Intra-patient tumour heterogeneity is a widely accepted cause of resistance to therapy, but the possibility to study this phenomenon is so far underexplored as the acquisition of multi region data sets is complex, costly and often ethically challenging. However, when these data are available, an additional set of bioinformatic challenges manifest, in order to optimise analysis of the combinatorial space spanned by different samples from the same patient and learn as much as possible from this valuable resource. So far, there is no high confidence established workflow to deal with these data.
In our current work, we developed two workflows to jointly call variants in samples from the same patient and show the significant improvement in performance over the current standard tumour-normal pair methods. In this analysis we use a simulated high depth whole genome sequencing (WGS) dataset as well as multi-region tumour samples from 8 patients (3x WGS; 5x whole exome sequencing) with on average 7 samples per patient. We show that in both our simulated dataset, as well as in the real-world data validated with targeted amplicon sequencing, our methods significantly improve the sensitivity to detect low allele frequency variants while still retaining high specificity. This in turn allows us to accurately call variants even in low tumour purity samples, which is often a major challenge with clinical samples.
This work is the first step to fill an unmet need of variant calling methods, taking the evolutionary connection of samples from the same patient into account. Accurate joint variant calling across multiple samples from the same patient has significant potential to impact our understanding of spatial and temporal heterogeneity, as well as tumour evolutionary trajectories.