Introduction

Alzheimer’s disease (AD) is the most prevalent form of dementia, accounting for up to 80% of all cases, and hallmarks of the disease include amyloid plaques and hyperphosphorylated tau tangles in the brain [1, 2]. There is no standard approach to diagnose AD and disease presence cannot be determined by a single test but is rather a multi-faceted and multi-disciplinary approach involving taking medical history, cognitive tests, amyloid-PET scans, and sometimes cerebrospinal fluid (CSF) samples for measurement of amyloid-beta or tau [2]. However, imaging and CSF tests may not always be possible due to cost, availability of equipment, and the invasiveness of procedures which may not be well tolerated [3]. Nevertheless, the ability to confirm amyloid-beta pathology in the brain will become increasingly important with the advance of disease-modifying therapies (DMTs) targeting amyloidbeta e.g. the now discontinued aducanumab [4], lecanemab [5] and donanemab [6], as one of the requirements for prescribing these DMTs is confirmation of brain amyloid burden [7]. Thus, there is a need for acceptable, scalable, and accurate diagnostic approaches to determine disease presence, severity, and response to any treatment.

Plasma biomarkers offer an opportunity as a cost- and time-effective tool that is minimally invasive for screening and diagnosis, stratification, monitoring disease progression, and assessing treatment response. Biomarkers that have been proposed include amyloid-beta (Aβ40, Aβ42, and their ratio), phosphorylated tau (p-tau181 and p-tau217), glial fibrillary acidic protein (GFAP), and neurofilament light (NfL) (reviewed in [1]). The sensitivity and specificity of these biomarkers is an area of active research. In particular, p-tau217 has been demonstrated to be a valuable biomarker for predicting cognitive decline and monitoring treatment efficacy in response to DMT [8, 9].

Although plasma biomarkers, and particularly p-tau217 [10], show great promise as clinical tools very little is known about non-disease-related factors that may influence the concentrations of these biomarkers in blood. Biomarker levels may vary between individuals due to demographic or comorbid factors (inter-individual variation), but they may also vary within an individual due to behaviour or biological processes (intra-individual variation). Factors of interest include demographic variables such as age and sex, but also behavioural factors such as activity, posture, and eating and drinking. One factor of particular interest is the time of day since many physiological variables in blood display 24-h rhythmicity. However, to date, the impact of time-of-day has not been taken into consideration for the implementation of plasma biomarkers for dementia. The importance of time of day for diagnostic samples has already been demonstrated in other clinical conditions. For example, for people living with severe asthma, sputum samples from morning clinics have significantly higher levels of eosinophils than samples from afternoon clinics [11] which may impact clinical decision-making.

Here, we explored in a heterogenous group of participants consisting of people living with mild clinical Alzheimer’s disease (PLWA), their caregivers, and cognitively intact older adults, whether plasma levels of biomarkers of dementia-related brain changes over the course of a 24-h day. The data were collected under laboratory conditions that are similar to real-life conditions, i.e. in the presence of sleep–wake, dark–light cycle, and meals.

Materials and methods

Participants

Demographics

Data were collected from participants who were enroled in one of two studies: (1) in cognitively intact older adults and (2) in PLWA, their study partner, and cognitively intact older adults. The study protocol and eligibility criteria have previously been described in detail [12,13,14]. Briefly, eligibility was assessed using pre-defined inclusion/exclusion criteria for each of the three study groups. PLWA had to be 50–85 years old with a confirmed diagnosis of prodromal or mild clinical AD, have an SMMSE (standardised mini-mental state examination (MMSE) [15]) score ≥ 23, be living in the community and be on a stable dose of any medication for dementia for at least three months. The diagnosis of prodromal or mild AD was based on clinical history, cognitive tests, and CT/MRI imaging. PLWA could participate in the study by themselves, or they could have a ‘study partner’ who must have known them for at least six months and could be their carer or a family member or friend. Study partners were ≥18 years old and had to have an SMMSE score ≥ 27. Cognitively intact older adults had to be aged 50–85 years, have an SMMSE score ≥ 27 (Study Two), and any comorbidities and concomitant medications must have been stable for the past three months. Cognitively intact adults were recruited via the Surrey Clinical Research Facility database. Potentially eligible PLWA and their study partners were identified via Surrey and Borders Partnership NHS Foundation Trust (SABP) memory services and were approached by one of the SABP team initially by telephone to discuss the study before being provided with the participant information sheet.

Ethics approval and consent to participate

Study One (cognitively intact older adults) received a favourable opinion from the University of Surrey Ethics Committee (UEC 2019 065 FHMS), and Study Two (PLWA, caregivers of PLWA, and cognitively intact older adults) received a favourable opinion from an NHS ethics committee (London—City & East Research Ethics Committee: 22/LO/0694). Study Two is registered as a clinical study on the ISRCTN (International Standard Randomised Controlled Trial Number) registry (ISRCTN10509121). The protocols were conducted in accordance with the Declaration of Helsinki and guided by the principles of Good Clinical Practice. All personal data were handled in accordance with the General Data Protection Regulations and the UK Data Protection Act 2018. Written informed consent was obtained from participants prior to any study procedures being performed. Participants were compensated for their time and inconvenience.

Procedures and measures

The full study protocols have been reported in detail elsewhere [12, 13]. Briefly, following a screening visit to assess eligibility, participants were monitored for up to 14 days at home using a variety of technologies to assess their sleep–wake patterns, environment, and cognitive function. They then attended the UK-DRI Clinical Research Facility at the University of Surrey for a 27-h residential session which included a full clinical polysomnography (PSG) recording during an extended 10-h period in bed. PSG was recorded using the Somnomedics SomnoHD system with Domino software (v 3.0.0.6; sampled at 256 Hz; SOMNOmedics GmbHTM, Germany) with an American Academy of Sleep Medicine standard adult montage. Habitual bedtime was determined from the information provided in the Pittsburgh Sleep Quality Index (PSQI) [14] and the PSG metrics have been previously reported for Study 1 [12].

During the residential session, participants remained in environmentally controlled bedroom environments with en-suite facilities. For PLWA and their study partners, the aim was to recreate their sleeping situation at home so they could either share a room in a double occupancy suite or be in adjacent rooms with an interconnecting door. During the afternoon/evening/morning hours, participants were free to pursue their own activities around scheduled procedures including sample collection, meals, questionnaire completion, and having PSG equipment attached.

Study one

Participants had two blood samples drawn 12 h apart via venepuncture at 19:46 ± 00:33 h and 07:53 ± 00:35 h (mean ± SD). The evening sample was 2.88 ± 0.80 h before lights off, and the morning sample was 0.68 ± 0.92 h after lights on. Dinner was scheduled ~5 h before habitual bedtime (at approximately 18:30 h) and breakfast ~2 h after habitual waking (at approximately 09:30 h). Thus, the evening sample was taken after dinner and the morning sample before breakfast.

Study two

Participants had an indwelling cannula sited and blood samples were drawn at three-hourly intervals relative to their habitual bedtime. Sampling began 9 h before habitual bedtime and continued until 15 h after habitual bedtime. Lunch was ~9.5 h and dinner was ~4 h before habitual bedtime; breakfast was ~1.5 h after habitual waketime. Meals varied in content due to individual preference, but the relative sizes of each meal were consistent between individuals.

Blood samples were collected into K2 EDTA Vacuettes which were centrifuged within 10 min of collection at 4 °C at 1620 × g for 10 min. The plasma fraction was separated and stored at −80 °C. Samples were shipped to the UK DRI Biomarker Factory, UCL, London where they were analysed using Simoa HD-X technology. The following biomarkers were measured in both studies using the Neuro 4-PlexE assay kit (Quanterix, Billerica MA): amyloid-beta 40 (Aβ40), amyloid-beta 42 (Aβ42), GFAP, and NfL. Tau phosphorylated at threonine 217 (p-tau217) was measured using the ALZpath Simoa assay and measured in Study Two only (ALZpath, Carlsbad, CA). Samples were measured blind in singlicate, and four internal controls made of pooled plasma were used to monitor any intra-and inter-plate variation. All coefficients of variation for internal assay controls were below 10%.

Data analysis

For each biomarker, the mean values at each timepoint were computed and also intraclass correlations (ICCs) were calculated using R Statistical Software (v4.2.2; R Core Team 2022).

To assess time of day effects we created two analysis sets. Analysis set 1: we combined the evening and morning samples from study 1 and study 2 which allowed for a comparison of evening vs morning samples. For study 2, we used the samples 3 h before and 9 h after habitual bedtime. Analysis set 2: this consisted of the samples collected at the 9 time points in study 2. For both analysis sets, a PROC MIXED linear model (SAS v9.4, SAS Institute Inc) was run in which participant was the random effect with factors time-of-day, group, and their interaction. A second PROC MIXED linear model was run on the two analysis sets to investigate any effects of covariates: age, sex, BMI, PSQI, and PSG apnoea–hypopnea index (AHI) in addition to the effects of time of day and group.

Results

Here we report data from 38 participants (Supplementary Table 1) whose comorbidities included hypertension, Type-2 diabetes, arthritis, hyperthyroidism, and asthma [12, 13]

For both studies combined, 90% of scheduled samples were obtained. For the nine timepoint comparisons (Study Two), the plasma levels for each biomarker at each timepoint (mean ± SD) for all participants combined, as well as separately for each group are presented in Table 1. The ICCs for all participants combined ranged between 0.84 and 0.97 for the different biomarkers and the ICC values were similar across groups. Table 2 provides a similar comparison for the two-time points (evening vs morning) comparison, and here the ICCs range from 0.76 to 0.93 for all participants combined. These ICC values imply that the between-participant variation is greater than the within-participant variation and that this is similar across groups.

Table 1 Plasma biomarker levels (pg/mL, mean ± SD) across a 24-h period.
Full size table
Table 2 Plasma biomarker levels (pg/mL, mean ± SD): evening vs morning.
Full size table

For the Study 2 dataset, with nine-time points, the model showed that there was a significant main effect of time for all biomarkers (p < 0.01) except GFAP (p = 0.065) (Table 3). Figure 1 shows plasma biomarker levels (LS means for the deviation from the mean) for all participants across 24-h. For plasma p-tau217 (LS-means) lowest values were observed in the morning, and shortly after wake time, after which levels rose to the highest values in the afternoon and evening. Thus, p-tau217 concentrations in the first two samples after wake time were significantly lower (p < 0.0001) compared to the evening (3 h before habitual bedtime) sample. For Aβ40, Aβ42 and NfL, peak levels occurred during the sleep episode and lowest levels in the morning hours. The magnitude of the diurnal variation (change in LS-means expressed as a percentage from the overall mean) was: 14.0% (Aβ40), 15.3% (Aβ42), 4.6% (Aβ42/ Aβ40), 10.6% (NfL), 17.0% (GFAP), and 15.8% (p-tau217).

Table 3 Summary of PROC MIXED analysis for plasma biomarkers over nine-time points.
Full size table
Fig. 1: Levels of plasma biomarkers (deviation from the mean LS-means ± SE) across a 24-h period: p-tau217, Aβ40, Aβ42, Aβ42/Aβ40, NfL, and GFAP.
figure 1

The grey shading indicates the habitual sleep episode.

Full size image

A significant effect of the group was observed for p-tau217 (p = 0.003) (Fig. 2) with the highest levels observed in PLWA, and the effect of the group approached significance for GFAP (p = 0.069). A significant group-by-time interaction was not observed for any of the biomarkers. For p-tau217, the magnitude of the diurnal variation in PLWA estimated from the LS-means (0.233 ± 0.044, LS-mean ± SE) was 27% of the difference between the mean values for cognitively intact adults (0.349 ± 0.164, LS-mean ± SE) and PLWA (1.215 ± 0.153, LS-mean ± SE).

Fig. 2: Levels of plasma p-tau217 (LS-means ± SE) across a 24-h period in PLWA, their study partners, and cognitively intact older adults.
figure 2

Blue symbols represent cognitively intact older adults, green symbols represent study partners, and orange symbols represent PLWA. The grey shading indicates the habitual sleep episode. ***Indicates a significant (p < 0.0001) difference in levels between the indicated time points in PLWA. The data for the Study partners and the cognitively intact older adults are displaced by 15 min so that the variance indicators of the various groups are visible.

Full size image

To further establish the effects of time we compared evening to morning samples using data from both studies 1 and 2. In this comparison, data were available for 38 participants for all biomarkers except p-tau217 (n = 21). PROC MIXED analysis on the two-time points only (Table 4) revealed significant effects of time, group, and group-by-time interaction for p-tau217 only.

Table 4 Summary of PROC MIXED analysis for plasma biomarkers over two-time points.
Full size table

When the covariates (age, sex, BMI, PSQI, PSG-AHI) were added to the model and applied to the nine-time points, the effects of time remained significant for p-tau217, Aβ40, Aβ42, Aβ42/ Aβ40, and NfL. For p-tau217 the effect of group remained significant and a significant interaction between time and group emerged. A significant effect of age was observed for GFAP (p = 0.036). No significant effects of sex, BMI, PSQI or PSG-AHI were observed for any of the biomarkers (Supplementary Table 2). For the two timepoint datasets (Supplementary Table 3), a similarly significant effect of age was observed for GFAP (p = 0.006) and for p-tau217 the significant effects of time, group, and group*time remained.

Discussion

Here, we show that levels of commonly used plasma biomarkers in dementia research including p-tau217, Aβ40, Aβ42, Aβ42/Aβ40, and NfL vary with time of day. This significant variation with time-of-day was observed despite the rather large ICC values (range 0.76–0.97), which indicate that the between-participant variation is greater than the within-participant variation. The ICCs reported here are in line with previous studies which investigated the longitudinal reliability of plasma biomarkers and observed ICC values between 0.66 and 0.78 [16]. Our observed impact of age on GFAP levels is in line with previous observations in people living with Parkinson’s disease where GFAP was shown to correlate with both age and MMSE [17].

The mean values of p-tau217 observed ranged between 0.32 and 0.62 pg/mL for cognitively intact participants and study partners, and between 1.1 pg/mL and 1.4 pg/mL for PLWA. These ranges are in line with those previously reported where <0.40 pg/mL indicated a negative p-tau217 result and >0.63 pg/mL a positive result [18].

We observed significant time-of-day variation for p-tau217, NfL, Aβ40, Aβ42, and Aβ42/Aβ40 with the effect approaching significance for GFAP, with the magnitude of diurnal variation ranging from 4.6% to 15.8% for the significant effects. Previous work has demonstrated that cerebrospinal (CSF) levels of amyloid-beta fluctuate with time of day [19,20,21]. The observed diurnal fluctuations for Aβ40 and Aβ42 were 2.6% and 0.4%, respectively, for amyloid-positive participants, with the highest values in the early afternoon and lowest values upon waking [20]. This compares to a 14.0% for Aβ40 and 15.3% for Aβ42 diurnal variation in plasma observed in the current study.

The exact shape of the diurnal variation varied across the biomarkers, but the lowest values were in general observed in the morning. For p-tau217 highest values were observed before bedtime and the lowest values upon awakening. For Aβ40 and Aβ42, we observed the highest values during the nocturnal sleep period and the lowest values upon waking. A previous study of Aβ40, and Aβ42 in CSF showed levels were lower in the morning with the highest values in the afternoon [20]. For NfL, the highest values were also observed during the sleep period with the lowest values in mid-morning and relatively stable levels in the afternoon/evening and morning. Larger sample sizes are needed to further determine the precise shape of this diurnal variation and differences therein across the biomarkers.

The factors underlying the observed diurnal variation remain to be identified. They could be related to circadian modulation of production, phosphorylation, and clearance from the brain or could be a response to behavioural changes/processes across the 24-h day including sleep, meals, or posture. In the latter case, simple behavioural constraints could remove the variance and samples could be taken at any time, whereas in the former case, samples should be taken within particular time windows or values should be corrected for the time of day. The observed differences in the shape and timing of the diurnal variation across the biomarkers make it unlikely that one common mechanism, such as changes in blood volume, or circadian or sleep-mediated clearance from the brain into the circulation drives all of this diurnal variation.

Although the time-of-day effects we observed may appear small, when they are placed in the context of disease or treatment monitoring, they become of clinical interest.

For example, plasma p-tau217 has recently become a biomarker of interest in AD research due to its sensitivity for discriminating for AD, its ability to predict cognitive decline, and its capacity to track response to DMT [8, 9, 22]. Of particular interest is a study in cohorts of Aβ positive individuals (n = 171) who were cognitively unimpaired [8]. In this study, cognition was assessed using the MMSE and the modified preclinical Alzheimer Cognitive Composite (mPACC) over a median of six years. Plasma p-tau217 was shown to be the strongest biomarker for predicting cognitive decline and also conversion to AD [8]. Of particular relevance to our findings is that longitudinal monitoring in those with Aβ-positive prodromal AD showed an increase in p-tau217 of 14.7% per year [23]. This is very similar to the magnitude of the diurnal variation (15.8%) observed in the current study. This change is also meaningful when we consider that in the TRAILBLAZER-ALZ clinical trial following treatment with donanemab for up to 72 weeks, plasma-tau217 levels declined by 23% [9] and GFAP levels decreased by 12%, whereas under placebo both biomarkers increased by 6% and 15%. These percentages are also within the range of the systematic effect of time of day observed in our study.

Of the plasma biomarkers assessed in the TRAILBLAZER Trial (GFAP, NfL, p-tau217, and Aβ42/Aβ40) only p-tau217 was positively and significantly associated with baseline amyloid plaques and global tau deposition. It is of interest that in our small sample only p-tau217 showed a significant group effect.

For now, our results suggest that time of day matters when considering sampling for plasma biomarkers of dementia for monitoring disease progression or treatment outcome. This time-of-day variation was observed despite the presence of confounding factors that would be present in the real world including a light/dark cycle, sleep/wake state, and meals. As such, samples obtained at an early morning clinic may provide different results to those taken in an afternoon or evening clinic. Time of day should be standardised or at least recorded when samples are collected whether for diagnosis or monitoring their clinical status longitudinally. Recent studies suggest that biomarker concentrations also vary by food intake [24]. For now, we recommend that reference limits for biomarkers related to neurodegenerative dementias are established in samples collected while fasting and in the morning, and that samples for dementia diagnostics are collected accordingly.