NeuroGeM, a knowledgebase of genetic modifiers in neurodegenerative diseases

Data collection

NeuroGeM is a comprehensive collection of literature data on genetic modifiers of NDs and associated genetic information from a variety of databases (Figure 1). The ND models include AD subclassified as AD_Aβ (amyloid-beta models) and AD_Tau (tau models), HD, PD, Spinocerebellar ataxia type 1, 3 and 7 (SCA1, SCA3, SCA7), Amyotrophic lateral sclerosis (ALS) and generic polyQ-induced disease in D. melanogaster, C. elegans, and S. cerevisiae. At the time of data compilation, all known high-throughput (HT) screens for modifiers carried out in the model organisms and a handful of low-throughput (LT) experimental results were included. Overall, NeuroGeM contains 87,864 experimental records (3,618 for modifiers and 84,246 for non-modifiers) from the 9 different disease models in the three different species (Table 1).

Importantly, we continually update the knowledgebase with newly published results. In addition, users can submit their own data upon request of a login and the uploaded data will be made accessible to all users after curation (see Additional file 1).

In order to provide comprehensive information on genetic modifiers, we integrated relevant data from other sources into NeuroGeM. All entries in NeuroGeM contain information on gene function/annotations and are accompanied by direct links to the relevant information on FlyBase for D. melanogaster (ver Feb 2012) [6], WormBase for C. elegans (ver WS230) [7] and SGD for S. cerevisiae (downloaded in Jan 2012) [8]. Each gene entered in NeuroGeM contains an ID that is identical to the primary ID of the gene in its respective genome database (FlyBase, WormBase, or SGD). These IDs will allow users to easily access other databases and avoid the effort required for ID conversion. NeuroGeM records have a link to the PubMed entry of the original study from which the records stem.

As protein-protein interaction networks allow identification of functionally associated proteins or important functional clusters [9, 10], NeuroGeM visualizes the protein interaction sub-network around a queried gene; each protein node in the network is color-coded according to the available experimental results deposited in NeuroGeM. For this feature, NeuroGeM utilizes the protein interaction data from STRING (ver 9.05) [11]. In order to facilitate the identification of genetic modifiers with the same function or involved in the same process, NeuroGeM provides an ontology-based search functionality that searches for genes by GeneOntology (GO) annotations and the hierarchical structure of GO terms [12].

As users might be interested in finding homologs of the genetic modifiers entered in NeuroGeM, we also integrated homologous gene data from NCBI HomoloGene (build 65) [13] and InParanoid (ver7) [14, 15]. Orthologs are defined as genes in different species that have evolved from a common ancestral gene, while paralogs are genes related by duplication within the same species that often have different functions. Homologs are either paralogs or orthologs (for details see NCBI HomoloGene (build 65) [13] and InParanoid (ver7) [14, 15]). The homology data covers not only the three model organisms but also H. sapiens and M. musculus, though no modifiers from these organisms are deposited in NeuroGeM yet. Cross-linking genes via homology should facilitate the expansion of modifier studies in other organisms; for instance, confirmation of important modifiers in higher organisms. Gene information, GO annotations and protein interaction data for human and mouse genes were also integrated to help users search for modifiers that are homologous to human and mouse genes of interest. (source: EMBL Rel 68 [16], HGNC downloaded in Jan 2013 [17]; EMBL Rel 68 [18]; and MGI downloaded in Dec 2012 [19]). As detailed information on genes and proteins is frequently updated in their source databases, NeuroGeM also provides links to those databases when available.

Database implementation

The database was implemented with a web interface compatible with common web browsers to provide access to researchers. The data is stored in a relational database using a MySQL 5.0.59 server. Data processing and HTML generation for displaying information are carried out using PHP 5.3.3. Javascript and AJAX technology are used to improve search functions. CytoscapeWeb [20] is employed to visualize protein-protein interaction networks. The current database is running on Redhat Linux 5.6 with an Apache server 2.2.3. All the data in NeuroGeM can be downloaded as plain text files.

Categorical search for studies that identified modifiers

On the front page of the NeuroGeM web site, research results are organized according to disease and model-organism categories: studies that identified modifiers were categorized according to model organism (n = 3) and diseases (n = 9 including two AD subtypes) (Additional file 2: Figure S1). Clicking on an organism icon or disease-specific icon directs the user to a list of publications that belong to that category. Clicking on a pie chart at the intersection of a specific model organism and disease type directs the user to a list of publications that searched for modifiers in the corresponding disease model and organism. Clicking on a specific publication then directs the user to a list of genes that have been tested in that study.

For instance, clicking on the pie chart at the intersection of D. melanogaster and HD (Additional file 2: Figure S1) lists 24 publications in which genetic modifiers of HD have been identified in D. melanogaster. Clicking on the PubMed ID 17984172 will open a web page that lists the 26 genes that have been examined in this specific study and provides a summary of the experimental records deposited in NeuroGeM for each of these 26 genes (see below for the detailed explanation of the record summary). Clicking on the gene name Nup44A will then direct the user to the genetic modifier information page of Nup44A, which contains the details of the results of different genetic modifier screens for this gene as well as additional information that is discussed in further detail in the “Genetic modifier information” section below.

Keyword-based search for specific genes

NeuroGeM also provides different search methods from a unified search box (Figure 2). As a user starts typing letters in the search box, NeuroGeM suggests matching keywords (gene name, synonyms, IDs, GO terms, etc.). If the user selects one of the suggestions, NeuroGeM will open this gene’s modifier information page (Figure 2, red arrow). If the user presses the ‘Search’ button instead, NeuroGeM will list all genes that contain the keyword in their names, synonyms, or IDs. Then, the user can open a specific genetic information page by clicking on one of the listed genes (Figure 2, blue arrow). For example, when typing ‘nup4’ in the search box, several genes that contain the keyword ‘nup4’ are suggested. Selecting the D. melanogaster Nup44A gene from the suggestions will open this gene’s page directly. On the other hand, entering ‘nup4’ and clicking the ‘Search’ button will list all genes that contain the keyword and show respective experimental record summaries. The user can open the Nup44A gene page by clicking on the gene name Nup44A of the listed genes.

Ontology-based search for related genes

The keyword-based search directs the user to a specific gene, but this feature is not appropriate when searching for functionally related genes, e.g. searching for genes involved in the cell cycle. For such a relation-based search, we provide the ontology-based search. To use the ontology-based search, the user has to type a GO term or GO ID in the search box. Just like the keyword search, suggestions will pop up and the user can select one of the suggested GO terms. Then, NeuroGeM searches for not only genes with the user-specified GO term but also genes with a related (child) term. This ontology-based search will assist users in the identification of modifiers that are associated with specific cellular functions or processes.

For instance, as shown in Additional file 3: Figure S2, the query of ‘cell cycle (GO:0007049)’ in D. melanogaster returns 656 genes that are involved in the cell cycle: 622 out of 656 genes have been evaluated experimentally, and 256 out of 622 genes have been identified as modifiers. Drosophila’s Nup44A gene is also shown in the gene list since Nup44A has GO annotations for regulation of mitotic cell cycle (GO:0007346) and regulation of meiotic cell cycle (GO:0051445), which are child terms of cell cycle. Clicking on the name of Nup44A will direct the user to Nup44A genetic modifier information page.

Genetic modifier information

At the top of the genetic modifier information page (Figure 2a), the number of experimental records that are available in NeuroGeM for a specific gene and its homologs are reported (record summary) for each experimental scale (“L” stands for LT, “H₂” for secondary HT, and “H₁” for primary HT). If experimental data that identifies a specific gene as a modifier has been entered in NeuroGeM, this fact is highlighted in red, light red if it is only one HT experiment, and dark red if it is at least two HT experiments or at least one LT experiment. Similarly, records that report that a gene is a non-modifier are colored in light and dark blue. For example, there are currently two records in NeuroGeM for Nup44A indicating that Nup44A (FBgn0033247) is a non-modifier in HD, one based on a LT and one on a primary HT experiment. In addition, there are two records of LT experiments indicating that Nup44A is a modifier in SCA1. By clicking on a specific record in the “record summary” at the top of the page (e.g. “2 0 0” under the column header “L H₂ H₁” for Nup44A and SCA1), the user is immediately guided to the details of that entry (Figure 3).

Below the report summary, detailed information of the gene is displayed on the left side (Figure 2b), including synonyms, alternative gene names, and IDs used in other databases as well as GO annotations. On the right side (Figure 2c), NeuroGeM displays the protein-protein interaction sub-network around the gene, in which interacting proteins are colored by the type and result of experiments that tested them as modifiers. Specifically, the left half of each node (protein) is colored according to the evidence for it being a modifier. The right half of each node is colored according to the evidence for it being a non-modifier. The same coloring scheme as for the record summary is used. Importantly, the protein-protein interaction network and the coloring are organism- and disease-specific (coloring by different diseases can be selected below the network). Users can navigate to other genetic modifier information pages by clicking on the corresponding nodes. Figure 2c shows the protein-protein interaction sub-network of Nup44A colored according to the results of screens for modifiers of HD in Drosophila. A look at the network immediately reveals that Nup44A, Nup75, and Rae1 were identified as non-modifiers, while Nup107 was identified as modifier of HD in Drosophila. By contrast, SmB, SmD1, Cbp20, and Nup154 were identified as modifiers in some experiments but not in others.

In the next section of the genetic modifier information page, experimental details are displayed (Figure 3). Experiments are categorized by the disease model, and the number of experimental records for each disease model is shown with respect to experimental scale by using the same coloring scheme as for the report summary. The reported experimental details include: (i) type of modification; indicates whether the experiment found the gene to be a suppressor, enhancer, or non-modifier, (ii) mode of action; reports whether the queried gene modified aggregation size/number or changed disease symptoms, (iii) disease induction; denotes which (mutant) gene was used to cause disease symptoms, and shows the mutant gene and its expression cassette information, (iv) modulation method; denotes how the expression of the queried gene was modulated (e.g. over-expressed, knocked out, repressed by RNAi), (v) experimental scale; denotes the scale of performed experiments (LT, primary HT, secondary HT; secondary HT stands for experiments to confirm the results obtained from primary HT experiments), (vi) measurement; denotes what was quantified to identify a modifier and (vii) cell type; denotes the cell lines or stocks utilized in the experiment. For instance, Nup44A was tested as modifier in a Drosophila model of SCA1 that was created by expressing Ataxin-1 with a polyQ expansion of 82 (Figure 3). The impact of the overexpression of Nup44A on the disease model was quantified based on changes in the severity of an eye phenotype. Nup44A was categorized as a suppressor, indicating that the over-expression of the Nup44A gene alleviated the severity of the eye phenotype.

Below the experimental details of a specific gene, its homologous genes are displayed with genetic information, protein interaction sub-network, and experimental details if available.

Search for orthologs of human and mouse genes

The current version of NeuroGeM does not contain any genetic modifiers in human and mouse (to be included at a later stage). Nevertheless, researchers studying genetic modifiers of NDs in the three model organisms are likely to be interested in the homologous genes of the modifiers in higher organisms such as human and mouse. Thus, we also integrated human genome information from the EMBL and HGNC databases and mouse genome information from the EMBL and MGI databases. The user can search for human and mouse genes using their gene names or EMBL IDs, and then obtain not only information on the queried genes but also information on their homologous genes in D. melanogaster, C. elegans, and S. cerevisiae. For example, Drosophila’s Nup44A gene and its homologous genes in other species are listed in Figure 2a and the information on those genes is displayed at the bottom of the Nup44A gene page (omitted in Figures 2 and 3). The user can also search for human and mouse genes homologous to Nup44A by their gene names (Seh1l and SEH1L) or EMBL IDs (ENSMUSG00000079614 and ENSG00000085415) in the unified search box.

Applications of data in NeuroGeM

In order to demonstrate the broad applicability of NeuroGeM and how it can provide new understanding, we performed a variety of meta-analyses using the genetic modifiers data from NeuroGeM. In this section, the results of the meta-analyses are discussed briefly. The detailed methods, results and discussion for the meta-analyses are available in the Additional file 1. Due to the abundance of both HT and LT experimental data from D. melanogaster, mainly results obtained from the meta-analysis of genetic modifiers of D. melanogaster are presented here. Results from the analysis of genetic modifiers in other model organism as well as the comparison of modifiers in all three model organisms are available in Additional file 1.

1. i)

   NeuroGeM can be used to identify biological processes that are enriched within genetic modifiers in a specific disease or in groups of diseases. Genes with annotations for these processes can be prioritized for testing in other model organisms or for drug screenings. A meta-analysis of the data deposited in NeuroGeM revealed that modifiers across species are often involved in protein folding (Figure 4a). However, they account for only 3% of all genetic modifiers. The analysis revealed that modifiers are equally often involved in cell cycle and splicing, accounting for 7% and 3% of all genetic modifiers, respectively. This analysis result suggests that researchers expecting to discover more genetic modifiers should focus their efforts also on genes involved in cell cycle and splicing, biological processes that are also often enriched in modifiers. As shown in Figure 4b, a correlation analysis of modifiers between diseases revealed that polyQ diseases (HD, generic PolyQ, SCA1, SCA3, and SCA7 in Figure 4b) share many genetic modifiers and non-modifiers which are not seen in AD models, which is consistent with a previous report [21]. Specifically, a strong anti-correlation is observed when comparing the modifiers and non-modifiers of AD_Aβ and SCA3. Many SCA3-specific genetic modifiers are involved in protein folding and splicing [22, 23], while AD_Aβ-specific modifiers are involved in protein synthesis [24] . Untested genes involved in modifier-enriched processes can be prioritized in future screens to confirm these trends and elucidate their mechanisms. See Additional file 1.

    

ii) NeuroGeM allows easy identification (by non computational experts) of genes that modify the neurodegenerative toxicity in several ND models. Hence, cross-disease comparisons can identify potential generic modifiers that then can be tested experimentally in other disease models in other organisms (rodents) or compared to human genetic data. A first search for generic modifiers revealed that the genes DnaJ-1, thread, Atx2, and mub are modifiers in 5 out of 7 ND models in D. melanogaster (Figure 4c). Interestingly, DNAJB4 and BIRC3, the mammalian orthologs of DnaJ-1 and thread, have recently been shown to reduce neuronal cell death when up-regulated in multiple mammalian ND models (see Additional file 1). Moreover, Atx2 has recently also been associated with an increased risk for ALS [25]. Further experiments are necessary to confirm this hypothesis that DnaJ-1, thread, Atx2, and mub are generic modifiers.

iii) Equally, NeuroGeM facilitates the identification of genes that only affect the phenotype of specific NDs. Though all NDs are caused by aggregates, they definitely show different pathophysiology. In this regard, genes capable of modulating disease phenotype only in a specific ND give us hints to understand the difference in disease progression. For instance, modifiers currently confined to AD_Tau in D. melanogaster include sgg and par-1 (Figure 4d). This finding is consistent with the specific importance of hyper-phosphorylation of the tau protein in AD, which is a process that may be accelerated by par-1 and sgg [26, 27]. In S. cerevisiae, modifiers are enriched in protein synthesis in AD (RTG3, TEC1, SPT21, PPR1, MBP1, SRO9, SLF1, and SLS1), protein folding in HD (HSP26, HSP42 and APJ1), and transport in PD (FUN26, YCK3, and GOS1), which is consistent with recent results (Additional file 1).

vi) Most importantly, NeuroGeM facilitates the identification of new, so far untested modifiers. Mapping of genes on the protein interaction networks allows identification of new, untested genetic modifiers based on guilt-by-association. We illustrate this idea on genes involved in anti-apoptosis (GO:0006916). We first obtained 24 anti-apoptotic proteins in D. melanogaster (Figure 5a), and among these genes debcl, Buffy, and thread are interconnected with each other in the protein network (Figure 5b). In order to investigate whether genes interacting with these anti-apoptotic modifiers could also be modifiers, we extended the sub-network by adding proteins that interact with the three proteins. This extension can be easily done, as NeuroGeM allows the user to navigate from one gene to another by clicking on a node in a network. The newly added genes are highly interconnected with each other. Detailed literature surveys of the genes connected to debcl, Buffy, and thread revealed that 5 out of 15 interactors (marked in green in Figure 5b) are modifiers or at least highly related to disease progression. For example, one of the interactors is Ark; inactivation of Ark (FBgn0263864), a key regulator of apoptosis, is known to suppress formation and ubiquitination of polyQ aggregates [107] (Please see Additional file 1 for the details about the other four interactors and potential modifiers). Hence, the search tools of NeuroGeM will facilitate the identification of new modifiers that are involved in specific pathways or cellular processes.

D. melanogastergenes	D. melanogasterND models2	Ref	H. sapiensorthologs	M. musculusorthologs	Ref3
DnaJ-1	ADTau, HD, PolyQ, SCA1, SCA3	[30–34]	DNAJB1, DNAJB4, DNAJB5, DNAJB13	Dnajb1, Dnajb4, Dnajb5, Dnajb13	[35–40]
thread	ADTau, HD, SCA1, SCA3, SCA7	[21, 32, 41, 42]	BIRC2, BIRC3	Birc2, Birc3	[43–45]
Atx2	ADTau, HD, SCA1, SCA3, SCA7	[21, 41, 42, 46]	Atxn2, Atxn2L	Atxn2, Atxn2L	[47–52]
Hsc70-3	HD, SCA1, SCA7	[32, 42]	HSPA5	Hspa5	[53]
Hsc70Cb	ADTau, HD, SCA3	[5, 54, 55]	HSPH1(HSP110), HSPA4, HSPA4L	Hsph1 (Hsp110), Hspa4, Hsp4l	[56, 57]
Rpd3	HD, SCA1, SCA7	[34, 42, 58]	HDAC1, HDAC2	Hdac1, Hdac2	[59–65]
14-3-3epsilon	HD, SCA1	[66, 67]	YWHAZ, YWHAB, YWHAE	Ywhaz, Ywhab, Ywhae	[67–73]
CG5537	HD	[74]	UPRT	Uprt	N/A
Hsf	HD, SCA3	[75]	Hsf2, Hsf4, Hsfx1, Hsfx2, Hsfy1, Hsfy2, Hsf5	Hsf2, Hsf3, Hsf4, Hsfy2, Hsf5	[76–78]
Nipped-A	HD, SCA3, SCA7	[42, 54, 55]	TRRAP	Trrap	[79, 80]
Sec61alpha	HD, SCA3	[81]	Sec61A1, Sec61A2	Sec61a1, Sec61a2	[82, 83]
Nup160	HD, SCA3	[55, 74]	Nup160	NUP160	[84–86]
CG1109	HD	[74]	WDR33(WDC146)	Wdr33 (Wdc146)	N/A
Snap	HD	[66]	NAPA, NAPB	Napa, Napb	N/A
smt3	HD	[74]	SUMO1, SUMO2, SUMO3, SUMO4	Sumo1, Sumo2, Sumo3	[87–91]
Mef2	HD	[66]	MEF2A, MEF2B, MEF2BNB, MEF2C, MEF2D	Mef2a, Mef2b, Mef2c, Mef2d	[92–97]
chic	HD	[98]	PFN4	Pfn4	[99]
Rpt1	HD	[74]	PSMC2	Psmc2	[100–102]
Sin3A	HD, SCA1, SCA3	[5, 32, 34]	Sin3A, Sin3B	Sin3a, Sin3b	[103]
Rheb	ADTau, HD	[74, 104]	Rheb, RhebL1	Rheb, Rhebl1	[105, 106]

¹Orthologs were obtained from NeuroGeM in which protein homolog groups of NCBI HomoloGene and InParanoid were integrated.

²Disease models in which the genes were identified as modifiers.

³Studies in which the mammalian orthologs were identified as modifiers.