Sleep, Pain, and Neurodegeneration: A Mendelian Randomization Study

Our aim was to determine whether the genetic liability to sleep and pain-related traits have a causal effect on risk of neurodegeneration in individuals of predominantly European ancestry. We selected five neurodegenerative disorders, namely, age-related macular degeneration (AMD), Alzheimer's disease (AD), amyotrophic lateral sclerosis (ALS), multiple sclerosis (MS), and Parkinson's disease (PD). Sleep duration (SD), short sleep (SS), long sleep (LS), chronotype (CHR), morning person (MP), insomnia (INS), and multisite chronic pain (MCP) were considered as exposures. We conducted Mendelian randomization (MR) using an inverse-variance weighted (IVW) method to compute causal effect estimates using latest available GWAS data sets. The MP phenotype was observed as the strongest risk factor for genetic liability to AMD (ORIVW = 1.192; 95% CI 1.078, 1.318, P = 0.0007). We observed suggestive evidence of risky effects of CHR on AMD (P = 0.0034), SS on AD (P = 0.0044), and INS on ALS (P = 0.0123). However, we failed to observe any role of pain. The results were robust on sensitivity analyses. Our study highlighted the role of MP as a risk factor for AMD.


INTRODUCTION
Patients with neurodegenerative diseases (NDDs) often experience disruptions in circadian rhythmic activities (1,2). Many patients with NDD and circadian disruptions also complain of painful symptoms of variable origins and intensities (3). Both sleep and pain could often be treated and, thereby, can help maintain a stable quality of life in the absence of any disease-modifying treatment for NDDs (4). A greater understanding of the etiological relationship between sleep, pain, and neurodegeneration could, thereby, enable better management of NDDs.
It is well-recognized that circadian dysfunction in old age is due to degeneration of the suprachiasmatic nucleus (SCN) in the anterior hypothalamus, directly connected to the lightsensing retina (5). Different NDDs further exhibit marked heterogeneity in manifestation of circadian disruptions, which could be attributed to loss of different neuronal subpopulations in the SCN. Clinically, patients with Alzheimer's disease (AD) often show sleep-wake rhythm disorder, and patients with PD show a reduction in the amplitude of the circadian rhythm (6,7). A limited number of longitudinal studies have demonstrated the potential influence of circadian disruptions on predisposition to AD, PD, and related markers of neurodegeneration (8,9).
Similar to the involvement of specific brain regions directly influencing circadian rhythms, several brain regions, also referred to as pain matrix, have been shown to be activated during pain perception (10,11). The pain matrix comprising the primary (S1) and secondary (S2) somatosensory cortices, insula, anterior cingulate cortex (ACC), amygdala, prefrontal cortex (PFC), and thalamus, further shows differential activation during acute and chronic pain (12). Aging is specifically known to increase the likelihood of chronic pain and may amplify the neurodegeneration process (13,14).
To date, the sparse number of large longitudinal studies and clinical trials has limited our progress in understanding the relationship between sleep, pain, and onset or progression of neurodegeneration, necessitating the need for searching alternative approaches for judging the causality. A two-sample Mendelian randomization (MR) is one such approach that employs instruments or proxy markers of risk factor in one population to judge causality of the risk factor with an outcome in an independent population (15)(16)(17).
So far, limited studies have employed a genetic instrumentbased approach to judge the etiological relationship between sleep, pain, and NDDs. A recent MR study showed an absence of the role of genetic liability to sleep duration (SD) in influencing predisposition to AD (18). On the contrary, another report showed an association of genetic liability with sleep efficiency with AD (19). A couple of studies showed increased risk of ALS due to daytime sleepiness (19,20). Considering the highly varied role of various behavioral biomarkers of circadian rhythm on neurodegeneration and potential overlapping etiology of sleep and pain, we adopted a highly comprehensive approach by exploiting the availability of genetic instruments for various markers of circadian rhythm, namely, SD (21), short sleep (SS) (21), long sleep (LS) (21), chronotype (CHR) (22), morning person (MP) (22), insomnia (INS) (23), and multisite chronic pain (MCP) (24), and NDDs, namely, AD (25), AMD (26,27), ALS (28), MS (29), and PD (30,31) to dissect the bi-directional relationship between sleep, pain, and neurodegeneration using two-sample MR approach.

Identification and Correlation Among Data Sets
We employed a two-sample MR study design using summary estimates to examine the lifelong effect of sleep and pain-related traits on genetic liability to neurodegeneration in European populations. We used latest available discovery cohorts of metaanalyses of GWAS data sets in the literature. We identified single nucleotide polymorphisms (SNPs) that influence circadian rhythm-related traits, including SD (21), SS (21), LS (21), CHR (22), MP (22), INS (23), and MCP (24) ( Table 1). We adopted a P cutoff of 5 × 10 −8 to select the genetic instruments. Concerning the outcome data sets, we used the discovery cohort of a recent meta-analysis of GWAS on AD (25), AMD (26), ALS (28), MS (29), and PD (30). Before judging the causal role of sleep and pain in predisposition to PD, we checked for any potential correlation between different sleep and pain-related traits and different NDDs. We specifically employed a cross-trait LD score regression (LDSC) method to evaluate the genome-wide correlation between traits (https://github.com/bulik/ldsc) (32).
Since the study analyzed secondary data (publicly available data) that contained information at the population-level (summary-level data), informed consent and ethical approval were waived off for this study.

Causal Effect Estimation
The prioritized SNP IDs and positions were synchronized with the NCBI GRCh37 assembly. We further checked for the validity of MR assumptions by excluding SNPs with F-statistics < 10 and loci known to be directly involved in neurodegeneration based on existing evidence from previously published literature.
As the selected genetic instruments could be correlated, we performed clumping of significantly associated SNPs on each GWAS data set with the clump_data function of the TwoSampleMR package (version 0.4.25) in R (version 3.6.1). We employed a clumping window of 10,000 kb and linkage disequilibrium (LD; i.e., r 2 ) cutoff of 0.001, and used the European population in the 1,000 Genome Phase 3v5 data set to identify the leading SNPs.
The leading SNPs were further checked for availability in the respective outcome data sets. When possible, if a specific SNP was not available, a proxy SNP (r 2 > 0.8) was used. We further computed the pooled variance (R 2 ) for the respective risk factor using effect estimates (β x ) and effect allele frequencies (EAFs) of individual genetic instruments, i.e., R 2 = 2 * β 2 x * EAF * (1-EAF). Detectable risky and protective effect estimates at 80% power were computed for each NDD as an outcome at various pooled variances explained by the genetic instruments (ranging from 0.25 to 7.5%) using the Mendelian Randomisation Power Calculator (http://cnsgenomics.com/shiny/mRnd). To compute the effect estimates at specific variance for a given outcome, we employed a sample size of each outcome data set, the proportion of patients in the same data set, and a threshold P of 1.42 × 10 −3 .
We used the inverse variance-weighted (IVW) effect method as the primary method to compute the causal effect estimates, as used previously (17). We computed the causal estimates as odds ratio (OR) per unit of standard deviation (SD) for continuous traits and ORs for the outcome per unit log-odds of categorical traits. We employed a conservative Bonferroni correction of the significance level to account for 35 independent tests, including forward and reverse MR (threshold P = 1.42 × 10 −3 , i.e., 0.05/35). Heterogeneity was judged using the Cochran's Qstatistic and I 2 for the IVW method along with Rucker's Qstatistic, and the intercept deviation test for the MR-Egger's method (17). All the scripts used for the primary MR analysis have been provided as part of the a R-based mrpipeline package (https://github.com/CGEatTuebingen/mrpipeline). We used a previously published data set to replicate the findings before employing the package for to this study (17). The mrpipeline package is currently under the developmental phase, with a plan to integrate external databases, including GWAS and tissue expression repositories in the future. We also performed an Mendelian Randomization Pleiotropy RESidual Sum and Outlier (MR-PRESSO) global test to evaluate horizontal pleiotropy (33). Lastly, we performed an MR Steiger test of directionality to validate the assumption that a given exposure causes an outcome using the TwoSampleMR package (version 0.4.25) in R (version 3.6.1). Direct analysis was done using PD as an outcome and reverse was done using sleep and pain-related traits as outcomes.

Sensitivity Analysis
Several approaches were employed to rule out the influence of potential pleiotropic variants on the overall results. We used multiple modern MR methods, including the MR-Egger, weighted median (WME), and weighted mode (MBE) methods, to check the reliability of the estimates, as used in previous studies (16,17). Since most of the recent metaanalyses of GWAS compute effect estimates by pooling UK Biobank (UKB) data sets with previously available data sets, and the existence of any overlapping samples in exposure and outcome datasets could bias the effect estimates toward the confounded observational estimates, we also used the NDD datasets without UKB samples, when required (27,31). We conducted MR in the reverse direction to check and confirm the directionality of the observed associations. We further employed a leave-one-out and leave-one-groupout cross-validation approach to rule out the influence of outlier variants known to be associated with confounders of the relationship between the respective exposure and outcome data sets. We specifically employed the Phenoscanner database (http://phenoscanner.medschl.cam.ac.uk) to identify genetic variants associated with potential confounders. However, in the absence of knowledge of potential confounders, we adopted a more conservative approach, and all genetic loci known to be associated with non-sleep-related traits were assumed to be pleiotropic loci. We identified such loci by searching for all genetic variants in high LD with genetic instruments prioritized for this study using r 2 > 0.9 for previously reported associations in European populations. We used visual approaches, including scatter plots and funnel plots, to identify outlier variants.
We also performed a sensitivity analysis by adjusting for potential confounders using a multivariable MR method. As and when appropriate, we adjusted for quantity of sleep, sleep preference for a given time of day, and pain, the phenotypes of interest investigated in this study. As multiple, highly correlated, and overlapping traits representing both quantity of sleep (LS, SS, INS, and SD) and sleep preference (CHR, MP) were available, we performed a variable selection procedure to select the optimal variable that represented each category. Such an approach prevented us from conducting an overadjustment and avoided loss of power inherent with multiple variable regression methods. We selected SD to adjust for the quantity of sleep, as a continuous variable is more informative than a binary trait. Similarly, we selected CHR as a variable of choice representing sleep preference. Specifically, the genetic associations of instruments with respective NDDs were regressed on the genetic associations with all the risk factors (SD, sleep pattern, and pain) in a single regression model using IVW method. Genetic instruments entered into the multivariable regression model were allowed to be associated with any of the risk factor under consideration.
We further evaluated the potential biological influence of different brain regions on their respective contribution to the causal effect estimate by analyzing gene expression data for available genetic variants from the Genotype-Tissue Expression Project (https://www.gtexportal.org).

Identification and Correlation Among Data Sets
Details of discovery GWAS data sets used for the causal analysis in this study are shown in Table 1. The minimum number of individuals available for a specific NDD ranged from 12,557 ALS cases to 71,880 AD cases, which are broadly in consensus with their respective prevalence.
The pairwise genetic correlation analysis of complete GWAS data sets failed to show correlation of any of the NDDs with sleep or pain-related traits (Supplementary Table 1). Expectedly, a highly significant correlation was observed among the traits representing SD (SS, LS, SD, and INS) and among those representing sleep pattern (CHR and MP). Notably, MCP was strongly correlated with all the markers of SD (rg ranging from 0.28 for LS to 0.59 for INS), suggesting a need for conducting a multivariable analysis adjusting for MCP when judging the independent association of sleep markers with NDDs or vice versa.

Causal Effect Estimation
The genetic instruments were identified that influence sleep and pain-related traits through latest publicly available meta-analysis of GWAS summary datasets ( Table 1). Overall, we identified 771 genetic instruments to check the bidirectional causality between sleep, pain, and neurodegeneration, with F-statistic for individual SNPs ranging from 28.2 to 422.5. The detectable effect estimates for different NDDs as outcomes at 80% power and a type-1 error rate of 1.42 × 10 −3 are further shown in Supplementary Table 2.
The data used for computation of causal effect estimates are provided in Supplementary Table 3. The causal effect estimates using various MR approaches and heterogeneity analysis measures used to judge the robustness of the estimates are provided in Table 2 for the direct causal estimates for NDDs as outcomes. We observed a highly significant causal effect of MP on genetic liability to AMD (OR IVW = 1.192; 95% CI 1.078, 1.318, P = 0.0007). Heterogeneity check confirmed the reliability of the observed association with absence of any heterogeneity in the distribution of effect estimates of individual genetic variants (I 2 = 0.0%, Cochran's Q-test P = 0.9288, Rucker's Q-test P = 0.9414, MR-PRESSO global test P = 0.8420). The distribution of individual SNP-level effect estimates and the effect estimates computed with different MR methods for the effect of MP on AMD is further shown as scatter and funnel plots in Figure 1. We observed a similar directionality of causal effect estimates using the WME method (OR WME = 1.126; 95% CI = 1.044, 1.214). We also observed a similar trend using a highly correlated but continuous trait, CHR on AMD (OR IVW = 1.269; 95% CI 1.083, 1.486, P = 0.0034). The directionality of findings was further confirmed by a significantly higher variance explained by genetic instruments for MP and CHR than that explained by the respective genetic instruments for AMD (P Steiger = 2.1 × 10 −98 and P Steiger = 1.65 × 10 −24 ). In contrast, we did not observe any direct role of pain on predisposition to AMD.
We further observed a suggestive risky causal effect of SS on genetic liability to AD (OR IVW = 1.256; 95% CI 1.081, 1.459, P = 0.0044). Heterogeneity check further confirmed the     However, we did not observe any role of pain in predisposition to AD. We also observed a suggestive risky causal effect of INS on genetic liability to ALS (OR IVW = 1.551; 95% CI 1.121, 2.145, P = 0.0123). On the other hand, we failed to observe any role of pain in predisposition to ALS.
We did not observe any direct role of sleep and pain-related traits in predisposition to MS. Similarly, our MR analysis failed to detect a role of the sleep and pain-related traits in predisposition to PD.

Sensitivity Analysis
Concerning direct MR, the association of SS with AD was lost after the exclusion of overlapping UKB samples (data not shown). In the reverse MR, PD showed suggestion of a strong protective effect against CHR and MP after the exclusion of overlapping UKB samples (data not shown). Reverse causal estimates for various sleep and pain-related traits using various NDDs as exposure are shown in Table 3. Our reverse casual check confirmed the directionality of the observed associations of MP and CHR with AMD, as we failed to observe any effect of AMD on MP and CHR. Our reverse causal check also confirmed the role of SS in predisposition to AD, as we failed to observe the causal effect of AD on SS. Interestingly, all the sleep-related traits except for SS were observed to be influenced by genetic predisposition to AD when employing non-IVW methods for judging causal effects of sleep-related traits on AD. Lastly, our reverse casual check confirmed the role of INS in predisposition to ALS. On the contrary, our findings suggested a causal role of genetic predisposition to ALS in LS with a consistent significant risk effect using the IVW, WME, and MBE methods.
We failed to observe the predominant influence of any of the single variants on causal the effect estimates of MP with AMD, as shown in Supplementary Table 4. Similarly, the observed associations of CHR with AMD, SS with AD, and INS with ALS were retained (Supplementary Table 5). Among SNPs used for causal effect estimation of MP and CHR with AMD, 46 and 51 were identified as potential pleitropic variants for respective estimations (Supplementary Table 6). However, exclusion of these SNPs did not influence the observed casual association of MP and CHR with AMD (OR = 1.202, 95% CI 1.055, 1.370; OR = 1.262, 95% CI 1.049, 1.520). On the contrary, associations of SS with AD and INS with ALS were lost, which could be attributed to the presence of a high proportion of pleiotropic SNPs in the genetic instruments for SS and INS.
The sensitivity analysis using the multivariable MR approach also yielded similar results with the retention of the association of MP and CHR with AMD (OR = 1.184, 95% CI 1.083, 1.284; OR = 1.162, 95% CI 1.060, 1.263) (Supplementary Table 7).
Concerning the influence of specific brain regions, we specifically identified a high proportion of SNPs influencing brain IVW, inverse variance-weighted method; WME, weighted median method; MBE, weighted mode method (NOME assumptions). expression in the cerebellum and basal ganglia region (Table 4). However, exclusion of these SNPs did not affect the overall causal association of CHR and MP with AMD. Similarly, we failed to observe the effect of any of the other brain regions on the observed associations. We also failed to observe any influence of brain region-specific expression on other observed associations (data not shown).

DISCUSSION
The use of GWAS data in MR-based approaches has opened up opportunities to assess and define clinically relevant signatures for a diverse spectrum of diseases. Our study supports the role of a person's underlying circadian rhythm in genetic predisposition to neurodegeneration. We found an association of genetically predicted MP trait with AMD. The correlated trait CHR also had a suggestive risk association with AMD. We also found suggestive evidence for a possible association of genetically predicted SS with AD, and INS with ALS. Surprisingly, however, our study found no evidence to support the association between pain and NDDs.
To date, evidence from observational studies has shown a remarkable heterogeneity in the association of different circadian traits with various NDDs. A recent study investigating the incidence of AMD in 108,225 participants observed that patients with INS were 33% more likely to have subsequent AMD (HR 1.33; 95% CI 1.18, 1.48) (34). Previously, an observational study on 57 patients with neovascular AMD and 108 controls found a significantly increased risk of neovascular AMD in patients sleeping <6 h compared to those sleeping 7-8 h (OR 3.29; 95% CI 1.32, 8.27) (35). Another study failed to detect an association with LS in 316 patients with neovascular AMD compared to 500 patients without AMD (36). However, the study did find an association of LS with geographic atrophy, an advanced form of AMD, in 61 individuals (presence of a discrete area of atrophy with a diameter of ≥175 µm). A recent observational study further reported that individuals who take an afternoon nap are 60% less likely to be diagnosed with late AMD (56 with late AMD vs. 1,204 without AMD) (37). As darkness is known to stimulate the secretion of melatonin from the pineal gland, our findings are in agreement with previous studies showing that increased melatonin synthesis could play a protective role in the pathophysiology of AMD (38). However, a recent randomized controlled trial (RCT) failed to show any beneficial effect of lowlevel night-time light therapy on the progression of AMD (39).
In contrast to previously reported findings from epidemiological studies, we failed to observe any association of INS, SS, and LS with AMD using the genetic data in this study. However, we observed that MP is more likely to be predisposed to AMD (OR 1.19; 95% CI 1.08, 1.32). Our study suggests that more prolonged exposure to daylight in such individuals could increase the risk for AMD. Our findings are in contrast to a recent meta-analysis of observational studies demonstrating the absence of an association between sunlight exposure and AMD (OR 1.12; 95% CI 0.76, 1.67) (40). One of the possible reasons for this discrepancy could be that only one of the 14 studies included in the meta-analysis was a cohort study. The only included cohort study was a 10-year follow-up study, which demonstrated that participants exposed to summer sun for more than 5 h a day were more likely to show increased retinal pigment (RR 2.99; 95% CI 1.18, 7.6) and develop early age-related maculopathy (RR 2.2; 95% CI 1.02, 4.73) in comparison to those exposed for <2 h per day (41). It has also been suggested that excessive light exposure may induce phototoxic damage to the retinal pigmental epithelium and possibly contribute to the gradual worsening of vision in AMD (42)(43)(44).
Compared to the impact of circadian rhythms on other NDDs, the role of sleep-related traits has been well-investigated in AD but with mixed findings. Previous studies have predominantly  (8). Indeed, we observed a strong causal role of SS in predisposition to AD (OR 1.26; 95% CI 1.08, 1.46). However, our results need to be treated with caution, as the association was lost after excluding the overlapping UKB samples from the AD data set, as demonstrated previously (18). It is also possible that the association was lost because of decrease in sample size, necessitating replication with larger AD data sets in the future. Sleep disturbances are also frequently observed in patients with ALS. Our MR analysis also suggested a possible causal role of INS in ALS (OR 1.55; 95% CI 1.12, 2.14). A previous observational study has demonstrated decreased sleep efficiency and fragmented sleep architecture in 59 patients with ALS (47). Another study reported the presence of sleep disturbances in more than 2/3 of 40 patients with ALS. The study further reported a diagnosis of INS in 65% of the patients (48). These results are in agreement with a previous study reporting a significantly higher prevalence of INS in 90 patients with motor neuron disease compared to 96 healthy controls (48.9 vs. 31.3%, p = 0.014) (49). In summary, reports of sleep disturbance among patients with ALS in small sample-sized observational studies and the suggestive causal role of INS in ALS in this study necessitate a need for conducting large-scale epidemiological studies.
Despite the consistent findings of excessive daytime sleepiness or altered sleep timing in patients with PD, our MR findings demonstrate the absence of any causal role of sleep-related traits in predisposition to PD (1). One possible explanation could be that dopaminergic treatment might have influenced the sleeping behavior of patients with PD, as excessive daytime sleepiness is known to be one of the common side effects of dopaminergic treatment. In such a scenario, causal analysis using biological markers of circadian rhythms such as core body temperature, cortisol, and melatonin rhythms, might potentially shed light on the true relationship between sleep-related traits and PD.
We also failed to observe any causal association of sleeprelated traits with MS, although sleep disturbance is a common symptom of MS (1). It is suggested that the sleep disorders observed in patients with MS could be a secondary cause of fatigue, a symptom that affects 9 of 10 patients with MS (50).
Among all NDDs, high prevalence of pain has been observed in patients with AD and PD (1). Assessment of pain in such patients of is often challenging because of associated cognitive and motor impairments (51). Nevertheless, the use of genetic instruments of pain on a general population shows that MCP does not play any causal role in AD and PD. A recent crosssectional study investigating pain in 100 patients with PD patients showed that pain is more prevalent in patients with advanced-stage PD than patients with early-stage PD suggesting pain to be a consequence of the disease rather than a cause (52). Moreover, pain is a broad concept, and inconsistencies in the measurement of number of available pain behavior rating scales often limit their application in clinical settings.
Our study has several strengths and limitations. We adopted a highly comprehensive approach involving the exploration of several sleep-related traits and pain with commonly prevalent NDD. We further employed multiple MR methods and heterogeneity and sensitivity analysis approaches to confirm the reliability of the observed associations. Concerning limitations, previous observational studies have shown that the impact of sleep and pain-related traits may be dependent on the stage of neurodegeneration or severity of an NDD (2). However, we could not conduct such a stratified analysis because of the nonavailability of an individual-level data set for respective NDD. Furthermore, pain is a highly complex trait, and the lack of genetic instruments specific for neuropathic and nociceptive pain may undermine the findings of this study. The possibility of nociceptive pain confounding the causal relationship between neuropathic pain and neurodegeneration cannot be ruled out. One critical assumption for MR is that the effect of a genetic instrument for the main exposure on disease outcome is mediated by its influence on the intermediate trait. As genetic variants associated with sleep (duration or pattern) are highly correlated with pain and other sleep-related traits (duration or pattern), we addressed the potential pleiotropic effect by conducting a multivariable analysis. Our findings of causal association between sleep pattern (CHR or MP) and AMD remained robust after adjusting for the potential pleiotropic effect of SD and pain. However, despite adopting a multivariable MR approach, the possibility of residual confounding due to our inability to simultaneously adjust for all the highly correlated SD-related traits (SD or LS or SS) cannot be ruled out.
Using genetic data, we provide strong evidence that being an MP is a causal risk factor for genetic liability to AMD. There is a necessity for conducting large-scale epidemiological cohort studies to confirm our findings. Additional research is also required to understand the biological pathways underlying these associations, including causal analysis with biochemical makers of sleep and correlated traits associated with sleep.

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author.

ETHICS STATEMENT
Ethical review and approval was not required for the study on human participants in accordance with the local legislation and institutional requirements. Written informed consent from the participants' legal guardian/next of kin was not required to participate in this study in accordance with the national legislation and the institutional requirements.