3; Fig. 12E). These changes were not associated with changes in the size of the PSD regions (PSD areaL: 0.024±0.002 µm2 (n=59), PSD areaD: 0.025±0.004 µm2 (n=28), p=0.93; Fig. 12F).
A common way to analyze the number of dendritic branches formed by neurons relies on the use of Sholl analysis. This analysis is prone to errors when used on non-processed confocal images of astrocytes, because the multitude of fine processes of these cells confers them a fuzzy look. Recent image analysis approaches allow information about these fine, filamentous processes to be extracted using coherence-enhancing diffusion filtering (Weickert and Scharr, 2002) and an iterative convolution with oriented filters (see Methods). This procedure allowed a detailed visualization of astrocytic processes in confocal images (Fig. 12G,H). The Sholl analysis, applied to the processed confocal images, showed that the number of astrocytic intersections with circles of increasing radii centered on the soma decreased during the D-phase (intersectionsL: 4,108±326 (n=26), intersectionsD: 2,844±272 (n=20), **p=4.7e-3; Fig. 12I,J). An alternative method, not requiring image binarization, relies on the use of a shearlet multiscale framework (Guo and Labate, 2007) and provides an optimal approximation of anisotropic features like astrocytic process in confocal images (Brazhe, 2018). This method uses spatial entropy (H) and complexity (C) measures to describe the orderliness and feature-preferred orientation of an image, respectively (Brazhe, 2018; Gavrilov et al., 2018). In our analysis, we normalized the values of spatial entropy and complexity measured in the area covered by the biocytin-filled neuron by the values obtained in the adjacent neuropil. Consistent with the Sholl analysis, the normalized entropy was smaller during the D-phase (L: 0.935±0.003 (n=29), D: 0.924±0.002 (n=16), *p=0.037; Fig. 12K). These results support the 3D axial STEM data and indicate that astrocytes remodel their processes continuously during the circadian cycle.
Discussion
The results presented in this manuscript provide direct evidence for the existence of a circadian modulation of excitatory synaptic transmission within the hippocampus (Fig. 1), due to structural and functional remodeling of the synaptic environment. Neurons and astrocytes both contribute to this phenomenon, though in different ways. Neurons alter the size of the functional pool of glutamate NMDA receptors at the synapse (Fig. 3). Astrocytes change the number and proximity of their fine processes with respect to excitatory synapses (Fig. 12). The remodeling of astrocytes alters the time course of glutamate clearance from the extracellular space (Fig. 6). Although this does not have notable consequences on glutamate receptor activation during sparse synaptic stimulation, it profoundly alters the temporal summation of recurring stimuli and long-term plasticity (Fig. 1,5).
Astrocytes, neurons and the values of an ever-changing brain
Modifications of the synaptic environment are critical for information coding in the brain, and provide an important substrate for circuit plasticity. The shape, size and turnover of dendritic spines can be controlled in an activity-dependent manner over time scales that range from seconds to hours, during development (Rakic et al., 1986; Holtmaat et al., 2005; Richards et al., 2005; Kwon and Sabatini, 2011), in response to LTP-inducing protocols (Engert and Bonhoeffer, 1999) and in the adult brain (Gilbert, 1998). Analogous activity-dependent phenomena of structural plasticity also occur in small protoplasmic astrocytic processes that are commonly found around synapses (Reichenbach et al., 2010; Lavialle et al., 2011). Accordingly, different groups have reported that these fine processes can extend, retract, bud and glide through the neuropil (Hirrlinger et al., 2004; Haber et al., 2006). The distance over which these movements take place are comparable to the distance that spines can grow or shrink (i.e. hundreds of nm) and are therefore physiologically relevant. One of the reasons these movements are functionally important is that small changes in the relative proximity of astrocytic processes and spines have important consequences on the lifetime of glutamate in the extracellular space, its uptake, and ultimately for the regulation of the strength and timing of information transfer across neurons. Accordingly, suppressing astrocyte motility has been shown to impair the stabilization and maturation of synapses (Nishida and Okabe, 2007).
At excitatory glutamatergic synapses, multiple molecular mechanisms can lead to changes in the proximity of astrocytic processes to spines, which in turn contributes to modify the time course of glutamate receptor activation. For example, we previously showed that activating PAR1 G-protein coupled receptors leads to proliferation and closer apposition of astrocytic processes to excitatory synapses, leading to reduced AMPA and NMDA receptor activation (Sweeney et al., 2017). Since PAR1 receptors are typically activated by serine proteases in the bloodstream, we concluded that this movement of astrocytes towards synapses might be indicative of similar phenomena of structural plasticity of the synaptic environment occurring during small hemorrhagic stroke. The experimental data described here indicate that astrocyte remodeling is not a rare phenomenon that only happens in experimental conditions mimicking pathological states, but can also occur daily under physiological conditions in the hippocampus. These recurring, circadian changes in the relative proximity of astrocytes and spines occur in concert with changes in glutamate receptor activation in pyramidal neurons, which might be exacerbated by concurrent changes in the release of the NMDA receptor co-agonist D-Serine from astrocytes (Papouin et al., 2017). Together, these phenomena allow the brain to regularly sculpt the fundamental units of its neuronal circuits and provide a new mechanistic understanding of the molecular events that control the circadian modulation of synaptic plasticity and memory formation in the hippocampus.
We can only speculate on why the brain invests its energies in constantly turning up and down the strength of its synaptic connections. One possibility is that this may be a necessary step in a long-term investment plan, whereby tuning down synaptic efficacy during the D-phase may bring the hippocampus into a low energy-demanding state, to parse energy consumption over longer periods of time (Attwell and Laughlin, 2001). Because of these fluctuations, hippocampal networks may be able to switch between functioning modes that best process different types of information at specific time points during the day. Our data may provide a mechanistic basis for Marr’s theory of the archicortex (Marr, 1971), in which the hippocampus alternates between times during which it stores information about patterns of neural activity representing events as they happen, and times during which this information is transferred to other regions of the brain, including the neocortex (Willshaw et al., 2015). Based on our experiments, for nocturnal animals like mice, the time to store information locally would be the L-phase, whereas the time to relay information to the neocortex would be the active D-phase.
LTP, melatonin and glucocorticoids
The first observation that LTP expression at hippocampal Schaffer collateral synapses varies at different times of the day dates back to 1983, when Harris and Teyler first showed that LTP at Schaffer collateral synapses in juvenile and adult rats was more robust in slices prepared during the L-phase (Harris and Teyler, 1983). Based on the fact that these results were obtained in reduced slice preparations, it was concluded that they did not depend on the activity of extra-hippocampal afferents but rather on the enduring effects of modulators that are lost very slowly from slice preparations, for example hormones. Similar conclusions were later obtained in hamsters (Raghavan et al., 1999), highlighting the need to determine the underlying causes.
One of the first candidate molecules that was thought to mediate these effects was the hormone melatonin, mainly synthesized during the D-phase in the pineal gland from the precursor serotonin (Vanecek, 1998). In acute hippocampal slices, exogenously applied melatonin blocks LTP induction without altering low-frequency synaptic transmission (Collins and Davies, 1997; Wang et al., 2005). These effects are unlikely to be mediated by melatonin binding directly to melatonin receptors, because few receptors are expressed in the hippocampus. Given that melatonin has close structural similarity to various antagonists of the glycine-binding site of NMDA receptors (Huettner, 1989; Salituro et al., 1990; Salituro et al., 1992), it has been suggested that melatonin could reduce LTP by antagonizing NMDA receptors. This hypothesis has been tested and ultimately disputed by Collins and Davies (Collins and Davies, 1997). If melatonin does not have any known membrane receptor target, how does it exert its biological actions? Because of its hydrophobicity, melatonin released from the pineal gland into the bloodstream can easily pass through cell membranes. This allows melatonin to modulate a variety of intracellular molecular pathways and even bind to nuclear receptors (Carlberg and Wiesenberg, 1995; BenitezKing et al., 1996; Rafii-El-Idrissi et al., 1998; Benitez-King et al., 2001). Naturally, some of these interactions could have a deleterious effect on LTP. The problem is that many inbred strains of laboratory mice, like C57BL/6J, have no detectable levels of melatonin in the pineal gland due to two independent mutations in the genes encoding the biosynthetic enzymes for melatonin: N-acetylserotonin and hydroxyindole-o-methyltransferase (Ebihara et al., 1986; Kasahara et al., 2010; Peirson et al., 2018). It is therefore unlikely that our circadian effects on astrocyte morphology and AMPA EPSC summation are mediated by melatonin acting on molecular targets aside from MT1/2 receptors.
Other hormones known to be produced with circadian rhythmicity are adrenal corticosteroids, which act by activating glucocorticoid and mineralocorticoid receptors (Chung et al., 2011). High-affinity (Type I) mineralocorticoid receptors and low-affinity (Type II) glucocorticoid receptors are abundantly expressed in the hippocampus and both affect synaptic plasticity and memory formation (Kim and Yoon, 1998). Because of their different steady-state affinity, Type I receptors are fully occupied at low glucocorticoid plasma levels, whereas Type II receptors become fully occupied only with higher glucocorticoid plasma levels (Pavlides et al., 1995; Chung et al., 2011). Physiological increase in alertness and arousal during the D-phase of the circadian cycle can activate both types of receptors. Although the exclusive activation of Type I receptors promotes LTP, the concomitant activation of Type II receptors suppresses it (Pavlides et al., 1995). Consistent with these findings, our experiments show that the detected loss of LTP in the D-phase is associated with activation of both Type I and Type II receptors. To the best of our knowledge, the molecular and cellular mechanisms leading Type I-II receptor activation to loss of LTP had not been previously determined.
Generalizability and functional implications
Mice have evolved as nocturnal animals because they have fewer predators at night, making this the prime time to search for food. This nocturnal pattern of activity is still detected in mice kept in captivity. Although most animal facilities offer ad libitum access to food, caretakers are generally not around at night and therefore this remains the time of least perceived danger. In both nocturnal and diurnal species, the production of melatonin peaks during the D-phase, due to the fact that light inhibits its production through the activation of retino-hypothalamic afferent projections to the SCN. In contrast, the timing of glucocorticoid production coincides with the active phase, which is inverted between nocturnal and diurnal animals. This is because light exerts different effects on the light-sensitive neurons in the SCN of diurnal and nocturnal species. In diurnal species, light reduces the firing rate of photosensitive neurons. By contrast, in nocturnal species, light increases the firing rate of these cells. Given that activation of the SCN inhibits the HPA axis through the release of vasopressin, light stimulates the release of corticosteroids in diurnal but not in nocturnal animals (Jiao et al., 1999; Smale et al., 2003). Therefore, to a first approximation, the production of corticoisteroids in humans and mice is locked to their activity phase (Dickmeis, 2009). Because in our experiments activation of corticosteroid receptors disrupts plasticity at Schaffer collateral synapses, it is tempting to speculate that synaptic plasticity might also be degraded in the human hippocampus through mechanisms similar to those described here but with opposite periodicity. However, since there are numerous molecular and behavioral differences and similarities between humans and mice, one needs to be cautious before jumping to the conclusion that L-phase of humans is analogous to the D-phase of a mouse. That said, our findings indicate that there are circadian changes in the molecular landscape of the hippocampus that enable this structure of the brain to switch to different functional modes. As a substantial reprogramming of oscillating transcripts occurs in temporal lobe epilepsy (Debski et al., 2017), our findings may have important implications to advance not only our understanding of circadian rhythmicity during physiological conditions but also to understand the pathogenesis and treatment of this and other chronic disease states.
Materials and Methods
Ethics statement
All experimental procedures were performed in accordance with protocols approved by the Institutional Animal Care and Use Committee at the State University of New York (SUNY) Albany and guidelines described in the National Institutes of Health’s Guide for the Care and Use of Laboratory Animals.
Mice
Unless otherwise stated, all C57BL/6 mice were group housed and kept under a 12:12 hr L/D-cycles. The lights were turned on at Zeitgeber time 0 (ZT0), which in our animal facility was 7:00 AM. The fact that the ZT time sets the origin ot the 24 hour period to the beginning of the L-phase allows comparisons among studies independently of the actual clock-time settings of different animal facilities. Food and water were available ad libitum throughout the circadian cycle. Unless otherwise stated, all dissections for the L-phase data were performed at ZT3.5. All dissections for the D-phase data were performed at ZT15.5.
In vivo chronic temperature recordings
To record the mouse body temperature, we implanted programmable micro-transponders (Cat# IPTT300; BDMS) under the skin of the neck of P12 mice. We took temperature measures hourly from P14–21 using a wireless hand-held reader for IPTT micro-transponders (Cat# DAS-8007-IUS, BDMS). Data analysis was performed in IgorPro 6.37 (Wavemetrics, Lake Oswego, OR) using custom-made software (A.S.).
Fluorescent in situ hybridization of clock gene transcripts using RNAscope
We dissected the brain of P14-21 C57BL/6 mice at ZT3.5 or ZT15.5, removed the olfactory bulbs, cerebellum and temporal lobes and fixed it with 4% PFA/PBS overnight at 4°C. The brain was then cryo-protected in 30% sucrose PBS at 4°C for 48 hr and stored in PBS for no more than a week. To prepare slices for RNAscope, we separated the two hemispheres, embedded them in agar and prepared 40 µm thick slices using a vibrating blade microtome (VT1200S, Leica Microsystems, Buffalo Grove, IL). The slices were post-fixed in 4% PFA/PBS for 30 min at room temperature (RT) and mounted onto Superfrost plus microscope slides. Mounted slices were used for fluorescence in situ hybridization (FISH) using an RNAscope multiplex fluorescent assay (Advanced Cell Diagnostics, Newark, CA) according to manufacturer instructions, using Mm-Per2-C1 and Mm-Arntl-C2 RNA probes and Opal 520 and Opal 570 dyes (Akoya Biosciences, Menlo Park, CA). DAPI Fluoromount G was used as the mounting medium (Cat# 0100-02; SouthernBiotech, Birmingham, AL). The presence of Per2 and Arntl transcripts was assessed using a confocal microscope (Zeiss LSM710) equipped with a Plan-Apochromat 63X/1.4NA oil objective. Image size was set to 1024×1024 pixels and represented the average of 8 consecutive frames.
Acute slice preparation and electrophysiology recordings
Acute coronal slices of the mouse hippocampus were obtained from C57BL/6 mice of either sex (P14– 21), deeply anesthetized with isoflurane and decapitated in accordance with SUNY Albany Animal Care and Use Committee guidelines. The brain was rapidly removed and placed in ice-cold slicing solution bubbled with 95% O2/5% CO2 containing the following (in mM): 119 NaCl, 2.5 KCl, 0.5 CaCl2, 1.3 MgSO4·H2O, 4 MgCl2, 26.2 NaHCO3, 1 NaH2PO4, and 22 glucose, 320 mOsm, pH 7.4. The slices (250 µm thick) were prepared using a vibrating blade microtome (VT1200S; Leica Microsystems, Buffalo Grove, IL). Once prepared, the slices were stored in slicing solution in a submersion chamber at 36°C for 30 min and at RT for at least 30 min and up to 5 hr. Unless otherwise stated, the recording solution contained the following (in mM): 119 NaCl, 2.5 KCl, 2.5 CaCl2, 1 MgCl2, 26.2 NaHCO3, and 1 NaH2PO4, 22 glucose, 300 mOsm, pH 7.4. We identified the hippocampus under bright field illumination using an upright fixed-stage microscope (BX51 WI; Olympus Corporation, Center Valley, PA). To record evoked currents, we delivered constant voltage stimuli (50–100 µs) to a bipolar stainless steel electrode (Cat# MX21AES(JD3); Frederick Haer Corporation, Bowdoin, ME) positioned in stratum radiatum, ∼100 μm away from the recorded cell. Whole-cell, voltage-clamp patch-clamp recordings were made using patch pipettes containing (in mM): 120 CsCH3SO3, 10 EGTA, 20 HEPES, 2 MgATP, 0.2 NaGTP, 5 QX-314Br, 290 mOsm, pH 7.2. All recordings were obtained using a Multiclamp 700B amplifier (Molecular Devices, San Jose, CA) and filtered at 10 KHz, converted with an 18-bit 200 kHz A/D board (HEKA Instrument, Holliston, MA), digitized at 10 KHz, and analyzed offline with custom-made software (A.Sc.) written in IgorPro 6.37 (Wavemetrics, Lake Oswego, OR). Patch electrodes (#0010 glass; Harvard Apparatus, Holliston, MA) had tip resistances of 5 and 3 MΩ when patching neurons and astrocytes, respectively. Series resistance was not compensated (∼20 MΩ for neurons, ∼10 MΩ for astrocytes) but was continuously monitored and experiments were discarded if this changed by >20%. Tetrodotoxin (TTX), 2,3-Dioxo-6-nitro-1,2,3,4-tetrahydrobenzo[f]quinoxaline-7-sulfonamide disodium salt (NBQX) and D,L-2-Amino-5-phosphonopentanoic acid (APV) were purchased from Hello Bio (Princeton, NJ; Cat# HB1035, HB0443, HB0251, respectively). (3S)-3-[[3-[[4-(Trifluoromethyl)benzoyl]amino]phenyl]methoxy]-L-aspartic acid (TFB-TBOA) was purchased from Tocris (Minneapolis, MN; Cat# 2532). All other chemicals were purchased from MilliporeSigma (Burlington, MA). All recordings were performed at RT.
Biocytin filling and confocal imaging
Biocytin 0.2–0.4% (w/v) was added to the intracellular solution used to patch astrocytes and neurons. Each cell was patched and filled for at least 20 min. The slices were then fixed overnight at 4°C in 4% PFA/PBS, cryo-protected in 30% sucrose PBS, and incubated in 0.1% streptavidin-Alexa Fluor 488 conjugate and 0.1% Triton X-100 for 3 hr at RT. The slices were then mounted onto microscope slides using Fluoromount-G mounting medium (Cat# 0100-01; SouthernBiotech, Birmingham, AL). Confocal images were acquired using a Zeiss LSM710 inverted microscope equipped with 488 nm Ar laser. All images were acquired as z-stacks using a 40X/1.4 NA oil-immersion objective. To visualize full cells, we stitched together z-stacks (1024×1024 pixels) collected by averaging four frames for each focal plane (1 μm z-step). The image analysis to measure the astrocyte coverage area was performed using Fiji (https://fiji.sc/). Briefly, we generated a maximum intensity projection of each image stack and manually traced the contour of the outer boundaries of the area of the neuropil covered by each astrocyte. Confocal images that contained more than one filled astrocyte were discarded from the analysis, because they introduced inaccuracies in the derived contours. The polar plots were obtained by centering the soma of each astrocyte at the center of a polar plot. The main primary branch of each astrocyte was oriented along the 270° angle. Average coverage traces were obtained using IgorPro (Wavemetrics, Lake Oswego, OR).
Protein-retention expansion microscopy and two-photon imaging
All experiments were performed according to the proExM protocol described by (Tillberg et al., 2016). Briefly, to measure the expansion factor (Fig. 10) we fixed hippocampal slices with 4% PFA/PBS overnight at 4°C, cryoprotected them in 30% sucrose PBS and stored them in PBS. The day before starting the proExM protocol, we incubated them with DAPI Fluoromount G overnight at 4°C (Cat# 0100-20; SouthernBiotech, Birmingham, AL). Slices containing biocytin filled astrocytes were fixed, cryo-protected, stored in PBS and labelled with streptavidin-Alexa Fluor 488 as described in the previous section. In both cases, each slice was then incubated with 200 µl of anchoring solution overnight at RT. The following day, the slices were gelled and digested with Proteinase K overnight at RT. Subsequently, they were expanded using three consecutive incubations with distilled water for 15 min each. Slices stained with DAPI were imaged as they expanded during each incubation period, using a 2X/0.06 NA air objective on an EVOS FL Cell Imaging System equipped with DAPI filter set (λex: 357/44 nm, λem: 447/60 nm; ThermoFisher Scientific, Waltham, MA). Images were collected at 0.1 fps and manually traced to measure slice perimeter and surface area using IMOD (https://bio3d.colorado.edu/imod/). The expanded gels containing biocytin-filled astrocytes were then covered with 2% agarose and submerged in distilled water before being imaged with a custom-made two-photon laser-scanning microscope. The two-photon laser scanning system (Scientifica, Clarksburg, NJ) was powered by a Ti:sapphire pulsed laser (Coherent, Santa Clara, CA) tuned to 760 nm and connected to an upright microscope with a 60X/1.0 NA objective (Olympus Corporation, Center Valley, PA). The green fluorescent signal was separated from the red using a 565 nm dichroic mirror and filtered using FITC filter sets (Olympus Corporation, Center Valley, PA). We averaged eight frames for each optical section (512×512 pixels) and collected z-stacks for a total distance of ∼200 μm (in 1.2 μm steps). The composition of the anchoring, gelling and digestion solution is reported below.
Western blot
Western blot experiments were performed on protein extracts from the hippocampus of mice of either sex aged P14–21, sacrificed at ZT3.5 or ZT15.5. Membrane proteins were extracted using the Mem-PER Plus Membrane Protein Extraction Kit (Cat# 89842; ThermoFisher Scientific, Waltham, MA) according to the manufacturer’s instructions using a mixture of protease and phosphatase inhibitors (10 µl/ml, Cat# 78441; ThermoFisher Scientific, Waltham, MA). The total membrane protein concentration was determined by spectrophotometry. Equal amounts of protein (100 µg) from each sample were resolved on 10 or 12% acrylamide gels. The proteins were transferred to PVDF membranes (Cat# P2563; MilliporeSigma, Burlington, MA) using a semidry blotting approach. The membranes were blocked with 5% nonfat milk in TBST pH 7.6, and probed with primary antibodies overnight at 4°C in 5% BSA in TBST, pH 7.6. Secondary antibody incubation was performed for 1–2 hr at RT in 5% nonfat milk in TBST, pH 7.6. Pre-adsorption experiments were performed using the control antigen provided by the primary antibody supplier according to the manufacturer’s instructions (1 µg/µg antibody). Membranes were probed with either rabbit anti-GLAST (1:1,000), anti-GLT-1 (1:1,000) or β-actin antibodies (1:1,000). Biotinylated horse anti-rabbit antibody was used as the secondary antibody for all blots (1:1,000 for GLAST and GLT-1, 1:5,000 for β-actin). We amplified the immuno-labeling reactions using the Vectastain ABC kit (1:2,000 for GLAST and GLT-1; 1:5,000 for β-actin) and the Clarity Western ECL system served as the substrate for the peroxidase enzyme (Cat# 1705060, Bio-Rad, Hercules, CA). For semi-quantitative analysis, protein band intensities were collected as 16-bit images using a digital chemiluminescence imaging system (c300, Azure Biosystems, Dublin, CA) at different exposure times (0.5–200 s). Each image was converted to an 8-bit image for image analysis using Fiji software (https://fiji.sc/). Only images collected at exposure times that did not lead to pixel saturation were included in the analysis. The intensity of each band was calculated as the mean gray value in a region of interest (ROI) surrounding each band in 3 images collected at different exposure times. All GLAST and GLT-1 band intensities were normalized to the band intensity of β-actin in the same lane. Only images collected at exposure times that did not lead to pixel saturation were used to quantify band intensities.
A list of antibodies and reagents, with their RRID, is reported below:
Electron microscopy and axial STEM tomography
Acute hippocampal slices processed for electron microscopy analysis were prepared as described for the electrophysiology experiments, from P17 mice. Slices were microwave-fixed for 13 s in 6% glutaraldehyde, 2% PFA, 2 mM CaCl2 in 0.1 N sodium cacodylate buffer and stored overnight at 4°C. After three washes in 0.1 N cacodylate buffer, we cut samples from the middle part of CA1 stratum radiatum, ∼100 μm away from the pyramidal cell layer. These samples were treated with 1% osmium tetroxide for 1 hr on ice, en bloc mordanted with 0.25% uranyl acetate at 4°C overnight, washed and dehydrated with a graded series of ethanol, and embedded in epoxy resins. Thin sections (70–90 nm) were counterstained with lead citrate and uranyl acetate and examined on a JEOL 1200 EX transmission electron microscope. Images were collected with a CCD digital camera system (XR-100, AMT, Woburn, MA). To visualize the arrangement of pre-synaptic terminals, post-synaptic terminals, and astrocytic processes, thick sections (∼1 μm) were cut from regions of CA1 stratum radiatum and electron tomograms were collected in a 300 kV electron microscope operated in the scanning transmission electron microscopy (STEM) mode, as described previously (Hohmann-Marriott et al., 2009; Sousa et al., 2011). A sample thickness of 1 μm – enabled by axial STEM tomography – provided sufficient sample depth to visualize features of interest in their entirety, such as synapses. In contrast to standard TEM tomography, conventional TEM tomography is limited to a specimen thickness of ∼400 nm and cannot be applied to such thick sections because the transmitted electrons undergo multiple inelastic scattering processes, resulting in images that are blurred by chromatic aberration of the objective lens. Axial STEM tomography is not affected by chromatic aberration because the objective lens that forms the electron probe is in front of the specimen. Dual-axis tilt series of selected sections were recorded using an FEI Tecnai TF30 TEM/STEM operating at 300 kV (1.5° tilt increment, tilt range from 55° to −55°, pixel size = 5.6 nm). Image registration, tomogram generation, tracing, surface area and volume measures were performed using IMOD 4.7 (http://bio3d.colorado.edu/imod/). In the tomograms, we identified astrocytic processes based on their lack of synaptic vesicles and post-synaptic densities and because they did not give rise to pre- or post-synaptic terminals. Orthoslices through the STEM tomograms showed that astrocytic processes contained glycogen granules, intermediate filament bundles and a more electron-lucent cytoplasm with respect to that of neurons. The astrocytic processes were traced for the entire thickness of the reconstructed volume (1–1.5 μm). We reconstructed all astrocytic processes and all PSDs in each block. The reconstructed volumes were converted into object files and imported into the open-source software Blender 2.76 (https://www.blender.org/). Astrocyte-PSD distance was measured in Blender using custom-made analysis software written in Python (Sweeney et al., 2017).
Tomographic holographic 3D microscopy
The measures of the cell membrane to cytoplasm ratio were obtained from acute hippocampal slices, prepared as described in the electrophysiology section of the methods, from animals aged P14–21 at ZT3.5 (L-phase) or ZT15.5 (D-phase). We used slices prepared at ZT3.5 for our osmolality test. Briefly, we prepared different batches of the slice recording solution, with different concentrations of NaCl (95, 107, 119, 131, 143 mM). We measured the osmolality of these solutions using a vapor pressure osmometer (Wescor 5600, Wescor, Logan, UT). Each slice was stored in one of these solutions for 1 hr, then it was fixed overnight with 4% PFA/PBS and cryo-protected in 30% sucrose PBS. The slices were then re-sectioned at 40 µm thickness using a vibrating blade microtome (VT1200S, Leica Microsystems, Buffalo Grove, IL) and mounted on a microscope slide using Fluoromount-G mounting medium (Cat# 0100-01; SouthernBiotech, Birmingham, AL). The slices were then imaged using the 3D Cell Explorer holographic tomographic 3D microscope (Nanolive, Ecublens, Switzerland) equipped with a 60X objective (NA=0.8), a class I laser source at λ=520 nm with a power output of 0.1 mW (sample exposure = 0.2 mW/mm2), and a USB 3.0 CMOS Sony IMX174 sensor providing 1024×1024 pixels at 165 fps (Cotte et al., 2013). The output data were acquired as image stacks of 512×512×96 voxels (0.186×0.186×0.372 μm) with a rate of tomogram acquisition of 0.5 s-1. This label-free non-phototoxic type of microscopy measures the refractive index (RI) in 3D samples, and allows users to visualize it using different levels of tolerance referred to as the refractive index gradient (IG). The acquired images were processed using Software for Tomographic Exploration of living cElls (STEVE) and Fiji (https://fiji.sc/). By using STEVE, we digitally stained the cells’ membrane (RI = 1.327–1.333; IG = 0.015–0.070) and cytoplasm (RI = 1.329–1.330; IG = 0.007–0.021). The RI and IG settings were varied across slices, but remained constant for our measures in the pyramidal cell layer and stratum radiatum of hippocampal area CA1. By using Fiji, we calculated the number of digitally stained pixels in three ROIs (200×100×40 pixels) positioned in the CA1 pyramidal cell layer and stratum radiatum of each slice. The center of the ROIs along the z-axis was positioned half way through the original z-stack acquired using the 3D Cell Explorer.
Computational Modelling
3D z-stacks of biocytin filled CA1 pyramidal neurons were acquired using a Zeiss LSM710 confocal microscope and saved as czi files. The czi files were imported into Fiji (https://fiji.sc/) using the Simple Neurite tracer v3.1.3 plugin (http://imagej.net/Simple_Neurite_Tracer), first developed by (Longair et al., 2011). The radius of the soma was calculated as the radius of an equivalent circle with the same area of the soma obtained from the maximum intensity projection of the confocal z-stacks. When using this plugin, we first picked the location of the center of the soma and then manually traced the 3D trajectory of different neuronal processes. Each process was tagged as axon, basal or dendritic branch so that they could be assigned different conductances. The tracings were converted into .swc and .asc files to be imported in the NEURON simulation environment v7.6 (Hines and Carnevale, 1997). All model and simulation files will be uploaded to the ModelDB database (https://senselab.med.yale.edu/modeldb/, acc. n. 257027). Also, the morphology will be uploaded to the neuromorpho.org database (www.neuromorpho.org), and an interactive “live paper” page will be setup in the Brain Simulation Platform of the Human Brain Project (https://www.humanbrainproject.eu/en/brain-simulation/live-papers/). To implement a CA1 pyramidal neuron model, we used the 3D accurate morphology reconstructed within this work (Fig.5D) equipped with the typical somatodendritic distribution of channels observed in rodents hippocampal CA1 pyramidal neurons (Migliore and Shepherd, 2002). The channel kinetics were identical to those used in (Migliore et al., 2018) (ModelDB acc.n. 244688). The AMPA synapse kinetic model was from (Tsodyks et al., 1998) (ModelDB acc.n. 3815), with tau_1= 4.5 ms, a peak conductance of 0.2 nS, and a tau_rec of 10 ms or 30 ms to model light and dark condition, respectively. The NMDA kinetics was taken from (Gasparini et al., 2004) (ModelDB acc.n. 44050) with the reverse (unbinding) rate β=0.005 ms-1 and a peak conductance of 0.7 nS or 0.55 nS to reproduce light and dark conditions, respectively. To mimic the train of synaptic activations caused by an extracellular electrical pulses we added 20 synapses randomly distributed in the apical dendrites between 50 and 300 μm from the soma, synchronously activated every 10 ms.
Data analysis
Electrophysiological recordings were analyzed within the Igor Pro environment using custom made software (A.S.). Synaptically activated transporter currents (STCs) were isolated as described previously (Diamond, 2005; Scimemi et al., 2009; Sweeney et al., 2017). Briefly, single and pairs of stimuli, 100 ms apart, were delivered every 10 s. We averaged 10 single and 10 STC pairs and subtracted single from paired STCs to isolate the current evoked by the second pulse. The single STC was then shifted in time by 100 ms (the inter-pulse interval), so that its onset matched the onset of the second STC. The current obtained by subtracting the time-shifted single STC from the paired STCs represented the facilitated portion of the STC (fSTC). This analytical strategy allowed us to get rid of the sustained potassium current that is superimposed on the STC but does not facilitate with repeated stimulation. In the unlikely event that a component of the sustained potassium current was still present after this subtraction, we removed it by subtracting a mono-exponential function with an onset time of 4 ms and an amplitude that was scaled to match the amplitude of the remaining potassium current (Scimemi et al., 2009). We used the fSTCs in control conditions and in the presence of a sub-saturating concentration of TFB-TBOA (1 μM) to calculate the time course of glutamate clearance by performing a deconvolution analysis of the STCs (Diamond, 2005; Scimemi et al., 2009; Sweeney et al., 2017). First, the STCs recorded under control conditions and in TFB-TBOA (1 µM) were binomially smoothed and fitted with the following equation (Nielsen et al., 2004): Second, we approximated glutamate clearance in TFB-TBOA (1 µM) with an instantaneously rising function decaying mono-exponentially, with the same time course of the STC. Third, we deconvolved this approximated glutamate clearance from the STC recorded in low TFB-TBOA to obtain the filter. Fourth, we deconvolved the filter from the STC recorded in control conditions to obtain the glutamate clearance waveform in control conditions.
Miniature events were detected using an optimally scaled template (Clements and Bekkers, 1997) adapted for Igor Pro (A.S.). Paired-pulse ratio (PPR) of EPSCs was calculated by subtracting single from paired EPSCs averaged across 20 traces, after first peak normalization. When delivering trains of stimuli, we interleaved single pulses with trains of four and five pulses. The 5th EPSC was isolated by subtracting the average response to four pulses from the average response to five pulses.
Image preprocessing for Sholl analysis
The Sholl analysis was performed on adaptively thresholded maximum intensity projections of filtered confocal z--stacks. Each plane of the z-stack was first filtered with a coherence-enhancing diffusion filter (Weickert and Scharr, 2002), and then we filtered the maximal projection intensity image as described below. Astrocytic processes are generally thin and elongated. To take advantage of this property, we used steerable oriented filters and stored in each pixel the maximal response to different filter sets. Any region near a round bright spot is “oriented” towards it, meaning that the maximal response is obtained with respect to the filter with similar orientation. We used a round isotropic Gaussian filter to find local intensity maxima and oriented filters to highlight the thin elongated structures. To suppress the algorithm from detecting false-positive filamentous structures in dark and noisy regions of interest we performed a randomization-based regularization consisting of many iterations in which we disturbed the input image with noise and then thresholded the responses. By averaging the results obtained over multiple iterations, we obtained an estimate of the probability that each pixel has an oriented structure brighter than its surroundings. The final mask is the product of the isotropic and anisotropic filter responses.
Image analysis of complexity and entropy
The complexity-entropy analysis allows quantification of the spatial properties of representative images with respect to their balance between randomness and structural order, triviality and complexity as previously described (Brazhe, 2018). In mathematical terms, this is described as location in the complexity-entropy plane or their distribution along a complexity-entropy spectrum. Highly ordered structures (e.g., a regular grating) have near-zero entropy and near-zero complexity. In contrast, completely disordered structures (e.g., independent and identically distributed Gaussian samples) have maximal entropy and small statistical complexity. Intermediate values of entropy are associated with higher values of complexity if the underlying pattern contains features with preferred orientation. For example, signals generated by systems with deterministic chaos result in a complexity-entropy spectrum with near-0.5 peak complexity and near-0.7 entropy (Lamberti et al., 2004; Rosso et al., 2007). The implementation code is available at DOI:10.5281/zenodo.1217636. Briefly, complexity and entropy measures can be based on a feature distribution of the analyzed patterns, which can be compared to equiprobable or singular feature cases. Accordingly, the relative entropy (H) was defined as the Shannon entropy: normalized by the entropy of an equally probable distribution (Smax), thus giving values in the range [0 … 1]. Here N is the number of features analyzed. The complexity measures were based on the notion of disequilibrium (Lamberti et al., 2004) and defined in terms of normalized Jensen-Shannon divergence from an equally probable distribution (Rosso et al., 2007): In this expression, J[P,Pe] is Jensen-Shannon divergence of some distribution P from an equally probable distribution Pe: and Jmax is obtained when the probability of just one feature is equal to 1, while the probability of all other features is zero. The defined probability-based definitions of entropy and complexity require building descriptive probability-like measures for spatial patterns, preferably allowing for multiscale resolution and local analysis. A spatial pattern can be described in terms of local edge orientations and scales, which can be achieved by discrete shearlet transform, which is analogous to wavelet transform with added orientation information. We treated normalized power of shearlet coefficients as probability densities of given spatial feature orientation and scale at a given location. Complexity and entropy calculations were only performed in regions containing non-zero pixels, while other areas were masked out.
Statistical analysis
Data are presented as mean±SEM unless otherwise stated. All experiments were performed on multiple mice of either sex. Statistical significance was determined by Student’s paired or unpaired t test or ANOVA, as appropriate (IgorPro 6.37). Differences were considered significant at p<0.05 (*p<0.05; **p<0.01; ***p<0.001).
Competing Interests
None to declare.
Author Contributions
Conceptualization: A.Sc.
Methodology: A.B., A.Sc.
Formal Analysis: J.P.M., L.Y.D., N.A., J.J.W., S.Z., S.S., A.A.S, R.M., A.Sc.
Investigation: J.P.M., M.A.P., L.Y.D., G.C.T., N.A., R.M., A.Sc.
Resources: R.D.L., A.K., A.Se., M.M., A.Sc.
Writing – Original Draft Preparation: A.Sc.
Writing – Review & Editing: All authors
Visualization: A.Sc.
Supervision: A.Sc.
Project Administration: A.Sc.
Funding Acquisition: This work was supported by the Presidential Award for Undergraduate Research and Initiative for Women Award to N.A.; RFBR COMFI (Grant Number 17-00-00412K) for joint research to A.Se. (Grant Number 17-00-00409) and A.B. (Grant Number 17-00-00407); NIH/National Institute of Biomedical Imaging and Bioengineering Intramural Research Program (Grant Number ZIAEB000046) to R.D.L.; NIH/National Institute on Drug Abuse (Grant Number R01DA04741001) to A.K.; EU H2020 program (Grant Number 785907, HBP SGA2) to M.M. and R.M.; NIH/National Institute of Neurological Disorders and Stroke (Grant Number R03NS102822), SUNY Albany and SUNY Albany Research Foundation to A.Sc.
Acknowledgements
We would like to thank Drs. Pablo Valdes and Ed Boyden for support with the proExM protocol and Drs. Juan Burrone, Christophe Bernard and Damian Zuloaga for valuable discussions.
References
- 1.↵
- 2.↵
- 3.↵
- 4.↵
- 5.↵
- 6.↵
- 7.↵
- 8.↵
- 9.↵
- 10.↵
- 11.↵
- 12.↵
- 13.↵
- 14.↵
- 15.↵
- 16.↵
- 17.↵
- 18.↵
- 19.↵
- 20.↵
- 21.↵
- 22.↵
- 23.↵
- 24.↵
- 25.↵
- 26.↵
- 27.↵
- 28.↵
- 29.↵
- 30.↵
- 31.↵
- 32.↵
- 33.↵
- 34.↵
- 35.↵
- 36.↵
- 37.↵
- 38.↵
- 39.↵
- 40.↵
- 41.↵
- 42.↵
- 43.↵
- 44.↵
- 45.↵
- 46.↵
- 47.↵
- 48.↵
- 49.↵
- 50.↵
- 51.↵
- 52.↵
- 53.↵
- 54.↵
- 55.↵
- 56.↵
- 57.↵
- 58.↵
- 59.↵
- 60.↵
- 61.↵
- 62.↵
- 63.↵
- 64.↵
- 65.↵
- 66.↵
- 67.↵
- 68.↵
- 69.↵
- 70.↵
- 71.↵
- 72.↵
- 73.↵
- 74.↵
- 75.↵
- 76.↵
- 77.↵
- 78.↵
- 79.↵
- 80.↵
- 81.↵
- 82.↵
- 83.↵
- 84.↵
- 85.↵
- 86.↵
- 87.↵
- 88.↵
- 89.↵
- 90.↵
- 91.↵
- 92.↵
- 93.↵
- 94.↵
- 95.↵
- 96.↵
- 97.↵
- 98.↵
- 99.↵
- 100.↵
- 101.↵