GBA variants in REM sleep behavior disorder A multicenter study

Objective To study the role of GBA variants in the risk for isolated REM sleep behavior disorder (iRBD) and conversion to overt neurodegeneration.

Isolated REM sleep behavior disorder (iRBD) can be considered a prodromal synucleinopathy, because >80% of patients with iRBD will eventually convert to an overt neurodegenerative syndrome associated with α-synuclein accumulation-Parkinson disease (PD), dementia with Lewy bodies (DLB), or multiple system atrophy (MSA) 1 -with a conversion rate of about 6% a year. 2 For unknown reasons, while some patients with iRBD convert rapidly, others can remain free from parkinsonism or dementia for decades. 3,4 Variants in the gene encoding for the lysosomal enzyme glucocerebrosidase, GBA, are strong and relatively common risk factors for PD 5,6 and DLB, 7 yet their role in MSA is unclear. [8][9][10] Patients with PD who carry GBA variants, as a group, tend to have higher rates of nonmotor symptoms, including REM sleep behavior disorder (RBD), cognitive impairment, hyposmia, and autonomic dysfunction. 11 GBA variants can be classified as severe or mild based on the type of Gaucher disease (GD) associated with the variant. 12 Accordingly, patients with severe GBA variants have a higher risk for PD, an earlier average age at onset (AAO), 5 and faster cognitive decline 13,14 compared to patients with PD with mild or no GBA variants.
Few studies with small sample size have examined the role of GBA in iRBD, including studies of 69, 15 171, 16 and 265 patients with iRBD, 15 all supporting an association between GBA variants and iRBD but with different risk estimates. It has been shown that in PD cohorts with available data on probable RBD (pRBD), GBA variants are more frequent in the group with pRBD. 15 However, there are no accurate estimates of the risk of iRBD among GBA variant carriers and there have been no studies separately analyzing severe and mild GBA variants. It is not clear whether GBA variants affect the rate of conversion from iRBD to overt synucleinopathies, as only 2 small sample size studies with contradicting results examined this hypothesis. In one, there was no association with the rate of conversion in 8 GBA variant carriers with iRBD 16 ; in the other, a faster conversion was shown for 13 GBA variant carriers with iRBD compared to noncarriers. 17 In this study, we analyzed GBA variants in a large, multicenter study including 1,061 patients with iRBD, more than double the sample than all previous studies combined, and 3,086 controls, all of European origin. We further examined the effects of severe vs mild GBA variants on risk of iRBD, reported AAO of iRBD, and the potential effects on conversion from iRBD to an overt neurodegenerative disease.

Population
The patient population included 1,061 individuals diagnosed with iRBD with video-polysomnography according to the International Classification of Sleep Disorders, version 2 or 3 criteria. 18 The recruiting centers and the number of patients from each center are detailed in table 1. Additional data were available for subsets of samples, including reported AAO of RBD (n = 594), age at diagnosis of iRBD (n = 599), eventual phenoconversion to an overt neurodegenerative disease (data available for n = 584, converted n = 218), and rate of phenoconversion (n = 217). The average follow-up period for GBA carriers was 4.8 years and for noncarriers it was 4.1 years (p = 0.22).
The data on these variables were collected in 2018. The control population included a total of 3,086 individuals, comprised of 1,317 in-house controls of European origin (confirmed by principal component analysis using available genome-wide association study [GWAS] data compared to data from HapMap v.3 and hg19/GRCh37) and an additional 1,769 previously published European controls in which GBA was fully sequenced and all the variants were reported (table e-1 details these controls and the reported GBA variants in each of the articles; github.com/gan-orlab/GBA_RBD/). The in-house controls had a mean age of 46.5 ± 15.0 years and included 46.6% men, compared to 60.5 ± 9.9 and 81% men in the patients, therefore when analyzing these populations, adjustment for age and sex was performed (see Statistical analysis and Results).
GBA sequencing and classification of GBA variants GBA was fully sequenced as described previously 19 and the full protocol is available upon request. In brief, we designed molecular inversion probes (MIPs) targeting the coding sequence of GBA and performed next-generation sequencing (NGS) post capture. Alignment, variant calling, and annotations were done as previously described 19 using a standard pipeline. Exons 10 and 11 were also sequenced using Sanger sequencing since the coverage of NGS of these exons was low. Table e-2 (github.com/gan-orlab/GBA_RBD/) details the probes used for the MIPs capture. Classification of GBA variants as severe or mild was performed as described Glossary AAO = age at onset; CI = confidence interval; DLB = dementia with Lewy bodies; GCase = glucocerebrosidase; GD = Gaucher disease; GWAS = genome-wide association study; iRBD = isolated REM sleep behavior disorder; MIP = molecular inversion probe; MSA = multiple system atrophy; NGS = next-generation sequencing; OR = odds ratio; PD = Parkinson disease; pRBD = probable REM sleep behavior disorder; RBD = REM sleep behavior disorder.
previously 5,12 based on the occurrence of these variants in the severe (type II and type III) and mild (type I) forms of GD. The p.E326K and p.T369M variants, which do not cause GD but have a comparable risk as that of the p.N370S variants in PD, 20,21 were therefore included in the mild variant group.

Statistical analysis
To examine the association between GBA variants and risk of iRBD and controls, we performed association tests (χ 2 or Fisher exact test), logistic regression adjusted for sex and age, and burden tests. To examine the association of GBA variants with risk of iRBD comparing all controls, we used χ 2 or Fisher exact tests since there were no available data on age and sex from the controls collected from the literature to perform adjusted logistic regression. We therefore also performed this association using only our in-house European controls, for which data on age and sex were available, using logistic regression model adjusted for age and sex. Of note, having younger controls may result in underestimation of the risk, as some of the young controls with GBA variants may develop iRBD or overt neurodegeneration in the future. Therefore, if the statistical adjustment is not complete, the risk estimations that were calculated could be slightly lower (i.e., false-positive results are not likely; rather, underestimated risk is likely). We  also performed burden tests using the R package SKAT. Association with AAO and specific types of GBA variants (severe or mild) was tested using the nonparametric Kruskal-Wallis test since the group of severe GBA variants included only 5 patients. The association with conversion was tested using a χ 2 test for the total number of conversions, and Kaplan-Meier survival analysis was performed to examine the rate of conversion. All statistical analyses were performed using R or SPSS v24 (IBM, Armonk, NY).
Standard protocols approvals, registrations, and patient consents All study participants signed informed consent forms, and the study protocol was approved by the institutional review boards.

Data availability
Anonymized data will be shared by request from any qualified investigator.

Results
GBA variants are associated with increased risk of iRBD with differential effects of severe and mild variants.
The variants in GBA identified in each of the participating centers are detailed in table 1, with a total of 17 distinct variants found in patients and controls (table 2).  ). When severe GBA variant carriers were compared to mild GBA carriers and noncarriers, a possible association was demonstrated, as the Breslow test was statistically significant and the Tarone-Ware test was at near statistical significance, while the log-rank test did not reach statistical significance (figure; Breslow p = 0.017, Tarone-Ware p = 0.051, log-rank p = 0.24).

Discussion
Our results confirm the association between GBA variants and increased risk of iRBD and suggest that severe and mild GBA variants have differential effects on risk, similar to previous reports in PD. 5 These results also suggest that patients with iRBD with severe GBA variants may have earlier AAO and may convert faster to overt neurodegenerative disease. However, the results on AAO and conversion should be considered as preliminary only and with caution, due to several limitations discussed below.
Three previous small sample size studies have examined the association between GBA variants and iRBD. 15,16,22 Two of these studies included full sequencing of the gene 15,16 and the third only examined 2 specific variants (p.N370S and p.L444P). 22 Due to their size, analyses of specific variants or types of variants, such as severe or mild, were not possible. The current study includes 2 of the previously published cohorts 15,22 and additional cohorts of European ancestry. With the larger sample size accrued, we were able to demonstrate a much larger risk in carriers of severe GBA variants. Given the small numbers of these variants and the wide range of the CIs, the risk estimates may be different in future, larger studies. The current results are in line with previous results from PD, which clearly demonstrated similar relationships between severe and mild GBA variants and risk for PD. 5 Previous studies have also suggested that the type of GBA variants may affect PD progression, 13,14 which is further supported by our preliminary findings on AAO and conversion of iRBD.
In recent years it has been demonstrated that the coding variants p.E326K and p.T369M, which do not cause GD, are risk factors for PD. 20,21,23 In DLB, the association between p.E326K and risk for the disease is clear, yet it is still unclear whether p.T369M is a risk factor for DLB. Only a few studies that examined p.T369M in DLB have been performed and in most of them there was no association. A multicenter study that included over 700 patients with DLB reported lack of association, and in a GWAS with over 1,700 patients with DLB, only the p.E326K variant was reported to be associated with the disease. 7 Conversely, recent data from 556 patients with DLB did suggest an association. 24 The lack of association in the current study in iRBD may also provide further support for lack of association of p.T369M with iRBD and DLB. However, it is important to keep in mind that the association of this variant with PD was only reported in much larger studies 20,21 due to its lower effect on risk compared to other GBA variants. Only much larger studies can determine conclusively whether p.T369M is associated with iRBD and DLB. There was a large difference between the frequency of p.T369M in our in-house controls (2.7%) and the controls from the literature (1.1%), perhaps due to population structure, but the combined frequency (1.7%) is comparable to that seen in the gnomAD European population (1.9%), rendering our results for this variant as likely unbiased.
Our study has several limitations. The possible association between GBA variants and rate of conversion reported here,   . Larger studies will be required to determine conclusively whether GBA variants are associated with the rate of phenoconversion. An additional potential limitation is that the measured duration from age at diagnosis or iRBD to conversion might not reflect the actual length of disease duration, as patients can remain unaware for many years about their dreamenactment behaviors, especially if they do not have a bed partner or if they do not have very active or violent dreams. The small number of severe GBA variants is also a limitation in the risk analysis, as it created a wide CI. Since the effect of severe vs mild variants is in line with previous studies in PD, it is likely that these risk estimates of iRBD are overall correct, yet the precise estimate might change in future, larger studies.
The mechanisms underlying the association between GBA variants, the enzyme encoded by GBA, glucocerebrosidase (GCase), and the development of neurodegeneration are unknown. 11 Several mechanisms have been proposed, including interaction of GCase substrates with α-synuclein, which may lead to its accumulation, 25 changes in the lysosomal membrane composition, which may lead to reduced autophagy and mitophagy, 26,27 accumulation of misfolded GCase and endoplasmic reticulum stress, 28 and others. The association with iRBD may suggest that studying these mechanisms in nondopaminergic neuronal models that are involved in RBD could lead to new discoveries and better understanding of these potential mechanisms.
Our results demonstrate that GBA variants are associated with increased risk of iRBD. These results may also suggest that severe and mild GBA variants may have differential effects on the risk, and possibly on AAO, of iRBD and its conversion to overt neurodegenerative disease. Due to the limitations mentioned above, the latter associations should be considered as preliminary with additional, larger studies on GBA in iRBD required to confirm or refute them. One important implication of the association between GBA variants and iRBD is the possibility to perform screening for iRBD in healthy GBA variant carriers. This may allow for even earlier detection of prodromal neurodegeneration and could be especially useful when home detection of iRBD will be made possible.

Acknowledgment
The