A Next Generation Multiscale View of Inborn Errors of Metabolism

Abstract

Inborn errors of metabolism (IEM) are not unlike common diseases. They often present as a spectrum of disease phenotypes that correlates poorly with the severity of the disease-causing mutations. This greatly impacts patient care and reveals fundamental gaps in our knowledge of disease modifying biology. Systems biology approaches that integrate multi-omics data into molecular networks have significantly improved our understanding of complex diseases. Similar approaches to study IEM are rare despite their complex nature. We highlight that existing common disease-derived datasets and networks can be repurposed to generate novel mechanistic insight in IEM and potentially identify candidate modifiers. While understanding disease pathophysiology will advance the IEM field, the ultimate goal should be to understand per individual how their phenotype emerges given their primary mutation on the background of their whole genome, not unlike personalized medicine. We foresee that panomics and network strategies combined with recent experimental innovations will facilitate this.

Keywords: human genetic disease; metabolism; network biology; omics.

Figure 1. Inborn errors of metabolism (IEM) are increasingly viewed as complex diseases

(A) IEM are not unlike common diseases as they often present as a spectrum of disease phenotypes that poorly correlate with the severity of the disease-causing mutations (genotype). The abandonment of the one gene-one disease idea implies that modifying factors such as environmental, epigenetic, and microbiome factors as well as additional genes contribute to the disease. It also means that IEM phenotypes are emergent properties of biological networks rather than the result of changes to single genes, metabolites or phenotypes alone. Thus we have to expand our understanding of the clinical expression of the IEM beyond a single gene level to that of a consequence of a set of molecular interactions (subnetwork). (B) IEM are being considered more and more alongside common disease as part of a spectrum of ‘errors’ in metabolism. In this spectrum, ‘classic’ IEM are on the on extreme and arise from a primary genetic variant influenced by modifier genes, while the common metabolic diseases are on the other extreme and are caused by multiple genetic variants with relatively small effect sizes. The variants in the primary disease locus in the context of the individual’s background such as genome, epigenome, and environmental exposures will ultimately determine the molecular state of the individual and an individual’s risk of disease and spectrum of phenotypic presentation. This view is highlighted by the Resilience project, whereby large scale genetic screening of general populations for a panel of rare disease causing mutations is hoping to uncover healthy individuals harboring rare genetic disease and the genetic modifiers that make them resilient to this disease.

Figure 2. Studying IEM like complex disorders through adopting multi-scale omics technologies and network approaches

We can study IEM as common disorders by taking advantage of approaches like multi-scale omics technologies and integrative network analysis and even by sharing datasets. (A) Omics data generated from samples collected in common populations of humans or model organisms can be integrated alongside public database information to generate predictive molecular networks. (B) We propose these networks can be repurposed and used as a reference or framework to associate the various IEM phenotypes, scored through multi-omics approaches on samples from IEM models and patients, to identify candidate genetic modifiers and modifying biology. For example, disease signature sets generated by various omics technologies on material derived from patients, patient-derived cells (iPSCs) or experimental model systems can be used to probe a reference network to reveal disease-associated subnetworks. (C) As these Bayesian networks have a causal predictive component they can be used to inform on key molecular drivers of the pathophysiology associated with the IEM. Genes within subnetworks can be nominated as key molecular drivers through statistical algorithms and functional and therapeutic insight can be derived through annotation of subnetwork gene members. Potential impacts of such network approaches to IEM include improving the presently poor correlation between disease severity and the primary mutated locus as well as overcoming the fundamental gap in our knowledge of disease modifying genes and biology.

Figure 3. A summary of different classes of mathematical modeling approaches that can be applied to biological data

Networks represent a way to uncover relationships in data that may help elucidate causal relationships among molecular traits and biological processes and derive mechanistic insights into the causes of disease and other phenotypes of interest. They may also enable predictions of phenotypes.

Figure 4. An example of a predictive molecular network, the mouse liver Bayesian network

(A) A predictive molecular network generated from genomic and hepatic gene expression data scored in several hundred offspring from different F2 crosses of inbred strains of mice. The utility of networks from common disease datasets to inform on IEM relies on demonstrating that IEM disease-oriented pathophysiology arises from molecular pathways that are not markedly atypical and actually reflect some extreme or alternate form of common physiology. We tested if this was the case by probing the network with two different IEM model derived disease signature sets (seed set). One signature set was derived from the liver transcriptomic data generated in a Gba KO conditional mouse, an experimental model for Gaucher Disease (GD, green nodes) and the second was from the liver transcriptomic data generated in an Acadl KO mouse, a fatty acid oxidation deficient experimental model (FAO, blue nodes). (B) A histogram of the shortest path calculation for 10⁴ randomly generated gene sets, of the same size as the disease signature sets, on the network in A. The arrow in the histogram represents the average shortest path of the GD signature gene set. The average shortest path for the FAO signature gene set was also significantly lower relative to random chance (data not shown). The low average shortest distance for the two disease signature sets relative to that of randomly derived gene sets indicates in network terms, a non-random, tight interconnection of genes in the network. In biological terms, this is suggestive that a significant part of the pathophysiology associated with IEM is indeed related to common physiology. (C) Key molecular drivers were nominated amongst the genes within the isolated GD subnetworks through statistical algorithms. The isolated Gba KO subnetwork highlights one nominated key driver, Cathepsin S (Ctss) in red.