Mouse Phenome Database (MPD)

INTRODUCTION

Phenomics, the comprehensive study of phenotypes, complements advances in knowledge from genomics, pharmacogenomics and other -omics approaches. Most human diseases are complex, polygenic and subject to modifications by environmental, genetic and epigenetic factors. A phenomics approach requires collecting comprehensive high-dimensional phenotypic data, integrating it with other types of diverse biological data, and evaluating it with computational tools and bioinformatic approaches.

MPD is the only database of its kind—amassing, annotating, integrating and maintaining quantitative phenotypic data for the laboratory mouse, and providing online tools to access and analyze those data. MPD is an established community resource and research tool enabling investigators to access thousands of data points from hundreds of strains, saving time and money by precluding the need to (re)characterize strains. MPD provides many tools to facilitate mouse research. For example, MPD enables users to:

• Compare strains to identify mouse models of human disease.

• Select optimal genetic backgrounds for new or existing mutations.

• Take advantage of the MPD Protocol Library containing detailed, illustrated procedures that have been validated by experts in their fields (and further, to view data generated from these protocols to confirm experimental results).

• Discover phenotype–genotype relationships and

• Validate and refine results from traditional mapping studies.

More powerful mapping tools are increasingly essential and indispensable in the analysis of complex traits and systems genomics to dramatically reduce the broad, unwieldy intervals of previously identified quantitative trait loci (QTL) containing hundreds (if not thousands) of genes, down to the gene level. For several years now, the community has had access to high-density SNP data for 16 inbred strains at over 8 million genome-wide locations through the NIEHS-Perlegen resequencing effort (1) and for 94 strains at 132K locations provided by the Mouse HapMap Project (2). These data were used to develop a novel high-density genotyping array, the Mouse Diversity Genotyping Array (3) (see also http://cgd/tools/diversityarray.shtml). These SNP data sets are available in MPD: Perlegen2 (4), Broad2 (5) and CGD2 (6), respectively. MPD has Mouse Diversity Array data for 123 inbred strains at 550K locations. In addition, the Wellcome Trust Sanger Institute recently released whole-genome sequence data for 17 inbred mouse strains and is generating transcriptome data (RNA-Seq) for these strains (7,8). Taken together with high-quality phenotype data, transcriptome and dense genotypic data enable the identification of regions of the genome that may be causal to a phenotype of interest through correlative analyses. Sanger SNPs, indels and structural variants will be added to MPD and made available in the coming months.

MPD history, standards, schema and other topics have been presented elsewhere (9–11). Here, we review a few salient points regarding MPD contents. The recommended basic study design is a strain survey format, e.g. 10 females and 10 males of a large number of genetically stable, reproducible strains (such as inbred strains). Acquired data sets are annotated and formatted to meet a set of defined MPD standards. Each data set is accompanied by a detailed protocol; health status and environmental parameters of the test animals; and any other information essential to understanding and evaluating the data. Data and documentation are then bundled together into an MPD ‘project’, which is given a descriptive title and annotated with submitter information, references and funding acknowledgements.

MPD phenotypic data sets are typically quantitative, although we do house categorical data such as MHC haplotypes and a growing collection of annotated images. Data from microarray gene expression studies are considered a special class of phenotype data and handled accordingly. Summary statistics are computed as part of the MPD accessioning process, where the unit of analysis is an MPD ‘measurement’ with strain (by sex) being the analysis group. Each measurement in a data set is classified and integrated in the MPD phenotype category structure. Male and female data are not combined. Data from different MPD measurements are not averaged together although these data may be displayed in the same plot for comparison (composite/consensus view). As part of the accessioning process, regression analysis is performed on all measurements (pair-wise) to identify compelling correlations, which may indicate common genetic determinants or overlapping biological pathways. Results are stored to optimize and support queries based on correlations. Strain genotypes are collected and stored in MPD so that phenotypic and genotypic data can be linked or visually juxtaposed, facilitating the ability to determine how allele-specific variations translate to variation in mouse phenotype.

Our last NAR update was published in 2009 (11). Here, we summarize new data and data types, and describe major improvements in MPD functionality. This update refers to Supplementary Data.

CONTENTS

‘MPD current contents’ are summarized in Table 1, and a snapshot of ‘What’s New’ is shown in Table 2. Users can search the database by our keyword search, or browse by strain, phenotype category, data submitter, interventions, study design and other convenient groupings (Supplementary Data, Navigation and Finding Data: S1–S8). Briefly, over 750 strains of mice are represented in MPD with phenotypic and/or genotypic data available. Strain names are kept updated with current nomenclature, and linkouts to other databases are provided. Table 1 provides an inventory of strains, strain panels and strain sources (also Search or Browse Strains: S1, S2, S3). Investigators associated with MPD as data contributors or as formal collaborators represent approximately 150 institutions from 14 countries supported by over 80 funding agencies and research foundations worldwide (Investigators: S4; Large-Scale Initiatives: S5).

FUNCTIONALITY

MPD provides many options to view, analyze and download data (MPD Toolbox Demo: S13). Since our last NAR update, we have made significant improvements and added new or expanded functionality to the MPD Toolbox. Here, we present a comprehensive example to demonstrate the power and utility of MPD and to highlight some of our new tools and features. We start with MPD phenotype data (IGF-1 levels at 6 months of age) and use new MPD tools to confirm and further narrow published mapping results (Yuan1) (28,29). This example spans Figures 1–3 and illustrates the following new or updated features:

• Improvement of various tools with new layouts, more options and links to other tools and information (Figure 1).

• Development of interfaces to facilitate association mapping using the UCLA EMMA-Correction Server (see below and Figure 2).

• Implementation of a user interface for working with gene expression data and providing tools for finding significant correlations to phenotype data using an innovative variance-based, data-screening feature to minimize false positives (Gene Expression: S12; Correlations Center S14) (Figure 3) and

• Implementation of a ‘workbook’ feature to help users pare-down and gather candidate genes with an option to generate an output of gene IDs for quick copy–paste into other bioinformatics resources (Figure 3).

MPD has been exploring and evaluating methods for genome-wide association studies (GWAS) that may be developed into general tools for MPD users. One approach that has received a lot of attention is the ‘efficient mixed-model method’ (EMMA), which corrects for inbred strain population structure thereby significantly reducing the number of false positive associations (30). GWAS results for selected MPD measurements (e.g. testing at least 20 strains) are available in collaboration with the UCLA Computer Science and Bioinformatics group (ZarLab) using EMMA and the Mouse HapMap for 94 inbred strains at 132K locations (2). Wild-derived strains are automatically excluded from these analyses (see more details in Figure 2). For ineligible MPD measurements (<20 strains), MPD provides an option to produce input-ready files specifically formatted for a real-time run on the UCLA EMMA Server (mouse.cs.ucla.edu/emmaserver). Users can choose to include wild-derived strains and are able to further manipulate the input file, e.g. to delete specific strains, and to select other SNP data sets for interrogation.

In addition to the highlighted examples (Figures 1–3), MPD has also implemented the following since our last NAR update:

• Correlations Center: centralizes, organizes and explains all MPD correlation operations on a single webpage; and provides convenient links to relevant tools (Correlations Center: S14).

• Private Analysis Portal: users enter their own experimental data and run real-time regression analysis across the entire MPD to identify statistically significant correlations to existing phenotype and gene expression data (Private Analysis Portal: S15) and

• Tool to generate ratio or difference vectors for user-selected MPD measurements: users can view plots and identify independent MPD measurements that are highly correlated to the de novo generated vector (MPD Toolbox Demo: S13; Correlations Center: S14).

HIGH-LEVEL OVERVIEW OF IMPLEMENTATION

The specifics of our implementation have not changed significantly since our previous update (11). MPD runs on a SUSE Linux/Apache 2 system using open source database, HTML generation and graphics generation software, with custom-written C modules used in performance-critical areas. Phenotype data sets are typically submitted by data contributors as Excel spreadsheets. Genotype and gene expression data sets are usually downloaded by us from open public resources. Ad hoc processing methods are used to integrate these data into our standardized framework. In conjunction with data contributors, MPD staff creates supporting documentation in HTML. We add content and perform testing and QA on an internal development node; then this instance is copied over for public release. MPD follows an agile development philosophy with rapid prototyping and frequent updates; version numbering is not used. For more documentation, see MPD ‘BioDBcore Statement’ (S19). Information for developers is also available (For Developers: S20).

FUTURE MPD DIRECTIONS AND AN INVITATION TO INVESTIGATORS

MPD will continue implementing useful linkages to other databases and populating the database with phenotype, genotype and gene expression data. We continue to refine MPD data classifications with relevant ontologies, optimizing search and navigation functionality as well as facilitating data integration within MPD and across databases. We are in the process of loading Sanger SNPs, indels and structural variants derived from whole-genome sequencing of 17 inbred mouse strains (7,8), and we will take advantage of the availability of the RNA-seq data being released by Sanger on this set of strains. RNA-seq data will complement existing MPD data and will enable queries to determine expression levels of different alleles of a gene and to detect post-transcriptional mutations or other genetic alterations.

Many phenotypic domains are currently represented in MPD; however, the acquisition of new data is open-ended with the goal of collecting data on a broader scope (and in some cases to a deeper level for more granular phenotypes or endophenotypes) as well as collecting data generated from new, more sophisticated phenotyping technologies. Although we have added a significant number of phenotypes since our last update (Table 2), we will continue working toward accessioning more data for inbred strains and other strain types as well as for new panels such as the Collaborative Cross (34–36). We would like to take this opportunity to make an appeal for clinically relevant data in toxicology and specific disease areas: cancer, digestive diseases, kidney disease, neurological disorders, drug addiction and infectious diseases. To expand the scope and maximize the utility of MPD, members of the global scientific community are invited to contribute their strain survey data or join us in a coordinated effort to seek funding that will support these studies. This spirit of collaboration has shaped and will continue to guide the future growth and development of MPD. Researchers interested in contributing data or in collaborating on new phenotyping projects should contact us at phenome@jax.org. Data submission guidelines are accessible through the MPD homepage (Contributing Data: S21).

CITING MPD AND USER SUPPORT

For general citation of MPD, this NAR article may be used. In addition, the following citation format is preferred when referring to MPD projects or using MPD data sets.

Investigator(s) name (year project posted) Project title. MPD project symbol (e.g. Smith1) and/or accession number (MPD:XXX). Mouse Phenome Database Website, The Jackson Laboratory, Bar Harbor, ME, USA. (URL: http://phenome.jax.org), date of download or access.

For more information visit our website and search on ‘citing’.

MPD provides user support through online documentation (S17, S18) and via email (phenome@jax.org). We welcome user input and suggestions, which can be submitted anonymously through our Suggestion Box links at the bottom of most MPD webpages.

SUPPLEMENTARY DATA

Supplementary Data are available at NAR Online: Supplementary Table 1.

FUNDING

The Jackson Laboratory and National Institutes of Health (grant numbers HG003057, HL66611, AG025707, AG038070, MH071984, DA028420). MPD houses and maintains data from research projects supported by over 80 funding agencies and research foundations worldwide (Funding: S22). Funding for open access charge: National Institutes of Health (DA028420).

Conflict of interest statement. None declared.