Computation as the mechanistic bridge between precision medicine and systems therapeutics

Abstract

Over the past 50 years, like molecular cell biology, medicine and pharmacology have been driven by a reductionist approach. The focus on individual genes and cellular components as disease loci and drug targets has been a necessary step in understanding the basic mechanisms underlying tissue/organ physiology and drug action. Recent progress in genomics and proteomics, as well as advances in other technologies that enable large-scale data gathering and computational approaches, is providing new knowledge of both normal and disease states. Systems-biology approaches enable integration of knowledge from different types of data for precision medicine and systems therapeutics. In this review, we describe recent studies that contribute to these emerging fields and discuss how together these fields can lead to a mechanism-based therapy for individual patients.

Figure 1

Information flow in precision medicine. Network analysis integrating clinical data and molecular data from genomic, transcriptomic, and proteomic experiments with prior biological knowledge can help develop a library of patient-specific networks that identify varying molecular characteristics associated with the disease. Traditionally, disease treatment is based mainly on clinical data, including clinical symptoms and pathological classifiers. The rapid increase of physiological, cell biological, and biochemical data and genomic characterization enabled by high-throughput technologies offers the opportunity to develop detailed classifiers based on patient-specific molecular characteristics. Computational network analysis allows for the combination of prior knowledge with patient data to obtain a more precise characterization of disease states and predict more precise therapy as central elements for precision medicine.

Figure 2

Cellular regulatory networks can explain rare adverse events. A disease module focused on long-QT syndrome connecting targets for two drugs (blue), the Src inhibitor dasatinib (a cancer drug) and loperamide (an antidiarrheal) that interacts with calmodulin (CALM1), through intermediate nodes to genes known to be associated with congenital forms of the long-QT syndrome. Such a subnetwork provides a plausible explanation of how these drugs used to treat different pathophysiologies produce the same adverse event. Figure from Berger et al.

Figure 3

Reconfiguration of a signaling network upon drug treatment. (a) Triple-negative breast cancer cells show enhanced activity of RAF-MEK-ERK. In the naive cells, c-Myc, a transcriptional repressor of multiple RTKs, is protected from proteasomal degradation by ERK-mediated phosphorylation. (b) Upon treatment with a MEK inhibitor (shown by X), ERK activity is reduced, resulting in decreased proliferation. ERK-mediated c-Myc phosphorylation also decreases, causing increased c-Myc degradation by the proteasome. As a consequence, several RTKs are overexpressed that reactivate MEK2, but not MEK1, within 24 h. (c) Eventually, the AKT and mTOR pathways are activated and the ERK pathway is restored, leading to resistance to MEK inhibitors. In this manner, a signaling network can be reconfigured after drug treatment. ERK, extracellular signal-regulated kinase; mTOR, mechanistic target of rapamycin.

Figure 4

The integration of mRNA coexpression networks and genotypic and clinical data allows the identification of disease-relevant modules and biomarker sets. Gene expression was measured in the liver and adipose tissue of more than 300 mice (blue) and a gene coexpression network was generated for each tissue. Clustering of the two coexpression networks revealed several modules that were filtered for enrichment of genes whose mRNA levels were in a causal relationship between metabolic quantitative trait loci and metabolic phenotypes (traits). The two highly overlapping top-ranked modules in the liver and adipose tissue were combined into a mouse metabolic disease module that was enriched in macrophage-related terms. It was used to identify new disease genes and a similar module in a gene coexpression network from human subcutaneous fat (orange). Ninety-eight percent of the genes in the human metabolic disease module significantly correlated with BMI. A set of cis-expression SNPs associated with the mRNA levels of these genes was shown to be significantly related to obesity phenotypes, allowing for the identification of an accessible biomarker set. BMI, body mass index; QTL, quantitative trait loci; SNP, single-nucleotide polymorphism.

Figure 5

Systems-therapeutics work flow for predictive dosing of individuals. A hypothetical work flow scheme depicting how genomic and other types of molecular interactions can be integrated into pharmacokinetic (PK)/pharmacodynamic (PD) models. Ordinary differential equation (ODE)–based models using network organization derived from molecular characterization of the patient can be used to develop predictors of patient-specific drug targets and drug doses. These models have to be built using parameters such as rate constants and concentrations of reactants. The parameters can be canonical, i.e., they are the same for all patients, or can be based on patient-specific measures, e.g., blood tests. Further parameters can be obtained by parameter fitting with the aim of generating an ODE model whose output resembles the patient’s disease state. Optimal solutions to restore the healthy state could be obtained by trial-and-error approaches, resulting in predictions for treatment regimens with maximal efficacy and minimal adverse effects.