No evidence linking sleep traits with white blood cell counts: Multivariable‐adjusted and Mendelian randomization analyses

Disturbances in habitual sleep have been associated with multiple age‐associated diseases. However, the biological mechanisms underpinning these associations remain largely unclear. We assessed the possible involvement of the circulating immune system by determining the associations between sleep traits and white blood cell counts using multivariable‐adjusted linear regression and Mendelian randomization.


| INTRODUCTION
Sleep is a high-dimensional phenotype, 1 is different for men and women, and is changing with increasing age. 2 Disturbances in single dimensions of habitual sleep are increasingly recognized as important risk factors contributing to the onset of age-related diseases, including metabolic, cardiovascular and neurodegenerative diseases, [3][4][5][6][7] COVID-19 and influenza susceptibility, 8,9 some types of cancer, [10][11][12] and mortality. 13Mendelian randomization, which uses genetic variants associated with the exposure of interest as instrumental variable, [14][15][16][17] indicated that short self-reported sleep duration and poor sleep quality play a causal role in the risk of cardiometabolic disease and dementia, 16,[18][19][20] and susceptibility to some infectious diseases, 9 thus making sleep of particular interest as target for disease prevention.However, the biological mechanisms underpinning the sleep-disease associations at a population level are still poorly understood and require additional research.
Several studies have suggested that disturbances in sleep (e.g.sleep deprivation) influence the immune system (reviewed in [21][22][23] ) and particularly in immune suppression, 12 although evidence for the role of immune disturbances in habitual sleep in large population studies is limited.Of particular interest, we previously observed in multicohort and Mendelian randomization analyses, that the presence of self-reported insomnia symptoms was associated with increased levels of the systemic inflammation marker alpha-1-glycoprotein. 24 Furthermore, gene-short-sleep interaction studies have identified several genes underlying sleep-lipid associations that encode for proteins are involved in immune regulation and autoimmune disease development. 25White blood cells are considered a long-term systemic marker of acute and chronic inflammation, and have been associated with, amongst others and including in Mendelian randomization studies, (atherosclerotic) cardiovascular disease [26][27][28] and dementia. 29White bloods cell can be categorized in five subclasses, namely lymphocytes (mainly involved in adaptive immunity), monocytes and neutrophils (mainly involved in innate immunity), and eosinophils and basophils (mainly involved in IgE-mediated immunity). 30hanges in white blood cell counts could therefore underly the observed associations between disturbances in habitual sleep and age-related disease.In previous observational studies, sleep duration 31 and sleep regularity 32 have been associated with white blood cell counts, but evidence favouring possible causality is lacking and studies were generally based on small samples requiring replication in larger samples.
In the present study, we aimed to assess the association between disturbances in various sleep traits and white blood cell counts using multivariable-adjusted regression and Mendelian randomization analyses.The combination of multivariable-adjusted regression and Mendelian randomization can provide additional support about potential causality. 33

| Study setting
The multivariable-adjusted regression analyses as part of the present study were conducted using existing and available data from the UK Biobank.The UK Biobank is a large prospective cohort study with over 500,000 participants between the age of 40 and 69 recruited between 2006 and 2010. 34Eligible individuals were invited when they lived within a 25-mile distance from the 22 assessment centers across England, Wales and Scotland.Self-reported data was collected by an interview with the study nurse and a touch screen questionnaire.Patients also underwent physical measurements, and (random) blood, urine and saliva samples were drawn.All participants provided written informed consent.The North-West Multi-center Research Ethics Committee (MREC) approved the UK Biobank study.The Patient Information Advisory Group (PIAG) from Wales and England approved access for information to invite participants.More information about the UK Biobank can be found on [https:// www.ukbio bank.ac.uk/ ].

| Study population
For the present cross-sectional study, we used only the data on sleep, blood cell counts and covariables collected at the baseline visit, and included all participants with data available on all white blood cells (N = 462,413).We excluded participants with a total white blood cell count below 4500 and above 11,000/μL (N = 39,655), which could be indicative for leukopenia and leukocytosis, respectively.Furthermore, we excluded participants with missing data on the sleep variables and covariables, and Mendelian randomization, sleep, white blood cells furthermore excluded participants with a self-reported daily habitual sleep duration below 4 h and above 14 h (N = 65,102).In the end, we had complete data available from 357,656 participants.

| Sleep traits
For the present study, we examined the following habitual sleep traits: sleep duration (data field 1160), sleep chronotype (data field 1180), sleeplessness (data field 1200) and daytime dozing (data field 1220).All data was collected through self-reported questionnaires.Self-reported sleep duration was reported based on the question 'about how many hours of sleep do you get every 24 hours?(please include naps)'.In addition of considering total habitual sleep duration as a continuous variable, we also studied short sleep duration and long sleep duration as separate individual sleep phenotypes.Short and long sleep duration we defined at the 20% lowest or highest age-, age 2 -and sex-adjusted residuals of total sleep duration, respectively, as done previously to correct the dichotomizing for the differences between men and women and in younger and older participants. 2,25Participants not defined as either having a short or long total sleep time were considered as reference.Self-reported insomnia symptoms were based on the question 'Do you have trouble falling asleep at night or do you wake up in the middle of the night?' and were categorized as either 'usually' having insomnia symptoms or as 'never/rarely/sometimes' (reference).Selfreported chronotype was reported as 'Evening person' and 'Morning person'.Daytime dozing (narcolepsy) was categorized as 'never/rarely' and 'Sometimes/often'.

| Blood cell counts
White blood cell counts were determined using fresh blood samples using a fully automated, and clinically validated, Coulter LH 750 Haematology Analyser (Beckman Coulter Life Sciences, USA).In detail, data from following white blood cells were collected and used for the present study: total white blood cell counts (data field 30000), lymphocytes (data field 30180), monocytes (data field 30190), neutrophils (data field 30200), eosinophils (data field 30210), and basophiles (data field 30220).All counts for the individual white blood cells were relative to the total amount of white blood cells.All procedures were performed with the guidelines as provided by the manufacturer.A more detailed description of the measurements and the study protocol can be found on the UK Biobank website (http:// www.ukbio bank.ac.uk).To increase comparability between the study results, we standardized the data from all blood cell counts (after log transformation) so that means were zero and standard deviations one.

| Other variables
Age was determined at the day of the visit to the assessment center.As a measure of socioeconomic status, we used the Townsend Deprivation Index (data field 22189), which was defined at the moment of enrolment, and is a composition of four different variables related to socioeconomic status: unemployment, non-ownership of a home, non-ownership of a car and household overcrowding. 35Importantly, the Townsend Deprivation Index is not linked to a specific individual, but instead linked to the postal codes from the UK Biobank participants and is therefore a reflection of overall socioeconomic status of the neighbourhood in which the participants are living.Smoking status (data field 20116) was based on a self-reported questionnaire, and reported as 'never', 'previous' and 'current'.Alcohol use (data field 1558) was also based on a self-reported questionnaire, and reported as different strata reflecting the amount of alcohol consumed by the participants (never, occasionally, few, regulative and most/all days of the week).Use of blood pressure-and cholesterol-lowering medication was based on a medication inventory at the moment of the visit to the assessment center (based on data fields 6153 and 6177).Body mass index (data field 23104) was assessed using the Tanita BC418MA body composition analyser (Tanita, Inc.Manchester, UK).Height was measured in standing position using a Seca 202 measuring rod (Seca Gmbh, Hamburg, Germany).History of cardiometabolic disease (notably type 2 diabetes mellitus, coronary artery disease and stroke; based on data fields 130708, 131296, 131298, 131304, 131306, 131360, 131362, 131364, 131366 and 131368) was defined based on the date of the first report stored with UK Biobank.This data could either be based on selfreport, hospital records and/or records from the general practitioner.In the event a case was recorded before the date of recruitment in UK Biobank, a participant was classified as having a history of the disease.Ancestry (data field 2100) was based on selfreport and classified as either European or non-European.

| Statistical analyses
All statistical analyses were performed using R (v4.1.2) statistical software (The R Foundation for Statistical Computing, Vienna, Austria).Characteristics were presented for the total study population as means (with standard deviation) or number (with percentage).
Cross-sectional analyses were performed using multivariable-adjusted linear regression models, adjusted for age, sex, body mass index, smoking status, alcohol use, Townsend Deprivation Index, use of cholesterol-or blood-pressure-lowering medication and history of cardiometabolic disease.Statistical analyses were done separately for all the sleep traits in the whole study population as well as separately for men and women to observe potential differences between the sexes.p-values for interaction were calculated by including an interaction term, on a multiplicative scale, between the sleep trait and sex in the multivariableadjusted regression analyses.In general, the regression analyses were done comparing the disturbance as the index group and the non-disturbed group as the reference.Total sleep duration was examined on a continuous scale and chronotype was examined using morning chronotype as the index group and evening chronotype as the reference group.Results are presented as a beta coefficient, representing the difference in SD between the index and reference group, with accompanying 95% confidence intervals.Sensitivity analyses were performed restricted to participants with European ancestry only.
Mendelian randomization analyses were done using a two-sample design on summary-based results from publicly-available genome-wide association studies.As genetic instruments, we used available data derived from European-ancestry participants from UK Biobank.Within UK Biobank (N = 410,000), analyses were corrected for familial relationships between the individuals using the kinship matrix.For each of the considered sleep traits, genome-wide significant SNPs (p < 5 × 10 −8 ) were selected as SNP-exposure associations, and then pruned to obtain independent instrumental variables by the TwoSampleMR and IeuGWASR packages, which uses the PLINK clumping method with clumping window of 10 Mb and linkage disequilibrium of r 2 < .001. 36otentially weak instrument bias was examined by the F-statistic, for which a threshold greater than 10 is conventionally considered sufficient for MR analysis. 37SNPoutcome associations of the selected SNPs for the sleep traits were derived from available data from the Blood Cell Consortium, 38 which included 563,085 participants from cohorts in which no data from UK Biobank was included.The exclusion of data from UK Biobank ensured completely independent samples for the SNP-exposure and SNP-outcome associations used as input data for the Mendelian randomization analyses.
The associations of the sleep traits with each study outcome from each database were estimated by the inverse-variance weighted (IVW) method, which combines the Wald-ratio estimates (estimated association of genetic variants with outcome divided by estimated association of genetic variants with exposure) for individual genetic variants by a fixed-effect meta-analysis with inverse-variants weights. 37,39Given that the IVW method assumes all genetic instruments are valid (e.g.1][42] The weighted-median estimator could still provide a consistent estimate of the causal effect even when up to 50% of the identified genetic variants are invalid IVs. 40The MR-Egger method does not require a zero horizontal pleiotropy effect, and could detect the pleiotropy by the intercept term (under the InSIDE assumption), which when different from zero indicates a bias in the IVW estimation. 41,42 3 |RESULTS

| Characteristics of the study population for the multivariable-adjusted regression analyses
The total study population, after excluding participants outside the normal clinical range for total white blood cell counts and with missing data, consisted of 357,656 participants.Of these, 44.4% were men and had a mean age of 56.5 (SD 8.1) years.A total of 5.2% of the study participants did report a non-European ethnicity.Participants were, on average, slightly overweight with a mean body mass index of 27.4 (SD 4.8) kg/m 2 .Men had a higher number of cardiometabolic diseases at the baseline examination and women had a higher frequency of sleep insomnia symptoms than men (31.6% vs. 23.6%,respectively) (Table 1).

| Multivariable-adjusted linear regression analyses
Results of the multivariable-adjusted linear regression analyses examining the associations between sleep traits and blood cell counts is presented in Table 2. Within the whole study population (N = 357,656), we did not observe associations between any of the sleep traits and any of the studied white blood cell counts, after the correction for the considered confounding factors.Observed effect sizes also very close to zero (<.01 standard deviations between the index and reference groups).In addition, we did not observe potential effect modification by sex (p-values for interactions >.05) nor did we find different results when restricted to participants from European ancestry only.

| Mendelian randomization analyses
Results from the inverse-variance weighted Mendelian randomization analyses are presented in Figure 1 (full results presented in Table S1).For none of the examined relations between sleep traits and blood cell counts, we observed evidence for possible associations.These results did not materially differ within the sensitivity analyses for Mendelian randomization weighted median estimator and MR-Egger regression (Table S1).In addition, we did not observe in the MR-Egger intercept test that the intercept deviated from zero in any of the examined relationships (p-values >.05).

| DISCUSSION
In the present study, we aimed to investigate the associations between different sleep traits, as a reflection of sleep health, and different white blood cell counts, considered a long-term systemic marker of acute and chronic inflammation, using multivariable-adjusted cross-sectional yses and Mendelian randomization.Based on data derived from over 350,000 participants from the UK Biobank, we did not find evidence for associations between the investigated sleep traits and blood cell counts using both methods embedded in an observational setting.It is therefore unlikely that disturbances in habitual sleep affect white blood cell counts.As a result, it is therefore unlikely that changes in white blood cell counts mediate the previously observed associations between disturbances in habitual sleep (e.g., short sleep duration and presence of insomnia symptoms on a regular basis) and age-related disease.
19][20]43 However, there is currently little insight into the biological mechanisms underpinning these associations.Without exception, the immune system has been postulated to play an important role in all of these diseases, and therefore the immune system is a likely candidate to mediate the associations between disturbances in selfreported habitual sleep and age-related disease.However, based on the analyses performed in the present study, we did not observe evidence favouring a possible link between disturbances in self-reported habitual sleep and different white blood cell counts.Previously, we observed in a multicohort analysis that presence of insomnia symptoms was associated with elevated levels of the inflammation marker alpha-1-glycoprotein using both multivariable-adjusted and Mendelian randomization analyses, 24 as a way to triangulate etiological findings and observational studies. 33In research done by others, a short sleep duration and daytime napping have been previously associated with elevated levels of c-reactive protein, [44][45][46] as another marker, which is also used in clinical practice, of low-grade systemic inflammation.However, the nature of these studies did not allow causal inference of the study results.In addition, disturbances in habitual sleep, particularly having a short habitual sleep duration, has been previously shown to increase the risk developing the autoimmune disease rheumatoid arthritis. 47It is worth mentioning that c-reactive protein and white blood cell counts independently predict mortality in older people (aged 85 years and over), as shown previously by our research group, 48 suggesting that, even though both white blood cell counts and c-reactive protein are measures of chronic inflammation, they apparently reflect different aspects.It could well be that more insights on the role of sleep aspects in chronic inflammation can be obtained using additional inflammatory markers, including the interleukins and complement factors.Given the increasing availability of high-quality (proteomics) data in large and well-characterized studies, this hypothesis can be studied further in future studies.For example, small-scale studies have shown that sleep loss was associated with higher secretion of interleukin 6 in monocytes, 49 more DNA damage and lower DNA repair in white blood cells, 50 and induced a senescence-associated secretory phenotype in leukocytes. 51evertheless, based on the present study results, sleep disturbances, on the level of the general population, are not likely to cause changes in white blood cell counts.
There are a number of strengths and limitations that need to be considered while interpreting the value of the observations done in the present study.First, the study population had a very large sample size and was therefore able to detect also small effects between habitual sleep and blood cell counts.The lack of observed associations are therefore unlikely to be the result of underpowered statistical analyses.Furthermore, without exception, the results are close to zero, suggesting the any potential effect (if existing) is very small and unlikely to be clinically and/ or biologically relevant.However, the current study used self-reported data on different sleep characteristics instead T A B L E 2 Multivariable-adjusted regression analyses for the associations between sleep traits and white blood cell counts in the overall UK Biobank sample (N = 357,656).more objective sleep data as from accelerometry and polysomnography.Although the use of questionnairebased sleep data likely increased measurement error and decreased statistical power to some extent, genetic variants for self-reported habitual sleep traits were successfully validated using accelerometer-derived data, 52,53 albeit that overall genetic correlation was modest. 53Furthermore, the phenotypic correlation between self-reported habitual sleep and accelerometer-derived sleep data is modest at most, 54 which suggests the phenotypes derived from both methods reflect different aspects of human sleep quality.Since the data on accelerometer-derived sleep data in UK Biobank were collected some years later, 53 we were not able to link these to the white blood cell count data used in the present study.For the Mendelian randomization analyses, we used instrumental variables that explained a minor proportion of the variance in the self-reported sleep trait.However, we exclusively used strong instruments (F-statistic >10) and results were explored using a wellpowered SNP-outcome data base. 38For the present study, we made use of data on white blood cell counts without having data available on the activation status of the immune cells; it can therefore not be ruled out that more sophisticated essays are required to demonstrate associations that are independent on the white blood cell counts, including single-cell assessments and cytokine production capacity upon, for example, LPS stimulation.However, it is unlikely that such measures will become available soon in sample sizes as large as in the present study.And last, the present study was performed in people from predominantly the European ancestry, and Mendelian randomization analyses were performed in Europeans only; the study results to other groups should therefore be done with caution.

Measures of sleep quantity
In summary, despite the availability of a large sample size from the general population, we did not find evidence favouring that disturbances in different sleep traits, as a reflection of sleep health, cause changes in the levels of specific white blood cells.It is therefore unlikely that the association between disturbances in habitual sleep traits and age-related disease are mediated by changes in white blood cell counts.More likely, disturbances in habitual sleep affect the risk of age-related disease through other pathways and/ or through other aspects of the immune system.

F I G U R E 1
Results from the Mendelian Randomization analyses between sleep traits and blood cell counts.Results derived with the inverse-variance weighted method for summary-level two-sample Mendelian randomization.Results presented as the unit difference in white blood cell count (with accompanying 95% confidence interval) in standard deviation per unit of the sleep trait.Apart from sleep duration (presented per hour increase) all other sleep traits compared the index group with a reference group.

Total sleep duration (beta, 95% CI) Short sleep duration (beta, 95% CI) Long sleep duration (beta, 95% CI)
Results are presented as the difference in the blood cell count in standard deviation between the index and reference group, or, in the case of total sleep duration, as the difference in blood cell count in standard deviation per hour increase in total sleep duration.Regression analyses adjusted for age, sex, Townsend Deprivation Index, smoking status, alcohol use, body mass index, history of cardiometabolic disease, use of cholesterol-lowering medication, use of blood pressure-lowering medication and non-European ethnicity. Note: