Novel associations of BST1 and LAMP3 with rapid eye movement sleep behavior disorder

Isolated rapid-eye-movement (REM) sleep behavior disorder (iRBD) is a parasomnia, characterized by loss of muscle atonia and dream enactment occurring during REM sleep phase. Since a large subgroup of iRBD patients will convert to Parkinson s disease, and since previous genetic studies have suggested common genes, it is likely that there is at least a partial overlap between iRBD and Parkinson s disease genetics. To further examine this potential overlap and to identify genes specifically involved in iRBD, we fully sequenced 25 genes previously identified in genome-wide association studies of Parkinson s disease. The genes were captured and sequenced using targeted next-generation sequencing in a total of 1,039 iRBD patients and 1,852 controls of European ancestry. The role of rare heterozygous variants in these genes was examined using burden tests and optimized sequence Kernel association tests (SKAT-O), adjusted for age and sex. The contribution of biallelic (homozygous and compound heterozygous) variants was further tested in all genes. To examine the association of common variants in the target genes, we used logistic regression adjusted for age and sex. We found a significant association between rare heterozygous nonsynonymous variants in BST1 and iRBD (p=0.0003 at coverage [≥]50X and 0.0004 at [≥]30X), mainly driven by three nonsynonymous variants (p.V85M, p.I101V and p.V272M) found in a total of 22 (1.2%) controls vs. two (0.2%) patients. Rare non-coding heterozygous variants in LAMP3 were also found to be associated with reduced iRBD risk (p=0.0006 at [≥]30X). We found no statistically significant association between rare heterozygous variants in the rest of genes and risk of iRBD. Several carriers of biallelic variants were identified, yet there was no overrepresentation in iRBD patients vs. controls. To examine the potential impact of the rare nonsynonymous BST1 variants on the protein structure, we performed in silico structural analysis. All three variants seem to be loss-of-function variants significantly affecting the protein structure and stability. Our results suggest that rare coding variants in BST1 and rare non-coding variants in LAMP3 are associated with iRBD, and additional studies are required to replicate these results and examine whether loss-of-function of BST1 could be a therapeutic target.


Introduction
Isolated rapid-eye-movement sleep behavior disorder (iRBD) is a prodromal synucleinopathy, as more than 80% of iRBD patients will eventually convert to an overt neurodegenerative syndrome associated with α -synuclein pathology. Typically, iRBD patients will convert to Parkinson's disease (about 40-50% of patients), dementia with Lewy bodies or unspecified dementia (40-50%), or, in much fewer cases, to multiple system atrophy (5-10%) (Iranzo et al., 2016;Postuma et al., 2019). While our understanding of the genetic background of dementia with Lewy bodies or multiple system atrophy is limited, the rapid development of various genetic methods during the recent decades has led to wealth of data on the role of common and rare genetic variants in Parkinson's disease. To date, there are 80 genetic loci found to be associated with Parkinson's disease risk discovered through genome-wide association studies (GWASs) (Nalls et al., 2019;Foo et al., 2020), and several genes have been implicated in familial Parkinson's disease (Gan-Or et al., 2018;Gan-Or and Rouleau, 2019;Blauwendraat et al., 2020).
In order to study the genetic background of iRBD and its conversion to α synucleinopathies, recent studies have examined whether Parkinson's disease-or dementia with Lewy bodies-related genes are also associated with iRBD. These studies have suggested that while there is some overlap between the genetic backgrounds of iRBD and Parkinson's disease or dementia with Lewy bodies, this overlap is only partial. For example, it was demonstrated that GBA variants are associated with iRBD risk, Parkinson's disease and dementia with Lewy bodies (Gan Or et al., 2015;Gan-Or and Rouleau, 2019), but pathogenic LRRK2 variants are found to only be associated with Parkinson's disease, and not with iRBD and dementia with Lewy bodies (Heckman et al., 2016;Bencheikh et al., 2018;Blauwendraat et al., 2020). We have recently reported that the familial Parkinson's disease and atypical parkinsonism genes PRKN, PARK7, GCH1, VPS35, ATP13A2, VPS13C, FBXO7 and PLA2G6 are not likely to be involved in iRBD (Mufti et al., 2020). Heterozygous variants in SMPD1 have been reported to be associated with Parkinson's disease risk (Gan-Or et al., 2013;Alcalay et al., 2019), yet no association was found with iRBD (Rudakou et al., 2020). Whereas variants in MAPT are associated with Parkinson's disease and APOE haplotypes are important risk factors of dementia with Lewy bodies, (Dickson et al., 2018;Li et al., 2018), neither are linked to iRBD (Gan-Or et al., 2017;Li et al., 2018).
Furthermore, there are independent risk variants of Parkinson's disease, dementia with Lewy bodies and iRBD within SNCA locus; specific variants in the 3' untranslated region (UTR) are associated with Parkinson's disease but not with iRBD, and other, independent variants at 5' UTR are associated with Parkinson's disease, iRBD and dementia with Lewy bodies (Krohn et al., 2020b). In the TMEM175 locus, there are two independent Parkinson's disease risk variants, but only one of them, p.M393T, has also been associated with iRBD risk (Krohn et al., 2020a).
Thus far, the role of most Parkinson's disease GWAS genes has not been thoroughly studied in iRBD. In the current study, we aimed to examine whether rare and common variants in 25 Parkinson's disease-related GWAS genes are associated with iRBD. The entire coding regions with the exon-intron boundaries as well as the regulatory 3' and 5' UTRs were captured and sequenced. We then performed different genetic analyses to investigate the association of rare and common variants in these genes with iRBD.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

Study population
This study included a total of 2,891 subjects, composed of 1,039 unrelated individuals diagnosed with iRBD (according to the International Classification of Sleep Disorders criteria, version 2 or 3) and 1,852 controls. Details on age and sex of patients and controls have been previously described (Mufti et al., 2020) and can be found in Supplementary Table 1. Differences in age and sex were taken into account as needed in the statistical analysis. All patients and controls were of European ancestry (confirmed by principal component analysis [PCA] of GWAS data compared to data from HapMap v.3 and hg19/GRCh37).

Standard protocol approvals, registrations, and patient consents
All study participants signed an informed consent form before entering the study, and the study protocol was approved by the institutional review boards.

Selection of genes and genetic analysis
The current study was designed and performed before the publication of the recent Parkinson's disease GWAS (Nalls et al., 2019), therefore, the genes for analysis were selected from previous GWASs (Nalls et al., 2014;Chang et al., 2017). A total of 25 genes were selected for analysis, including : ACMSD, BST1, CCDC62, DDRGK1, DGKQ, FGF20, GAK, GPNMB, HIP1R, ITGA8,   LAMP3, MAPT, MCCC1, PM20D1, RAB25, RAB29, RIT2, SETD1A, SLC41A1, STK39, SIPA1L2, STX1B, SYT11, TMEM163 and USP25. These genes were selected based on the . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 28, 2020. . presence of one or more of the following: quantitative trait loci, expression in brain, potential interaction with known Parkinson's disease-associated genes and involvement in pathways implicated in Parkinson's disease, such the autophagy-lysosomal pathway, mitochondria quality control and endolysosomal recycling. The 25 genes were fully captured (coding sequence and 3'and 5'-untranslated regions) using molecular inversion probes (MIPs) designed as previously described (O'Roak et al., 2012). The full protocol is available upon request. Supplementary  (McKenna et al., 2010), and ANNOVAR for annotation (Wang et al., 2010). The Frequency of each variant was extracted from the Genome Aggregation Database (GnomAD) (Lek et al., 2016).

Quality control
Quality control (QC) was performed using PLINK v1.9 (Purcell et al., 2007). We excluded variants that deviated from Hardy-Weinberg equilibrium in controls with a threshold set at p=0.001, and those identified in <25% of the reads for a specific variant. We also filtered out variants with genotyping rate lower than 90%. The same genotyping rate cut-off was used for exclusion of individual samples. Threshold for rate of missingness difference between patients and controls was set at p=0.05, and variants below this threshold were excluded from the analysis. To be included in the analysis, the minimum genotype quality score was set to 30. We . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 28, 2020. . used two coverage thresholds for rare variants (minor allele frequency [MAF] <0.01), >30X and >50X, and all analyses were repeated using these thresholds. For the analysis of common variants, coverage of >15X was used.

Statistical analysis
To test whether rare heterozygous variants (defined by MAF<0.01) in each of our target genes are associated with iRBD, we performed sequence kernel association test (SKAT, R package) (Lee et al., 2012) and optimized sequence kernel association test (SKAT-O) on different groups of variants: all rare variants, potentially functional rare variants (including nonsynonymous, frame-shift, stop-gain and splicing), rare loss-of-function variants (frame-shift, stop-gain and splicing), and rare nonsynonymous variants only. In addition, we further tested whether rare variants that are predicted to be pathogenic based on Combined Annotation Dependent Depletion (CADD) score of ≥ 12.37 (representing the top 2% of potentially deleterious variants) are enriched in iRBD patients. To test the association between biallelic variants and iRBD risk, we compared the frequencies of carriers of two vary rare (MAF<0.001) nonsynonymous, splice-site, frame-shift and stop-gain variants between patients and controls using Fisher's exact test.
Bonferroni correction for multiple comparisons was applied as necessary. We tested the association between common variants (MAF>0.01) in the target genes and iRBD risk using logistic regression adjusted for age and sex using PLINK v1.9. Linkage disequilibrium between the discovered variants and the respective GWAS top hits was examined using the non-Finnish European reference cohort on LDlink (https://ldlink.nci.nih.gov/) (Machiela and Chanock, 2015).
Effects of common variants on expression was viewed using the genotype-tissue expression database (GTEx -https://www.gtexportal.org). We further performed in silico structural analysis . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 28, 2020. . of BST1 to test whether the rare coding variants that were found to be associated with iRBD in our analysis could potentially affect the enzyme structure and activity. The atomic coordinates of human BST1 bound to ATP-γ-S were downloaded from the Protein Data Bank (ID 1isg). The steric clashes induced by each variant were evaluated using the "mutagenesis" toolbox in PyMol v. 2.2.0.

Data availability
Data used for the analysis is available in the supplementary tables. Anonymized raw data can be shared upon request.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 28, 2020. .

Coverage and identified variants
The average coverage of the 25 genes analyzed in this study was >647X (range 73-1162, median 790). An average of 95% of the target regions were covered with >15X, 93% with >30X and 90% with >50X. The average coverage of each gene and the percentage of the nucleotides covered at 15X, 30X and 50X are detailed in Supplementary Table 3. Finally, there were no differences in the coverage between patients and controls. A total of 1,189 rare variants were found with coverage of > 30X, and 570 rare variants with > 50X (Supplementary Table 4). We identified 125 common variants across all genes (Supplementary Table 5) with a coverage of >15X.

Rare heterozygous variants in BST1 and LAMP3 are associated with iRBD
To examine whether rare heterozygous variants in our genes of interest may be associated with iRBD risk, we performed SKAT and SKAT-O tests, repeated twice for variants detected at depths of coverage of >30X and >50X (see methods). Supplementary Table 4 (Table 1). The Bonferroni-corrected p-value threshold for statistical significance was set at p<0.001 after correcting for the number of genes and depths of coverage.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 28, 2020. .
We found a statistically significant association of rare heterozygous functional variants in BST1 (SKAT p=0.0004 at >30X and p=0.0003 at >50X for rare functional variants), found more in controls than in iRBD patients. This association is mainly driven by the nonsynonymous variants p.V85M (rs377310254, found in five controls and none in patients), p.I101V (rs6840615, found in seven controls and none in patients), and p.V272M (rs144197373, found in 10 controls and two patients). Overall, these variants were found in 22 (1.2%) controls vs. 2 (0.2%) patients. Another statistically significant association was found between rare variants in LAMP3 gene and reduced iRBD risk in SKAT-O analysis. This association is driven by two noncoding variants (one intronic [location -chr3:182858302] and one at the 3' UTR of LAMP3 [rs56682988, c.*415T>C]) found only in controls (15 and nine controls, respectively). In order to further examine whether these variants indeed drive the association in both BST1 and LAMP3, we excluded them and repeated the analysis (SKAT and SKAT-O), which resulted in loss of statistical significance for both genes (Supplementary table 6). There were no additional statistically significant associations of the remaining genes with iRBD after correcting for multiple comparisons (p<0.001).

Structural analysis of BST1 variants suggests that loss-of-function may be protective in iRBD
To investigate the potential impact of the three BST1 nonsynonymous variants (p.V85M, p.I101V and p.V272M) on the structure and activity of the enzyme, we performed in silico mutagenesis and evaluated potential clashes with surrounding residues. Figure 1 depicts the structure of BST1 with the respective locations of the three nonsynonymous variants that drive . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 28, 2020. . the BST1 association detected in our analysis. The structure of human BST1 was solved by X-ray crystallography in complex with five substrate analogues (Yamamoto-Katayama et al., 2002).
All structures revealed a homodimeric assembly, with the catalytic clefts facing the cavity at the interface of the two chains ( Figure 1A).
The sidechain of p.V85M points towards the hydrophobic core of the protein, behind a helix facing the nucleotide binding site. The amino-acid change from valine to the bulkier sidechain of methionine results in clashes with other residues in the core, for all rotamers ( Figure   1B). This variant would therefore likely destabilize the enzyme active site and potentially unfold the protein. The sidechain of the variant p.I101V is located underneath the active site towards the hydrophobic core. Although the amino-acid change from isoleucine to the smaller sidechain of valine does not create clash ( Figure 1C), it reduces the packing in the core, which could also destabilize the enzyme. Finally, the p.V272M variant is located in a helix at the C-terminus of the protein that forms symmetrical contacts with the same helix in the other chain of the dimer.
The p.V272M variant would create clashes with sidechain and main-chain atoms located in the other chain of the dimer ( Figure 1D). As p.V272M resides at the dimer interface of the enzyme and probably helps maintaining the two subunits together, this variant would most likely lead to the disruption of the dimer. Overall, all the disease-associated nonsynonymous variants in BST1 (p.V85M, p.I101V, and p.V272M) appear to be "loss-of-function", suggesting that reduced BST1 activity may be protective in iRBD. This is supported by the top Parkinson's disease GWAS hit in the BST1 locus, the rs4698412 G allele, which is associated with reduced risk of Parkinson's disease (Nalls et al., 2019). This allele is also associated with reduced expression of BST1 in blood in GTEx (normalized effect size =-0.07, p=1.5e-6), suggesting that reduced expression might be protective.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 28, 2020. .

Very rare bi-allelic variants are not enriched in iRBD patients
In order to examine whether bi-allelic variants in our genes of interest are enriched in iRBD, we compared the carrier frequencies of very rare (MAF<0.001) homozygous and compound heterozygous variants between iRBD patients and controls. To analyze compound heterozygous variants, since phasing could not be performed, we considered carriers of two very rare variants as compound heterozygous carriers, with the following exceptions: 1) when variants were physically close (less than 112 base pairs [bp]; probes' target length) and we could determine their phase based on the sequence reads, and 2) if the same combination of very rare variants appeared more than once across samples, we assumed that the variants are most likely to be on the same allele. We found five (0.5%) iRBD patients and seven (0.4%) controls presumably carriers of bi-allelic variants in the studied genes (Table 2, p=0.731, Fisher test).

Association of common variants in the target genes with iRBD
To test whether common variants in our target genes are associated with iRBD, we performed logistic regression (using PLINK v1.9 software) adjusted for age and sex for common variants (MAF>0.01) detected at coverage depth of >15X. A nominal association was observed in 12 variants across all genes (Supplementary table 5), but no association remained statistically significant after Bonferroni correction for multiple comparisons (set at p<0.0005).
Of the variants with nominal associations, one variant in the ITGA8 3' UTR (rs896435, OR=1.15, 95% CI = 1.01-1.32, p=0.04) is the top hit from the most recent Parkinson's disease . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 28, 2020. . GWAS (Nalls et al., 2019), and two other ITGA8 3' UTR variants are almost in perfect LD (D'=1.0, R 2 >0.99, p<0.0001) with rs896435. Four variants in the 3' UTR of RAB29 were almost in perfect LD (Supplementary table 5) and are associated with expression of RAB29 in multiple tissues in GTEx, including the brain. Three MAPT variants were in partial LD with Parkinson's disease GWAS hits in the MAPT locus and were associated with expression of multiple genes in multiple tissues in GTEx, demonstrating the complexity of this genomic region.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 28, 2020. .

Discussion
In the current study, we studied a large cohort of iRBD patients by fully-sequencing and analyzing 25 Parkinson's disease-related GWAS genes and their association with iRBD. Our results identify BST1 and LAMP3 as novel genes potentially associated with iRBD. Based on in silico models, the three nonsynonymous BST1 variants that drive the association with iRBD may be loss-of-function variants, suggesting that reduced BST1 activity may reduce the risk of developing iRBD. The variants driving the association of LAMP3 are in noncoding regions and could be regulatory. These hypotheses will require confirmation in functional studies in relevant models. While some common variants were nominally associated with iRBD, none of them remained statistically significant after correction for multiple comparisons.
BST1, also called CD157, is a glycosyl phosphatidylinositol (GPI) anchored membrane protein initially found in bone marrow stromal cells and is essential for B-lymphocyte growth and development. It has an extracellular enzymatic domain that produces cyclic ADP-ribose (cADPR). This metabolite acts as a second messenger that can trigger Ca 2+ release from intracellular stores (Ishihara and Hirano, 2000), a process that plays a role in cellular function and death. Specific features of calcium homeostasis have been suggested to be responsible for the specific vulnerability of dopaminergic neurons in Parkinson's disease (Chan et al., 2009), yet whether BST1 is involved in calcium homeostasis in human neurons is still unclear, as most work was done in non-human models. Another mechanism by which BST1 may be involved in Parkinson's disease is immune response and neuroinflammation, which are likely important in the pathogenesis of the disease (Wang et al. 2015). BST1 serves as a receptor which regulates leukocyte adhesion and migration, and plays a role in inflammation (Malavasi et al., 2006).
However, its potential role in microglia activation and neuroinflammation is yet to be . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 28, 2020. . determined. Our in-silico analysis suggested that the BST1 variants found mostly in controls are loss-of-function variants. We can therefore hypothesize that these variants may reduce immune response and lead to a reduced risk of iRBD, and that inhibition of BST1 could be a therapeutic target for iRBD and Parkinson's disease treatment or prevention.
LAMP3 encodes the lysosomal-associated membrane protein 3, which plays a role in the unfolded protein response (UPR) that contributes to protein degradation and cell survival during proteasomal dysfunction (Dominguez-Bautista et al., 2015). Furthermore, LAMP3 knockdown impairs the ability of the cells to complete the autophagic process, and high LAMP3 expression Our study has several limitations. First, despite being the largest genetic study of iRBD to date, it may be still underpowered to detect rare variants in GWAS Parkinson's disease-related genes, as well as common variants with a small effect size. Therefore, we cannot completely rule out the possibility that rare and common variants in these genes may contribute to iRBD risk. A second limitation is the younger age and the differences in sex distribution between iRBD patients and controls, for which we adjusted in the statistical analysis as needed. Another potential limitation is the possibility that there were undiagnosed iRBD patients among the control population. However, since iRBD is found in only ~1% of the population (Postuma et al., . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 28, 2020. . 2019), the effect of having undiagnosed iRBD patients in the controls would be minimal, given the large sample size.
To conclude, our results suggest two novel genetic associations with iRBD; an association with rare functional variants in BST1, and with rare non-coding variants in LAMP3.
All the association-driving coding variants found in BST1, mainly in controls, appear to potentially cause loss-of-function, suggesting that reduced BST1 activity may reduce the risk of iRBD. Further studies would be required to confirm our results and to examine the biological mechanism underlying the effect of disease-associated variants in both LAMP3 and BST1. The absence of evidence of association between rare and common variants in the remaining genes and iRBD risk suggests that these genes either have no effect in iRBD or have a minor effect that we could not detect with this sample size. Environmental factors and environment-gene interactions are likely to play a major role on iRBD, and larger studies that include carefully collected epidemiological data and more extensive genetic data such as whole-exome or wholegenome sequencing will be required to clarify these issues.
. CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Competing interests
JYM received fees from Takeda, Eisai and Paladin Pharma for consultancies in unrelated fields.

ZGO received consultancy fees from Lysosomal Therapeutics Inc. (LTI), Idorsia, Prevail
Therapeutics, Inceptions Sciences (now Ventus), Ono Therapeutics, Denali, Deerfield, Neuron23 and Handl Therapeutics. None of these companies were involved in any parts of preparing, drafting and publishing this review.

Supplementary material
Supplementary is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 28, 2020. . . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 28, 2020. . I  s  h  i  h  a  r  a  K  ,  H  i  r  a  n  o  T  .  B  S  T  -1  /  C  D  1  5  7  r  e  g  u  l  a  t  e  s  t  h  e  h  u  m  o  r  a  l  i  m  m  u  n  e  r  e  s  p  o  n  s  e  s  i  n  v  i  v  o  .  C  h  e  m  i  c  a  l   i  m  m  u  n  o  l  o  g  y  2  0  0  0  ;  7  5  :  2  3  5  -5  5  .   K  r  o  h  n  L  ,  Ö  z  t  ü  r  k  T  N  ,  V  a  n  d  e  r  p  e  r  r  e  B  ,  O  u  l  e  d  A  m  a  r  B  e  n  c  h  e  i  k  h  B  ,  R  u  s  k  e  y  J  A  ,  L  a  u  r  e  n  . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted June 28, 2020.  P  o  s  t  u  m  a  R  B  ,  I  r  a  n  z  o  A  ,  H  u  M  ,  H  ö  g  l  B  ,  B  o  e  v  e  B  F  ,  M  a  n  n  i  R   ,  e  t  a  l  .   R  i  s  k  a  n  d  p  r  e  d  i  c  t  o  r  s  o  f  d  e  m  e  n  t  i  a  a  n  d   p  a  r  k  i  n  s  o  n  i  s  m  i  n  i  d  i  o  p  a  t  h  i  c  R  E  M  s  l  e  e  p  b  e  h  a  v  i  o  u  r  d  i  s  o  r  d  e  r  :  a  m  u  l  t  i  c  e  n  t  r  e  s  t  u  d  y  .  B  r  a  i  n  2  0  1  9  ;   1  4  2  (  3  )  :  7  4  4  -5  9  .   P  u  r  c  e  l  l  S  ,  N  e  a  l  e  B  ,  T  o  d  d  -B  r  o  w  n  K  ,  T  h  o  m  a  s  L  ,  F  e  r  r  e  i  r  a  M  A  ,  B  e  n  d  e  . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 28, 2020. .  . CC-BY 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 28, 2020. .