APOE ε4, Alzheimer’s disease neuropathology and sleep disturbance, in individuals with and without dementia

This retrospective cohort study used data obtained for participants in the Brains for Dementia Research (BDR) Programme (https://​www.​brainsfordementi​aresearch.​org.​uk/​) and held on the UK Brain Banks Network (UKBBN) database. The database holds demographic and neuropathological details of donated brains, processed and assessed according to detailed and comprehensive post-mortem protocols, as well as clinical assessments undertaken prior to post-mortem as part of the BDR project established in 2007. This links 5 brain banks across the UK (London, Oxford, Newcastle, Bristol and Manchester) with common protocols for consent, tissue handling and quality indicators. Volunteers were recruited via posters, radio adverts, presentations to groups/clubs and signposted via Alzheimer’s Research UK and Alzheimer’s Society charities. The population comprises many healthy participants with a positive family history of dementia and also participants with a diagnosis of dementia [57]. See Table 1 for inclusion and exclusion criteria.

Sleep disturbance was measured by component K of the neuropsychiatric inventory (NPI-K) [58]. These score responses provided by an informant, caregiver or study partner including increased latency, increased wake time after sleep onset, wandering, early morning wakening, excessive daytime sleep and sleep-wake cycle disturbance. A global score was obtained by multiplying frequency and severity domains (see Table 2).

Each participant had undergone post-mortem analysis of CERAD neuritic plaque stage, Thal Aβ plaque stage and Braak NFT stage allowing for calculation of the National Institute on Aging-Alzheimer’s Association ABC Score [56]. The combination of A, B and C score determine the extent of AD neuropathological change, designated “Not”, “Low”, “Intermediate” and “High”.

The primary outcome measure was determined by multivariate linear regression with NPI-K Sleep Disturbance Global Score as the dependent variable. Crude and adjusted analyses were performed including dummy variables reflecting APOE ε4 allele copy number, with 0 as reference. Covariates were introduced to an adjusted model to control for APOE ε2 allele number, age, gender, CDR sum of boxes (CDR-SOB) and neuropsychiatry inventory measures of depression and anxiety. Dummy variables were created to reflect NIAA-AA ABC neuropathological stages. All regression models were checked for multicollinearity with variance inflation factors < 10. The study is powered at 80% to detect an effect size of f² = 0.088 (n = 202, α = 0.05).

As a post hoc sensitivity analysis, these regressions were repeated in groups stratified by NIAA-AA ABC score and CDR status.

Initial database search yielded n = 728 BDR cases, of which n = 202 fulfilled our criteria for analysis (See Fig. 1).

Selected participants had a mean age of 84.0 years (SD = 9.2), 51.0% were male, mean CDR 1.8 (SD = 1.3) and mean Mini-Mental State Examination (MMSE) score 14.0 (SD = 11.8). Baseline demographics of the study population stratified by APOE-ε4 allele count (non-ε4 carriers n = 96, ε4 heterozygotes n = 95, ε4 homozygotes n = 11) are shown in Table 3. There were statistically significant differences in AD ABC stage, mean MMSE and mean CDR scores.

Crude sleep disturbance scores in the cohort stratified by ε4 are shown in Table 4. There were statistically significant increases in all neuropsychiatry inventory measures of sleep disturbance between those with 2 vs 0 alleles. Severity, frequency and caregiver distress domains were also significantly higher in those with 2 vs 1 allele. There were increased caregiver distress scores only in those with 1 allele compared with 0.

Positive trends between ε4 allele number and increasing mean NPIK sleep disturbance score were across the full-cohort irrespective of stratification by CDR status, ABC score of neuropathological change and clinical classification (Fig. 2).

Full multivariate linear regression revealed a statistically significant effect of APOE ε4 homozygosity on global scores of sleep disturbance (β 2.53, p=0.034) controlling for AD pathological status, ε2 carrier status, age, gender, depression, anxiety and CDR-SOB status. A positive trend was found for ε4 heterozygosity in both crude (β 0.67, p=0.221) and adjusted (β 0.41, p=0.471) analyses, although neither reached statistical significance (see Table 5).

Further, multivariate regression testing additional models were performed post hoc (Supplementary Material Table 1). Whilst the effect size estimates for APOE ε4 status within these models differed, overall trends and statistical significance remained unaltered. To further assess the independent effects of APOE ε4 status, further multivariate regressions were performed after stratification of the cohort by neuropathological change and CDR status (Tables 6 and 7). Positive trends between APOE ε4 status and sleep disturbance were seen in all stratified groups. Sleep disturbance was significantly associated with ε4 heterozygosity in the group without clinical dementia (CDR 0/0.5) (β 1.28, p=0.024) and with ε4 homozygosity in the relatively cognitively impaired group (CDR 1/2/3) (β 2.95, p=0.045).

APOE ε4 is thought to influence AD pathology [9] through enhanced Aβ deposition [61], tau phosphorylation and neurotoxicity [62, 63], all of which may lead to sleep abnormalities [64]. For example, in participants with AD, APOE ε4 allele status influences CSF measures of tauopathy, itself associated with night-time behaviour disturbance [65]. We have found that ε4 influences sleep independently of Aβ and tau stage. Hence, as well as its impact on these hallmark features of AD, APOE ε4 may also influence AD risk through other neurotoxic pathways and/or loss of neuroprotective functions [9] that have the potential to influence sleep quality (Fig. 3).

In a study of 85 patients with AD, APOE ε4 homozygosity compared with heterozygosity was associated with significantly reduced post-mortem CSF melatonin (32pg/ml ± 8 vs 71 ± 7, p=0.02) [66]. Reduced melatonin has been linked to sundown syndrome in dementia [67, 68], and replacement improves symptoms according to systematic review [69]. However, melatonin is up to five times higher in healthy adults compared with those with AD [66, 67], and it is plausible that APOE ε4 may be associated with reduced melatonin via secondary mechanisms linked to AD severity as opposed to direct effects.

APOE ε4 was reported to be directly associated with OSA and symptomatic sleep disordered breathing [31], possession of this allele being associated with an approximate doubling of risk for apnoea-hypoxic index > 15 [OR 2.0 (1.2–3.5)] in adults and separately in children [32]. Two main mechanisms for a causal relationship were proposed. Firstly, an ε4-associated increase in respiratory/sleep centre tau or amyloid burden may drive centrally mediated sleep disordered breathing [31]. Alternatively, (or additionally) ApoE4 has a central role in lipid metabolism [70] mediating lipoprotein to cell surface receptor binding, increasing plasma low-density lipoprotein (LDL) levels and accelerating atherogenesis [71]. Centrally mediated sleep disordered breathing is recognised in a wide range of cerebral pathologies [72], including cerebrovascular pathology that might be exacerbated by possession of ε4. A further plausible mechanism, given that APOE ε4 predisposes to metabolic syndrome [73] and increased insulin resistance [74] would be through secondary increased obesity; however, ε4 carriers on average have a lower body mass index than do ε3 or ε2 carriers [75, 76]. A systematic review found no support for a causal association between APOE ε4 allele and OSA [OR 1.13 (0.86–1.47)] [77], but the authors commented that the studies were heterogeneous, may not have accommodated important gene-gene interactions and may have been underpowered.

Previous work linked possession of ε4 with accelerated age-related cortical thickness loss [78, 79]. This itself was associated with self-reported sleep disturbance in healthy community dwelling adults [80] and reduced objectively measured total sleep time, random eye movement, N2 and N3 stages of sleep in alcohol-use disorder [81].

Baseline activity within the Default Mode Network (DMN)—the distributed network of brain regions more active during rβ and characterised by high functional connectivity—is greater in APOE ε4 carriers than in non-carriers [82‐84]. This overactivity was hypothesised to inhibit brain structures stimulating sleep initiation as described within the ‘failure to inhibit wakefulness’ hypothesis of sleep onset [85, 86]. Whilst the mechanism for this is uncertain, the findings extend to young adults aged 20–35, underlining a potential active role for ε4 outside of established AD pathology [83].

This study is subject to several limitations. Firstly, the sample, whilst deeply characterised, was limited by small numbers in the APOE-e4 homozygous group and relatively small numbers within each neuropathological group within the heterozygous group limiting power to detect full effects. Linked to this, p values of ‘statistically significant’ findings were close to 0.05. Reassuringly, however, the sleep disturbance signal was positively correlated with allele number. Participants in the BDR cohort are mostly from less socially and economically deprived parts of the UK [57] and are not therefore fully representative of the general population in the UK. Medication history was not comprehensive enough for inclusion in our analysis and may represent a significant confounder with prescribed sleep medications ameliorating or masking symptoms. Systemic illness that could have impacted on sleep disturbance may not have been detected post-mortem.

The use of the neuropsychiatry inventory sleep disturbance score as principal outcome measure is a further potential weakness. Whilst broad and encompassing a heterogeneous range of disorders, it relies on caregiver report and is therefore potentially subject to bias, e.g., false negative reports of subtle changes. However, it is well-validated, widely used and its reliance on a semi-objective caregiver as opposed to subjective personal reports also has advantages. Outcome scores for this study were obtained within 12 months of death. At the more distant end of this scale, pathological changes could have evolved between clinical data collection and autopsy; however, recall bias from retrospective data collection is eliminated.

Strengths include the categorisation of participants and quantification of AD changes based on the gold standard of post-mortem neuropathology with application of strict exclusion criteria, allowing for the effects of AD pathology to be largely determined in isolation. Data collected as part of the BDR project is standardised and collected as part of a detailed and comprehensive protocol. The population also reflected a range of AD pathology, with 38.2% of the study population recording ABC Scores of ‘Not’ or ‘Low’.