Developmental Patterns of Sleep Slow Wave Activity and Synaptic Density in Adolescent Mice

Abstract

Study Objective:

In humans sleep slow wave activity (SWA) declines during adolescence. It has been suggested that this decline reflects the elimination of cortical synapses, but this hypothesis has never been tested directly.

Design:

We focused on mouse frontal cortex and collected data from early adolescence (∼postnatal day 20, P20) to adulthood (P60) of (1) SWA; (2) expression of synapsin I, a presynaptic marker; and (3) number of dendritic spines in layers I-II.

Setting:

Basic sleep research laboratory.

Patients or Participants:

YFP-line H mice (n = 70; P15-87, all males) and GFP-line S mice (n = 14; P17-60, 8 females) were used for EEG recording. Forty-five YFP mice (P19-119, 12 females) and 42 GFP-S mice (P20-60, 14 females) were used for in vivo 2-photon imaging and ex vivo confocal microscopy, respectively. Other YGP mice (n = 57, P10-77) were used for western blot analysis of synapsin I.

Interventions:

N/A.

Measurements and Results:

As in humans, SWA in mice declined from early adolescence to adulthood. Synapsin I levels increased from P10 to P24, with little change afterwards. Mean spine density in apical dendrites of layer V pyramidal neurons (YFP-H) showed no change from P20 to P60. Spine number in layers I-II apical dendrites, belonging to layer III and V pyramidal neurons (GFP-S), increased slightly from P20 to P30 and decreased from P30 to P60; smaller spines decreased in number from P20 to P60, while bigger spines increased.

Conclusions:

In mice, it is unlikely that the developmental decrease in SWA can be accounted for by a net pruning of cortical synapses.

Citation:

de Vivo L; Faraguna U; Nelson AB; Pfister-Genskow M; Klapperich ME; Tononi G; Cirelli C. Developmental patterns of sleep slow wave activity and synaptic density in adolescent mice. SLEEP 2014;37(4):689-700.

INTRODUCTION

During non-rapid eye movement (NREM) sleep, which in humans and rodents accounts for ∼80% of sleep, the cerebral cortex generates slow waves, one every second or so, which travel along the cortical surface and are especially prominent over frontal regions.^1–4 SWA, defined as the EEG power between 0.5 and 4 Hz during NREM sleep, provides a quantitative measure of slow waves. Intracellular recordings have shown that SWA in the EEG is generated by virtually all cortical neurons engaging in a slow (< 1 Hz) oscillation, consisting of a depolarized up state, when neurons show sustained firing, and a hyperpolarized down state, characterized by neuronal silence.^5–8 Converging evidence derived from theoretical considerations,⁹ large-scale computer simulations,¹⁰ and experiments in animals and humans^4,11,12 indicates that the amplitude and slope of EEG slow waves is related to the number of neurons that enter an up state or a down state near-synchronously, and that synchrony is directly related to the number and efficacy of synaptic connections among them.

Several recent studies in humans have shown that during development absolute values of SWA follow an inverted U curve, with a progressive increase during childhood, a peak around 8 years of age, and a rapid decline during adolescence. For instance, a longitudinal study documented a steep decline in sleep SWA (1-4 Hz) and theta (4-8 Hz) activity, especially between age 11 and 16.5 years.¹³ The authors compared their EEG data with published longitudinal declines in MRI-estimated cortical thickness, and found that changes in SWA tracked the thinning of 5-layer cortex, while theta activity paralleled the earlier maturational thinning in 3-layer cortex. High-density EEG recordings also indicate that the location of maximal SWA undergoes a shift in topography, from a peak in posterior regions to a peak in frontal regions, which seems to parallel brain maturation.¹⁴ Moreover, during adolescence, SWA decreases the most over the brain regions that show the largest decrease in gray matter.¹⁵ Changes in cortical synaptic density as revealed by a few ultrastructural studies^16–18 also follow an inverted U curve, with a peak before adolescence followed by a decline. Finally, developmental changes in brain glucose consumption,^19,20 which is a function of synaptic density/strength,²¹ also follow an inverted U curve. Thus, the hypothesis has been put forward that developmental changes in the sleep EEG reflect changes in synaptic density, and more specifically, that the decline in sleep SWA during adolescence reflects the net elimination, or pruning, of cortical synapses.^13,22 However, MRI cannot assess synapses directly, and reduced cortical thickness could be due to many other factors.^23,24 Moreover, electron microscopy data in humans have been obtained in a limited number of samples, especially during the critical years that span early adolescence.

In an attempt to directly and systematically compare the developmental change in EEG power spectra with that of synaptic number, we focused on the frontal cortex of adolescent mice. Rodent adolescence spans the period from weaning (∼P21) to sexual maturity (P50-60), and can be further subdivided in a “periadolescence” period (P34-46) around the onset of puberty (at ∼ P40 in rats and mice), preceded by early adolescence (P21-34), and followed by late adolescence (young adults, P46-59). This age classification and the overall strength of the rodent model for the purpose of comparison or extrapolation to human development have been validated extensively.^25–29 We collected data of sleep SWA, cortical expression of synapsin I, a presynaptic protein present in virtually all synapses of the mouse cortex,³⁰ and number of dendritic spines in two types of cortical pyramidal cells.

MATERIALS AND METHODS

EEG Recordings

YFP-line H mice (n = 70; P15-87, all males) and GFP-line S mice (n = 14; P17-60, 8 females) (Jackson Laboratory, Bar Harbor, Maine; all bred in-house) were implanted for chronic polysomnographic recordings under isoflurane anesthesia (1-1.2%). These 2 lines of transgenic mice were chosen because they express a yellow (YFP) or a green (GFP) fluores-cent protein in different subsets of cortical pyramidal neurons,³¹ allowing the analysis of both SWA and dendritic spine density. All electrodes were soldered to flexible wires (#NUF30-4046, Cooner Wire, Chatsworth, CA) and placed over the right and left frontal (anteroposterior, AP, +1 mm from bregma; medio-lateral, ML, 1 mm), parietal (AP -2 mm; ML 2 mm) cortices and one over cerebellum (AP -1 mm from lambda) as reference. No electrode was implanted on left frontal cortex of GFP mice, to allow better imaging of dendritic spines after sacrifice. Two coated stainless steel wire electrodes (#AS636, Cooner Wire) were inserted into the nuchal muscle for electromyogram (EMG) recording. Electrodes were affixed to the skull and insulated with dental cement. Following surgery, mice were housed individually in sound-attenuating, environmentally controlled recording chambers (12-h light/12-h dark cycle, lights on at 08:00, 25°C ± 1°C, food and water ad libitum). All electrodes were gathered into a flexible cable and connected to the Multi-channel Neurophysiology Recording system (Tucker-Davis Technologies, TDT, Alachua, FL). EEG and EMG signals were collected continuously at a sampling rate of 256 Hz (digitally filtered between 0.1-100 Hz). For sleep staging, signals were processed by custom-made Matlab scripts (Mathworks, Natick, MA) using standard TDT routines and subsequently converted into European Data Format (EDF) with Neurotraces software (www.neurotraces.com, Fort Lauderdale, FL). Twenty-four hour polygraphic recordings used for analysis started ≥ 5 days after surgery, with the exception of mice that underwent surgery before P20, in which analysis of EEG recordings started 3 days after surgery to obtain data as early as possible. Data were scored offline for NREM sleep, REM sleep, and wake by visual inspection of 4-sec epochs (SleepSign; Kissei Comtec, Irvine, CA). EEG power spectra of consecutive 4-sec epochs (FFT routine, Hanning window) were calculated for the frontocerebellar derivation within the frequency range of 0.0-30 Hz. Absolute SWA (0.5-4.0 Hz) and SWA relative to NREM high frequencies (15-30 Hz), a previously described normalization,^32,33 were computed. Normalization of SWA using high frequencies helped to compensate for possible changes in signal strength across different animals, due to technical issues, and to reduce data noise. Absolute high frequency power (15-30 Hz) did not consistently change across development (Figure S1, supplemental material).

For all mice having surgery before P20, special care was taken to make sure they received proper nutrition, following the recommendation of the veterinary staff. Specifically, in addition to the normal food pellets (given to allow chewing/ gnawing), these mice received the breeder diet (8626 Teklad Mouse breeder diet, Harlan laboratories, Madison, WI), which is softer and has a higher fat content. Two pellets of breeder diet were softened in water and replaced daily with fresh softened pellets until there was clear visual evidence of the solid pellets being consumed. Each mouse was given special bedding material made from pulped cotton fiber to maintain warmth when single housed (NestlestsT, Ancare, Bellmore, NY). For the youngest animals a portion of the bedding from the home cage was transferred to the recording cage so that a nest was available immediately after surgery.

All animal procedures followed the National Institutes of Health Guide for the Care and Use of Laboratory Animals; facilities were reviewed and approved by the IACUC of the University of Wisconsin-Madison and were inspected and accredited by AAALAC.

Synaptoneurosome Preparation and Western Blotting for Synapsin I

YFP-H mice (all males) were rapidly anesthetized with isoflurane and decapitated at the same time of day (12 noon). Heads were cooled in -20°C methyl butane, and the whole brain was removed. In the first set of experiments, the entire left cortical hemisphere was dissected and used to prepare synaptoneurosomes from P10, P14, P20, P25, P29, P35, P40, P50, and P68 mice (3-5 mice/group). In the second set of experiments, left and right frontal lobes were dissected and pooled from P22, P25/26, P28, P32, P35, P40, P46, P51, P60, and P77 mice (3-5 mice/group). Cortical samples from each mouse were immediately frozen on dry ice and stored at -80°C. Samples were then homogenized in ice-cold homogenization buffer (10 mM Hepes [Sigma], 1.0 mM EDTA [Promega], 2.0mM EGTA [BioProducts], 0.5 mM DTT [Promega], 0.1 mM PMSF [MP Biomedicals], Protease Inhibitor Tab [1 tab/10 mL, Roche], 100 nM microcystin [Alexis], Nuclease free water [Promega] using a glass/glass tissue homogenizer [Kontes]). A fraction (∼10%) of the homogenate from each sample was boiled in 10% SDS for 10 min and stored unprocessed at -80°C. The other fraction of the homogenate was centrifuged at 2000 g for 1 min at 4°C, and the supernatant collected and passed through two 100-μm-pore nylon mesh filters (Small Parts Inc.), then through a 5-μm-pore filter (Millipore), and centrifuged at 1000 g for 10 min at 4°C. Pellets were suspended in boiling 1% SDS, boiled for 10 min and stored at -80°C. After protein concentration was determined by BCA assay (BCA, Pierce), equal amounts of synaptoneurosomes from each animal were first separated by Tris-HCl gel electrophoresis (Bio-Rad) in 1X Tris, Glycine, SDS Buffer (BioRad), transferred to 0.45-μm pore size nitro-cellulose membranes (BioRad) in 1X Tris base, Glycine, Methanol blotting buffer, and immunoblotted with anti-PSD-95 (1:250, BD Biosciences Pharmingen) and anti-Tubulin (1:1000, Chemicon) antibodies to confirm the enrichment for synaptic proteins in synaptoneurosomes. Afterwards, equal amounts of synaptoneurosomes from each animal were separated by 4-15% Tris-HCl gel electrophoresis and transferred to 0.45-μm pore size nitrocellulose membranes as above, and immunoblotted with anti-synapsin I (1:1000, Imgenex, this antibody recognizes both synapsin Ia and Ib as a duplet of 78kDa). All blots were first blocked for 1 h at room temperature in freshly prepared 5% nonfat dry milk (Biorad) in 1X TBS (Fisher) with 0.1% Tween-20 (Sigma) (TBST-MLK), incubated with the primary antibody overnight with agitation at 4°C, rinsed in ddH₂O, and then incubated with the secondary antibody (horseradish peroxidase-conjugated anti-rabbit IgG antibody, Millipore) for 1 h with agitation at room temperature in TBST-3% MLK. After the final washes, immunoreactive bands were visualized using chemiluminescence (ECL- Prime kit, GE Healthcare) and captured by Typhoon 9410 Variable Mode Imager (Amersham). ECL signal intensities were quantified using the ImageQuant software (Amersham). Optical densities were calculated for each band of interest after performing background correction by subtracting the value of the region immediately above the band of interest in the same lane. No normalization using housekeeping proteins was used, because most if not all proteins enriched in synaptoneurosomes are likely to change their expression during development, especially in the youngest experimental groups. Instead, to insure reliability, in both experiments (whole cortex and frontal lobes) 3-4 independent blots were run for each mouse, and results were then pooled. Synapsin I levels were expressed as % change relative to the youngest experimental group (P20 for whole cortex, P22 for frontal lobes). In each blot, a 0 value was assigned to the P20 synapsin I band, and the protein levels for all other time points in the same gel were expressed as % change relative to P20. Afterwards, for each age, % changes across blots were averaged. For this reason, in Figure 2 the P20 value equals 0 and there is no standard error.

Figure 2.

Expression of synapsin I across development. Cortical protein levels of synapsin I (whole cortex) increase sharply from P10 to P25 and then stabilize. Bars indicate % of change in synapsin I expression normalized to P20 (mean ± SEM, 3-5 mice/age group, n = 28 mice).

Two-Photon Imaging in YFP-H Mice

YFP-H mice (Jackson Laboratory, Bar Harbor, Maine) were housed individually and maintained on a 12 h light/12 h dark cycle (lights on at 08:00) with food and water available ad libitum. Overall, 45 mice were imaged (12 females). Six animals underwent combined EEG recordings and 2-photon (2P) imaging. All animals were consistently imaged between 2 and 4 h after lights on. Skull thinning for 2P laser scanning microscopy was performed in frontal cortex.³⁴ Transcranial imaging was performed under isoflurane anesthesia (∼0.8-1 % in 100% O₂) using a 2P microscope (Prairie Technologies, Madison, WI) with a Ti:Sapphire laser (Cameleon Coherent, Coherent Inc., Santa Clara, CA) tuned to the excitation wavelength for YFP (920 nm). Stacks of image planes were acquired using a water-immersion 60X objective zoom (Olympus, LUMPlanFI/IR, 0.8 numerical aperture; 3X optical zoom). Randomly chosen dendritic branches were imaged at ∼50 to ∼200 μm from the pial surface (step size = 0.70 μm). Previous studies^35,36 showed that these dendritic segments likely belong to different parent neurons, due to the high density of labeled cells in this mouse strain. The average number of dendritic segments scored was 2.4 ± 1.64 per animal. All dendritic protrusions clearly (5 pixels = 0.65 μm) emanating laterally from the dendritic shaft were counted, classified as spines or filopodia (long and thin protrusions without a bulbous head³⁵), and analyzed separately. In all analyses the analyzer was unaware of experimental condition (i.e. age group). The identification of spines used for analysis was performed manually. NeuronStudio (CNIC, Mount Sinai School of Medicine) was used to archive the detection and classification of manually detected spines.³⁷

Confocal Imaging in GFP-S Mice

A total of 42 GFP-S mice were used (15 females): 14 pre-adolescent mice (P20), 15 adolescent mice (P30), and 13 adult mice (P60). Animals were housed individually in environmental controlled chambers, with a 12 h light/12 h dark cycle (lights on at 08:00), water and food ad libitum. Pre-adolescent mice were weaned at P17 and provided with soft food and cotton nesting material to ensure adequate feeding and temperature control.

Tissue Preparation and GAD-67 Immunofluorescence

GFP-S mice were rapidly (< 1 min) and deeply anesthetized with 3-5% isoflurane, and transcardially perfused with saline, followed by a solution of 4% paraformaldehyde freshly dissolved in 0.1M phosphate buffer (PB, pH 7.4). Brains were postfixed in the same fixative overnight, and then sliced coronally on a Vibratome (thickness of 60 μm). Sections were mounted on glass slides, air dried, and covered with Vecta-Shield H-1000 mounting medium. For immunofluorescence, sections from 6 mice at P20, P30, and P60 were incubated with 10% normal goat serum (NGS) in 0.1M PB (1 h) followed by primary monoclonal antibody anti-glutamic acid decarboxylase (GAD-67, MAB5406, Millipore; 1:500) in 0.1M PB (2 h at room temperature and overnight at 4°C). The following day, sections were rinsed in PB, incubated with NGS10% (20 min) and with anti-mouse Alexa Fluo 568 (Invitrogen, 1:300; 1 h). Sections were kept free floating in a solution of 0.03% sodium azide in 0.1M PB until the day of imaging. All images, including those for spine analysis (see below) were obtained with a Praire confocal microscope equipped with 488 nm and 561 nm lasers. For GFP/GAD-67 colocalization studies, images were taken with a 20X water objective lens (NA = 0.95), raster size 512 × 512, voxel size of ∼1 × 1 × 1 μm, average of 4 frames, speed of 4 μs per pixel and a pinhole of 1 Airy Unit.

Spine Analysis

Images for dendrite and spine analysis were taken with a 60X oil objective lens (NA = 1.42) and a voxel size of 0.1 × 0.1 × 0.3 μm. Laser power and PMT gain were optimized in each acquisition in order to obtain a full dynamic range for the GFP intensity, being careful to avoid pixel saturation. PMT gain was always kept under 700 V. In each animal, 9 images from 3 to 6 different sections were acquired in layer I-II of 2 subregions of the medial prefrontal cortex (mPFC), namely anterior cingulate cortex (ACg) and prelimbic cortex (PrL).³⁸ Only the brightest GFP positive dendritic branches were imaged. An average ± SD of 20.9 ± 6.5 dendritic segments per mouse was imaged. In mice that underwent electrophysiological recordings, imaging took place in the mPFC of the non-implanted hemisphere (left hemisphere). NeuronStudio was used to obtain measures of dendritic length and diameter, spine density, and spine diameter.³⁷ Due to the high density of dendrites and axons present in the acquired images, automated spine detection with NeuronStudio was not possible. All spines were identified manually, and spine density expressed as number of spines per unit of dendrite surface area (μm²), computed as described by Dumitriu.³⁹ Normalizing the number of spines by the local amount of plasma membrane compensate for the presence of dendrites with different length and different diameter.

RESULTS

Decline in Slow Wave Activity from Early Adolescence to Adulthood

YFP-H mice (n = 70) were implanted for polysomnographic recordings and continuously recorded for 1-3 weeks. A detailed analysis of sleep/wake patterns in these YFP-H mice was published recently.³³ Briefly, a preference to sleep during the day was already present from early adolescence (∼P20), with all animals showing an average ∼2:1 ratio of sleep during the light period relative to the dark period, although there was inter-individual variability. Throughout adolescence (P21-P59) total sleep remained constant, while REM sleep declined steeply in early adolescence (P20-P30). Moreover, even in the youngest mice (∼P20), SWA was a reliable marker of sleep pressure during baseline, as described in adult mice: it declined in the course of sleep during the day and increased after time spent spontaneously awake at night.

A group of GFP-S mice (n = 14) was also implanted for chronic polysomnographic recordings in order to characterize SWA changes during adolescence in this strain. GFP-S mice were only studied at 3 time points, P20, P30, and P60, but overall, sleep/wake changes from early adolescence to adulthood showed the same general trend as in YFP-H mice. Younger mice (P20) showed only a slight preference to sleep during the light period relative to the dark period (light/ dark = 1.2 ± 0.5, mean ± SD), but preference to the light/ dark cycle increased with age (light/dark, P30 = 1.97 ± 0.2; P60 = 2.09 ± 0.4). REM sleep duration decreased from P20 (182 ± 61 min) to P30 (134.5 ± 20 min) and to P60 (112 ± 10 min; Kruskal-Wallis test P = 0.031, Dunn multiple comparison test P20 vs. P60, P < 0.05), while total sleep time remained similar across all ages (mean ± SD, in min; P20 = 637 ± 50; P30 = 596 ± 56; P60 = 652 ± 22, Kruskal-Wallis P = 0.22). As expected, and consistent with what observed in YFP-H mice, SWA activity declined during the light phase (the major sleep phase in mice), in all P30 and P60 GFP-S mice (average % of change in SWA in the last hour of the light period compared to SWA in the first hour after sleep onset ± SD: P30 = -28 ± 14.6, P60 = -26.2 ± 12.2). By contrast, at P20, 3 of 5 mice showed higher SWA at the end of the light phase compared to the beginning (% SWA change = +11.8 ± 48), consistent with their weak light/dark entrainment.

We then focused on the developmental changes in SWA in both mouse strains. Raw EEG traces during sleep and wake from 3 representative animals for each strain are shown in Figures 1A and B, which also show the overall 24-h temporal profile of SWA and locomotor activity (bottom panels). In both strains there was a progressive decline in the amplitude of the EEG signals with age, which was most apparent for the slow waves of NREM sleep. This decline is summarized in Figures 1C and D, which show that normalized NREM SWA (24-h mean) decreased as a function of age in both the frontal (-0.4341 per day, R² = 0.30) and parietal derivation (-0.4452 per day, R² = 0.27), with an overall decline of approximately 50% from early adolescence to adulthood. However, consistent with human studies,^13,40 Figures 1C and D also show that there was large interindividual variability in SWA values in both frontal and parietal cortex, especially in the younger animals, suggesting that the SWA maturational pattern is complex and may not depend exclusively on age. Of note, during NREM sleep the frontal EEG power in the theta (4-7 Hz) and alpha (8-12 Hz) range also showed a developmental decline (theta: -0.101 per day R² = 0.242; alpha: -0.048 per day R² = 0.378), although not as steep as for SWA. By contrast, during either wake or REM sleep there was no developmental decline in frontal SWA, theta, or alpha activities. In fact, the EEG power in the alpha range increased from early adolescence to adulthood in both wake and REM sleep (REM alpha: 0.027 per day R² = 0.150; Wake alpha: 1.20 per day R² = 0.334; Figure S2, supplemental material).

Figure 1.

Changes in EEG signal and mean SWA across age. (A) and (B) show frontal EEG traces (10 sec) recorded from representative YFP-H and GFP-S mice during NREM sleep, REM sleep, and wake at P20 (early adolescence), P30 (middle adolescence), and P60 (adulthood). The bottom panels show the 24-h time course of NREM slow wave activity (SWA, expressed as % of 24-h mean SWA, 4-sec epochs) and locomotor activity (video-based motion detection, 4-sec epochs) for the same animals (light phase corresponds to hours 0-12). (C) and (D) show NREM SWA levels (24-h mean SWA normalized by high frequencies 15-30 Hz) across age from frontal and parietal cortex. Each symbol refers to a single animal, and each mouse contributed only one SWA value, collected ≥ 5 days after surgery. Black dots represent YFP-H mice only used for EEG recordings, while gray dots refer to YFP-H mice (n = 6) implanted for simultaneous EEG recordings and in vivo 2P imaging; in these animals 24-h SWA values were obtained the day following the imaging session. Red circles refer to GFP-S mice (n = 9) used first for EEG recordings, and then sacrificed the next day to perform confocal imaging. Additional mice are shown as reference but were not used to determine the regression line: YFP-H mice (n = 4) and GFP-S mice (n = 5) that were early weaned are shown with open black and red circles, respectively; in these animals SWA values were collected 3 days after surgery (see Methods for details). X refers to older adult YFP-H mice that were recorded well beyond the adolescent developmental window that is the focus of this study.

Increase in Expression of Synapsin I before Onset of Adolescence, with Little Change Afterwards

In the first set of experiments, synapsin I levels were measured in the entire left hemisphere of male YFP-H mice starting at P10. As shown in Figure 2, synapsin I expression increased the most between P10 and P20, with an additional small increase between P20 and P25 and little change afterwards. In the second set of experiments, we focused on the frontal lobes, since a detailed analysis of SWA changes across age had been done for that region (Figure 1C). Overall, the results were consistent with the findings in whole cortex, with only small (≤ 10%) differences in synapsin I expression between early adolescence (P22) and adulthood (P77) (data not shown). Thus, the developmental expression of a ubiquitous presynaptic marker, synapsin I, does not seem to closely mimic the decline in SWA seen throughout adolescence. More specifically, the early drop in frontal SWA (P20 and P32) occurred while synapsin I levels in the same region were stable. Similarly, both frontal and parietal SWA progressively declined from early adolescence to adulthood, while synapsin I levels in whole cortex were stable across the same period.

Changes in Spine Density across Adolescence in the Apical Dendrites of Layer V Pyramidal Neurons (YFP-H mice)

Spines typically receive only one asymmetric synapse,^41,42 and more than 95% of all spines in the neocortex are associated with synapses.^43–45 Thus, the number of spines is a good approximation of the total number of excitatory synapses on a dendrite. We measured spine density in frontal cortex of YFP-H mice whose age ranged from P19 to P119 (Figure 3A). In this strain, a few pyramidal neurons in layer V are labeled, allowing their apical dendrites to be imaged with 2P microscopy in layer II (Figure 3B). As shown in Figure 3C, there was large interindividual variability in the mean spine density even within mice of the same age, and this variability did not significantly decrease by the end of adolescence. Overall, average spine density did not significantly decline during early or late adolescence (R² = 0.049, P = 0.166). By contrast, filopodial protrusions were more numerous in the youngest mice (P ≤ 21) (R² = 0.279, P = 0.0005).

Figure 3.

Spine density in vivo two photon microscopy. (A) Schematic representation of the region imaged in a frontal section of mouse brain. (B) Examples of dendritic processes and spines as they appear at the 2P microscope at P20 and P30: note the higher number of filopodia in the younger animals. Scale bars = 10 μm. (C) Average values of spine density per mouse across the adolescent period. Gray dots refer to 6 mice that were also implanted for EEG recordings. The bar graph in the right corner shows average spine density (average ± SD) grouped by age intervals: black < P21, red P22-30, green P31-40, yellow P41-60. No significance difference in spine density between groups was revealed (one-way ANOVA on ranks F = 1.99, P = 0.16). (D) Average values of filopodia density per animal. A significant difference in the < P21 group versus the P41-60 one emerged running a Dunn method pairwise comparison after significance in Kruskal-Wallis ANOVA (P < 0.05; H = 14.70, P = 0.0005). Gray dots come from mice that were also implanted for EEG recordings.

Changes in Spine Density and Size across Adolescence in Pyramidal Neurons of Layers II-III and V (GFP-S mice)

GFP-S mice express green fluorescent protein in a sparse subset of cortical neurons in layers II-III and V, starting at P15. A previous study⁴⁶ found that in the visual cortex of these mice some GFP positive cells can be spinous interneurons, and that in layers I-III about 90% of these interneurons are GABAergic. Since data were not available for medial prefrontal cortex (mPFC), our area of interest, we first assessed the percentage of GFP positive cells in mPFC that also express GAD-67, a GABAergic marker. We counted 270 GFP positive cells (6 mice), of which only 3 (0.9% ± 0.7%, mean ± SEM) were positive for GAD-67 (Figure 4A). Thus, virtually all dendritic spines identified in layers I-II of mPFC belong to pyramidal neurons (Figure 4B).

Figure 4.

Age-related changes in spine density and size in GFP-S mice. (A) Schematic representation of a frontal section of mouse brain, showing the sub-regions of the medial prefrontal cortex (mPFC) that were imaged: Prelimbic cortex (PrL) and Cingulate cortex (Cg). Bottom panels show an example of one GFP positive neuron expressing GAD-67. Scale bar = 20 μm. (B) shows one GFP positive pyramidal cell in layer III of mPFC. Scale bar = 50 μm. Images in (C) are examples of layer II cortical dendrites from the same GFP mouse at P30: note that spine density varies considerably within the same layer. Scale bar = 10 μm. (D) Average values of spine density per unit of area at P20 (n = 14), P30 (n = 15), and P60 (n = 13). Values represent mean ± SD. ANOVA P = 0.041. (E) Cumulative distributions of spine density per unit of area (μm²) for the entire population of dendrites at P20 (271 dendrites), P30 (347 dendrites), and P60 (261 dendrites). Kolmogorov-Smirnov test (KS test), P20 vs P30, P = 0.001; P30 vs 60, P = 0.011; P20 vs P60, P = 0.032. (F) Box plot showing the distribution of density for the entire population of dendrites at different ages: upper and lower side of boxes represent 75^th and 25^th percentile of spine populations, respectively. (G) Average values, per animal, of filopodia per unit of area. One-way ANOVA P = 0.0023, Bonferroni multi-comparison test P20 vs P30, P > 0.05; P20 vs P60, P < 0.01; P30 vs P60, P < 0.05. (H) Examples of dendrites with numerous filopodia (arrowheads) in a preadolescent mouse (left panel) and in an adult mouse with no filopodia (right panel). Scale bar = 10 μm.

Spine density varied considerably in layers I-II dendrites, even within the same mouse (Figure 4C). This range of variability led to average values of spine density that changed almost twofold in mice of the same age. As a result, mean spine density showed a slight increase at P30 compared to P20 and P60 (Figure 4D). Values of spine density from dendrites collected from mice of the same age were then pooled together and the distributions of the P20, P30, and P60 populations were compared. A group of dendrites with higher spine density emerged at P30, shifting the upper part of the cumulative distribution of this group significantly towards the right (Figure 4E). Thus, the data suggest that the number of dendrites with high spine density increases from P20 to P30, and then decreases from P30 to P60 (Figure 4E and F). Filopodia were counted separately. Their density was high in preadolescent mice and decreased significantly in the late phase of adolescence (Figure 4G and H).

We then conducted the analysis of spine diameter and found that the distributions of pooled data differed significantly from each other (Figure 5A), with P20 and P30 showing a higher number of small spines than P60 (Figure 5B). By contrast, the proportion of big spines increased progressively with age (Figure 5B).

Figure 5.

Changes in spine diameter across adolescence in GFP-S mice. (A) Cumulative distributions of spine diameter population differ at P20 (6415 spines), P30 (8789 spines), and P60 (5759 spines). KS test P20 vs P30, P = 0.011; P30 vs P60, P = 0.0038; P20 vs P60, P = 0.0000004. (B) Normalized frequency distributions of spine diameters for apical dendrites. The dashed lines show the 25^th and 75^th percentiles (0.47 and 0.62 μm, respectively) of the entire spine population (20963 spines): under the 25^th percentile (small spines), the proportion of small spines at P60 is lower than at P20 and P30 (KS test P = 0.03), whereas above the 75^th percentile (big spines), the proportion of big spines increases from P20 to P60 (KS test P20 vs P30, P = 0.0003; P30 vs P60, P = 0.017).

DISCUSSION

NREM SWA is believed to reflect the number of cortical neurons that enter near-synchronously the up state or the down state of the slow oscillation, as well as the overall number and strength of the synaptic connections among them.^4,9–12 Thus, the more neurons are strongly connected and fire together, the larger is SWA. In humans, SWA reaches a peak in early childhood, and then decreases sharply during adolescence.⁵³ Studies in mice and rats suggest that the rise of SWA during early development is linked to the rapid increase in the number and strength of cortical synapses. Specifically, electron microscopy (EM) studies have shown that a synaptogenic “explosion” occurs between P10 and P15 in rat barrel cortex, with the total number of cortical synapses increasing 2-3 fold.^47,48 Increases of similar extent during the second postnatal week have also been described using EM in the mouse somatosensory cortex,⁴⁹ and a more recent study in mouse barrel cortex found that stel-late cells of layer IV lack spines almost completely until P8, show a 70-fold increase in spine density between P8 and P9, and a 250-fold increase between P8 and P13.⁵⁰ Early EEG recordings in neonatal rats have shown that the total power in the EEG signal in all behavioral states increases after P9,⁵¹ cortical activity increases sharply at P11-P12,⁵² and after P12 the total power in the EEG signal during quiet (slow wave) sleep is consistently higher than in active (REM) sleep and wake.⁵¹ Thus, the sudden increase in cortical synapse number is likely to account for the rapid increase in EEG power after P9 and, shortly afterwards, for the increase in EEG power during quiet sleep relative to wake and active sleep, which is related to a greater level of cortical synchrony in quiet sleep.

In the current study, we did not assess the link between explosive synaptogenesis and rise of SWA, due to the difficulty in obtaining stable chronic EEG recordings in very young, pre-weaned mice. Instead, we focused on adolescent mice, to test the long-standing hypothesis that the decline in SWA during this critical period of development is due to synaptic pruning, i.e., to a decrease in cortical synaptic density.^22,53 Our first finding was that, as in humans, NREM SWA in mice declines during adolescence, and does so starting at around ∼P19, the youngest age at which we could collect stable chronic EEG recordings. Of note, due to the deterioration of the EEG signal, most mice, especially the youngest, were recorded for no more than a week, and each animal contributed only one time-point to the developmental profile of SWA shown in Figure 1, chosen to be five days after surgery to be consistent across animals. Thus, this is not a bona fide longitudinal study, because each mouse was not recorded long enough to span a large part of adolescence.

We found that during NREM sleep the decline in EEG signal was not restricted to SWA but encompassed the theta and alpha range, consistent with previous studies in humans that found that the EEG power in most or all frequencies of the NREM sleep EEG declines throughout adolescence.^{13–15,40,54,55} We also found that the developmental decline in SWA was specific for NREM sleep, since in wake and REM sleep the EEG signal in the same low frequency range (0.5-4 Hz) did not decrease. Some studies in human adolescents have reported a broad decline in power also during REM sleep,^40,55 while in others the decline was limited to REM theta power.¹⁵ Very few studies have measured developmental changes in EEG power during wake, due to the difficulties in controlling for the level of arousal and attention. In one case the overall EEG power declined between 6 and 17 years of age, while fast alpha activity (9.5-12.5 Hz) during wake increased with age.⁵⁶ This latter finding may be consistent with our results in mice, but the interpretation of the results is difficult, because a link between alpha activity and reduced alertness has been suggested in both humans⁵⁷ and mice.³³

In adolescent mice, we found no strong evidence for a decline in spine density that would parallel the decline in SWA. Specifically, using in vivo 2P imaging, we found no significant change in spine number from P20 to P60 in apical dendrites of frontal cortex layer V pyramidal neurons. Moreover, using confocal microscopy to assess the apical dendrites of frontal cortex layer III and V pyramidal neurons, we found a slight increase in spine number from P20 to P30 and a decrease from P30 to P60. Overall, these findings are consistent with those from other anatomical studies in mouse somatosensory cortex. One study, for instance, found that the density of total synaptic contacts progressively increases in all cortical layers from P16 to P32, and decreases by about 18% only in older adults (P > 100).⁴⁹ Another study found that between P20 and P90 spine density remains stable in layer I, and increases constantly in layers II-III.⁵⁸ Other anatomical studies in rat cortex also showed that pruning is restricted to specific dendritic domains and occurs early, long before the time of our recordings. For instance, layer V short (callosal) pyramidal cells prune their apical dendrites during the second postnatal week.⁵⁹ During the same week layer V tall-tufted pyramidal cells also lose filopodia-like segments from their apical dendrites, but overall spine density increases, and continues to do so until P36, after which it stabilizes with no decline.⁶⁰ Finally, in layers II/III pyramidal neurons some selective pruning occurs between P9 and P17 close to the soma, but new branches are added distally, with no changes in the total numbers of tips.⁶¹

Since spine analyses were limited to a specific cell type and a specific dendritic segment, we also measured the expression of synapsin I, a presynaptic protein present in all excitatory and inhibitory synapses of adult mice.³⁰ We found that synapsin I protein levels in mouse frontal cortex increase from P10 to P24, and show little if any change afterwards, consistent with a previous study showing that in the whole mouse brain synapsin I levels progressively increase from P5 until reaching a plateau at ∼P30, with no decline afterwards.⁶² Thus, Western blot analysis of synapsin I expression also does not provide strong evidence for a decline in spine density that would parallel the decline in SWA. Western blot analysis, however, is not very sensitive, and small changes in synapsin I levels may have gone undetected with this technique. Moreover, whereas in the adult brain there is strong evidence, from both light and electron microscopy, that synapsin I localizes to more than 90% of presynaptic elements, similar information is lacking for early stages of development. Therefore, we cannot rule out the possibility that at least a small number of functional synapses lacking synapsin I protein exists in the young brain.

In contrast to spines, both our in vivo and ex vivo analysis indicated that the number of filopodial protrusions seems to parallel the SWA decline during adolescence, decreasing progressively from P20 to P60. However, filopodia are considered non-functional precursors of spines, and at least in the hippocampus, their developmental decline is associated with the appearance, not the disappearance, of functional synapses into spines.⁶³ Thus, it is unlikely that the decline in filopodia can account for the change in SWA signal during adolescence.

Synaptic strength increases during wake and decreases during sleep in the adult mouse cortex, and molecular, electrophysiological, and structural studies show that these sleep/ wake homeostatic changes in synaptic efficacy are closely linked to the homeostatic changes in SWA.⁶⁴ Yet, spine turnover across the 24 h cycle appears to be very limited in the adult mouse cortex,³⁴ suggesting that changes in synaptic efficacy most likely occur without significant changes in the number of synapses. One possibility, therefore, is that a similar situation happens during adolescence, i.e., the developmental decline in SWA is driven by a decrease in synaptic efficacy without any major change in synaptic density. If this were the case, however, there should be some evidence for a generalized decrease in spine size during adolescence, because it is well established that the size of the spine head is directly proportional to the strength of its synapse. Specifically, the spine head volume is positively correlated to the area of the post-synaptic density, to the number of postsynaptic AMPA receptors, and to the number of presynaptic docked vesicles.^65–69 Our analysis of spine diameters using confocal microscopy did not provide any evidence that spine size could account for the overall decline in SWA during adolescence: while the number of smaller spines decreased from P20 to P60, the number of bigger spines gradually increased during the same time period. The stable levels of synapsin I as measured by Western blots also seem inconsistent with a major decline in postsynaptic strength, although changes in postsynaptic strength can certainly occur without presynaptic changes.^70,71 It should be emphasized, however, that our analysis of spine size was limited to a specific type of cortical neurons and relied on confocal microscopy. In the future, more sensitive and comprehensive methods, such as 3D-EM reconstruction of spine volumes in large cortical areas,⁷² should be used to establish whether there is a link between developmental SWA changes and changes in spine size.

If not changes in synaptic number or size, what could account for the decline in sleep SWA during adolescence? One important factor that has been suggested²² may be the functional optimization of cortical circuits, which occurs throughout adolescence, well after all major anatomical wiring has occurred. This “sculpting” or “refinement” phenomenon may involve the activity-dependent addition and survival of some synapses and the elimination of others, without major changes in the number and/or size of synapses, leading to a process of synaptic rearrangement that increases the specificity of connections.^73,74 For example, a recent study in the mouse visual cortex found that after eye opening, synaptic connections between neurons with similar visual responses increased in number, while at the same time the connections between visually unresponsive neurons decreased. The result was that the local connectivity reorganized extensively, even if the overall connectivity rate did not change.⁷⁵ At the functional level, sculpting is well documented in all sensory cortices. For instance, at the macroscopic level the somatotopic maps of the rat body become adult-like around P20, at the beginning of adolescence, but receptive fields at that time are still larger than in adulthood.⁷⁶ Similarly, the maturation of visual acuity in mice is not complete until ∼P45, well after the completion of the anatomical visual maps,⁷⁷ and an adult-like tonotopic map of sound frequency emerges at ∼P22 in rat primary auditory cortex, but the refinement of tonotopicity continues well into adulthood.⁷⁸

Our results suggest that in mice the decline in sleep SWA during adolescence is unlikely to be accounted for by synaptic pruning. Whether this conclusion can be extended to humans remains to be demonstrated. However, even in humans it has been pointed out that the developmental trajectories of gray matter, glucose metabolism, and synaptic density do not match exactly, especially during adolescence.²⁴ In fact, there are only two EM studies in humans by Huttenlocher and colleagues, based on a very limited number of subjects, one with no data from year 7 to year 16,¹⁶ and the other showing a decline in prefrontal cortex from year 10 to year 21 (only one subject/year) and no decline in auditory and visual cortex.¹⁷ A more recent study that used the classical Golgi staining to measure synaptic density in layer III and V of human prefrontal cortex shows an overall decline between childhood (∼year 5) and adulthood (> year 35), but provides no supporting evidence for a decline from ∼ year 8 to ∼ year 20.¹⁸ Using Western blot analysis in human prefrontal cortex, another recent study found a decline in synaptophysin, a marker of synaptic vesicles, between age 10 and 16, but during the same time PSD-95, a postsynaptic marker that also reflects the strength of excitatory synapses, remained elevated.⁷⁹ Thus, more experimental evidence is needed before concluding that signifi-cant synaptic pruning occurs during adolescence in humans. The available evidence is that the decline in sleep SWA during adolescence correlates with a decline in gray matter,¹⁵ but whether this is due to synaptic pruning remains unclear, because synapses occupy only a small portion of the gray matter, and structural MRI studies have shown that during adolescence the decline in gray matter is associated with an increase in white matter, which, most notably in the frontal lobes, accounts for a net brain growth.⁸⁰ Thus, increased myelination at the border between gray and white matter could account for the decrease in gray matter as seen by MRI.⁸¹ In this context, it is worth noting that during adolescence the gray matter/white matter ratio also correlates with SWA,¹⁵ and in young adults, SWA correlates with the white matter of the corpus callosum, but not with gray matter.⁸²

In conclusion, the present results indicate that in mouse cortex the observed developmental decrease in sleep SWA cannot be accounted for simply by a net pruning of synapses. Thus, other causes for the decline in SWA during adolescence must be considered, by extending the investigation to both gray and white matter, and especially by assessing the role of cortical refinement.

DISCLOSURE STATEMENT

This was not an industry supported study. The study was supported by NIMH (1R01MH091326 to Drs. Tononi and Cirelli). Dr. Tononi has consulted for Philips Respironics and has been involved in a research study in humans supported by Philips Respironics. This study is not related to the work presented in the current manuscript. The other authors have indicated no financial conflicts of interest.

SUPPLEMENTAL MATERIAL

Figure S1

Absolute high frequency power does not change across development. Age related values of absolute NREM high frequency power (15-30 Hz), Frontal (left) and Parietal (right). Each symbol refers to a single animal, and each mouse contributed only with one value, collected ≥ 5 days after surgery. Black dots represent YFP-H mice only used for EEG recordings. Red circles refer to GFP-S mice (n = 9) used first for EEG recordings, and then sacrificed the next day to perform confocal imaging. YFP-H mice (n = 4) and GFP-S mice (n = 5) that were weaned early are shown with open black and red circles, respectively; in these animals SWA values were collected 3 days after surgery (see Methods for details). X refers to older adult YFP-H mice that were recorded well beyond the adolescent developmental window that is the focus of this study.

Figure S2

Changes in EEG spectra across age. Age related change in the EEG spectra (24 h mean), Frontal (left) and Parietal (right), is shown for specified frequency bands and vigilance states. (A) normalized NREM Theta (4-7/15-30 Hz) (Frontal: -0.101 per day R² = 0.242; Parietal: -0.077 per day R² = 0.140). (B) normalized NREM Alpha (8-12/15-30 Hz) (Frontal: -0.048 per day R² = 0.378; parietal not significant). (C) normalized REM Alpha (8-12/15-30 Hz) (Frontal: 0.027 per day R² = 0.150; Parietal: 0.0924, R² = 0.330). normalized Wake Alpha (8-12/15-30 Hz) (Frontal: 1.20 per day R² = 0.334; Parietal: 0.071, R² = 0.415). Line represents the linear regression for frequency band vs. postnatal age for each panel. It was determined using only the values from black and red solid circles, which represent YFP-H mice used for EEG recordings and GFP-S mice used first for EEG recordings and then for confocal imaging, respectively. Additional mice are shown as reference but were not used to determine the regression line: YFP-H mice (n = 4) and GFP-S mice (n = 5) that were early weaned are shown with open black and red circles, respectively; in these animals SWA values were collected 3 days after surgery (see Methods for details). X refers to older adult YFP-H mice that were recorded well beyond the adolescent developmental window that is the focus of this study.