Identification of sleep changes as potential risk factors for ALS
We reviewed published articles on sleep problems in ALS and identified changes in sleep: daytime sleepiness, long sleep latency, frequent awakenings and low sleep efficiency that revealed in self-reported questionnaires and polysomnography as potential exposures for ALS [9–11]. Another potential exposure that we considered was abnormal sleep duration, which is reported to increase the risk of developing of other neurodegenerative diseases [14, 15].
Genome-wide Association Studies (GWASs) Summary Statistics For Sleep Traits
We searched PubMed for the summaries of most resent and largest available GWASs (up to November 2019) of the sleep issues identified above. The traits we chose were daytime sleepiness, sleep latency, sleep efficiency and sleep duration. We also used the number of sleep episodes to represent frequent awakenings. The four largest and most recent GWASs of each trait in adult participants were chosen to provide genetic variation for the downstream association analyses. The characteristics of the GWASs we chose are described in the following text and Additional Table 1. Data on daytime sleepiness, sleep efficiency, number of sleep episodes and sleep duration were gathered from GWASs based on the UK Biobank, which gathered information from volunteers of European ancestry between 40 and 69 years of age in the years 2006–2010 [16].
IVs for daytime sleepiness were identified from a GWAS that included up to 452,071 individuals and assessed the severity of daytime sleepiness using the question “How likely are you to doze off or fall asleep during the daytime when you do not mean to?”, answered as a continuous variable on a scale from 1 to 4 points [17].
IVs for sleep efficiency and number of sleep episodes were extracted from one GWAS that included accelerometer recordings from 85,670 participants for up to 7 days [18]. Sleep efficiency was calculated by dividing actual sleep time by the time between the start of the first inactivity session and the end of the last inactivity session; the result was expressed as a percentage (%). The number of sleep episodes was determined by counting sleep episodes within the time window defined as the sleep period; the result was stated as a count value. Individuals who had an average of more than 30 or fewer than 5 episodes were excluded.
IVs for sleep duration were found in a GWAS including up to 446,118 individuals. Sleep duration, reported as a continuous variable, was assessed on the basis of self-reported questionnaires and activity monitoring [19]. The unit of sleep duration was hours.
IVs for sleep latency were drawn from a meta-analysis of GWASs involving 4,242 individuals of European ancestry from seven cohorts [20]. In all GWASs included in this meta-analysis, sleep latency was measured using the Munich Chronotype Questionnaire. Sleep latency on days off was used in the analyses for those cohorts. The unit of sleep latency was minutes.
The process of trait identification and the choice of IVs for each trait are shown in Fig. 1.
GWAS Summary Statistics For ALS
Genetic association data for ALS were obtained from a publicly available GWAS that included 20,806 ALS patients and 59,804 controls of European ancestry [21]. All individuals classified as ALS patients met the standard for a probable or definite diagnosis of ALS according to the El Escorial criteria (Brooks, 1994). The implementation of the assessment was performed by a neurologist specializing in ALS. The characteristics of the GWAS of ALS is described in Additional Table 1.
Selection Of Instrumental Variants
For each potential risk factor, independent genetic variants (single nucleotide polymorphisms (SNPs)) were chosen as IVs according to the following principles: 1. The SNPs are not in linkage disequilibrium, defined as r2 < 0.001; 2. The genome-wide significance of the SNPs met the threshold (P < 5*10− 8) for the corresponding risk factors. SNPs that were not available for ALS data were replaced with proxies (r2 > 0.9) from the online website SNiPA. If an SNP was not available for ALS data and had no proxy that was, that SNP was excluded from downstream association analyses. Because there were no SNPs for sleep latency that met the inclusion criteria, this trait was excluded from the subsequent analysis.
MR Analysis
The MR approach was based on 3 assumptions: 1. The genetic variants used as IVs for the potential risk factors are associated with the target disease. 2. The genetic variants are not associated with any confounders. 3. The genetic variants are associated with the target disease only through the risk factor and not through any alternative causal pathway (Fig. 2). We used 2-sample MR, a method of MR analysis, to analyze the causal effect of daytime sleepiness and night sleep changes on ALS [22].
In the main analyses, we summarized the ratio estimates for individual genetic variants using the conventional fixed-effect inverse-variance-weighted (IVW) method [23]. Simple medians and weighted medians were used as sensitivity analyses to confirm the main findings [24].
Pleiotropy was evaluated based on the intercept calculated by MR-Egger regression [25]. To investigate the influence of outlying and/or pleiotropic genetic variants, we performed a leave-one-out analysis in which each genetic variant was omitted in turn.
Four sleep issues suspected to be risk factors were calculated in our study. We used Bonferroni correction to test for multiple comparisons. The corrected significance threshold was 0.0125 (0.05 divided by 4). However, 0.0125 < P < 0.05 was also considered suggestive evidence for a potential association.
The results are expressed as the odds ratio (95% confidence interval) for a 1-point or 1-SD increase per genetic prediction in each risk factor. All analyses were performed in R Studio (Version 1.2.1335).