Pleiotropic Effects of Variants in Dementia Genes in Parkinson Disease

Background: The prevalence of dementia in Parkinson disease (PD) increases dramatically with advancing age, approaching 80% in patients who survive 20 years with the disease. Increasing evidence suggests clinical, pathological and genetic overlap between Alzheimer disease, dementia with Lewy bodies and frontotemporal dementia with PD. However, the contribution of the dementia-causing genes to PD risk, cognitive impairment and dementia in PD is not fully established. Objective: To assess the contribution of coding variants in Mendelian dementia-causing genes on the risk of developing PD and the effect on cognitive performance of PD patients. Methods: We analyzed the coding regions of the amyloid-beta precursor protein (APP), Presenilin 1 and 2 (PSEN1, PSEN2), and Granulin (GRN) genes from 1,374 PD cases and 973 controls using pooled-DNA targeted sequence, human exome-chip and whole-exome sequencing (WES) data by single variant and gene base (SKAT-O and burden tests) analyses. Global cognitive function was assessed using the Mini-Mental State Examination (MMSE) or the Montreal Cognitive Assessment (MoCA). The effect of coding variants in dementia-causing genes on cognitive performance was tested by multiple regression analysis adjusting for gender, disease duration, age at dementia assessment, study site and APOE carrier status. Results: Known AD pathogenic mutations in the PSEN1 (p.A79V) and PSEN2 (p.V148I) genes were found in 0.3% of all PD patients. There was a significant burden of rare, likely damaging variants in the GRN and PSEN1 genes in PD patients when compared with frequencies in the European population from the ExAC database. Multiple regression analysis revealed that PD patients carrying rare variants in the APP, PSEN1, PSEN2, and GRN genes exhibit lower cognitive tests scores than non-carrier PD patients (p = 2.0 × 10−4), independent of age at PD diagnosis, age at evaluation, APOE status or recruitment site. Conclusions: Pathogenic mutations in the Alzheimer disease-causing genes (PSEN1 and PSEN2) are found in sporadic PD patients. PD patients with cognitive decline carry rare variants in dementia-causing genes. Variants in genes causing Mendelian neurodegenerative diseases exhibit pleiotropic effects.

AD is the most common form of dementia and is characterized pathologically by the accumulation of amyloid plaques and neurofibrillary tangles. Amyloid protein precursor (APP), presenilin-1 (PSEN1), and presenilin-2 (PSEN2) mutations cause autosomal dominant forms of early-onset AD (Cruts et al., 2012). Approximately 220 pathogenic mutations in the PSEN1 gene have been reported in AD patients worldwide, whereas 27 and 16 pathogenic mutations have been described in the APP and PSEN2 genes, respectively (Cruts et al., 2012). Rare functional variants in APP (Schulte et al., 2015) and pathogenic mutations in PSEN1, and PSEN2 have been also reported in PD patients (Takao et al., 2002;Jimenez-Escrig et al., 2004;Puschmann et al., 2009;Niwa et al., 2013). We recently reported the presence of leucinerich repeat kinase 2 (LRRK2), p.G2019S mutation in members of two multigenerational families with AD and a suggestive association of variants in the PTEN-induced putative kinase 1 (PINK1) gene with AD (Fernández et al., 2017). These results suggest a genetic overlap between familial AD and PD.
There is also an overlap of neuropathology in the brains of AD and PD patients. Approximately 50-60% of the sporadic and familial AD patients with pathogenic mutations in APP, PSEN1, or PSEN2 genes exhibit widespread α-synuclein pathology (Meeus et al., 2012a). Abnormal cortical amyloid-beta (Aβ) deposition is present in 60% of PD patients with dementia (Kotzbauer et al., 2012), the burden of Aβ plaques inversely correlates with cognitive status in PD cases with dementia (Irwin et al., 2013) and the progression of the dementia in PD correlates with Lewy body (LB) and cortical AD-type pathology (Compta et al., 2011). In addition, AD-like changes in cerebrospinal fluid (CSF) biomarkers (Aβ levels) have been reported in PD patients (Terrelonge et al., 2015). Taken together, these results suggest that abnormal APP processing and Aβ accumulation occurs in PD. Here, we evaluate the genetic variation of APP, PSEN1 and PSEN2 genes in PD patients.
FTD may cause up to 10% of all cases of dementia and is the second most common cause of early-onset dementia (<65 years of age) (Ratnavalli et al., 2002). The three most common genetic causes of FTD are mutations in the genes MAPT and granulin (GRN), and expansions of a hexanucleotide repeat (GGGGCC) in the C9orf72 gene. C9orf72 expansions cause 5-12% of all FTD and 10-35% of familial FTD. MAPT or GRN mutations are found in 1-10% of all FTD and 5-25% of familial FTD. MAPT variants associated with FTD also increase the risk of developing PD (Pastor et al., 2000;Benitez et al., 2016). C9orf72 expansions have been found in PD patients in some (Baizabal-Carvallo and Jankovic, 2016) but not all studies (Harms et al., 2013). Parkinsonism precedes the cognitive and behavioral symptoms of FTD by several years in patients with mutations in GRN (Baizabal-Carvallo and Jankovic, 2016). Up to 41% of FTD patients with GRN mutations exhibit parkinsonism (Josephs et al., 2007). In addition, GRN pathogenic mutations have been reported in PD patients (Brouwers et al., 2007;Rovelet-Lecrux et al., 2008). Here, we explore the role of variants in the GRN gene in PD patients.
PD is the most common neurodegenerative movement disorder, affecting ∼1-2% of people over 60 years of age (Wright Willis et al., 2010). Clinically, PD patients exhibit bradykinesia, rest tremor, rigidity, and disturbances in balance. PD is characterized neuropathologically by the presence of α-synuclein-positive neuronal inclusions, commonly referred to as LBs and Lewy neurites, as well as neuronal loss in the substantia nigra. Genetically, familial autosomal dominant PD is caused by highly penetrant mutations in the alphasynuclein (SNCA) and LRRK2 genes, whereas autosomal recessive PD is caused by mutations in the PARK2/PARKIN, PINK-1, and PARK7/DJ-1 genes (Petrucci et al., 2014). Recently, a GWAS identified at least 41 loci associated at genomewide significant level with disease risk in individuals of European ancestry (Chang et al., 2017). The most statistically significant signals associated with PD are common variants located close to SNCA, MAPT, and GBA genes (Chang et al., 2017).
Dementia is one of the most common non-motor symptoms in PD (Foltynie et al., 2004;Emre et al., 2007). The prevalence of dementia in PD patients at any stage of disease ranges from 22 to 48% (Foltynie et al., 2004;Emre et al., 2007). Several risk factors for dementia in PD include severity of parkinsonism, the presence of non-motor symptoms, older age, male sex and presence of cognitive symptoms at PD diagnosis (Aarsland and Kurz, 2010). Relatives of PD patients exhibit a higher risk of dementia than relatives of control subjects, (Rocca et al., 2007) supporting a role of genetic factors in the development of dementia in PD. However, the genetic contributors to dementia in PD have yet to be fully established. Longitudinal studies have shown that the APOE ε4 allele, the H1 haplotype in MAPT and mutations in the GBA gene are associated with a more rapid cognitive decline in PD (Morley et al., 2012;Collins and Williams-Gray, 2016;Liu et al., 2016). In contrast, a cross-sectional study showed that PD carriers of the LRRK2,p.G2019S mutation exhibit lower rates of dementia (Srivatsal et al., 2015). However, it is not clear how frequently mutations in dementia-causing genes occur in PD patients or whether variants in these genes relate to dementia in PD patients.
Here, we performed a systematic screening of known dementia-causing genes (APP, PSEN1, PSEN2, or GRN genes) in 821 PD cases and 423 controls from North America in addition to 553 PD patients and 550 healthy controls from Spain.

Subjects
Sequencing Methods

Single Variant Analysis
Association analysis between PD cases and in-house controls was performed using logistic regression, assuming an additive model for allelic effects. The analysis was adjusted for age, gender, and PCs (using the first two principal components) to correct for potential population stratification using Plink1.9 (Purcell et al., 2007;Chang et al., 2015). The minor allele frequency (MAF) of each variant in PD cases was also compared with the MAF described in the Non-Finnish European subgroup (NFE) from ExAC (Lek et al., 2016). Only coding regions with high-quality (PASS filter) variants reported in ExAC were included in our analyses.

Gene Based Analysis
The gene-based association in PD cases and in-house controls was performed using SKAT-O, which utilizes the R package SKAT (Wu et al., 2011). In addition, the burden of rare protein-altering variants in PD cases from the WUSTL and PPMI cohorts were compared with the burden observed in NFE samples from ExAc by collapsing the counts of all missense, nonsense and frameshift variants in each gene with a MAF < 0.01 and then, Fisher's exact test with Yates correction was applied. ExAC cannot be regarded as a pure control dataset. However, several studies have used this resource as a proxy for reassessing the effect of rare variants in Mendelian genes in the general population in different complex diseases (Roberts et al., 2015;Walsh et al., 2017).

Cognitive Impairment Assessment and Analysis
Cognitive impairment was assessed by the Folstein Mini-Mental State Examination (MMSE) and the Montreal Cognitive Assessment (MoCA) in the WUSTL and PPMI datasets respectively. For the combined analyses, z scores were derived by converting the mean raw MoCA and MMSE scores and standard deviation (SD) to the standard normal distribution with mean 0 and SD 1. Multiple regression analyses were performed in the residuals adjusting by age at diagnosis, age at dementia assessment, gender, study site and APOE carrier status as covariates (PROC GLM, SAS). All validated variants were included in the model independently of their clinical interpretation. Global cognitive impairment was defined by an MMSE ≤ 25 according to the recommendation the International Parkinson and Movement Disorder Society (MDS) Task Force (Dubois et al., 2007). The recommended MoCA cut-off for PD of <26 was applied (Dalrymple-Alford et al., 2010). Patients with a diagnosis of dementia [PPMI (N = 5) and WUSTL (N = 9)] were excluded from the analyses to avoid disproportionate leverage on the statistical models (Thaler et al., 2012). Nominally significant p-value threshold was set at 0.05. Multiple-test correction cutoff for the single-variant analysis using Bonferroni correction for 4 tests is 1.3 × 10 −2 . All statistical tests are two-sided unless otherwise stated, and were performed using Statistical Analysis System (SAS Institute Inc) or GraphPad Prism 5.0.

APOE Analysis
The effect of APOE allele and genotype in PD risk was tested comparing the frequency in PD cases and controls. Comparisons were made using the X 2 method. Multiple regression adjusting by age at diagnosis, age at dementia assessment and gender was performed to evaluate the effect of APOE ǫ4 allele on MoCA or MMSE scores (PROC GLM, SAS).

WUSTL Cohort
Twenty-one coding variants were found among the screened genes (Supplementary Table 2). 14.3% (3) variants were novel, 4.8% (1) were reported as known pathogenic variants, 19% (4) were classified as pathogenic nature unclear and 33% (7) were reported as non-pathogenic. The three novel variants were PSEN1 (p.P303L), PSEN2 (p.C358R), and GRN (p.A29V). Both the PSEN1 and PSEN2 variants were found in late-onset PD individuals whereas the GRN variant was found in one control (Supplementary Table 3A). The PSEN1 p.A79V mutation (a known AD pathogenic mutation) was found in three PD cases: An early-onset PD case (44 years old at onset) and two lateonset PD cases (75 and 64 years old at onset, respectively). None of the three carriers reported PD or AD family history. Neurological evaluation of these three patients at an age of 46, 86, and 82 years revealed no evidence of dementia after 2, 11, and 18 years of disease onset, respectively. The carrier of the "probable pathogenic" PSEN2, p.S130L variant exhibited an AAO of 45 years and rapid progression of cognitive impairment (Supplementary Table 3B).

UN Cohort
Ten coding variants were found in 4.9% of cases from the Spanish PD cohort. No novel variants were found (Supplementary Table  6). Nine variants were present in the AD/FTD mutation database. 50% (5) of them were considered non-pathogenic, 30% (3) of them were of pathogenic nature unclear, and 10% (1) of them (PSEN2 p.V148I) was considered to be pathogenic (Cruts et al., 2012). The PSEN2 p.V148I carrier is an early onset PD case (25 years at onset) with a tremor-dominant parkinsonism and positive family history of PD but dementia-free at last assessment after 12 years of PD. Seven (1.3%) carriers of the PSEN2 p.S130L variant were found in Spanish PD patients. Three had positive PD family history, three had EOPD, three had psychiatric comorbidities and two with dementia PD (Supplementary Table  7).

APOE Association With Status and Cognitive Test Performance
APOE has previously been associated with cognitive impairment in PD patients (Parsian et al., 2002;Morley et al., 2012;Tsuang et al., 2013) but APOE effect on PD risk is still controversial (Federoff et al., 2012). Here, no association was found between different APOE alleles (ε4 or ε2) and PD case-control status (Supplementary Tables 8A,B). The APOE ε4 allele was not associated with lower MoCA scores among PD patients in the PPMI cohort (p = 0.56). However, consistent with a previous report, the presence of the APOE ε4 allele was associated with a lower MMSE score (p = 9.0 × 10 −3 ) in WUSTL PD patients (Parsian et al., 2002).

Effect of Variants in Dementia-Causing Genes on Cognitive Tests Performance
Several rare variants in the GBA gene have been associated with lower MMSE scores in PD patients (Liu et al., 2016). We hypothesized that variants in the dementia-causing genes would affect performance on cognitive tests. Interestingly, 4.9% of all PD patients in the PPMI cohort carrying rare variants in dementiacausing genes exhibit significantly (p = 3.0 × 10 −2 ) lower MoCA scores than non-carriers PD patients ( Figure 1A). Similarly, PD patients carrying rare variants in dementia-causing genes (7.9% of all patients) exhibit significantly (p = 2.0 × 10 −3 ) lower MMSE scores than non-carriers in the WUSTL cohort ( Figure 1B).
Cognitive impairment was assessed with a different cognitive test in each cohort included in this study (MMSE for WUSTL and MoCA for PPMI). Therefore, we perform a combined analysis with age at PD diagnosis, age at which the dementia test was performed, APOE status, and cohort included as covariates. PD patients carrying rare variants in dementia-causing genes exhibited lower scores on cognitive tests than non-carrier PD patients (p = 2.0 × 10 −4 ), independent of age at PD diagnosis, age at evaluation, APOE status or cohort ( Table 2).

DISCUSSION
The genetic architecture of dementia in PD has not yet been fully established. Genetic variants in APOE and MAPT have time-dependent effects on cognition, which vary with disease stage: MAPT appears to have its greatest impact on cognitive decline in early PD, whereas APOE may have a more pronounced effect late in the course of the disease (Collins and Williams-Gray, 2016    explained only up to 1.4% of PD patients with cognitive decline (Liu et al., 2016). Our results show that 6.7% of PD patients carry rare coding variants in dementia-causing genes and exhibit lower scores on cognitive tests in two independent cohorts compared to non-carriers. The risk of developing dementia varies according to the duration of PD and age at onset (Rocca et al., 2007). However, the results of a covariate-adjusted model confirm that the association between rare variants in dementia-causing genes and lower scores in cognitive tests appears to be independent of age, cohort, disease duration, or APOE status. The APOE ε4 allele has been associated with a higher prevalence of dementia in PD (Morley et al., 2012;Tsuang et al., 2013). The small number of PD cases with dementia in each study, the significant heterogeneity of odds ratios between studies, and evidence of publication bias limits the confidence of the APOE and dementia in PD association (Huang et al., 2006). In this study, the APOE ε4 allele was not associated with risk of developing PD in none of the cohorts (Supplementary  Tables 8A,B). However, in the WUSTL cohort, APOE ε4 carriers exhibited lower scores on cognitive tests, but the PPMI cohort did not replicate these findings. Therefore, further studies are needed to clarify the role of APOE in PD cognitive impairment (Parsian et al., 2002).
The wide variation reported in the prevalence of cognitive impairment (CI) in PD across studies may be due to the cognitive tests employed (Goldman and Litvan, 2011;Burdick et al., 2014). The MMSE and MoCA are the most commonly used tests to assess CI in PD. However, controversy remains regarding the sensitivity of MMSE in assessing cognition in PD patients (Burdick et al., 2014). Even though the MoCA is more sensitive for detecting cognitive changes in PD patients (Hoops et al., 2009), the MMSE was the cognitive test that showed the effects of GBA and LRRK2 mutations on CI in PD patients (Srivatsal et al., 2015;Liu et al., 2016) and is the cognitive test recommended by the International Parkinson and Movement Disorder Society Task Force (Dubois et al., 2007). Nevertheless, here we show that rare variants in dementia-related genes affect both MMSE and MoCA scores in two heterogeneous PD populations (p = 2.0 × 10 −4 ; Table 2).
In addition, 0.3% of all PD patients screened in this study carry known AD pathogenic variants in PSEN1 and PSEN2 genes (Supplementary Tables 2, 6). A frequency of 0.3% is comparable to the percentage of PD patients carrying the most common pathogenic mutation (LRRK2 p.G2019S) known for sporadic PD (0.4-1%) (Healy et al., 2008;Correia Guedes et al., 2010). None of the PD patients carrying pathogenic mutations in PSEN1 and PSEN2 exhibited signs of dementia at their last clinical assessment. Unfortunately, CSF biomarkers or Pittsburgh compound B (PiB) imaging were not available from these carriers to help determine their preclinical status. The PSEN1 p.A79V mutation is associated with a broad range of AAO in AD and DLB patients (Cruchaga et al., 2012a;Meeus et al., 2012b). PSEN1 p.A79V is found in non-demented carriers at an AAO as late as 78 years (Kauwe et al., 2007) and, in some multigenerational AD families, it does not segregate perfectly with disease status (Cruchaga et al., 2012a). However, the PSEN1 p.A79V mutation increases the Aβ42 level and Aβ42/Aβ40 ratio in vitro (Kumar-Singh et al., 2006;Kauwe et al., 2007) but unlike other PSEN1 mutations that cause massive Aβ42 deposition, the average Aβ42 brain deposition in sporadic AD patients exceeded those observed in the brains of p.A79V carriers (Kumar-Singh et al., 2006). In addition, non-demented p.A79V carriers also exhibit changes in CSF Aβ42 levels and the Aβ42/40 ratio with no evidence of β-amyloid deposition using PiB imaging (Kauwe et al., 2007). Thus, PSEN1 p.A79V may be acting through a mechanism other than increasing Aβ deposition. Additional mutations in the PSEN1 gene including p.G217D, p.V272A, and p.L420R have been associated with dementia and parkinsonism (Takao et al., 2002;Jimenez-Escrig et al., 2004;Puschmann et al., 2009;Niwa et al., 2013). Recently, it was reported that a de novo PSEN1 mutation is responsible for an early-onset parkinsonism with cognitive impairment (Carecchio et al., 2017). In addition to the clinical overlap between dementia and parkinsonism in carriers of PSEN1 mutations, there is extensive Lewy body pathology in early-onset AD carriers of the PSEN1 p.S170F (Snider et al., 2005) and PSEN1 p.A431V mutations (Qiao et al., 2017) suggesting an interaction between PSEN1 dysfunction and α-synuclein aggregation.
We also found the PSEN2 p.V148I mutation in a Spanish PD patient but not in 550 age-matched Spanish controls. The PSEN2 p.V148I mutation was originally reported in a Spanish patient with late-onset AD (AAO = 71 years) (Lao et al., 1998) but, its pathogenicity has been questioned based on the absence of effect on either Aβ42 levels, Aβ40 levels, or the Aβ42/40 ratio in vitro (Walker et al., 2005). However, some variants that show no effect on Aβ42 levels or the Aβ42/Aβ40 ratio (Walker et al., 2005) affect calcium signaling in cultured skin fibroblasts from mutation carriers (Li et al., 2006). Increasing evidence suggests a role of the PSEN2 p.S130L variant in PD and dementia. Here, we found a total of nine PD patients and one control carrying the "probably pathogenic" PSEN2 p.S130L variant. The PSEN2 p.S130L variant has been reported in patients with late-onset AD and mild bradykinesia (Tomaino et al., 2007;Lohmann et al., 2012). Two siblings of a PSEN2 p.S130L carrier AD patient were diagnosed with PD (Tomaino et al., 2007). Recently, PSEN2 p.S130L was reported in an individual with idiopathic PD with dementia (AAO = 73 years) (Schulte et al., 2015). Another mutation in PSEN2 (p.V191E) also was found in one late-onset PD patient (AAO = 75 years) with cognitive decline (Meeus et al., 2012b). Interestingly, a Swedish PD family who carry a de novo α-synuclein p.A53T mutation also carried the PSEN2 p.R163H variant (Puschmann et al., 2009). Carriers of both mutations develop early-onset dementia (Puschmann et al., 2009). All these findings support our data that pathogenic variants in the presenilin genes are present in a small proportion of sporadic PD patients and contribute to α-synuclein aggregation.
We found an enrichment of rare variants in the GRN gene in PD patients compared to the general population (ExAC NFE). The frequencies of the p.R433W and p.R478H variants were higher in the WUSTL cohort than in the NFE ExAC cohort. Interestingly, the variant p.R433W was reported in neuropathologically confirmed LBD cases (Meeus et al., 2012b). Mutations in the GRN gene occur in LBD patients (Meeus et al., 2012b). A heterozygous deletion removing exons 1 to 11 of the GRN gene was reported in an 83 year-old PD patient (Rovelet-Lecrux et al., 2008) and, the IVS0 + 5G>C mutation was reported in a 56 year-old PD patient (Brouwers et al., 2007). In addition, parkinsonism occurs in some FTD patients and is more common in those patients (up to 41%) with GRN haploinsufficiency (Josephs et al., 2007). Interestingly, GRN overexpression in the substantia nigra protected nigrostriatal neurons in a mouse model of PD (Van Kampen et al., 2014). GRN seems to play an important role in multiple neurodegenerative diseases including PD, likely due to its function as a neurotrophic factor and its recently uncovered lysosome function (Tanaka et al., 2013).
Low frequency and rare mutations in the GBA gene, which encodes the lysosomal enzyme β-glucocerebrosidase-1, consistently relate to CI in PD and to Lewy body dementia (Nalls et al., 2013). α-synuclein is mainly degraded by lysosomes (Cuervo et al., 2004) and lysosomal dysfunction may contribute to de novo aggregation of α-synuclein and impaired autophagic degradation of cytosolic aggregates (Bourdenx et al., 2014). LBs and Lewy neurites may seed around impaired lysosomes and grow in size by continuous deposition of lysosomalderived un-degraded material as the disease progresses (Dehay et al., 2013). Thus, considering that presenilin and granulin are lysosomal proteins (Sannerud et al., 2016;Kao et al., 2017), it is logical to suggest that variants in the presenilin or granulin genes may exacerbate the cognitive impairments in PD by affecting lysosomal function and facilitating cell-to-cell transfer of proteopathic seeds in the progression of synucleinopathies.
The relatively small size of this study limits the statistical power, which could be the reason we failed to find significant associations between the PD cases and the in-house controls. However, the inclusion of the large NFE ExAC cohort with similar genetic background minimized this limitation. The data used in this study were not generated using a single sequencing method. However, none of the approaches used were expected to have 100% sensitivity for variant detection. Although these technical limitations could have marginal effects on estimates of rare variant frequency and odds ratio values, we do not expect them to alter the key conclusions of this study. Further studies are needed to confirm the role of variants in dementia genes in the cognitive impairment found in PD patients.

CONCLUSION
Our study shows that rare variants in several dementia-related genes are enriched in PD patients compared with normal controls. The PD patients with these variants exhibited lower cognitive performance than PD patients without these variants. Moreover, known dementia-causing mutations are found in PD patients.

AUTHOR CONTRIBUTIONS
BB and CC conceived and designed the study. JP, AD, UD, SO-C, and PP acquired and analyzed the clinical data. LI, BB, MF, JB, and MD-F acquired the genetic data. LI, BB, and CC performed the statistical analysis and interpreted the genetic data. LI and BB wrote the draft of the manuscript and JP, AD, PP, and CC provided critical comments on the draft of the manuscript. All authors read and approved the final version of the manuscript.