Here’s an example of something that we’re all going to see more of in the coming years: the computational approach to biochemical pathway discovery and (potentially) new therapies. In this case, the authors are looking at some pretty intractable tumor types (type 3 and type 4 medulloblastoma), which is a good place for discovery in that anything would probably be an improvement over what we have now. These are the most heterogeneous forms of the disease (which disproportionately strikes children), and have by far the worst prognosis.
This is a drug repositioning effort, looking for new uses of known pharmaceuticals. That’s a difficult field – it’s very appealing in theory, and you hear a lot about such work, but reducing it to useful practice is much less common. In this case, the authors are taking an intensive systems biology approach. There are large databases on signaling pathways in various forms of cancer, and on the gene expression profiles of known drugs. This paper tries to bridge these with genomic data collection in medulloblastoma patients (deep sequencing, DNA copy number, DNA methylation state, mRNA expression profiling).
The computational part involves trying to model possible signaling networks based on all these data sources. And I can see how this would be a job for a machine, because the number of possibilities is huge, and examining them systematically by hand is just not feasible. The algorithm keeps adding to each possible network and trying to match these up to the expression profiles, optimizing the scoring (without, presumably, just throwing huge numbers of new members to the network just to run up the digital score!) Coupling this with the profiles of the known drugs suggested possibilities for useful off-target effects.
A good reality check is that the approach did rediscover known important signaling networks in medulloblastoma. It also binned patients into different categories based on the sequence data obtained from biopsy samples (which is particularly important in patient populations like these – just saying “Type 3 MB” makes it sound like a much more real category than it is. It also predicted a number of known chemotherapy agents as positives, which is a good sign. Overall, 12 of the top 100 drugs predicted by this method showed marked effects on medulloblastoma-derived cells in culture, which is far over what you’d expect by chance (and better than some previous efforts in this area). Interestingly, what came out the other end of this process were several cardiac glycosides, such as digoxin. It’s known from retrospective studies that patients treated with these agents seem to have lower incidence of some cancer types (although perhaps greater risk for others, although this is unclear).
These compounds had marked effects on the viability of medulloblastoma cells, with activity ranging down to nanomolar levels (!) Control cells didn’t reach those effect levels even up to 100 micromolar concentrations, so it looks like they’re on to something. This activity carried over to patient-derived tumor implant models in mice, with significant survival benefits even versus radiation therapy (up to the point of apparent cures in some animals). Now, digoxin is not the easiest thing in the world to dose, because of those well-known cardiac effects, but the blood levels involved suggest that there may be a therapeutic window. Profiling the cells involved suggests that the EKR/AKT signaling network is being affected, with increased apoptosis and midochondrial dysfunction. Interestingly, such effects turn out to have already been implicated with some of the cardiac glycosides versus melanoma tumor lines.
Now, while I like the idea of computationally repurposing drugs, I should note that digoxin had also already been discovered in 2011 as a possible prostate cancer therapy by a more traditional screen-all-the-drugs study. The current paper does not seem to reference this result. And that’s one of the drug-repurposing challenges: there are only so many approved drugs out there, and by this point many of them have been tested against many diseases. I would be interested to see how well this sort of systems-biology method compares against straight screening (the 12% hit rate mentioned above would seem to indicate that it does have advantages). The general idea of working out important pathways by systems biology and computation has a lot of promise as well, but these are early days. I’m more interested in these prospects than that of digoxin as an antitumor agent itself, since the literature alone could have pointed someone towards that.