Virtual Liver May be Cure for Drug Failures in ClinicalTrials

Kalyanasundaram Subramanian had lost count of the number of times pharma firms had asked him: "Do you have a good model for predicting drug toxicity?" It was early in the last decade, when he was working for the firm Entelos, a modeling and simulation company near the Silicon Valley. Enetelos and its competitors had clever computer models to predict the efficacy of drugs, but pharma companies were interested in toxicity as well.

Toxicity was the reason why so many drugs failed in trials, and regulators had been gradually putting pressure on the pharma companies to weed out the toxic drugs. The ability to predict drug toxicity was a great business opportunity, a really good solution had eluded the cleverest minds in the industry.

Subramanian returned to India in 2002, joined a Bangalore startup called Strand Life Sciences, and started thinking about the problem. Strand had been set up by four professors at the Indian Institute of Science (IISc), and was trying to ride the wave of In Silico biology. The company took up the project in earnest seven years ago.

This year, Strand started one of the few commercial services in the world to predict animal liver toxicity by using a computer model, but its results are backed up by its own lab experiments. Three leading pharma and biotech firms have given Strand drug candidates to test, and they will use the company's predictions to prioritise them for clinical development.

"Even a 15-20% improvement in efficiency before clinical trials can help pharma companies make huge savings," says Subramanian. Drug failures in clinical trials have become a serious issue for pharmaceutical companies, who have been trying to reduce the rising costs of drug development. Liver and kidney toxicity are the most important reasons why drug candidates fail during trials. These two organs are involved in detoxification of the body. All water-soluble compounds are directly eliminated through the kidney and the liver breaks down and prepares the rest for elimination. As it breaks down these compounds, the liver also exposes itself to highly reactive molecules that are part of this break-down process. Nearly 50% of drug failures in the world are due to liver toxicity. Every seventh liver failure is due to an adverse drug reaction.

And safety data from trials are not always the whole story. "Current statistics imply that the main reason why drugs fail is lack of efficacy but this is misleading because drugs are often given in ineffective doses because of safety concerns " says Paul Watkins, director of the Hamner-University of North Carolina Institute for Drug Safety Sciences.

Getting a handle on liver toxicity can provide multiple benefits to drug companies. They can shortlist drug candidates quickly, and tweak the dangerous ones to reduce their toxicity. However, the liver is also the second most complex organ in our body after the brain. It has over ten thousand molecular pathways that also interact with one another (a pathway is a series of reactions that finally produce a substance that is used by the body in some form). Biologists do not understand the liver well enough to look at a drug and predict its behaviour in the organ.

Some of the toxicity becomes apparent when tested on cells in the lab, a few others become evident during animal testing, many more during the four phases of clinical trials, and a significant number after the drug is launched. Regulators these days are particularly concerned about idiosyncratic toxicity that affects one in thousand or ten thousand people. "Using computer models can reduce animal testing that is becoming very difficult these days," says CM Gupta, former director and now emeritus scientist of the Central for Drug Research Institute in Lucknow.

Strand approached the problem in a unique way. First it combined computer models with data from lab experiments. "I realised from the beginning that pure modeling will not work well," says Subramanian. Secondly, it used different modeling techniques as well. Other commercial liver toxicity models use a database of drug structures that are toxic and trains the computer to make predictions. The Strand liver model uses biology to run simulations that represent the liver function. It captures the important metabolic pathways in the liver that are known to be involved in toxicity like fat processing, cholesterol formation, bile salt metabolism and so on. By modeling them, it can predict three major drug toxic outcomes: cell death, fat deposition and stoppage of bile flow into the duodenum.

Strand combines this computer models with experiments on liver cells in the lab, which it has set up in the city. In 2010, when the computer model was good enough to begin services, Strand spent a year looking for an Indian Clinical Research Organisation (CRO) that could do the experiments. After that it hired a Silicon Valley company to do the lab experiments and validate its technology. As this was too expensive, it set up its own lab inside the Centre for Cellular and Molecular Platforms in Bangalore using funding from the Department of Biotechnology. Strand tested 10 molecules with the Silicon Valley company, 25 with the DBT-funded project, and is now commercially testing 26 molecules for pharma companies.

Since Strand uses biological mechanism to predict toxicity, its approach will be useful to predict the toxicity of completely new drugs. Other researchers are now trying this approach but they are in academic centres. One of the most prominent is the Virtual Liver Network in Germany funded by the government, which looks at modeling liver function in excruciating detail. The US Environment Protection Agency is also funding a project to model the liver at similar depth. Both use a bottom-up approach, looking at the liver function in totality rather than those that are important in toxicity. The Hamner Institute of Health Sciences in North Carolina, like Strand, is developing a model using a top-down approach, beginning with pathways that are important in toxicity and then slowly working its way down. It product, called Dili-Sim, has been funded by the FDA and 11 pharma companies, but it is not planned as a commercial version.

Strand could take use this product in multiple ways. It can be used, as is being done now, to deliver value-added services; even if new methods of toxicity testing emerge - like stem -cell-based tissue engineering - the Strand methods can be used to understand the reasons for toxicity. Strand can package and sell the software model as a commercial product, along with the biological assays (tests). It could also be the beginnings of another important business: clinical services. "Once we incorporate human toxicity into the model, we could use this product to begin a clinical services business," says Vijay Chandru, CEO of Strand. That would be one year away. The company may need to open up its platform at some stage to get full value. "The current model of Strand cannot become part of regulatory decision-making," says Watkins. Strand intends to keep it that way for the moment.