This web page was produced as an assignment for an undergraduate course at Davidson College.
Novel mathematical methods allow the interactions of drugs to be characterized and organized and allow unknown interactions to be predicted.
An interactome is a 3 dimensional map that characterizes the interactions between various proteins within an organism. (Cusick et al. 2005). Knowledge of the organization of the interactome can inform our understanding of various biological processes and the proteins that are associated with them. (Cusick et al. 2005). A complete interactome, naturally, would require an accurate characterization of every possible interaction between every protein in an organism. Unfortunately, such a complete interactome remains essentially a pipe dream, especially as the human interactome is concerned. (Menche et al. 2015). Despite this, the incomplete interactome that we do understand has plenty of applications. For instance, mapping onto the interactome the specific subset of genes a particular disease impacts can produce a disease module. (Menche et al. 2015). Comparing these between diseases can help to characterize the relationship between them, knowledge which may present new treatment strategies. (Menche et al. 2015).
Such a strategy is not limited to disease, however. It could theoretically be applied to anything that impacts multiple proteins in unique ways. In fact, a deeper understanding of the way drugs interact with each other has the opportunity to uncover new and more effective therapeutic strategies for various conditions. (Jia et al. 2009). Despite this hope, unforeseen adverse effects of drug combinations have hampered the progress of clinical trials. (Lopez and Banerji 2017). Thus, a systematic method for characterizing the impact of various drug combinations would enable further and faster progress in this area of medical science. Caldera, et al used the interactome to characterize the relationship between various drugs (again, based on the specific proteins they interact with). They also experimentally investigated the specific impact of drugs, either individually or paired with another drug, on the shape of human cells in order to quantify the effect of combining said drugs.
The authors first assembled their interactome, which consisted of 309,355 interactions between 16,376 proteins, and their drug library, which contained 267 chemical compounds. They mapped onto the interactome the perturbation module of each drug, which consists of every protein that the drug interacts with. They found that 92% of the perturbation modules were significantly more localized than would be predicted by random chance, and more localized modules contained proteins that were more similar. They also found that drugs with closer/overlapping perturbation modules were more similar to each other (in terms of function and disease association) than drugs with more distant perturbation modules. They also found that the closeness of perturbation modules may be related to the side effects of a drug. For instance, anti-protozoal drugs, which have psychoactive side effects, overlapped with drugs that stimulate the central nervous system.
The authors then began to investigate the interactions between perturbations. To accomplish this, they developed a model of 78 morphological features that, together, can define the shape of a cell. They then exposed human epithelial cells to every drug individually and every combination of two drugs to determine the effect each drug or combination has on the shape of the cells (based on changes in the 78 features). Using this data, along with some surprisingly intuitive vector-based math, they can quantify the interaction between drugs based on changes in their individual phenotypes. They used these quantified interactions to develop the perturbome, which organizes each drug based on its interaction with other drugs.
From the perturbome, they discovered that only 5% of the 35,909 potential interactions were observed, in other words 95% of drug combinations produced either no change from the individual phenotypes or a simple addition of phenotypes. Negative interactions, where one or both phenotypes were diminished by the combination, were the most frequent, representing 44% of all interactions. Emergent interactions, where new phenotypes were present in the combination, were next most frequent at 31%. Positive interactions, where one or both phenotypes were enhanced by the combinations, were the least frequent, representing 24% of the interactions. They also found that 95% of interactions were unidirectional, where only one drugs phenotype was impacted by the combination. They also found that the perturbome can be divided into a small core of densely connected, strong perturbations and a loosely connected periphery consisting mostly of weak perturbations.
Finally, they used various computational and statistical methods to further explore the perturbome and its relationship to various molecular properties. For example, they used machine learning, in addition to various annotations of drug properties, to predict the interaction between two drugs, using the perturbome to establish the algorithms accuracy. Interestingly, they found that the distance of the drugs’ module on the interactome was significantly related to the type of interaction they displayed. Emergent interactions were associated with larger distances, while positive interactions were associated with smaller distances. They also explored the impact of 12 molecular determinants on the interaction between drugs.
In general, this study represents a significant leap forward in our understanding of the interactions between drugs, which may allow us to predict the benefits or side effects of drug combinations before testing them. However, while this method of characterizing drug interactions has been demonstrated to be robust in terms of the effect of drugs on cell shape, in would be interesting to apply it to other properties of cells, such as changes in gene expression. That being said, the authors have made public the computation pipelines they employed in this paper, so I expect to see many applications of it in future.
Anthony Eckdahl is a senior biology major at Davidson College. Contact him at: aneckdahl@davidson.edu
References
Caldera M., F. Müller, I. Kaltenbrunner, M. P. Licciardello, C.-H. Lardeau, et al., 2019 Mapping the perturbome network of cellular perturbations. Nature Communications 10: 5140. https://pubmed.ncbi.nlm.nih.gov/31723137/
Cusick M. E., N. Klitgord, M. Vidal, and D. E. Hill, 2005 Interactome: gateway into systems biology. Human Molecular Genetics 14: R171–R181. https://pubmed.ncbi.nlm.nih.gov/16162640/
Jia J., F. Zhu, X. Ma, Z. W. Cao, Y. X. Li, et al., 2009 Mechanisms of drug combinations: interaction and network perspectives. Nature Reviews Drug Discovery 8: 111–128. https://pubmed.ncbi.nlm.nih.gov/19180105/
Lopez J. S., and U. Banerji, 2017 Combine and conquer: challenges for targeted therapy combinations in early phase trials. Nat Rev Clin Oncol 14: 57–66. https://pubmed.ncbi.nlm.nih.gov/27377132/
Menche J., A. Sharma, M. Kitsak, S. D. Ghiassian, M. Vidal, et al., 2015 Uncovering disease-disease relationships through the incomplete interactome. Science 347. https://pubmed.ncbi.nlm.nih.gov/25700523/
Steve T., 2009 Medicine 01. https://www.flickr.com/photos/13519089@N03/4746653392
© Copyright 2020 Department of Biology, Davidson College, Davidson, NC 28036.
