Haplotype of the astrocytic water channel AQP4 modulates slow wave energy in human NREM sleep

Cerebrospinal fluid (CSF) flow through the brain parenchyma is facilitated by the astrocytic water channel aquaporin 4 (AQP4). Homeostatically regulated electroencephalographic (EEG) slow waves are a hallmark of deep non-rapid-eye-movement (NREM) sleep and have been implicated in the regulation of parenchymal CSF flow and brain clearance. The human AQP4 gene harbors several single nucleotide polymorphisms (SNPs) associated with AQP4 expression, brain-water homeostasis and neurodegenerative diseases. To date, their role in sleep-wake regulation is unknown. To investigate whether functional variants in AQP4 modulate human sleep, nocturnal EEG-recordings and cognitive performance were investigated in 123 healthy participants genotyped for a common eight-SNP AQP4-haplotype. We show that this AQP4-haplotype is associated with distinct modulations of NREM slow wave energy, strongest in early sleep and mirrored by changes in sleepiness and reaction times during extended wakefulness. The study provides the first human evidence for a link between AQP4, deep NREM sleep and cognitive consequences of prolonged wakefulness.


Introduction
Glial-dependent cerebrospinal fluid (CSF) flow through the brain parenchyma, by some termed the glymphatic system, facilitates a circulation of nutrients and removal of waste by generating a convective flow of CSF and interstitial fluid (1,2). Evidence suggests that the pathway relies on three main processes: Firstly, bulk flow of CSF through perivascular spaces is generated by arterial pulsations from the heartbeat and respiration (3,4).
Secondly, movements of CSF from the perivascular space into the brain parenchyma relies on the water channel aquaporin 4 (AQP4), which is highly expressed on astrocytic vascular endfeet (1,5). Mice lacking AQP4 show a strong reduction in parenchymal CSF influx (5,6) and increased interstitial beta-amyloid depositions (7), which is ameliorated by sleep deprivation (8). Thirdly, the inward flow of CSF through AQP4 channels mainly occurs during non-rapid eye-movement (NREM) sleep (9) and in preclinical studies, glymphatic flow is known to be positively correlated with slow wave production (10).
Emerging data support the existence of sleep driven CSF movements and clearance in the human brain.
Increased levels of intracerebral tau and ß-amyloid have been observed in healthy adults after sleep loss (11,12), and recently sleep driven parenchymal CSF pulsations in the fourth ventricle were demonstrated in the human brain (13), providing the first evidence for human glymphatic mechanisms. To date, however, no studies have described the link between AQP4 and human sleep-wake regulation or investigated whether genetic modulation of AQP4 may have restorative effects on cognitive functions after sleep loss.
The gene encoding AQP4 is located on chromosome 18 (18q11.2-q12.1) (14) (Figure 1). Several thousand single nucleotide polymorphisms (SNPs) in the non-coding regions of AQP4 have so far been identified and their function(s) in the normal and diseased human brain is an active area of research. Human AQP4 SNPs have been shown to impair cellular water permeability and water homeostasis in vitro (15). Moreover, an array of studies have associated human AQP4 SNPs with neurological disorders including Alzheimer's disease progression (16), vascular depression phenotype (17), leukariosis (18), outcome after traumatic head injury (19), edema formation (20,21) and the risk of stroke (22). These findings suggest a link between AQP4 and the development of brain diseases associated with waste deposition and fluid movements. Recently, a single variant within AQP4 was associated with a 15-20% change in AQP4 expression (23).  (23), we hypothesized that the high AQP4 expressing variant (HtMa; black) presents with improved glymphatic flow compared to the low AQP4 expressing HtMi variant (red). Assuming that NREM slow waves are the endogenous regulator of glymphatic flow, the HtMa variant require less slow wave energy (SWE) to initiate glymphatic flow than the HtMi variant, establishing an innate glymphatic-sleep feedback loop. (B) Physical map of the AQP4 gene and the location of the 8 haplotype SNPs. The three SNPs genotyped in this sample are marked with full red arrows. Dark green blocks: coding exons, Light blue blocks: 5′-and 3′-untranslated regions. (C) Linkage disequilibrium (LD) block among the AQP4 SNPs in the investigated haplotype. The pairwise LD coefficients (r 2 ) of SNPs in the LD block are color-scaled in red tones with dark red indicating perfect LD (r 2 = 1). (D) Table shows bases at the 8 different SNP locations in the AQP4-gene for the two haplotypes HtMa (76.7%) and HtMi (19.7%), and their respective frequencies in the CEU and TSI populations, representative of the investigated Swiss cohort. 3.7% of CEU and TSI are predicted to be carriers of rare haplotype variants (42). Frequencies in investigated study population were close to the prediction (HtMa: 75.7%; HtMi: 23.5; others: 0.7%). (E) Visualization of study design common for all subjects included from six separate studies. After a minimum three-day inclusion period with monitored bedtimes and no caffeine intake, all study participants underwent an adaptation night in the laboratory before baseline sleep, 40 hours prolonged wakefulness and a recovery night, adding up to more than 1950 hours (123 x 2 x 8h) of included sleep EEG recordings. Subjective sleepiness ratings and the ~10 min psychomotor vigilance test (PVT) were performed at three-hour intervals.
Here, we aimed to investigate the role of AQP4 in human sleep-wake regulation. We hypothesized that if NREM slow waves are the endogenous regulator of CSF brain pulsations then a reduced expression of AQP4 should be associated with a compensatory increase in deep NREM sleep ( Figure 1A). To investigate this association, a haplotype spanning AQP4 was genotyped in a sample of 123 healthy participants from controlled sleep deprivation studies. Data from all-night electroencephalographic (EEG) recordings in baseline-and recovery-sleep, as well as measurements of subjective sleepiness and global alertness throughout 40 hours of prolonged wakefulness were analyzed ( Figure 1E).

The AQP4-gene harbors an eight-SNP haplotype associated with AQP4 expression
Initial examination of SNPs in the AQP4-gene revealed a conserved haplotype spanning the entire gene with two common variants ( Figure 1B). This haplotype consists of eight SNPs, including rs335929 implicated in cognitive decline in Alzheimer's patients (16) and rs162008 demonstrated to reduce AQP4 expression (23).
Based on overall variance in EEG slow wave energy (SWE; 0.75 -4.5 Hz) and the a priori aim to detect an effect size of at least 5%, power analysis suggests a required sample size of 78 (39 per group) as sufficient (see methods). The AQP4-haplotypes were compared by means of dominant analysis. Fifty-two subjects were carriers of the minor allele (HtMi) and 71 individuals were homozygous for the major allele (HtMa). The two groups did not differ in demographic characteristics (Table S1 & Table S2), presented with similar sleep architecture and had a normal response to sleep deprivation (Table S3).

Figure 2. AQP4 haplotype modulates EEG energy in the slow wave range
Comparison of EEG-energy (EEG-power x time) across baseline and recovery nights in the slow wave range (0.75-4.5 Hz) within the AQP4 haplotype variants HtMa homozygotes (black) and HtMi-carriers (red). To minimize false positive results, EEG data was analyzed by a hypothesis driven fixed sequence procedure that only revealed significant effects of AQP4 in the whole night slow wave band, which was significantly increased in the AQP4 HtMi carrier group when compared to HtMa homozygotes (A; 'genotype':

AQP4-haplotype modulates slow waves in NREM sleep
The sleep EEG is genetically determined, with NREM sleep exhibiting up to 90% heritability (24), making it one of the most hereditary human traits described. To investigate whether the AQP4-haplotype modulates homeostatic sleep-wake regulation, EEG energy in predefined frequency bands in NREM sleep in baselineand recovery nights and the evolution of subjective sleepiness as well as cognitive performance measures were compared between the AQP4-haplotypes by a fixed sequence procedure (25) (see Methods). EEG SWE, which is a combined measure of sleep intensity and duration, and one of the best validated markers of sleep propensity in humans (26), was defined as the primary outcome variable. EEG quantification revealed that the HtMicarriers produced more SWE than the HtMa homozygotes ('genotype': P < 0.03; Figure 2A). This effect was not observed in the spindle range or any other frequency band (data not shown), nor in REM sleep ( Figure S1).
The demonstrated AQP4-haplotype modulation of SWE documents an association between the intensity of deep NREM sleep and the expression of the AQP4-water channel, a relationship that may be central for CSF driven brain pulsations. Given how recent preclinical evidence show that the intensity of slow waves is directly linked to glymphatic influx (10), and that the complete removal of AQP4 in mice results in brain impairments after sleep deprivation (8), this suggest an important role of AQP4-mediated clearance during sleep. The results match our initial hypothesis suggesting that in an attempt to compensate for a reduced AQP4 expression (23), AQP4 HtMi-carriers have a stronger parenchymal CSF flow and increased SWE in NREM sleep ( Figure 1A).
Interestingly, the AQP4-haplotype modulation was similar in baseline-and recovery nights ('genotype x night': F 1,120 = 0.37; P > 0.55; h p 2 = 0.31%), proposing that the AQP4 modulation is present both under normal sleep conditions and following the sleep homeostatic challenge.
To localize the AQP4-dependent effect in the slow wave range, bin-wise frequency analysis was performed and revealed significantly higher energy in the 0.75 -2 Hz range for the HtMi carriers than the HtMa homozygotes ( Figure 2 B-C).

EEG markers of sleep homeostasis in NREM sleep are modulated by the AQP4-haplotype
To investigate whether the AQP4-haplotype modulates the well-known homeostatic decline of slow waves across sleep, EEG SWE was quantified across the first four NREM sleep episodes in baseline-and recovery

The AQP4-haplotype modulates subjective and objective responses to prolonged wakefulness
To probe whether the AQP4-haplotype modulation of SWE has cognitive consequences, we investigated psychomotor vigilance performance (PVT) and subjective sleepiness (Stanford sleepiness scale, SSS) across prolonged wakefulness in the two genetic groups (see Figure 1E). Subjective sleepiness ratings revealed that HtMi-carriers coped slightly better with sleep deprivation than HtMa homozygotes and showed a smaller increase in sleepiness ratings from day 1 to day 2 ('haplotype x day': P < 0.04; Figure 4A). Importantly, median response speed on the PVT mirrored the effects on subjective sleepiness with the HtMi carriers reducing their  speed slightly less than HtMa homozygotes ('haplotype x day': P < 0.04; Figure 4B), despite comparable speeds on day 1. Effects for lapses of attention were visually similar yet did not reach significance ( Figure   4C). These data suggest that alterations in AQP4-dependent parenchymal CSF flow also have cognitive consequences, unveiled during sleep deprivation. Our data extends the recently described connection between slow waves, neuronal activation and CSF flow in the 4th ventricle (13), by showing that AQP4 haplotype in turn modulate NREM slow wave energy and the restorative effect of sleep on cognitive functions. These observations may indirectly suggest that AQP4 dependent fluid flow within the neuropil is regulated by EEG slow waves during NREM sleep.

Conclusion
Our data highlights that subjects carrying the low AQP4-expressing HtMi variant have enhanced SWE mainly in early NREM episodes and cope slightly better with prolonged wakefulness. Given that SNPs associated with the HtMi variant also affect the cognitive decline in Alzheimer's patients (16), our data provides novel evidence for the existence of sleep dependent AQP4-driven CSF pulsations in the human brain consistent with proposed glymphatic mechanisms. It also supports the hypothesis that sleep slow waves are part of the regulatory machinery of parenchymal CSF flow. Further studies investigating the AQP4-haplotype and its association to sleep-associated brain functions are warranted.

SNPs of the AQP4 gene and haplotype analysis
To investigate genetic modulations of AQP4, common variants in the AQP4-gene were explored using the dbSNP database build 152 (https://www.ncbi.nlm.nih.gov/snp/). The initial search in the 1000 genome project database within dbSNP revealed 32 SNP's with a global minor allele frequency (MAF) above 5%.
Only 16 of these were common (MAF above 20%) in the European population data representative of the investigated Swiss cohort (CEU and TSI populations). Linkage disequlibrium (LD) analysis showed that eight of the 16 SNPs (rs162007, rs162008, rs63514, rs455671, rs335931, rs335930, rs335929, and rs16942851) form a distinct haplotype with SNPs in high LD (r 2 > 0.8) ( Figure 1C). Further LD analysis of the haploblock revealed that these eight SNP's form two common variants of the haplotype: a major haplotype (HtMa) with a 76,7% incidence and a minor haplotype (HtMi) with a 19.7% incidence (for further details see Figure 1). Based on the 1000 genome database, this haplotype is widely detected across the European, American, east and south Asian populations.

Genotyping of APQ4 SNPs
Genomic DNA extracted from 3 ml fresh EDTA-blood (wizard R Genomic DNA purification Kit, Promega, Madison, WI) was used for genotyping. The rs335931, rs335929 and rs16942851 polymorphisms of AQP4 were chosen as tag-SNP's to represent the haplotype and were determined using Taqman® SNP genotyping Assay (Life Technologies Europe B.V.; see also Table S2). Allelic discrimination analysis was performed with SDS v2.2.2 software (applied Biosystems, Foster City, CA, USA.) All genotypes were replicated at least once for independent confirmation.
The MAFs of the genotyped variants were in accordance with MAFs predicted by the dbSNP database (Table   S2). All three SNPs were in Hardy-Weinberger equilibrium. Pairwise LD coefficients (r 2 ) were calculated between rs335929, rs16942851 and rs335931 confirming high linkage disequilibrium (r 2 > 0,95). The two haplotypes and allele frequencies are shown in Figure 1.
Due to the well-established association between Alzheimer's disease risk and Apolipoprotein E (APOE) genotype (27), we checked the distribution of APOE genotypes (rs429358 and rs7412) among the AQP4 haplotype groups using the same Taqman® SNP genotyping approach. The analysis revealed that the distribution was similar in HtMa-homozygytes and HtMi-carriers (p > 0.47; Table S1).

Study population
To examine the impact of the genetic haplotype of AQP4 on the sleep EEG, we investigated data from 134 healthy participants of six previously published sleep deprivation studies (28)(29)(30)(31)(32)(33) All studies were conducted under strictly controlled conditions in the sleep lab of the Institute of Pharmacology and Toxicology at the University of Zürich, Switzerland using similar protocols and methodology ( Figure 1E). Two carriers of rare haplotypes as well as 9 older participants with an age above 60 years were excluded from the analysis. The No difference in the distribution of the AQP4-haplotype between the six studies was observed (p > 0.21). In studies that included the administration of one or more treatments (28)(29)(30), only data from the placebo-arm were analyzed.
Study participants were right-handed healthy volunteers with a medical history free of neurological and psychiatric disorders. They were drug-and medication abstinent and reported being good sleepers with regular bedtimes and no shift-or nightwork. No participants passed through time zones or consumed excessive amounts of alcohol or caffeine in the two months prior to study-enrollment. Before inclusion, participants underwent a screening night in the sleep laboratory to check for undiagnosed sleep disorders or low sleep efficiency (<85%) (see Table S1 and S3).
The study protocols were approved by ethics committee of the Canton of Zurich for research on human subjects. Written consent was obtained from all participants before the experiments.

Sleep study protocol
The six study protocols were very similar and were performed as follows: In the final 3 days leading up to the sleep studies, participants were required to keep a strict 8-hour/16-hour sleep schedule and to refrain from caffeine (coffee, tea, cola drinks, chocolate and energy drinks) and alcohol intake. Compliance with these requirements was verified by actigraphy from a wrist-activity monitor, sleep-wake diaries and determination of saliva caffeine as well as breath alcohol levels upon arrival in the sleep lab.
The sleep studies consisted of a block of four consecutive nights (see Figure 1): First and second nights served as adaptation and baseline-nights, respectively. The subjects were then kept awake for 40 hours (i.e. for two days, skipping one night of sleep) until bedtime on the fourth night, where they were given a 10h sleep opportunity for recovery. During the period of prolonged wakefulness, the participants were constantly supervised by members of the research team and engaged in studying, playing games, watching films, and occasionally taking a walk outside the laboratory.
The data from the C3M2 derivation are reported. In both conditions, the analyses were restricted to the first 8 hours (480 min) after lights-off.

EEG analyses
Four second EEG spectra (fast Fourier transform routine, Hanning Window, frequency resolution 0.25 Hz) were calculated with MATLAB (MathWorks Inc., Natick, MA), and EEG power spectra of 5 consecutive 4 second epochs were averaged and matched with the scored sleep stages.
The first four NREM episodes were defined according to current standards (35).

Cognitive testing and sleepiness ratings
The psychomotor vigilance test (PVT) is a simple reaction time task implemented in e-Prime software (Psychology Software Tools Inc., Pittsburgh, PA), in which subjects are instructed to press a button as quickly as possible with their right index finger when they see a digital millisecond counter that starts to scroll in the center of the computer screen (36). Nineteen individuals were excluded from analysis because they performed a different, non-computerized version of the task, resulting in a sample size of 104 subjects for the PVT task analyses. Moreover, because a subset of participants underwent neuroimaging on the second day of sleep deprivation, some performance measures are missing on day 2. Subjects received oral instructions and performed a training session prior to study start. For each PVT trial, 100 stimuli were presented (random interstimulus intervals: 2 -10 s). Two extensively validated PVT variables were quantified (37,38): 'lapses of attention' (defined as the percentage of trials with reaction times longer than 500 ms) and median response speed (based on inverse reaction times). Immediately prior to all PVT assessments, a validated German version of the Stanford Sleepiness Scale was administered (39). The sleepiness ratings of all 123 subjects were included in the analyses.