Fasting (3–4 h) capillary ammonia concentrations were measured during the fourth week using an Ammonia Checker (Menarini Diagnostics, Firenze, Italy). The pre-warmed foot was punctured with a 25G needle, 20 μl of blood obtained via a capillary tube and transferred to an ad hoc reagent strip. Ammonia concentrations were then determined by a reflectance meter (Huizenga et al., 1995).

Animals were habituated to handling at least 4 days prior to surgery, which was performed during the fourth week. Under general anesthesia, they were implanted with electroencephalography/electromyography (EEG/EMG) electrodes and a microdialysis guide cannula. They were anesthetized with isoflurane (IsoFlo®Vet 100%, Abbott Laboratories Ltd, England) (5% induction, 2% maintenance) and placed in a stereotaxic device. Before the surgery they were injected with buprenorphine (Temgesic®, Indivior UK Limited, Slough, UK, 0.05 mg/kg, s.c.). After exposing, cleaning and disinfecting the skull bone, two gold-coated screws were fitted into the skull for frontal-parietal epidural bipolar recording of the EEG, while two silver wire electrodes were inserted into the neck muscles for EMG recording. A unilateral guide cannula (CMA11 Guide Cannula; CMA/Microdialysis, Stockholm, Sweden) was implanted targeting the basal forebrain cholinergic area (Porkka-Heiskanen et al., 2000) for subsequent insertion of the microdialysis probe. Electrodes and guide cannula were fixed to the skull with acrylic dental cement. After surgery, animals were individually housed in open Plexiglas boxes, connected to the recording cables through swivels allowing them to move freely. Rats were left for adaptation and recovery for at least 5 days.

Each rat underwent adenosine sample collection by in vivo microdialysis, and EEG/EMG recording during three consecutive days [pre-baseline (pre-BL), baseline (BL), and sleep deprivation (SD) day] (Figure 1). A microdialysis probe (CMA11 Microdialysis probe; 14 mm shaft length, 2 mm dialyzing membrane length, 0.24 mm diameter; CMA/Microdialysis) was inserted through the guide cannula into the basal forebrain (Porkka-Heiskanen et al., 2000) and inlet and outlet tubing (1.5 m pieces of fluorinated ethylene propylene tubing; CMA/Microdialysis) connected to the probe the day before the beginning of the experiment. Artificial cerebrospinal fluid (NaCl 147 mM, KCl 3 mM, CaCl₂ 1.2 mM, MgCl₂ 1 mM, pH 7.2) was continuously perfused at a flow rate of 1 μl/min. Thirty minute dialysate samples were automatically collected for 11 h during the light period from 09:30 to 20:30 h and stored at −20°C.1.

EEG/EMG recordings were sampled at 277.78 Hz with a resolution of 8 mV/bit. Signals were amplified (gain EEG: 5,000; EMG: 2,000), analogically filtered (EEG: 0.3 Hz high-pass, 100 Hz low-pass; EMG: 10 Hz high-pass, 100 Hz low-pass), monitored on-line during the experiments and stored on a computer equipped with the Spike 2 software (version 5.19; CED, Cambridge, UK) for off-line analysis.

On the third day, animals were sleep-deprived for 6 h, starting at 11:00 h (Figure 1), by gentle handling (Franken et al., 1991), which includes presentation of new objects into the cage and tactile stimulation, such as light touches of a brush/hand. After sleep deprivation, microdialysis sampling and EEG/EMG recording were continued and animals left undisturbed to allow for recovery sleep (Figure 1).

At the end of the recovery period, rats were sacrificed with a lethal dose of pentobarbital (Mebunat 60 mg/kg). The location of the tips of the probes was marked by injecting color ink through a modified microdialysis probe. Brains were removed and immediately frozen. Twenty micrometer coronal sections were cut using a cryostat, mounted on a glass plate and stained with toluidine blue. The location of each probe was verified by visual inspection under a light microscope, using a rat brain atlas (Paxinos and Watson, 1998). Only those rats with probe locations in the close vicinity of the target area, including the horizontal diagonal band of Broca, substantia innominata, magnocellular preoptic area, lateral preoptic area and the basal nucleus of Meynert, were included in subsequent analyses (Figure 2).

The dialysate samples were analyzed with a microbore high-performance liquid chromatography (HPLC) system coupled to a fluorescent detector (Waters 2,475), as previously described (Kalinchuk et al., 2015). The mobile phase, consisting of ammonium acetate 50 mM, tetrabutyl-ammonium hydrogen sulfate 0.2 mM, EDTA 1 mM in 15% of MeOH, pH 5.6, was passed through the column at a flow rate of 1 ml/min. Ten microliters of the first 30 min sample of every hour from BL and SD days were injected through a Kinetex® C18 column (100A, 150 × 4.60 mm, particle size: 5 micron, Phenomenex, USA), coupled with a pre-column (KJO-4282, Phenomenex, USA). Adenosine was detected at a wavelength of 460 nm and its concentrations were determined by comparing sample peak areas with those of standards (AMP, adenosine, cAMP, cdAMP, Sigma-Aldrich) using LabSolutions software (Shimadzu). Detection limit: 0.1 nM; linear range of the assay: 0.1–50 nM; coefficient of variation < 10%. Any outliers were determined and treated according to a modified Thompson tau Method.

One of the main features of this microdialysis technique is that the sample is collected through a membrane, which cannot be penetrated by large molecules, i.e., enzymes (Porkka-Heiskanen et al., 1997). Thus, the degradation of adenosine in the collected microdialysis samples is completely dependent on the stability of the molecular structure, not on the degrading enzymes (as opposed to what happens in tissue samples, where adenosine is instantly metabolized). The recovery rate of adenosine, measured from known concentrations of adenosine solutions, is of the order of 10%, depending on flow rate, membrane, temperature and the concentration of the original solution.

A subset of the dialysate samples (n = 69) were analyzed in 5 control and 5 hyperammonaemic animals using the AbsoluteIDQ® p180 targeted metabolomics kit (Biocrates Life Sciences AG, Innsbruck, Austria) on a Waters Xevo TQ-S mass spectrometer (MS) coupled to an Acquity UPLC system (Waters Corporation, Milford, MA, USA; Davies et al., 2014). We took two samples per each animal during BL day: one at ZT3.5 and another one at ZT8, which correspond to the time points of beginning and middle of SD during SD days. This was done to evaluate metabolites during SD (i.e., BL vs. SD analysis). Then we took three samples per each animal during SD: at ZT4, ZT5.5 and ZT7.5. These correspond to the beginning (2 h after beginning of SD when, according to our experience, changes start to occur, i.e., adenosine rise), middle and the end of the SD. Finally, two samples of recovery period were taken from each animal: at ZT8.5 and ZT9.5 to evaluate metabolites immediately after finishing SD (during the first hour of recovery), and later on, closer to the end of light period. The dialysate samples (10 μl) were prepared according to the manufacturer's instructions adding several stable isotope–labeled standards to the samples prior to derivatisation and extraction. Sample order was randomized and all samples were run on a single 96-well plate including 3 levels of quality controls. Up to 183 metabolites from 5 different compound classes (acylcarnitines, amino acids, biogenic amines, glycerophospholipids and sphingolipids) can be quantified using either LC/MS or flow injection analysis/MS. Metabolites were selected if the % valid or % semiquantitative values were above 60% (average 76%). In addition, putrescine (35%), acetylcarnitine (C2) (40%), DOPA (46%), and SDMA (49%) were included since their values were above zero but below the lower limit of quantification.

All EEG traces were digitally high-pass filtered (1.4 Hz cut-off). EEG recordings were then semi-automatically scored in 4 s epochs using AUTOSCORE script, version 1.7 (Rytkönen et al., 2011) and manually checked using the SLEEPSCORE script, version 1.01 (CED). Vigilance states were scored according to standard criteria (Kalinchuk et al., 2015). Briefly, NREM sleep was recognized by high-amplitude EEG associated with slow waves in the delta range (0.5–4 Hz) and low-voltage EMG; REM sleep was recognized by low-amplitude, high-frequency EEG associated with the absence of EMG except for whiskers and ears twitches, as well as EEG theta activity (4–9 Hz); wakefulness was recognized as low-amplitude, high-frequency EEG activity with high-voltage EMG. The time spent in wake/NREM/REM was calculated over the total recording time of BL day, and hourly values compared between the two groups. The number of state-specific bouts per hour and the bout length distribution over the 24-h BL day were also obtained. State-specific time during the 16 h of recovery of the SD day were compared to the corresponding values at baseline, on an hourly basis.

Data are presented as means ± SD or SEM. Baseline ammonia and point adenosine/metabolite values were compared by the Student t or Mann-Whitney U-test, as appropriate. Baseline adenosine/metabolite values and differences between post-sleep deprivation and baseline values were analyzed by repeated measures ANOVA (factors: time and group); p-values were corrected for multiple comparisons. Differences (sleep deprivation-baseline) were chosen over ratios (sleep deprivation/baseline) because of the presence of zero values for wake/NREM/REM in several time bins. Empty cells, if any, were treated by standard missing values replacement and analyses were performed both prior to (missing out the pertinent time points) and after replacement. Multivariate analysis of the metabolite data was performed by principal component analysis (PCA) and orthogonal partial least squares discriminant analysis (OPLS-DA) validated by permutation testing (p ≤ 0.05), using SIMCA-P v12.0 software (Umetrics, Sweden). Differences in individual metabolite levels were analyzed in R version 3.1.2 using the linear models and ANOVA methods in the stats package. Linear models were fitted to the group (control or ammonia treated) and time (n = 7), with the animal as covariate. Significant differences for group, time and their interaction were determined using 2-way ANOVA. P-values were corrected for multiple comparisons according to the Benjamini-Hochberg False Discovery Rate (FDR). Metabolites were considered significant at a FDR cut off <0.05. For testing statistical significance, missing values were not taken into account.

Control and hyperammonemic rats spent comparable periods of time in wakefulness, NREM and REM sleep (Figure 3A). Similar state-specific time distributions over the 24 h were observed, with slight variations, particularly in REM sleep (Figure 3B). However, these did not impinge on total REM sleep time. The number of state-specific bouts per h and their length distribution were also comparable in control and hyperammonemic rats.

Following sleep deprivation, a significant decrease in wakefulness and increase in NREM and REM sleep compared to baseline values was observed in both groups [Figure 4 (n = 8 control and n = 6 hyperammonaemic animals) and Figure 5]. Direct comparisons of the sleep time increase/wakefulness decrease showed similar average values in the control and hyperammonemia groups [wakefulness: −151 ± 9 min (−26%) vs. −142 ± 14 min (−24%); NREM: 105 ± 6 min (+35%) vs. 97 ± 9 min (+33%); REM: 46 ± 4 min (+58%) vs. 46 ± 6 min (+60%)].

In a more detailed analysis, the differences between baseline and experimental day values were compared separately for each vigilance state in both control and hyperammonemic rats. This analysis revealed that the control animals exhibited a normal NREM recovery phase, lasting approximately 9 h with a gradual decline in time spent in NREM sleep (Figure 4A), while in hyperammonemic animals the primary recovery phase was shorter (as demonstrated by the statistical analysis described in Table 1), lasting only about 5 h (Figure 4B). In the first part of the recovery phase (h 9–16), the hyperammonemic animals exhibited significantly more wakefulness and less NREM sleep compared to the control animals (Table 1).

Out of 300 samples, 28 were lost for adenosine measurement. Average adenosine levels were lower (although not significantly lower) in hyperammonaemic compared to control animals at baseline (0.500 ± 0.064 nM vs. 0.683 ± 0.103 nM). Average adenosine levels rose during sleep deprivation, reaching similar values in hyperammonaemic and control animals (0.688 ± 0.103 nM vs. 0.740 ± 0.095 nM). Thus, the adenosine response to sleep deprivation was significantly greater and longer in hyperammonemic compared to control animals (Figure 6; time: F = 2.4, p = 0.012; group: F = 4.9, p = 0.047; time^*group: F = 0.5, p = 0.829).

Some metabolites were detectable (28 of 183, 15%) in the brain dialysate using targeted LC/MS metabolomics, largely comprising amino acids (n = 16) and biogenic amines (n = 8). PCA of the 28 metabolites showed a clear distinction between hyperammonemic (n = 5) and control animals (n = 5) in PC1 [amount of variance in the × matrix explained by PC1 (R²X) (cumulative) = 0.440, estimate of the predictive ability of the model (Q2) (cumulative) = 0.376]. A plot showing the mean score on PC1 (± SEM) for each sampling time point (n = 7) for the controls and hyperammonemic animals separately (n = 5 for both groups) is presented in Figure 7A. A clear difference between the controls and hyperammonemic animals across the time course (baseline; during sleep deprivation; after sleep deprivation) was observed. OPLS-DA models, validated by permutation analysis, showed separation between the control and hyperammonemic rats (Figure 7B). The p(corr) loading plot (Figure 7C) showed reduced concentrations of all of the amino acids (n = 16) except glutamate, as well as decreased taurine, t4-hydroxy-proline and acetylcarnitine levels in hyperammonemic animals. By contrast, putrescine and glutamate concentrations were higher in hyperammonemic animals. ANOVA analyses of the metabolite concentrations showed significantly reduced levels of 12 amino acids (alanine, arginine, glutamine, lysine, methionine, ornithine, phenylalanine, proline, serine, threonine, tyrosine, valine), taurine, t4-hydroxy-proline and acetylcarnitine in the hyperammonemic animals (FDR-adjusted p < 0.05). Only putrescine (FDR-adjusted p = 5.2E-10) and glutamate (FDR-adjusted p = 0.07) showed increased dialysate levels in the hyperammonemic animals compared to the controls. Of the 28 metabolites measured, only putrescine showed a significant group^*time interaction. The p(corr) values for the OPLS-DA model are presented in Table 2.

For each animal used for the targeted metabolomics analysis (n = 5 in each group), seven time series samples were analyzed [2 at BL 3.5 and 8.0 ZT (Zeitgeber Time); 3 during SD 4.0, 5.5 and 7.5 ZT and 2 during recovery (R) following SD 8.5 and 9.5 ZT]. Figure 8 presents the mean (±SEM) concentrations of the 28 metabolites at BL, SD, and R in the control and hyperammonemic animals. Analyses comparing dialysate metabolite levels at BL, SD, and R revealed fewer significant differences, partly because of low sample number/samples below the limit of detection. As a result, the metabolites arginine, glutamate, glycine, histidine, phenylalanine, proline, serine, tryptophan, tyrosine, DOPA, and SDMA could not be analyzed.

Significant changes over time in relation to the different conditions (baseline/sleep deprivation/recovery) were observed for the metabolites acetylcarnitine, t4-hydroxyproline, PC ae 30:1 and carnosine (Figure 8). In addition, while glutamine and methionine increased during sleep deprivation in control animals, they did not in hyperammonemic animals (Figure 8).

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