The Optimum Combination of Drugs

For children suffering from acute lymphoblastic leukemia (ALL) and for their relatives, the dozens of common drugs that are available today are a great source of hope. For doctors and researchers, they also present a challenge because the drugs used to treat this rare disease don’t have the same impact on every child. In addition, combinations of up to five active substances are often used, and it’s not always possible to predict exactly how they’ll interact with each other. How do you identify the therapy that works best from so many different options?

Researchers at the University Children’s Hospital Zurich have spent years trying to find answers to this complex question. Successful results are now starting to emerge, thanks to innovative methods that capture microscopic images of patients’ cancer cells with a high throughput and use artificial intelligence to evaluate them. Drug response profiling (DRP) records on a large scale whether and how individual drugs or combinations of them work on patients. The goal is to produce bespoke therapies based on these established “fingerprints”.

In a large-scale study based on data collected from 2016 to 2023, a team led by Fabio Steffen, who heads the functional precision oncology division at the University Children’s Hospital, has now investigated which drug-based therapies for ALL work and how – especially in children whose treatment had previously been unsuccessful or who had already suffered relapses. The study, which involved data from 340 patients across 17 European countries, was also a logistical challenge for the researchers and the doctors treating the patients. Samples and data from the hospitals had to be recorded, delivered, analyzed and fed back to the doctors at the hospitals virtually in real time – as the basis for making clinical decisions.

To enable the experts in Zurich to handle this flood of data from so many affected patients and combinations of drugs, also linked to genetic information, they use highly efficient computer-aided methods, combined with state-of-the-art technology in the laboratory: a pipetting robot injects the drugs in the form of tiny droplets onto the prepared tumor samples; after 72 hours a fluorescence microscope records images of the cells. For all of the samples, machine learning is then used to record how many cancer cells have been destroyed with which medication.

Acute lymphoblastic leukemia (ALL) is a malignant disease of the hematopoietic system in which immature lymphocyte precursor cells grow in an uncontrolled manner in the bone marrow and in the blood. This represses the production of normal blood cells and can lead to symptoms such as fatigue, susceptibility to infections, a tendency to bleed and bone pain. Treatment usually involves intensive, multiphase chemotherapy that is often supplemented by targeted medication. In certain cases, stem cell transplantation or immunotherapy may also be considered. In children, the prognosis today is very good with cure rates of around 90 percent, but they depend very much on the child’s age and the specific ALL disease. In adults, the chances of being cured are lower overall, but modern, targeted therapies are helping to improve survival rates for this group of patients, too.

In total, this has generated around 135,000 individual data records that have revealed key findings. For example, various tyrosine kinase inhibitors have been shown to be promising drugs for targeting cancer cells. The biochemist Fabio Steffen presents a vivid image to explain how they work: they operate like a police officer who brings the traffic back under control at a defective traffic light, which is only displaying green because of the disease, and stops the leukemia cancer cells from multiplying.

The study also revealed that, as was hoped, drug response profiling was able to deliver results within one to two weeks so the therapy could be modified in the appropriate way. The researchers, who are always looking for patterns in the data, were also interested in what are known as functional DRP twins – these are patients whose cells react to a treatment in a very similar way, even though their genetic characteristics don’t necessarily suggest that this would be the case.

Around two-thirds of patients who received DRP-based therapy responded to it. “This rate can be deemed a success when you consider this disease is very resistant to treatment,” says Steffen. The researcher adds that we shouldn’t jump to conclusions because the therapies and timescales are all very different. But for individual drugs, “thanks to drug response profiling, we’re seeing the first indications of an actual improvement in the clinical benefit”, especially when a drug is used as a bridging therapy to minimize the number of leukemia cells in the bone marrow ahead of subsequent stem cell treatment or immunotherapy.

Steffen is convinced that this method is a valuable tool that can help doctors in hospitals make decisions when they’re treating cancer patients. “The combination of data-driven precision oncology and AI presents huge potential,” he says, “and drug response profiling is an important piece of the jigsaw for managing complex oncology cases.” Following the success of this research, the University Children’s Hospital Zurich is now expanding the DRP program to include other types of cancer, such as brain tumors and bone tumors.

If this method is to become routine in the future, it will of course also need a scalable data infrastructure as well as intensive research and an expansion in expertise. This is why the experts are teaming up with the BioVisionCenter at UZH to devise methods that will enable them to process and use the vast amounts of data on a terabyte scale more easily and efficiently in the future. This means that more children might then be able to benefit from targeted therapy in years to come.