Robustness and lethality in multilayer biological molecular networks

Xueming Liu, Enrico Maiorino, Arda Halu, Kimberly Glass, Rashmi B. Prasad, Joseph Loscalzo, Jianxi Gao, Amitabh Sharma

TL;DR

This paper introduces a novel framework for understanding how interconnected gene, protein, and metabolic networks contribute to biological robustness. By simulating cascading failures across these layers, they identify influential genes and demonstrate that targeting metabolic disease genes causes disproportionately severe damage to the metabolic network, highlighting the importance of multilayered analysis.

What Problem Does This Paper Address?

Biological systems are remarkably robust, maintaining stable function despite various internal and external perturbations. While previous research has explored robustness in single molecular networks (like gene regulatory or protein-protein interaction networks), real biological systems are far more complex, involving intricate interactions between these different molecular layers. Understanding how these interdependencies influence a system's overall robustness and how perturbations cascade through them has been a significant challenge. Specifically, existing models often treat networks as homogeneous or ignore the complex, non-random ways different molecular layers are connected, leading to an incomplete picture of cellular dynamics and disease mechanisms.

Methods

The authors developed a multilayer biological network model by integrating three key components:

  • Gene Regulatory Network (GRN): Captures how genes influence each other's expression.
  • Protein-Protein Interaction (PPI) Network: Represents physical interactions between proteins.
  • Metabolic Network: Describes chemical reactions involving metabolites.

These layers are connected by:

  • Gene-Protein links: Genes are linked to their corresponding protein products.
  • Protein-Metabolite links: Proteins (as enzymes) influence metabolic reactions.

They then devised a cascading failure simulation to assess robustness. When a gene is perturbed (e.g., knocked out), it can lead to the dysfunction of its target genes in the GRN. This, in turn, causes the corresponding proteins to become dysfunctional in the PPI network. If these dysfunctional proteins are critical for maintaining the connectivity of the PPI network, other proteins can become dysfunctional. Finally, if a metabolite loses a sufficient fraction of its supporting proteins, the metabolic network can fail.

To quantify a gene's importance, they introduced an "influence score", calculated by measuring how much the final functional network size decreases when a specific gene is perturbed. This is compared to an "uncoupled" scenario where only the PPI network is considered. They also developed an analytical framework to predict these cascading failures and functional network sizes, which is computationally more efficient than simulations.

Datasets

The study utilized several publicly available datasets to construct the multilayer network:

  • Gene Regulatory Networks: Curated from FANTOM5 database (general and tissue-specific networks for forebrain, lymphocytes, and lung).
  • Protein-Protein Interaction (PPI) Network: Integrated from multiple databases, specifically using the human interactome built by Cheng et al.
  • Metabolic Network: Constructed by curating interactions from the STITCH database and mapping to the Human Metabolome Database (HMDB).
  • Disease Gene Lists: Essential genes from the Database of Essential Genes (DEG) and cancer genes from the Cancer Gene Census (CGC). Genes associated with dyslipidemia, hypertension, and type 2 diabetes were compiled from GWAS studies.

Key Insights and Ideas

  • Interlayer couplings are crucial: The study strongly suggests that the interactions between different molecular layers (GRN, PPI, metabolic) are not just additive but fundamentally alter how perturbations propagate and impact system robustness. The "coupled" model significantly outperforms "uncoupled" models in identifying important genes.
  • Metabolic disease genes are disproportionately damaging: Perturbing genes known to be associated with metabolic diseases has a more severe impact on the metabolic network than random gene perturbations, highlighting a specific vulnerability.
  • Real biological networks are surprisingly robust: The constructed multilayer network shows comparable or even greater robustness than various randomized network models, suggesting that biological networks have evolved specific structures to withstand failures.
  • Influence score is a powerful predictor: The proposed "influence score" is more effective than simple degree measures in identifying essential and disease-related genes, indicating it captures a more nuanced understanding of a gene's role in system integrity.
  • Analytical framework for efficiency: Developing an analytical method to predict cascading failures offers a computationally efficient alternative to large-scale simulations, making robustness analysis more accessible.

Key Results

  • The coupled multilayer model achieved significantly higher average precision scores (APS) for prioritizing essential (5.05% higher) and cancer genes (48.94% higher) compared to the uncoupled model.
  • Perturbing metabolic disease genes (e.g., for dyslipidemia) caused demonstrably more damage to the metabolic network than random perturbations, as shown by lower functional node fractions and statistically significant p-values in targeted attacks.
  • The real biological multilayer network was found to be more robust than several randomized models, including those with assortative rewiring in the GRN or assortative rewiring of gene-protein connections.
  • The analytical framework accurately predicted the functional network sizes in both synthetic and biological multilayer networks, with correlation coefficients around 0.9918.
  • Genes ranked by the influence score were enriched in essential and cancer genes at higher fractions compared to genes ranked by their out-degree.

Limitations and Future Work

  • Simplifying assumptions: The model makes simplifying assumptions, such as neglecting feedback loops from the metabolic layer to the PPI layer and assuming that a gene's perturbation directly affects all its targets.
  • Data noise: Biological interaction networks are inherently noisy. While the study showed some resilience to noise, further investigation into its impact is warranted.
  • Model complexity: The authors suggest incorporating more detailed molecular interaction parameters (e.g., protein binding affinities) to create weighted networks.
  • Dynamic modeling: Extending the model to incorporate stoichiometric and constraint-based techniques for a more dynamic depiction of the human metabolic machinery is a promising future direction.
  • Additional molecular mechanisms: Future work could also include other crucial molecular mechanisms like methylation, noncoding RNA regulation, and post-translational modifications.
  • Feedback mechanisms: Investigating feedback mechanisms, such as metabolic control of gene expression, could lead to more realistic network couplings.

Summary updated

Abstract

Robustness is a prominent feature of most biological systems. Most previous related studies have been focused on homogeneous molecular networks. Here we propose a comprehensive framework for understanding how the interactions between genes, proteins and metabolites contribute to the determinants of robustness in a heterogeneous biological network. We integrate heterogeneous sources of data to construct a multilayer interaction network composed of a gene regulatory layer, a protein-protein interaction layer, and a metabolic layer. We design a simulated perturbation process to characterize the contribution of each gene to the overall system's robustness, and find that influential genes are enriched in essential and cancer genes. We show that the proposed mechanism predicts a higher vulnerability of the metabolic layer to perturbations applied to genes associated with metabolic diseases. Furthermore, we find that the real network is comparably or more robust than expected in multiple random realizations. Finally, we analytically derive the expected robustness of multilayer biological networks starting from the degree distributions within and between layers. These results provide insights into the non-trivial dynamics occurring in the cell after a genetic perturbation is applied, confirming the importance of including the coupling between different layers of interaction in models of complex biological systems.
BibTeX
@inproceedings{liu2020robustness,
  author = {Xueming Liu and Enrico Maiorino and Arda Halu and Kimberly Glass and Rashmi B. Prasad and Joseph Loscalzo and Jianxi Gao and Amitabh Sharma},
  title = {{Robustness and lethality in multilayer biological molecular networks}},
  booktitle = {Nature Communications},
  year = {2020},
  doi = {10.1038/s41467-020-19841-3},
}
Figure