New computer model assigns drugs to protein synthesis disruptors in hereditary diseases and cancer

Genetic diseases caused by shortened proteins can be specifically treated by so-called nonsense suppression therapies—Drugs that prevent protein translation from stopping prematurely. A new computer model developed by scientists at the Institute of Research in Biomedicine (IRB Barcelona) and the Centre for Genomic Regulation (CRG) could use this information to predict which therapies are likely to work best for some genetic diseases as well as cancer.

Details on the model, called RTDetective, can be found in a new article published in Natural genetics entitled “Genome-wide quantification and prediction of pathogenic stop codon readthrough by small molecules.” The developers believe the tool could be helpful in the design, development and effectiveness of clinical trials of drugs called nonsense suppression therapies.

To understand these drugs, some background information is needed on the truncated protein translation due to premature termination codons. This phenomenon is associated with about 10–20% of hereditary diseases, including some types of cystic fibrosis and Duchenne muscular dystrophy. It is also an important mechanism by which tumor suppressor genes are inactivated in cancer.

Nonsense suppression therapies effectively target the problem by helping cells ignore, or “readthrough,” the instructions to stop that appear during protein production. Previous studies show that cells with higher readthrough rates produce more full-length or near-full-length proteins. However, many clinical trials of nonsense suppression therapies likely use ineffective patient-drug combinations. This is because a drug's effectiveness in promoting readthrough depends not only on the nonsense mutation, but also on its environment.

According to the authors, this was one of the key findings from “quantifying the readthrough of approximately 5,800 human pathogenic stop codons by eight drugs.” Data for this study came from patient reports submitted to large public databases such as ClinVar and the Cancer Genome Atlas. By understanding the impact of local sequence context, they were able to “develop models that predict the readthrough efficacy of the best performing drugs with very good performance across the genome.”

Sequence context turned out to be important for another reason. According to other results reported in the paper, while a drug may work well on one premature stop codon, it may not work on another within the same gene because of the local sequence. “We show that bypassing this obstacle depends strongly on the immediate environment,” said Ignasi Toledano, first author of the study and a joint PhD student at IRB Barcelona and the Centre for Genomic Regulation. He used roads as an analogy, explaining that “some mutations are surrounded by well-marked detour routes, while others are full of potholes or dead ends. This is how you see the ability of a drug to bypass obstacles and work effectively.”

Training computer models requires a lot of data. To train RTDetective, the scientists tested thousands of drug-stop codon combinations, resulting in over 140,000 individual measurements. They then used the algorithm to predict how different drugs would likely perform against each of the 32.7 million possible stop codons that can be generated in human RNA transcripts. Among other things, RTDetective predicted that at least one drug could achieve a readthrough of more than 1% for just over 87% of all possible stop codons, and a readthrough of 2% in almost 40% of cases.

These are promising numbers, according to the research team. They could mean potential relief for patients with conditions such as Hurler syndrome, a severe genetic disorder caused by a nonsense mutation in the IDUA gene. Forms of the disorder, formerly known as gargoylism, can be characterized by features such as developmental delays, cognitive decline, joint stiffness and shorter life expectancy. Studies show that as little as 0.5% readthrough is enough to generate a functional protein that mitigates the severity of the disease. In the study, RTDetective predicted that at least one of the drugs tested could achieve a readthrough of over 0.5%.

“Imagine a patient is diagnosed with a genetic disorder. The exact mutation is identified through genetic testing and then a computer model suggests which drug is most appropriate. This informed decision-making is the promise of personalized medicine that we hope to realize in the future,” said Ben Lehner, PhD, group leader at the CRG and the Wellcome Sanger Institute and one of the study's lead authors. In addition, “with this approach, when a new, end-to-end drug is discovered, we can quickly build a model for it and identify all the patients most likely to benefit,” he added.

Next, the researchers want to confirm that the proteins produced after administration of nonsense suppressor therapies are functional. This is important to establish the clinical applicability of RTDetective's predictions. They will also investigate other strategies that can be used in combination with the therapies to increase their effectiveness, particularly in cancer.