Iñigo Ayestaran is a first year PhD student at the Department of Oncology at the University of Cambridge. Before that however, he was a visiting scientist in the Institute of Computational Biology at Helmholtz Zentrum in Munich, where he investigated markers of cancer resistance together with Ana Galhoz under the supervision of Michael Menden. Their recent publication Identification of Intrinsic Drug Resistance and Its Biomarkers in High-Throughput Pharmacogenomic and CRISPR Screens has just been published, and we have now visited the author to explain the significance of his work.
Congratulations to your recent publication! Could you please summarize what you have done in this work, and in which way this is relevant for drug discovery?
Iñigo Ayestaran (IA): Cancer resistance is a huge practical problem in the clinic, but now there are new data sources available which allow us to do novel types of analyses. In this case, we tried to identify rare drug resistance biomarkers by looking at large pharmacogenomic screens. These screens consist of >1,000 cancer cell lines treated with hundreds of cancer drugs each, which is much more information than had been around only recently. We also have information about mutations and copy number alteration for each cell line, allowing us to find specific genomic alterations that correlate with drug response. Previous work has shown that many drug sensitivity markers (known and novel) can be identified this way.
Identifying drug resistance biomarkers is difficult though in practice, if the resistance marker is relatively rare across all cell lines. This is unfortunately very frequently the case in practice, so we needed to develop a novel, indirect approach for identifying cell lines with a resistance biomarker that also works in this case. The principle of the method is that we look for cell lines that should be responding to a certain drug (because they contain a sensitivity biomarker), but show an unusual lack of response. These cell lines (which we call ‘UNusually RESistant’, or UNRES cell lines) are likely to contain some unique genetic marker that is causing the cells to not respond to drug treatment. By systematically analysing two large pharmacogenomic screen datasets, we identified multiple UNRES cell lines containing both known and putatively novel resistance biomarkers. Combining this information with publicly available CRISPR essentiality screens, we can now come up with new hypotheses about how to overcome resistance, for example by using drug combinations.
So how does this go beyond what has been known before, how does it advance the field?
IA: Until now, drug resistance in large pharmacology screens was only studied directly. Because the screens are generally designed to identify sensitivity markers, the power to detect resistance markers is more limited. This means, as mentioned earlier, that only very common resistance biomarkers could be identified (e.g. TP53 mutations causing resistance to nutlin-3a treatment).
For rarer biomarkers though, the data is not so clear cut. Let’s take a well known example of a rare resistance biomarker with clinical relevance: gefitinb is an EGFR inhibitor used to target cells with driver EGFR mutations. If we consider any cell line that doesn’t respond to gefitinib, it could be because of two reasons: 1. the cells have a wild type EGFR that is not driving cell growth, or 2. the cells are EGFR mutant, but have an additional mutation (such as EGFRT790M) that prevents the drug from binding to its target. These two scenarios (also known as non-responder cells and resistant cells, respectively) are experimentally indistinguishable in the context of a large pharmacology screen. As a result, it’s impossible to identify these rare biomarkers directly.
Our approach is the first one that is able to distinguish between the two scenarios, and indeed highlights the well established EGFRT790M as a resistance biomarker, even if it only appears in one single lung adenocarcinoma cell line.
What was the most difficult part of this work, and why?
IA: Whenever we identify an UNRES cell line and its potential resistance markers, we ended up comparing very small numbers of cell lines. In most cases, we dealt with only one or two cell lines with potential markers. Furthermore, since each cell line has been extensively annotated in terms of single nucleotide variants and copy number alterations, the list of putative resistance biomarkers ends up being fairly long, with most markers probably not being biologically relevant. Refining this list of potential biomarkers is not easy, and we ended up integrating this information with other resources, such as CRISPR essentiality screens, queries in scientific literature and curated cancer gene lists to whittle down the gene lists obtained, in order to move towards something which can be used in practice for the design of future cancer therapies.
Aside from the biological insight obtained from the biomarker identification, we spent quite some time estimating the number of false positives upon the identification of putative UNRES cell lines, since it involved a chain of hierarchical statistical tests: Firstly, the identification of sensitive cell lines and, subsequently, that of UNRES cell lines. In this aspect, it was great to have Ana Galhoz, a mathematician, working with us on the project and she had great input in particular into this aspect of the analysis. We were able to come up with an upper bound for the overall False Discovery Rate of our analysis, both analytically and through simulations, whilst ensuring suitable statistical tests in each step of our framework. When dealing with rare events in such large datasets we realised it is absolutely crucial to have a robust estimation of what percentage of results might not be relevant, since otherwise we might end up with a high number of false positive results, which would lead to wrong directions for future research. This is what we of course tried to avoid as much as possible.
How can the results you have obtained be used by others?
IA: In terms of biological insight, our analysis has flagged multiple cell lines that show unexpected behaviour when treated with certain drugs. This opens the possibility to study these specific cell lines in order to clarify the mechanisms of drug resistance, for example via CRISPR screens in a treatment vs control setting. Unravelling the precise resistance mechanisms and potential vulnerabilities is the next step for designing effective drug combination strategies. And, of course, the principle can also be transferred to the in vivo situation of treating patients in the future, with all the usual caveats that would apply when trying to transfer an approach from the cell line to the patient level.
From the methodological side, we have described an analysis approach that could also be applied to other large pharmacology screens in search of UNRES cell lines in the future, and all the code used for the analysis in our paper is available on Github. So this is another way in which we aimed to make the findings of our work as useful to the public as possible.
Thank you for this conversation, and all the best for the next steps in (and after) your PhD!
Further reading: Identification of Intrinsic Drug Resistance and Its Biomarkers in High-Throughput Pharmacogenomic and CRISPR Screens, https://doi.org/10.1016/j.patter.2020.100065.