Free the Data: One Laboratory s Approach to Knowledge-Based Genomic Variant Classification and Preparation for EMR Integration of Genomic Data

Transcription

1 SPECIAL ARTICLE OFFICIAL JOURNAL Free the Data: One Laboratory s Approach to Knowledge-Based Genomic Variant Classification and Preparation for EMR Integration of Genomic Data Lora J.H. Bean, Stuart W. Tinker, Cristina da Silva, and Madhuri R. Hegde Department of Human Genetics, Emory University, Atlanta, Georgia Communicated by Richard Cotton Received 9 January 2013; accepted revised manuscript 3 June Published online 11 June 2013 in Wiley Online Library ( DOI: /humu ABSTRACT: Current technology allows clinical laboratories to rapidly translate research discoveries from small patient cohorts into clinical genetic tests; therefore, a potentially large proportion of sequence variants identified in individuals with clinical features of a genetic disorder remain unpublished. Without a mechanism for clinical laboratories to share data, interpretation of sequence variants may be inconsistent. We describe here the two components of Emory Genetics Laboratory s (EGL) in-house developed data management system. The first is a highly curated variant database with a data structure designed to facilitate sharing of information about variants identified at EGL with curated databases. This system also tracks changes in variant classifications, creating a record of previous cases in need of updated reports when a classification is changed. The second component, EmVClass, is a Webbased interface that allows any user to view the inventory of variants classified at EGL. These software tools provide a solution to two pressing issues faced by clinical genetics laboratories: how to manage a large variant inventory with evolving variant classifications that need to be communicated to healthcare providers and how to make that inventory of variants freely available to the community. Hum Mutat 34: , C 2013 Wiley Periodicals, Inc. KEY WORDS: clinical genetics; mutation; variant classification; bioinformatics; mutation database Introduction Identification of a disease gene opens the door for clinical testing for individuals at risk for having a gene mutation. After a newly identified gene is reported in the literature, additional reports of disease-associated genetic variants may not be published for years, if ever. Current technology allows clinical laboratories to rapidly translate research discoveries into clinical genetic tests with data from only a small cohort of patients; therefore, a potentially large proportion of sequence variants identified in individuals with clinical features of a genetic disorder are detected by clinical laboratories and are often not published or made available. Additional Supporting Information may be found in the online version of this article. Correspondence to: Madhuri R. Hegde, Emory University, Human Genetics, 615 Michael Street, Atlanta, GA mhegde@emory.edu Clinical laboratories rely on two bodies of knowledge when interpreting sequence variants: data on the mutation spectrum in affected individuals and data on normal variation within the gene in the general population. The first body of knowledge, data from published studies, is widely available through the peer-reviewed literature; however, insufficient indexing, use of outdated nomenclature, and frequent updates to the human genome reference sequence present challenges to exhaustive review. Published data on disease associated genetic changes are curated by organizations such as the Human Genome Mutation Database (HGMD; and numerous locus-specific databases (LSDBs; for examples see Although valuable, these resources have limitations. Access to the most recent data in HGMD is available only through a paid subscription and the completeness and quality of LSDBs varies widely. The availability of data on genetic variation in the general population is just as vital to interpretation of sequence variants as that of data from affected individuals. The availability of population genetic variation data from large-scale sequencing projects such as the HapMap Project [2005; Altshuler et al., 2010], 1000 Genomes (www. 1000genomes.org), and more recently, the NHLBI GO Exome Sequencing Project (evs.gs.washington.edu/evs/) has fundamentally changed our ability to interpret sequence variants. Data generated within clinical laboratories by sequencing of affected individuals, their family members, and interpreted through review of published information represent a third body of knowledge vital for the interpretation of sequence variants. The interpretation of nucleotide changes identified by clinical sequencing is determined using the guidelines set by the American College of Medical Genetics and Genomics (ACMG) [Maddalena et al., 2005; Richards et al., 2008]. There are currently five terms commonly used by clinical laboratories to describe the significance of sequence variants: (1) pathogenic, (2) likely pathogenic, (3) variant of unknown clinical significance, (4) likely benign, and (5) benign. This classification system is steadily replacing older terms such as deleterious, disease-causing, single-nucleotide polymorphism (SNP), and polymorphism. The most appropriate interpretation of the clinical significance of sequence variants is constantly changing as new knowledge is gained. Most clinical laboratories recognize the importance of the data they hold. While some laboratories freely share this data, other laboratories consider such data intellectual property, vital to their business model. Those willing to share face several barriers. Inhouse data management systems often use nonstandard electronic storage methods, variables, software, and annotations. Additionally, data storage systems may be designed only to capture data associated with a particular case with limited or no ability to review C 2013 WILEY PERIODICALS, INC.

2 or manipulate summary data. Although many nascent attempts are underway, there currently is no centralized, publically available, curated clinical database that facilitates easy sharing of data between clinically laboratories. Software packages such as the freely available Leiden Open Variation Database (LOVD; and Universal Mutation Database ( provide easy, user-friendly mechanisms to share sequence variant data; however, data must be transferred from in-house databases to the application [Beroud et al., 2005; Fokkema et al., 2011]. The LOVD software has been extensively and successfully by groups with disease-specific interests such as The International Society for Gastrointestinal Hereditary Tumours ( and The Leiden Muscular Dystrophy Pages, Leiden University Medical Center ( Other emerging projects include the ClinVar ( effort from the National Center for Biotechnology Information (NCBI) and The Human Variome Project (HVP; Transfer of clinical laboratory data to any or all of these endeavors requires commitment of time and resources. This commitment can be significant for the clinical laboratory if sequence data are not stored in a structured, easily manipulated format. We describe here the two mature components of Emory Genetics Laboratory s (EGL) in-house developed data management suite, EmGen. The first, EmBase, our laboratory s approach to a highly curated clinical grade variant database with a data structure designed to easily transfer data to an electronic medical record (EMR) or publically available database to facilitate open sharing of variants identified in genes tested at EGL. In addition, this system tracks changes in variant classifications, generating notifications for the laboratory about which cases predate a reclassification and are, therefore, in need of a review and possible report update. To further facilitate this process, we have developed the second component, EmVClass, a Web-based tool that allows any user access to variants seen at EGL and the variant s current classification. A review of the classification for or an initiation of a discussion about a particular variant can be requested through a simple request form. Our approach provides a viable solution to making data from clinical laboratories available in two ways: first, information about sequence variants identified in a clinical laboratory are freely available to the scientific community, via either direct Web access to our inventory of variants or, when possible, transfer of variant information to a curated database; and second, changes in variant classification are tracked so that updates can be communicated to healthcare providers and ultimately integrated into an EMR. Methodology EmGen To facilitate the management of sequence data, we have developed a variant database and clinical decision support system (CDSS). Patient and sample identifiers are transferred from EGL s EMR, and sequence variants are recorded into our internally developed variant database and CDSS entitled EmBase and published on our Website using our Web tool entitled EmVClass. EmBase and EmVClass belong to a suite of custom-designed software known collectively as EmGen. EmBase EmBase provides the environment to track variant definitions (functional classification, size, and coordinates) and associate those changes with the individuals in whom they were identified in a structured environment allowing for efficient tagging of internally and externally derived annotations, genomic mappings, standardized variant nomenclature development, expert review, and ultimately, clinical classification. Patient identifiers are pulled into EmBase from EGL s EMR via a custom software application. Current HUGO Gene Nomenclature Committee (HGNC) gene names are ported to the local instance of EmBase from and updated quarterly. Local gene transcript reference sequences and exon structures are obtained from the NCBI Nucleotide/GenBank as.gbk files using the NCBI Entrez Programming Utilities ( nuccore). Sequence variation data, matched on genomic coordinates, are ported to the local instance of EmBase from the Human Gene Mutation Database (version ), dbsnp Build 137 (source data hosted by University of California, Santa Cruz Center for Biomolecular Science & Engineering [ and the Exome Variant Server version ESP6500 ( Predictions for the effect of amino acid changes on protein function are calculated and stored for all possible basepair substitutions at each coding nucleotide for all genes in Em- Base using SIFT version 4.04 ( and PolyPhen2 version ( Variant data from EGL s next-generation sequencing (NGS) and exome analysis pipelines are pushed to their annotated variant record in EmBase via matching on their Human Genome Variation Society (HGVS) variant nomenclature. EmBase and the aforementioned data resources all reside on a Microsoft Windows Server running Microsoft SQL Server 2005 (version ) relational database management system (RDMS) with the exception of the local instance of HGMD, which was ported to Oracle s MySQL RDBMS version EmBase is written in Microsoft Access 2007 as an Access Data Project (ADP) with a SQL Server 2005 RDMS backend and distributed as an ACCDR runtime package with domain user/sspi-trusted connection application and database level security. EmVClass The Web-based client interface EmVClass application ( genetics.emory.edu/egl/emvclass/emvclass.php) is written in PHP version and served by Apache version 2.0 for Linux. EmVClass provides a de-identified published view of EGL sequence variant data by pulling variants and their interpretations from Em- Base. No user registration is required to access data on this Website. A Web form is used to collect contact information from users when a response from EGL is requested. EmVClass has been reviewed by our institution s legal counsel. The application and the data posted have been approved by the legal counsel of our institution. Results The EmBase software application supports the collection of Sanger-confirmed sequencing variants identified through clinical testing at EGL in a highly structured data environment. Data in EmBase are maintained at the gene level, the variant level, and the patient level to manage internal workflow and reporting processes, as well as to allow the dissemination of EGL s variant knowledge base in a de-identified manner to healthcare providers and the genetics community at large (Fig. 1) HUMAN MUTATION, Vol. 34, No. 9, , 2013

3 Figure 1. Data flow. Solid line automatic or prompted data transfer. Dashed line manual data transfer. EMR electronic medical record, NGS next-generation sequencing, EVS exome variant server, and HGMD human genome mutation database. Gene Level Gene structures are built by assigning genomic positions to coding regions of clinically relevant NM mrna reference numbers as reported in HGMD or appropriate published reports (Supp. Fig. S1A and B). EmBase provides tools for the advancement of historic variant syntaxes to current transcripts and/or can maintain variants across multiple transcripts. Variant Level EmBase automatically queries the HGMD, dbsnp, and the Exome Variation Server (EVS), for potential matches based on genomic coordinates (Supp. Fig. S1C). Exact matches are confirmed by the analyst. HGMD and dbsnp annotation field are autopopulated once the accession number is added by the analyst. EVS annotations, as well as precomputed SIFT and Polyphen predictions, are automatically imported to the variant record. Historic nomenclature may be added to a searchable aliases field. For example, the MUTYH gene has multiple reported transcripts with differences in amino acid numbering. The variant record for pathogenic variant c.1220g>a (p.g407d) on the current reference transcript NM also displays the aliases c.1178g>a (p.g393d) on transcript NM and G382D from historic nomenclature. In addition, EmBase integrates EGL s NGS data allowing the user to determine if a Sanger-confirmed variant has been identified via NGS. For variants observed by NGS, vital run data (e.g., instrument used, analysis software used, coverage, score, base calls, etc.) HUMAN MUTATION, Vol. 34, No. 9, ,

4 Figure 2. Distribution of EGL variant classifications. are presented for each sample in which the variant was observed. Since common variants identified by NGS are not routinely Sangerconfirmed and entered into the variant record, this connection to NGS data (confirmed and unconfirmed) allows a more accurate estimate of in-house observed allele frequencies than the Sanger sequencing data alone. The use of structured data elements allow for dynamically produced HGVS nomenclatures as well as genomic coordinates based on embedded transcript maps. Total hands-on time to enter a new variant record in EmBase with all associated information requires less than 5 min. Literature references or other notes detailing why a particular classification is made can be added to a variant notes field. Patient Level At the patient level, the analyst reviewing sequence data enter observed variants in EmBase by either selecting an existing variant record or creating a new variant record (Supp. Fig. S1D). For each variant, zygosity is recorded. Once the analyst completes and certifies the case, the case enters the queue for director review. Variant classifications are reviewed and certified by a laboratory director. EGL does not routinely list benign variants on clinical reports (these are available upon request); therefore, at the patient level, benign changes are flagged as no report. Upon certification of the case by a laboratory director, the patient record contains all information needed to select critical pieces of information needed to generate a report from within EmBase or push information to an EMR, such as the EGL clinical reporting system, or create an HL7 file for use by outside systems. Reclassification of Variants The reclassification of a variant affects both variant level and patient information. Variant review, classification, and reclassification audit trails are maintained to prompt the need and timing of variant review as well as track those cases affected by variant reclassification and prompt for the generation of an updated or amended report. We have developed two reporting formats for reclassification of variants. If new information has become available since the time of the original report (e.g., data from EVS), an updated report is issued. If our improved bioinformatics processes locate information that was available at the time of the original report but not found and utilized (e.g., an unrecognized published report), an amended report is issued. Summary Data EmBase currently contains over 7,500 fully curated and extensively annotated unique variants found within over 400 genes associated with EGL s traditional Sanger single gene sequencing, as well as Sanger-confirmed variants identified by NGS gene panels, and whole-exome sequencing tests (Fig. 2). EmVClass EmVClass is EGL s Web portal to the Emory variant inventory. Data elements from EmBase are displayed in a de-identified manner. For each variant detected by EGL, the gene name, exon number, current transcript, HGVS nucleotide change, predicted protein change, variant aliases, EGL classification, and last reviewed date are presented. All personal identifiers associated with a variant are removed prior to the presentation of the data. Note that protected information extends to allele frequency (i.e., how many times a variant has beenobserved)aswellasdataoncombinationsofvariantsseeninan individual. These data, although readily available to the laboratory in EmBase, are not publically presented in EmVClass, to protect the privacy of individuals with rare genetic variants. The EmVClass search function requires entering either a gene name or transcript number (Supp. Fig. S2A). A search may be 1186 HUMAN MUTATION, Vol. 34, No. 9, , 2013

5 Figure 3. Classification strategy to manage follow-up of variants between the EmVClass user interface and EmBase. narrowed by entering a nucleotide or protein change, location, or class (e.g., SNP, del, and ins). This search function was specifically designed for perfect and partial match queries. For example, the current transcript number for the ACADM gene is NM A search for 16.4 in the transcript field displays not only NM , but also NM , NM , and NM as potential matches. The search function capitalizes on the transcript-aware feature of EmBase, matching on both current and historic transcripts. For example, searching for the classically named common ACADM mutation, p.k304e returns the record for this pathogenic variant under the recommended nomenclature p.k329e. After variants matching the initial query are returned (Supp. Fig. S2B), the user may continue to drill into the data by entering full text searches such as the classification (e.g., pathogenic). Once a user identifies a variant of interest in EmVClass through the search tool, the system allows the user to prompt a review for variants of unknown significance or open a dialog about the classification of a variant potentially resulting in a request for an updated or amended report (Fig. 3). Users of EmVClass are not required to register; however, users do have the opportunity to provide their contact information when asking for a response from EGL via a Web form that requests the user s name, contact information, and question. It is anticipated that this functionality may be used by other research facilities and clinical laboratories as well as healthcare providers who may wish to know more about how a classification was generated. Since the launch of EmVClass in July 2012, nearly 2,000 unique user page views have been recorded. Of these, approximately 70% were from within the United States, approximately 20% were from countries outside the United States, and the remaining approximately 10% could not be localized from their IP addresses (Supp. Fig. S3A and B). Less than 10 requests for variant reviews have been submitted via from the EmVClass Web page. However, numerous individuals contacting EGL by telephone indicate that they have already visited the Website, suggesting that the EmVClass tool is being used to meet immediate needs. Development Costs The development of the EmBase and EmVClass applications took two data application specialists approximately 1 year to complete. Development included an investment of about $20,000 in servers and computers. This time and effort includes the task of transferring historical laboratory data from former data structures to the current structure. The effort needed to transfer the EmBase and EmVClass applications to other laboratories would be dependent on the effort needed to incorporate historical data. Once launched, the EmBase application has reduced the effort required to record and interpret sequence variants. Discussion Our ability to interpret sequence variants will continue to increase dramatically through large-scale sequencing of normal individuals and those affected with genetic diseases. Many genetics professionals agree that providing patients who have undergone genetic testing with updated interpretations of their findings is ideal; however, there are social, ethical, legal, and practical barriers to large-scale efforts to reexamine previously identified variants on a regular basis [Hirschhorn et al., 1999; Hunter et al., 2001; Pyeritz, 2011]. The ACMG sequencing guidelines suggest that the laboratories should issue amended reports when new knowledge is gained. Applications such as EmBase and EmVClass are critical first steps in breaking through barriers to recontacting healthcare providers when information about genetic test results change and to ultimately integrate HUMAN MUTATION, Vol. 34, No. 9, ,

6 original, as well as updated, patient genetic testing results into their EMR. The process of issuing updated clinical reports when the classification of a sequence variant is changed is laborious; however, the EmBase system helps to streamline this task by displaying and tracking previous clinical cases in need of review following a variant reclassification. Using this system, such cases can be approached in a dedicated, planned manner making the process as efficient as possible. Classification changes likely to have the highest clinical impact (e.g., changing from pathogenic to benign) can be easily identified and given highest priority. The EmVClass Web interface adds an additional facet to prioritization for investigation and variant reclassification by allowing healthcare providers to request a reclassification. The greatest strength of EmBase and EmVClass is the highly structured nature of the data collected. These systems capitalize on current genomic resources to allow the identity of a variant to be constant through multiple genome builds and transcript updates. The EmBase system is designed to provide variant data in any form required by curated gene or disease-specific databases. In addition, data stored in this manner provide the basis for generating clinical reports directly from the system or pushing critical data elements to less sophisticated EMR systems in a simple widely used format, such as HL7. In the future, true integration in an EMR system would allow instant notifications regarding variant reclassification and the availability of clinical trials or research programs. Information with EmBase can be easily displayed, as demonstrated through EmV- Class, or transferred to curated databases or larger variant collection projects such as ClinVar or the HVP. A limitation to widespread use of any application such as EmBase and EmVClass is that individuals viewing the data must be aware of how variant classifications were made. Unlike cases reported in the peer-reviewed literature, the scope of the data available on EmVClass is limited. EGL uses a rigorous and conservative process when determining variant classification; however, when using data from a source such as EmVClass in the care of a particular patient, the laboratory reporting the variant should be contacted to get full information on how classifications were reached, as well as the accuracy and timeliness of the information provided. We provide our guidelines on our Website. In addition, EGL has the advantage of having extensive expertise in the areas of cytogenetics, biochemical genetics, and molecular genetics. We are in the process of developing an additional application, EmPhenotype. EmPhenotype is an online phenotype data collection system for case phenotypes. Data submitted from the referring healthcare provider will be used to carry out phenotype-assisted analyses. We anticipate this tool being particularly useful for NGS gene panels and whole-exome sequencing cases. Prompted (checklist) phenotypes as well as free text will be mapped to Human Phenotype Ontology terms and genes of interest that will have immediate and long-term benefits as a part of the CDSS (EmBase). We will continue to use this system to match sequence variants to other structured datasets, such as pharmacogenetic and clinical trial databases as they become available. Our long-term plans also include a fully developed clinical laboratory reporting system that will generate and transmit clinical reports in a structured fashion. We have demonstrated that sequence variants and interpretation of their clinical significance generated by a single laboratory can be openly shared using the straightforward, custom-designed clinical grade EmBase and EmVClass applications. Building on publically available and other easily accessible genomic databases, these applications have made management of sequence variant data within the laboratory more efficient and have improved quality and consistency of interpretation. Our experience suggests that implementing systems such as EmBase and EmVClass would not impose a time or financial burden on laboratories, rather the opposite is true. References The International Hapmap Consortium A haplotype map of the human genome. Nature 437: AltshulerDM,GibbsRA,PeltonenL,DermitzakisE,SchaffnerSF,YuF,BonnenPE, de Bakker PI, Deloukas P, Gabriel SB, Gwilliam R, Hunt S, Inouye M, Jia X, Palotie A., et al Integrating common and rare genetic variation in diverse human populations. Nature 467: Beroud C, Hamroun D, Collod-Beroud G, Boileau C, Soussi T, Claustres M UMD (Universal Mutation Database): 2005 update. Hum Mutat 26: Fokkema IF, Taschner PE, Schaafsma GC, Celli J, Laros JF, den Dunnen JT LOVD v.2.0: the next generation in gene variant databases. Hum Mutat 32: Hirschhorn K, Fleisher LD, Godmilow L, Howell RR, Lebel RR, McCabe ER, McGinniss MJ, Milunsky A, Pelias MZ, Pyeritz RE, Sujansky E, Thompson BH, Zinberg RE Duty to re-contact. Genet Med 1: Hunter AG, Sharpe N, Mullen M, Meschino WS Ethical, legal, and practical concerns about recontacting patients to inform them of new information: the case in medical genetics. Am J Med Genet 103: Maddalena A, Bale S, Das S, Grody W, Richards S Technical standards and guidelines: molecular genetic testing for ultra-rare disorders. Genet Med 7: Pyeritz RE The coming explosion in genetic testing is there a duty to recontact? N Engl J Med 365: Richards CS, Bale S, Bellissimo DB, Das S, Grody WW, Hegde MR, Lyon E, Ward BE ACMG recommendations for standards for interpretation and reporting of sequence variations: revisions Genet Med 10: HUMAN MUTATION, Vol. 34, No. 9, , 2013