Fig. 1.

Detection of somatic variants in 1461 postmortem human brains. (a) Brain regions sampled within the study. (b) The proportion and number of individuals in each cohort. (c) Unfiltered variant allele fraction (VAF) with between 10% and 35% against relative exome sequencing depth. Those that were present before and after filtering are shown (red and blue respectively). (d) Variant detection pipeline. Section I: Exons are shown in red, with intergenic and intronic regions as a black line. II: Regions of high genomic complexity and common structural variants (determined from population databases and previous studies) were removed (yellow line/gray box). III: Relative sequencing depth of each exon is shown in blue above the relevant exon. Bases in which the sequencing depth was below 30 (as depicted by the red dashed line) in an individual were removed. These regions are then shown by gray boxes on the schematic exome and were also removed. IV: Finally, regions in which copy-number variants (gains or losses) were called from array genotyping¹⁰ were also removed from the overall panel. An example plot of the array genotyping in which a copy-number gain has been detected is shown. Again the corresponding region was removed from the exome depicted by a gray box on the exome panel. After these steps, remaining regions were subsequently subjected to analysis by deepSNV and a binomial test against the mean VAF for heterozygous variants (47%). (e) Schematic representation of the putative somatic alleles in the data set. A distribution of VAF in the whole data set is shown (pink histogram). Putative somatic alleles were those in which the VAF was greater than base error rate (as determined from deepSNV [green box and linked inset]), and those that also differed from the binomial threshold (<1 × 10⁻⁵) compared with an assumed VAF of 47% for heterozygosity. CJD Creutzfeldt–Jakob disease, PD-DLB Parkinson disease and dementia with Lewy bodies