The reported research is part of a research line that was initiated when I started my postdoc in the lab of Erik Larsson Lekholm in the beginning of 2015. We brainstormed about the relative absence of published reports on negative selection in the cancer genome and discussed different approaches to the problem. Detecting negative selection implies searching for mutations that are expected to occur, but at some stage during tumor evolution get removed because they’re harmful for the cancer cell. As these mutations have the potential to identify specific vulnerabilities in a tumor, identifying them has translational importance. We approached the problem from different angles and during the next 2 years we learned that we were not only searching for mutations that are not there but also for signals that are apparently almost absent. Apart from some evidence of negative selection acting on high impact mutations in a set of hemizygously deleted essential genes, we concluded that, in general, negative selection signals are weak in the cancer genome1,2.
When presenting a poster about our findings at the 2017 EACR cancer genomics conference, I met Martin Miller (CRUK, University of Cambridge) and he asked my opinion about negative selection acting specifically on somatic mutations that result in neoantigen formation. The idea is that cancer cells that present these small peptides to immune cells get eliminated, leading to a negative selection pressure on the underlying mutations and a depletion of the resulting neoantigens. Excited about the concept and encouraged by my earlier PhD experience with the immunology field, I started digging into the literature and found out that neoantigen depletion seemed quite well accepted by the field and signals were described in e.g. colorectal and kidney cancer. This raised several questions: why didn’t we see a general signal of negative selection in these cancers? What about other cancers that are known to respond to immunotherapy, like melanoma and lung cancer? …
As neoantigens are small mutated peptides that can only be presented to T cells when they bind to the cancer cell’s HLA molecules, we first determined for each nucleotide in the exome whether a somatic mutation could potentially result in the formation of HLA binding peptides and hence potential neoantigens. By applying this HLA binding annotation on a large set of somatic mutation data from The Cancer Genome Atlas, we discovered an important correlation between a somatic mutation’s DNA sequence context and the HLA-binding capacity of the translated peptides. These correlations lead to false positive selection signals when synonymous mutations or mutations in non-expressed genes are improperly used as a background mutation rate reference, which may have influenced other studies. Based on this knowledge we developed 2 new selection metrics which suggest an overall absence of neoantigen depletion signals in the treatment-naive cancer genome.
This lack of detectable neoantigen depletion could indicate the presence of efficient immune evasion mechanisms early during tumor evolution, resulting in an early relief of selection pressures on neoantigen forming mutations. Alternatively, the absent signals could also be related to low neoantigen prediction accuracy of HLA affinities and a resulting lack of sufficient statistical power. As HLA affinities are increasingly used to study immunogenic selection processes, we believe that the insights from our study have important implications for proper interpretation of other studies.
During this project I received an EMBO fellowship to spend 2 fruitful months in the laboratory of Martin Miller in Cambridge. I am very grateful to Martin and EMBO for this opportunity. Finally, I would like to thank Erik for a scientific inspirational time in his lab and all my fantastic colleagues in Gothenburg for the great collaboration we had in the past years.
Jimmy Van den Eynden