Pathways to Parkinsonism Redux: convergent pathobiological mechanisms in genetics of Parkinson's disease

Introduction

It is nearly 200 years since James Parkinson famously described the symptoms of his eponymous disorder (1). Since then, our understanding of Parkinson's disease (PD) has developed hugely. We now recognize that many of the classic motor symptoms such as bradykinesia stem from the loss of dopaminergic innervation of the striatum due to the progressive loss of dopamine neurons in the substantia nigra pars compacta (SNpc). Reflecting this, dopamine replacement therapy remains the primary treatment option. Postmortem analysis has shown that neuron loss can also occur in other brain regions and also revealed the presence of Lewy bodies. These and related structures, Lewy neurites, have become the pathologically defining feature of PD and represents insoluble and aggregated proteins, particularly α-synuclein, packed into intracellular inclusions. However, how these inclusion bodies form and indeed the pathogenesis of PD remain a mystery.

The epidemic of post-encephalitic parkinsonism in the early twentieth century and the cases resulting from inadvertent exposure to MPTP suggested that PD could be due to environmental factors (2). It was not until 1996 that genetics was given serious consideration as a causative factor with the first reporting of a family showing Mendelian segregation of PD. Using linkage analysis, a genetic factor was mapped to chromosome 4 and the locus labeled PARK1 (3). Soon after, an A53T mutation was found in the SNCA gene that encodes the α-synuclein protein (4) leading to a surge in interest in PD genomics that continues to the current time. Nineteen loci that segregate with familial forms of the disease have been reported (Table 1), collectively suggesting that monogenic PD accounts for 5–10% of all PD cases. This aspect of causation in PD has been extensively reviewed elsewhere (5,6)

Perhaps more disruptively, genetic studies centered on idiopathic cases have successfully identified genetic risk factors under the common disease common variant (CDCV) model. In this review, we will focus on how these genetic findings have shaped our understanding of sporadic PD pathophysiology and draw attention to a number of interesting pathways that result from these analyses.

Genome-wide Association Studies in PD

The CDCV hypothesis suggests that there is a cumulative effect of multiple genetic variants, each of low penetrance, to the risk of an apparently idiopathic disease, such as PD. By looking at ∼1 million polymorphisms in matched case and control cohorts, genome-wide association studies (GWAS) use linkage disequilibrium to nominate genetic regions that impart moderate risk of disease and are present at appreciable frequency in the population (7). In 2009, two independent groups working on a Caucasian and Japanese population sample set reported between them five loci associated with PD (8,9). These included loci containing the genes SNCA, MAPT, LRRK2, BST1 as well as the locus PARK16. Collectively, these studies support the CDCV hypothesis in sporadic PD. Additionally, as SNCA and LRRK2 are genes for Mendelian forms of parkinsonism, GWAS provide strong support for the concept that sporadic PD is mechanistically related to inherited disease.

Further analysis suggested that polymorphisms in the MAPT and BST1 loci were exclusive to European and Japanese populations respectively, highlighting the heterogeneity between disparate ethnic groups. For the MAPT locus, the H2 haplotype that confers risks for disease in Caucasians is absent in Asian populations. BST1 did reach genome-wide significance in a European ancestry population in a PD meta-analysis (10). By combining five Caucasian-based PD GWAS data sets, six previously nominated risk loci including BST1 were confirmed whereas a further five loci were identified for the first time. The total population-attributed risk (PAR) of these 11 loci was determined to be ∼60% (10). These results demonstrate how meta-analyses can be used to find common variants that make small contributions to disease pathogenesis by increasing the resolution of genetic risk (11). The latest and most extensive meta-analysis, incorporating all European ancestry data sets, looked at over 7.8 million variants in a discovery set of >13 000 cases and 95 000 controls. (12). Following validation in the replication set, Nalls et al. found 28 risk variants in 24 loci that are associated with PD (Table 2). Six of these loci were novel whereas four others (GBA, GAK-DGKQ, SNCA and the HLA region) were shown to contain two independent risk variants.

It is critical to note at this point that the single-nucleotide polymorphisms (SNPs) reported to be associated with PD most are unlikely to be the actual functional variant but rather are in linkage disequilibrium with a polymorphism that does affect risk. This means that although many loci are assigned gene names, the true causative gene(s) remains formally ambiguous. The improved resolution offered by exome and whole genome sequencing technologies may help resolve this issue in the future. Also important to note is that most GWAS-nominated variants do not change amino acid sequences of the encoded proteins and, therefore, likely work by alterations in gene expression. Expression quantitative trait mapping (eQTL) has therefore been useful in prioritizing candidate genes (13).

With these caveats, there are some loci for which there are exceptionally strong single candidates. For example, within the reproducibly associated region on Chr4, the SNCA gene is particularly strong given its prior genetic and pathological association with PD. To date, over 800 SNPs within the SNCA gene have been reported with nearly half showing a significant association with sporadic PD (http://pdgene.org/view?gene=SNCA). Collectively, these variants may have a substantive effect as a PAR of 12% for the SNCA region has been reported (14). Mechanistically, polymorphisms in the SNCA, particularly around the 3′ untranslated region, may affect RNA processing and ultimately protein expression (8). Additionally, polymorphisms within the microsatellite region Rep1 located in the promoter region ∼10 kb upstream from the SNCA gene start site alters susceptibility to PD (15). Expansion of Rep1, a polymorphic mixed-dinucleotide repeat, has been shown to increase α-synuclein expression in human brains (16,17) and transgenic mouse models (18). Increased expression of α-synuclein is known to cause parkinsonism in a dose-dependent manner as demonstrated by individuals harboring SNCA copy number variant (19,20). As duplication and triplication, respectively, result in a 50 and 100% increase in a-synuclein expression, it is feasible that chronic increased α-synuclein expression in the order of as little as 10% may increase the risk of developing PD in the range of that nominated by GWAS (21).

The second significant variant to come out of the meta-analysis maps to the MAPT locus on chromosome 17, with the top SNP being rs17649553. The PAR attributed to this locus is 10%, second only to SNCA (14). Although MAPT has been associated with Progressive Supranuclear Palsy (PSP) (22), another parkinsonism disorder, its identification as the second highest-ranked risk factor for PD was somewhat surprising. In contrast to the association of the H1 haplotype with PSP (23), for PD it is the H2 haplotype. This is consistent with the lack of association between MAPT haplotype and PD in Japanese samples (9), as discussed earlier. The MAPT locus is large, containing several genes and the H1/H2 haplotypes do not recombine, meaning that it is plausible that another gene rather than MAPT is associated with PD. However, given that MAPT mutations are associated with frontotemporal dementia with parkinsonism and that tau pathology is sometimes observed upon postmortem analysis of idiopathic PD or some familial mutations (24,25), MAPT remains the most logical candidate for PD risk in this locus. Whether the variants affect expression of MAPT, as is likely for SNCA, or other aspects of gene regulation such as splicing, is not yet clear.

Along with SNCA and MAPT, a third gene that had been associated with inherited PD, LRRK2, was also identified in the original PD GWAS (8,9). Interestingly, the associated SNPs were located in the 5′ region of the LRRK2 locus, again suggesting effects on transcript processing and expression. LRRK2 is a kinase, and neuronal toxicity associated with increased LRRK2 kinase activity has been documented (26,27). The G2019S mutations, which elevates kinase activity, account for ∼1% of idiopathic PD cases in Caucasian populations (28), probably due to incomplete, age-dependent penetrance (29). Therefore, although not proven, it is possible that increased LRRK2 expression, leading indirectly to a modest overall risk by increasing net kinase activity, might be relevant to disease risk. In contrast to SNCA and MAPT, there may be risk associated with variants that change amino acids, likely to be rare individually but cumulatively important in a large gene such as LRRK2. For example, in Asian populations, four exonic variants (G2019S, G2385R, R1628P and A419V) were found to be risk factors associated with PD susceptibility in a LRRK2 centered meta-analysis (30).

Numerically, GBA could be argued to be the most exciting risk factor gene identified as mutations in this gene are more common than SNCA or LRRK2 (31). Additionally, variants in the GBA locus confer an odds ratio of 1.84 for disease risk, the highest of all PD-associated genes. Supporting a general link to α-synuclein pathology, GBA is also a risk factor for dementia with Lewy bodies at an even higher odds ratio of 8.28 (32). GBA encodes a lysosomal hydrolase, glucocerebrosidase (GCase), which converts the lipid glucocerebroside into glucose and ceramide. Loss of GBA function results in abnormal lipid accumulation and in an autosomal recessive manner presents as the lysosomal storage disorder Gaucher disease characterized by the presence of glucocerebroside-laden lysosomes in macrophages (33). Over 300 GBA mutations have been reported with N370S and L444P being the most common. Both Gaucher patients and heterozygous carriers have an increased likelihood of developing PD with 2–10% of PD patients, depending on ethnicity, harboring GBA mutation (34). Carriers of a single GBA mutation have an odds ratio of 5.4 for developing PD (35). Some mutations, like E326K, appear to be exclusive to PD patients (36). A heterozygous E471G mutation in SCARB2 has also been reported as a modifier for Gaucher disease (37). Interestingly, SNP rs6812193, located upstream of SCARB2 gene and in the locus shared by FAM47E, is a reported risk factor for idiopathic PD (12,38). SCARB2 encodes for lysosomal integral membrane protein type 2 (LIMP-2), a receptor for GCase (39). Therefore, collectively, these data show that GCase and associated genes represent a strong risk factor for PD likely due to a partial loss of function.

As discussed earlier, as well as loci including Mendelian genes, there are an additional 20 loci that confer a variable amount of risk to PD development (Fig. 1). To go through each of them would become rather repetitive and, at many loci, there are no candidates that can be strongly prioritized. However, we will briefly discuss some of the loci that have been replicated across studies.

The HLA locus, specifically the HLA-DQB1 allele, was reported to be associated with PD prior to GWAS (40). However, the first SNP in this region to reach genome-wide significance was situated in HLA-DRA (41). Subsequent associations were reported in HLA-DRB5 (10) and HLA-DRB1 (42). The most recent meta-analysis returned the HLA-DQB1 allele as a risk factor for PD (12). The reasons for the uncertainty around which exact allele influences risk is likely due to the HLA region being extremely polymorphic and also displaying a high density of closely located genes and complicated patterns of linkage disequilibrium (43). It is also plausible that multiple HLA loci are associated with PD (44). Of note, the HLA region harbors multiple risk alleles that alter susceptibility to Multiple Sclerosis (45). Nonetheless, the data suggest that one or more genes in the HLA region influence PD risk.

The PARK16 locus contains a risk factor with a relatively strong effect, with an OR of 1.3 for rs957211 in the original Japanese GWAS (9). Reflecting the locus designation, rather than a single named gene, there are three candidate genes in a single linkage disequilibrium block (NUCKS1, RAB7L1 and SLC41A1) with two additional genes nearby that might also be candidates (SLC45A3 and PM20D1/FLJ32569). Therefore, unlike the loci that also contain Mendelian genes, there is no clear single candidate in this region. Using an eQTL approach might slightly narrow the number of candidates, but there is signal at both NUCKS1 and RAB7L1, and some associations between cytosine methylation and risk allele (12). Therefore, for PARK16, there is no clear single gene that explains disease risk, and the locus may be complex with epistatic interactions between more than one gene.

Similar situations occur at other GWAS loci for PD. On chr4p, there is a risk locus including the gene encoding cyclin G-associated kinase, GAK, and Diacylglycerol Kinase, DGKQ, that are all reasonable candidate genes. To date, no evidence has been presented showing that any of the genes have eQTL or other associations. The one candidate that might be most logical is GAK, in that mutations in a homologous gene, DNAJC6, that encodes for auxilin are associated with an early onset form of parkinsonism (46). Along similar lines, there is signal from GWAS around the gene encoding GTP cyclohydrolase (GCH1), a key enzyme in the synthesis of dopamine that is a known cause of dopa-responsive dystonia (47). GCH1 is reportedly a substantive risk factor for PD with heterozygous variants conferring an odds ratio of 7.5 for PD development (48). Enrichment of GCH1 variants in PD patients has been confirmed in a second smaller study (49).

Implicated Molecular Pathways

These results show that even sporadic PD has a significant genetic component, but that this is oligogenic in nature and each locus imparts a small to modest contribution to lifetime risk of disease. Although there is ambiguity about which gene is the functional effector at many loci, it is reasonable to infer that there are one or two genes per locus and that they should share some underlying biological connections. It is noteworthy that for familial PD, factors that govern mitochondrial functioning and clearance are given considerable weight, especially in the recessive diseases related to PINK1/parkin. For sporadic PD, the complement system and innate immunity are highlighted, along with aspects of regulation of vesicular transport. There are likely even relationships between these major themes, as protein degradation and its regulation is a component of several groups of these genes. We will therefore discuss which of these pathways to parkinsonism are the most promising in the context of tying together multiple nominated PD risk genes.

Concluding Remarks

While genetic analysis of PD risk in both familial and sporadic cases has provided us with a wealth of information, in a sense we currently have jigsaw puzzle pieces without knowing the full picture that we are trying to reconstruct. However, it is clear that there are specific sets of interactions between implicated genes for some forms of PD. For example, in recessive parkinsonism, there are strong ties between Parkin and Pink1 (130), whereas we have proposed a link between Lrrk2, Rab7l1 and Gak by virtue of the three proteins being in the same complex (118). As discussed earlier, there are ways to extend these analyses out further to include other genes for PD and related disorders.

By working through the examples given here, we propose two generic pathways that might contribute to PD pathogenesis and, interestingly, these may indicate both cell-autonomous and non-cell-autonomous mechanisms. Neuroinflammation would involve both neurons and microglia (and potentially other immunologically active cells) whereas the autophagy-lysosome system, dependent as it is on microtubules, would lead to more restricted damage pathways. These are not meant to be mutually exclusive competing hypotheses as it is clear that some proteins, α-synuclein in particular, might be involved in both synaptic function (hence neuronal) but also in triggering neuroinflammation via TLR signaling. The autophagy pathway is likely important in both neurons and non-neuronal cells, further blurring these clean lines. Overall, it is our view at this time that PD pathways include both neuronal and non-neuronal cells and that the risk of PD is complex because both of these cell types can be impacted.

Thus far, GWAS has proved to be a powerful and successful tool for identifying common genetic variants associated with increased risk of PD development.

However, the danger with this approach is that because the GWAS linkage peaks are broad, several of the nominated candidates in the paragraph above may not actually be the causal genes. Where a given biological process is a broad category, the probability of overlap by chance alone is raised. Clarification of the actual functional gene at each locus is therefore required before re-assessing biological pathways.

Both eQTL mapping and epigenetic changes have been observed for a number of GWAS loci that might clarify some regions (12), but these might be further refined by improved technical approaches such as replacing current array data sets with RNA-Seq or by adding in new data sets. The examples of protein complex identification above (118) show how useful unbiased, ideally genome-wide functional screens can be. One could imagine adding other proteomics approaches, both in human brain and as screening tools for interactions, to refine candidates. Finally, siRNA-based screens against PD-relevant phenotypes, which might include immune regulation, autophagy or microtubule transport, could provide additional resolution at specific loci.

One of the biggest remaining challenges to the field will be to move from candidate genes to actual functional variants. A limitation in assessing the CDCV hypothesis is that we usually do not have full genomic coverage, relying instead on assayable proxy markers for genetic variation. Re-sequencing in high depth the PD loci, perhaps in combination with exome (or whole genome) sequencing, will likely yield resolution at some loci. It will still be important to test candidate variants in functional assays, again highlighting the importance of having validated assays that are specific for PD relevant processes.

Overall, there has been a huge and surprising acceleration in our understanding of pathways leading to parkinsonism in the past decade. We hope that this pace will continue in the next few years and start to impact how we view, and eventually treat, this devastating disorder.

Funding

This research was supported by the Intramural Research Program of the NIH, National Institute of Aging.