Paired tissues were taken from a total of nine individuals, with age ranging from 23 years to 45 years ( Supplementary table S1). For four individuals (post-mortem, road accident victims), tissue sections from two different parts of the brain viz. frontal cortex (FC) and corpus callosum (CC), were procured from NIMHANS Brain Bank, Bangalore, India. For the other five individuals, peripheral blood and saliva was taken from healthy volunteers representing circulatory cell types with high turnover and therefore high likelihood of spontaneous somatic variations. The project was approved by institutional human ethics committee of CSIR-Institute of Genomics and Integrative Biology and adhered to the international ethical guidelines (Declaration of Helsinki). For brain tissues, DNA was isolated using Omniprep Genomic DNA isolation kit (G-Biosciences, USA) and DNA from two other cell types viz. leukocytes (from blood) and epithelial cells (saliva) were isolated using Qiagen DNA kit (Qiagen, USA) and Oragene Saliva kit (DNA Genotek, Canada), respectively using manufacturer recommended protocols.

Exome capture for the isolated DNA was done using Illumina TruSeq Exome capture kit (62 Mb) and further exome libraries were sequenced (100 bp paired end) using Illumina Hiseq 2000 (Illumina, USA). Exome sequencing was done using manufacturer recommended protocols. A total of ~1.5 billion reads were generated for all the 18 samples. Supplementary figure S1 represents the overall pipeline followed for the analysis of the data and further text explains each step in detail. Raw data was checked for per base quality score and reads having 80% bases with phred quality score 30 (Q30) and greater were carried forward for downstream analysis and rest were discarded. Also, last few low quality bases were trimmed from all the reads (4–10 bases, depending upon the sample quality). This was done using Fastx (version 0.0.13) and FastQC (version 0.10.0). About 9–14% of data was removed from each sample. After checking quality of the data in various aspects, reads (Read 1 and Read 2 for each sample) were aligned to the reference genome (hg19) using BWA (version 0.6.1) ²⁰ allowing for maximum 2 mismatches. More than 98% percent of the data was aligned to reference for each sample. Data was also checked for PCR duplicates and the same were removed. Only reads with mapping quality (MQ) more than 40 were taken forward for further analysis. The sequence depth for the samples ranged between 91×–120× (average 100×) for the FC-CC samples and 25×–86× (average 51×) for the blood saliva samples ( Supplementary figure S2). Data has been deposited in the NCBI Sequence Read Archive and can be found at accession number SRP045655.

Varscan2 ²¹ (version 2.3.5) somatic module along with Samtools (version 0.1.18) ²² was used to call somatic variations from all the paired samples (CC vs. FC and blood vs. saliva). Firstly .bam files were processed using Samtools for making .mpileup files which were further processed through Varscan2 to call somatic variations using somatic SNP module. Following parameters were considered while calling variations:

Minimum coverage to call somatic variation at a locus was kept at 8 reads

After calling variations, the data was checked for read bias. In this filter we excluded variations for which all the reads with the variant allele were read only from one direction (either F1R2 or F2R1) using in house developed perl scripts. Variations with more than 90% reads from one strand were removed. Annotation of all the variations called was done using Annovar (version 2012-03-08) ²³ and VcfCodingSnps (v1.5). Sites falling in regions with 87% and above identity with another genomic region were also removed from the data. Sites were termed as somatic sites if they had different genotype in two tissues and somatic p-value was less than 0.05. Overall out of a total 371 somatic sites (with p<0.05) in brain 93.8% had at least 4 reads for the variant allele. All the sites confidently called as same genotype in both tissue types (p<0.05) were termed as germline variations. Further, somatic variations with supporting reads for the variant allele >10% of total reads were considered as heterozygotes and those with <5% reads supporting variant allele were called homozygotes. All the somatic sites with their details are provided as CSV files under ‘Data availability’. Percentage of somatic variations across different sample-pairs was calculated by the following formula: (number of somatic variations/number of total variations) × 100. Number of somatic sites were defined as sites with different genotype between two tissues of the same individual. and Nnumber of total variations was refers to, all sites that had varying genotype from the reference genome (hg19) in two tissues of the same individual. We also used MuTect ²⁴ (version 1.4.4) variant caller to call somatic variations from exome sequencing data for the brain samples. Details of the concordance between the two platforms is described in Supplementary table S2. It was found that up to 78.8% of the somatic sites called by Varscan2 were also called as somatic sites by MuTect. Concordant sites between the two software also showed an enrichment of G:C>T:A transversions with FC harboring heterozygotes. Pathway analysis of somatic variations was done using Gene Set Enrichment Analysis ²⁵ (release 2.2.0). The total 359 genes harbouring somatic variations from all brain samples (from varscan2 dataset) were included for pathway analysis. Pathway analysis for the concordant sites also showed similar results.

Western blotting for NeuN: Total tissue lysates from FC and CC were made using radio-immuno precipitation assay (RIPA) buffer from 5 samples (3 samples viz. Brain_152, Brain_156 and Brain_202 that were sequenced and extra 2 samples viz. Brain_174 and Brain_119 that were not sequenced). Two extra samples were considered to emphasize that this contrast in NeuN between FC and CC tissue is true across all samples and is not specific to only the ones that were sequenced. Protein estimation was done using bicinchoninic acid assay method (BCA method) and 30 micro gm (ug) of protein was used for western blotting for NeuN, which was selected as marker for detecting neuronal cell bodies in FC and CC. Anti-NeuN monoclonal antibody (ab104224, Abcam), raised in mouse was used at a dilution of 1:1000. This antibody gives 2–3 bands between 46–48 kDa as per manufacturer data-sheet, which we also observed. GAPDH (2118L- Cell Signalling, 1:2000) was used as loading control. For this experiment a 5% stacking gel and a 12% resolving gel were used (both with 30% acrylamide, all reagent were from Sigma-Aldrich, USA). Images were analyzed by Image J 1.48 version.

Immuno-Fluorescence for NeuN: The same antibody was used to do immuno-fluorescence in both the tissues. Immunofluorescence for NeuN was performed on 8 µm sections of FC or CC. Briefly, tissue sections were fixed in chilled acetone, permeabilized with 0.1% Triton X-100, blocked in 1.5% FBS, incubated with Anti-NeuN monoclonal antibody (1:500, 16–18hrs at 4°C) and anti mouse Alexa 488-conjugated secondary antibody (1:200, 1hr at RT), mounted with DAPI containing solution before taking images using a confocal microscope (at 63×) [Zeiss-510 Meta, Carl Zeiss, Zen-2009 software (December 2010 release)].

8-OHdG, oxidative DNA damage marker, was measured in lysates of FC and CC by competitive ELISA (Cayman, USA). Briefly, 3 µg lysates were incubated with conjugate and antibody for 18 hrs and developed using Ellman’s reagent and absorbance readings was taken at 405 nm and results were expressed in pg/3 µg for each lysate.

We randomly selected 20 sites from Brain_156 (15% of the total sites) due to availability of sufficient DNA and this pair having the highest number of somatic sites, for validation using Illumina MiSeq Low Sample protocol (LS). Amplicons ranging from 200 bp to 350 bp for 14 sites were generated using PCR. The PCRs were done for 35 cyles with annealing temperatures at either 62 or 58 degrees C (primer details are provided under ‘Data availability’). Further all the amplicons from each tissue-type were pooled together in equimolar ratios, resulting in 2 freshly prepared libraries. These pooled amplicons were then processed using LS protocol of MiSeq sample preparation kit. About 12 million total reads were obtained from the two libraries. Quality check of trimming 3’ ends of the reads and filtering reads with less than 80% bases with Q30 phred quality score was applied on the raw data. Further, frequency of the variant allele (as observed in HiSeq data) was checked in both the tissues (FC and CC) in MiSeq data. A site was considered validated if the variant allele was supported by minimum of 100 reads along with at least two fold differences in the variant allele frequencies between the two tissues. On an average read depth of 5000× was obtained across all sites captured.

The number of high-confidence somatic variation calls seen for each pair ranged between 32–132 for the FC-CC pairs while it varied between 13–35 for the blood-saliva pairs ( Figure 1). The percentage of somatic variations ranged between 0.1%–0.48% for the brain sample-pairs (FC-CC) and between 0.03%–0.17% for the blood-saliva pairs. The number of germline variations across all samples was comparable ( Supplementary figure S3). The observed somatic variations do not show any bias for the position of the variant loci. The distribution was 37–65% in CDS, up to 57% in 3’ and 5’ UTRs and a minor proportion from other regions such as non-coding RNAs, splice sites etc. A comparison of these distributions between germline and somatic variations is represented in Supplementary figure S4.

As much as 64% of the somatic variations in the brain samples would lead to non-synonymous changes at the protein level ( Figure 2A). Whereas, for the germline variations the trend was towards higher synonymous variations as expected ( Figure 2B). For the blood-saliva pairs such a consistent trend towards non-synonymous variations for the somatic sites was not observed ( Supplementary figure S5). We analyzed the possible effect of the somatic sites at the amino-acid level for all samples and did not find a consistent bias for any particular amino acid change ( Supplementary figure S6).

Having observed somatic variation in the brain with a trend towards higher non-synonymous changes, we analyzed the level of transversion over transition amongst the somatic sites. Up to 87% of the total somatic variations found in FC-CC pairs were G:C>T:A transversions ( Figure 3A)! The germline variations (for all samples) matched the expected and reported distribution where A:T>G:C (36%) and G:C>A:T (38%) transitions were the most common types of changes ( Figure 3B). In Figure 3C we have represented the enrichment of G:C>T:A transversion events as a proportion of somatic variations, with respect to the germline variations. The blood-saliva pairs did not show such consistent positive enrichment for the same class of transversion events ( Supplementary figure S7).

Surprisingly, for the G:C>T:A transversion sites, 70–100% of GT and CA heterozygotes were present in the frontal cortex and the corresponding homozygotes were observed in the corpus callosum ( Figure 4). For all other sub-classes of somatic variations in brain and blood-saliva pairs the distribution of reference homozygotes versus heterozygotes did not show such a consistent and overriding bias ( Supplementary figure S8).

The genes harboring the somatic variations in all the brain samples were analyzed for canonical pathways. It was observed that the genes were enriched for axon guidance pathways and neuronal processes related pathways ( Figure 5) with p-values ranging from 0.04 to 4.9×10 ^-8 (FDR corrected). When analyzed by individual samples, three out of four samples also showed similar enrichment – but for Brain_152 although the relevant genes were found in the dataset but due to small number of variations, enrichment could not be established. It has been shown that DSBs in neuronal cells tend to occur in long genes involved in neuronal functions ²⁶ . We also find a similar trend in our data: 46% of the pathway genes harboring somatic variations were long (>100kb) compared to total genes (18% are >100kb, p<0.002).

The most common cause of G:C>T:A transversions is mis-pairing of G to A (instead of G to C) due to modification of deoxy guanosine (dG) to 8-hydroxy-2’-deoxy-Guanosine (8-OHdG) mediated by oxidative/metabolic stress ^{27– 29} . We found significantly higher levels of 8-OHdG in the frontal cortex samples, when compared to the corpus callosum of the same individuals ( Figure 6), which corresponded with the abundance of neuronal cells in the frontal cortex ( Figure 7). Thus an increased accumulation of 8-OHdG in the frontal cortex (compared to the corpus callosum of the same individual) DNA might result in localized DNA variations with a bias towards G:C>T:A transversions. Using immunoblotting and immuno fluorescence staining on tissues against the neuron specific marker NeuN we confirmed that the majority of the cells in the FC samples were neurons whereas, CC samples were almost devoid of neurons ( Figure 7) – implicating a direct correlation of abundance of neuronal cells with accumulation of 8-OHdG leading to G to T somatic transversions in the FC samples. Further effect of postmortem could be ruled out as one of the contributing factors, as time of collection of tissues, storage conditions and DNA isolation protocol were exactly the same for both the tissues.

Amplicon sequencing using MiSeq (Illumina) was performed for validation of the somatic variations. Out of the total somatic sites across all brain samples 20 sites from sample number Brain_156 were chosen for validation. We have randomly selected 20 sites from 156 (15% of the total sites) due to availability of DNA and this pair having the highest number of somatic sites. Data for 14 sites was obtained and 10 out of these 14 sites showed difference in variant allele frequencies between the two tissues (CC vs FC) in accordance with the HiSeq data (71% validation rate) ( Supplementary Table S4). Rows marked in red shade are the ones that got validated between two platforms. It should be noted that although the allele frequencies of the variations in the validation set were not exactly the same as in the HiSeq data but the trend (same tissue harboring the higher amount of variant allele) was same. In general, low variant allele frequencies were obtained in MiSeq data as compared to HiSeq data. This could be attributed to very high read depth per site in MiSeq data. A trend of varying allele frequencies with read depth was observed, as in, lower the read depth higher the allele frequencies (and vice versa), thereby explaining the high frequencies in HiSeq data (with average read depth of 100×) as compared to MiSeq data (with average read depth of 5000×).