The Cube: An end and a beginning in biomedical research

At the same time, research discoveries made in mice remain difficult to efficiently translate to human medicine for a variety of reasons. Mice are not little humans of course, and it’s important to understand the differences and design experiments accordingly. Human genetic complexity is another important factor that has been challenging to model in research settings. There are also frustrating gaps between the vast amounts of mouse data being generated in laboratories and the human data acquired in healthcare settings. What, then, can be done to help smooth the path between research discovery and clinical progress?

An ambitious type 2 diabetes proof of concept

As a first step in answering the question, The Jackson Laboratory (JAX) undertook an ambitious project using mice to investigate a common, complex human disease with no cure: type 2 diabetes. Known as “The Cube Initiative proof of concept,” the name of the project serves as a scientific homage to the idea of a cube that combines mouse and human information within a data construct. Its goals were to show how to work with mice to best model human disease, determine what tools and data standards would help narrow the existing mouse<->human gaps, and create a foundational, unifying technology architecture that would underlie the continued development of a mouse-human data interface. After three years that included a variety of challenges, not least of which were pandemic-induced delays, as of this writing the Cube proof of concept is now nearing completion.

During a mini-retreat in Bar Harbor, Maine, in late 2022, the Cube team presented their data and findings at that time, which emphasized the sheer size of their “proof-of-concept” effort. It’s simultaneously impressive and indicative of the massive scale of data needed to realize a more complete human-mouse research ecosystem. Perhaps the central number is the 94 terabytes of data generated, but other statistics add to the story. The work included 19 new mouse-based studies plus imported data from 24 human studies, 2,670 mice from 35 different strains, 4,385 RNA sequencing samples, 11 types of genomic data, 48,000 body weights, and much more. In all, 3.6 million data files were generated. And while many additional details will emerge in peer-reviewed publications still in process, some important takeaways already merit discussion.

The three bears

One of the challenges associated with type 2 diabetes involves human genome-wide association studies, or GWAS. Over time, hundreds of locations in the human genome have been associated with type 2 diabetes risk, with the number now well over 400 and continuing to rise. How do you find the susceptibility signal in so much genetic noise, especially when variables such as behaviors (diet, exercise) and environment also play important roles in disease onset? Research with mice that reduces the variables provides an important starting point.

The JAX researchers used their knowledge of inbred mouse strains to set up an experiment they called the “three bears.” They fed so-called “western diet chow,” which contains far more fat and sugar than typical mouse chow, to three different strains: NZO, C57BL6/J (B6), and CAST. The mouse strains kind of mimic different people—most of us need to pay attention to diet to maintain a healthy weight, but some seem to be able to live on soda and chips and not gain an ounce while others gain weight all-too-easily. When subjected to identical diets and environments, NZO mice are highly prone to obesity and diabetes, CAST individuals tend to stay lean and active and are highly resistant to diabetes onset, and B6 mice are in between.

The differences in both people and mice are, of course, due to genetics. In the mice, the genetics are very well known, however, and diet, exercise and other environmental factors can be carefully controlled. The payoff? The genetic differences and the metabolic pathways they affect can be thoroughly analyzed for the factors that make the three strains’ responses to the same conditions so distinct. The basic data from the three bears western diet experiments bore out the expected trends in weight, activity and blood sugar anomalies. Diving deeper, the researchers found metabolic changes, tissue specific dietary effects, and altered immune signaling that differed from strain to strain. For example, the lean CAST mice have fundamental differences in fat cells that allow them to burn, rather than store, more energy from the high fat/high sugar diet. Further, the pancreases of diabetes-prone NZO mice were found to have more inflammation when fed this western diet. These strain differences have helped highlight important genes and processes that can be targeted for therapies for obesity or diabetes. Stay tuned for the important specifics to emerge in forthcoming publications.

Of mice and humans

Using mice to learn about the underlying biology and mechanisms of disease is only the beginning of the path to the clinic, of course. Investigating where the biology is analogous between humans and mice and how research findings can be applied to human disease is the next step. For example, Cube researchers analyzed mouse genomics data generated through the Cube project in comparison with available human data sets, and they found that regulatory elements within the genome are indeed conserved between mouse and human. What is more, diet-responsive regulatory elements found in mouse adipose and liver tissues provide important clues for which of the elements are associated with what genes that play a role in susceptibility—or resilience—to adverse dietary changes.

Merging data across species remains difficult, however. In recent years, researchers have established a framework, called “FAIR” (Findable, Accessible, Interoperable, Reusable), to make the huge volumes of data being produced machine-actionable, but many challenges persist. Indeed, even after data is found and accessed, different data formats and standards between species, institutions, experimental platforms and more can make interoperability elusive. Part of the Cube effort at JAX therefore involved carefully analyzing the problems, both intra- and inter-institutionally, and finding the solutions needed to develop a common data model and more fully implement FAIR data practices.

Based on their work, a JAX-wide team of computational scientists, researchers and other experts created and implemented a computational system called BioConnect. Described as a utility, a service platform and an ecosystem for facilitating data sharing, curation and analysis, BioConnect integrates data across species and experiments to allow for interpretation, translation and extrapolation of experimental results for any gene or drug. It also enables seamless cross species comparison of genetic and genomic data from different disease biology experiments. And, while it was developed in the context of the type 2 diabetes proof-of-concept project, BioConnect can be applied to any complex disease and represents a key data science resource for future research at the intersection of mouse and human data.

A subset of the Cube team at JAX also worked with glue solutions, inc., a data visualization firm, to bring genomic big data to life through the development of a suite of tools called “glue genes.” glue genes make visualizations possible for genomics and other biological data to reveal patterns and insight otherwise hidden therein.

Data science in focus

Although the Cube type 2 diabetes project is officially complete, it will continue to influence JAX’s efforts for the foreseeable future. A large-scale data science initiative at JAX, led by the Laboratory’s first chief data science officer, Paul Flicek, D.Sc., will build upon the foundation of BioConnect and other products of the project. JAX researchers are looking to expand cross-species experimental and analytical capabilities and extract far more benefit from the vast amounts of data being generated at JAX and throughout the biomedical research and clinical communities. Progress in the field used to be described as “bench to bedside,” but the path forward is far more iterative. It will be an interface between experimental/research and clinical/human data that will provide faster experimental validation of clinical findings as well as clinical translation of research discoveries. Achieving the vision will require large-scale collaborations and cooperative effort, but JAX is poised to help lead the way.