Data Sharing in the Undiagnosed Diseases Network

Introduction

The Undiagnosed Diseases Network (UDN) is structured to make advances in multiple disciplines related to the fields of genomics and undiagnosed diseases. The activities of the seven clinical sites, coordinating center, and two sequencing centers that constitute the UDN Network are shown in Table 1.

In the following sections, we describe the history and evolution of the network in the context of the importance of data sharing and breaking down barriers to discovery.

History and Aims

A request for applications from the NIH Common Fund for the Coordinating Center (CC) to the UDN was released in November, 2012 (http://grants.nih.gov/grants/guide/rfa-files/RFA-RM-12-020.html), and the request for sites was released on January 29, 2013 (http://grants.nih.gov/grants/guide/rfa-files/RFA-RM-13-004.html. The UDN CC was selected several months prior to the selection of the new Clinical Sites, which allowed for careful study and codification of the processes used and refined by the NIH UDP and selection for those suitable for Network-wide adoption and refinement. The request for a sequencing core was announced on August 7, 2013 (http://grants.nih.gov/grants/guide/rfa-files/RFA-RM-13-018.html), for which two centers were selected (Medical College of Wisconsin and Baylor).

The UDN is composed of a CC, seven UDN Clinical Sites (six new sites plus the foundational NIH UDP), and two Sequencing Centers (Table 1).

The program goals are to provide improved patient access to state-of-the-art diagnostic methods, by expanding the available expertise and facilities serving patients with these unusual disorders and to accelerate discovery and innovation in diagnosing and treating these patients.

Data Sharing

The success of the UDN will depend on the collection and subsequent sharing of well-curated clinical and research data both within and outside of the network. These activities are grounded in a common Data Sharing and Use Agreement (DSUA). The DSUA is a legal agreement between institutions, in this case, signed by the designated institutional officials from each of the members of the UDN (including affiliated institutions), which establishes what data may be shared, the ways in which the information in the data set may be used, and how the data will be protected. The UDN DSUA was carefully reviewed in order to ensure consistency with the IRB protocols and consent forms, as well as relevant sections of the UDN Manual of Operations, for example, UDN Publications Policies. Coming to agreement about the content of the DSUA required several rounds of interaction with technology transfer offices of the member institutions (including affiliates), an activity that was led by the CC.

The UDN DSUA has features that are not typical of such agreements. For instance, in the context of the UDN, data sharing is integral to our ability to see participants, rather than simply facilitating secondary data analysis. To illustrate this point, consider that initial applications are received centrally by the CC through the online application, the UDN Gateway: the transfer of these data to the clinical sites at which these applications will be evaluated depends upon the existence of the DSUA. Another feature that distinguishes the UDN DSUA is that it explicitly enables the sharing of personally identifiable information, which is essential due to the nature of the research being conducted within the UDN. Furthermore, many DSUAs are constructed as pairwise agreements in which Site A agrees to share data with Site B: the UDN DSUA covers multidirectional data sharing among all components of the network. Finally, the UDN DSUA explicitly requires sharing of the researcher-facing data elements contained within the UDN gateway, which have been agreed-upon by the UDN Steering Committee. Examples of these data elements include medical history information, human phenotype ontology-encoded phenotypes, and sequencing data.

UDN participants consent to have their data shared, in accordance with the UDN informed consenting process. Given that the UDN has adopted a central IRB model (with the central IRB being located at NHGRI), there is a great deal of consistency in the informed consent documents. The consent documents contain identical information, save for necessarily site-specific language (e.g., HIPAA authorization language). Each participant in the database will be associated with a UUID (Universal Unique Identifier), which will be used as the primary identifier for all data associated with that participant. Role-based access and physical security controls that are aligned with the sensitivity of the data at each point of use and access will be employed.

The UDN data sharing philosophy is consistent with the goals of the NIH Data Sharing Policy (NIH Data Sharing Policy, http://grants.nih.gov/grants/policy/data_sharing/). The NIH states, “Data should be made as widely and freely available as possible while safeguarding the privacy of participants, and protecting confidential and proprietary data.” Thus, the UDN DSUA also addresses external data sharing. In particular, the DSUA allows for the coordinating center to provision deidentified study data (HHS - Office for Civil Rights - HIPAA; http://www.hhs.gov/ocr/hipaa/) in the database of genotypes and phenotypes (dbGaP) or other controlled-access data repositories.

As has been covered in numerous forums (Knowledge Exchange, http://repository.jisc.ac.uk/5662/1/KE_report-incentives-for-sharing-researchdata.pdf), there are few incentives for an individual investigator to share data: it is an additional step in the research process and may be perceived as reducing the career benefits of having gathered the data. The UDN encourages data sharing via multiple approaches, including the provision of infrastructure, policy, and data standards to support data sharing; ensuring that data sharing would be an automatic consequence of Network operations; and by fostering a culture of collaboration, openness, and mutual trust. The previously mentioned governance is key to maintaining this culture, as Network decision-making is transparent and collaborative.

Collaborations with External Groups

UDN data may be shared with investigators who are not currently part of the UDN, if the initiatives are complementary and consistent with UDN goals. This is true on an individual participant and aggregate UDN population basis. For example, if there are useful experts in a patient's phenotype outside the UDN, the data would be shared with them on an as-needed basis.

There will also be data-sharing initiatives on a larger scale. Once a candidate gene for a condition has been identified, finding the second case to support the hypothesis has been traditionally word of mouth and involves serendipity. That said, there have been several “matches” made this way, leading to a second case identified. The UDN is interested in collaborating with, and contributing to, external organizations to facilitate gene discovery and the identification of additional cases. To this end, in addition to collaborations within and between the UDN sites and workgroups, the Network has also been linking to other groups focused on similar goals and activities.

Data resulting from UDN efforts will be deposited in dbGaP, maintained by the National Center for Biotechnology Information at the NIH. Data may also be shared with other controlled-access databases, registries, and repositories, such as PhenomeCentral (a collaborative effort of a number of rare disease research communities), the NIH Global Rare Diseases Registry, and potentially with other condition-specific registries, such as those listed in Orphanet.

The UDN is a member of Matchmaker Exchange's collaborative effort to address the common challenge of exome and genome sequencing in both the research and clinical settings. The UDN steering committee has voted to contribute data to the Matchmaker Exchange network via PhenomeCentral. PhenomeCentral is the ideal vehicle since it is a repository for secure data sharing working in the rare disorder community, and is also part of Matchmaker Exchange. It is hoped that the UDN and its participants will benefit from PhenomeCentral's remote matching API for finding genotypically and phenotypically similar patients within the Matchmaker Exchange network.

Data Sharing for Innovation in Genomic Analysis

While one critical goal of the UDN is to identify the causes of disease in the genomes of the individuals that are studied, another complementary goal is to contribute to the science of genome interpretation. In the rush of enthusiasm surrounding the remarkable advances in human genetics, it is easy to forget that interpreting individual genomes is very hard, and remains error-prone. This reality is especially important to keep in mind given that one explicit aim of genetic diagnostics is to inform treatment options. In this context, mistakes in genome interpretation can result in real clinical harm. For these reasons, the UDN is committed to evaluating and refining best practices in the evolving science of interpreting patient genomes. Beyond broad sharing, there are other cross-institution initiatives, such as sharing approaches to analysis, which will move the field forward via developing the building blocks necessary to support the science of genome interpretation.

Although some mistakes are made in groups 1 and 4, these outcomes are generally straightforward. The main point to make about group 4 is the critical need to reanalyze the genomes regularly, given the rapid pace of new gene discovery. It is also a priority for the field to develop a view of whether there are any predominant explanations for some negative results, such as pathogenic variants that are refractory to identification by next-generation sequencing, regulatory pathogenic variants that are difficult to identify as causal, or oligogenic models. The UDN hopes to address these questions at least in part through comparisons of whole-exome and whole-genome sequencing.

Groups 2 and 3 are where considerable methodological developments are needed to permit the most appropriate conclusions to be drawn in these more challenging settings. This begins by sharing phenotype and sequencing data among the UDN sites and throughout the Matchmaker Exchange as a way to identify unrelated individuals with the same phenotype and variants in the same genes. Data sharing at this level also facilities the sites doing their own large-scale analyses that would otherwise not be possible. The transparency of analysis protocols allows for the identification of best methods and practices [Brownstein et al., 2014]. A percentage of exomes/genomes carry multiple candidate genes for their phenotypic condition. Resolution in this case depends on finding additional individuals and functional assessment of the multiple candidates, as well as detailed phenotyping to assess whether expected phenotype correlates of each of the candidates are observed. The UDN affords us the opportunity for appropriate genomes to receive this full interpretation.

Additionally, when considering the outcome of many interpreted genomes, we are often left with interesting candidates on the basis of a bioinformatic signature, but not enough is known about them to draw conclusions. For example, in an analysis of 103 trios, it was reported that 28.2% of the 103 patients unresolved by a recessive disorder have a “hot zone” (defined in Petrovski et al. [2013]) de novo variant, versus 6.0% of the 728 control trios (P = 3.0 × 10⁻¹⁰). 19.4% of the 103 patients unresolved by a recessive disorder have a hot zone de novo variant in an essential gene, versus 2.1% of the 728 control trios (P = 8.8 × 10⁻¹¹). This translates to a 90% excess [Zhu et al., 2015]. These results emphasize that even when there are no clearly causal pathogenic variants, bioinformatics analyses demonstrate unequivocally the presence of real risk factors. Optimizing these bioinformatic predictors of pathogenic variants is therefore a priority for the field, as is learning the contexts when suggestive genomic findings should be shared with patients and their care providers.

The novel approach here is focused on what is necessary to resolve these candidate variants. By aggregation of candidates such as hot zone variants and sharing phenotypic, genotypic, and functional work across institutions, discoveries will be made that otherwise would not be possible.

Patient Participant—Driven Data Sharing

In addition to data sharing by UDN researchers with those outside of the network, UDN patients and families may lead their own data sharing efforts, as illustrated by the work of Matthew Might, who is the UDN Coordinating Center's patient and family advisor. His son was the first suspected case of a disorder caused by a mutation in NGLY1 [Enns et al., 2014]; Might (2012) chronicled his son's journey in a blog post, which led to the identification of a second case. To date, there are over 35 confirmed cases of NGLY1 deficiency [Might, 2015].

Conclusions

Following its first year, the UDN is poised to make great strides in the fields of genomics and informatics. The network is developing tools and best practices that are being shared and utilized by the genomics and informatics communities and beyond. The UDN is pioneering techniques and approaches that will advance the fields of genetics and genomics. Most of the phenotypes enrolled in the UDN are incredibly rare and unique. Sharing phenotype and sequencing data may allow for identification of other patients with the same phenotype, smoothing the path to solving the molecular basis of such phenotypes. The commitment to data sharing is pioneering and an example of where the field is moving. It is hoped that this will result in new diagnoses for undiagnosed patients, and overall improvements in health care, and enhanced understanding of the biology of disease. Exciting partnerships with Matchmaker Exchange and PhenomeCentral will also facilitate discovery.

Disclosure statement: DBG is a consultant with LabCorp in their diagnostic exome service. The other authors have no conflict of interest.