Analysis of somatic mutations in 131 human brains reveals aging-associated hypermutability

Abstract

We analyzed 131 human brains (44 neurotypical, 19 with Tourette syndrome, 9 with schizophrenia, and 59 with autism) for somatic mutations after whole genome sequencing to a depth of more than 200×. Typically, brains had 20 to 60 detectable single-nucleotide mutations, but ~6% of brains harbored hundreds of somatic mutations. Hypermutability was associated with age and damaging mutations in genes implicated in cancers and, in some brains, reflected in vivo clonal expansions. Somatic duplications, likely arising during development, were found in ~5% of normal and diseased brains, reflecting background mutagenesis. Brains with autism were associated with mutations creating putative transcription factor binding motifs in enhancer-like regions in the developing brain. The top-ranked affected motifs corresponded to MEIS (myeloid ectopic viral integration site) transcription factors, suggesting a potential link between their involvement in gene regulation and autism.

Fig. 1.. Mutations discovered across cohorts of brains.

A) Summary of coverage and detected somatic point mutations across cohorts using bulk brain samples. Mutation burden across cohorts was comparable when excluding hypermutable brains with more than 101 of mutations (named in the plots). Three colors represent institutions that processed the brain samples: Yale (blue), Harvard (orange), Lieber (red). B) The substitution spectra of detected mutations across all brains are comparable. C) Fractions of supported and not supported calls by assigning to haplotypes are comparable across brains, indicating the same accuracy of calls for hypermutable (indicated by arrows) and non-hypermutable brains. Due to the short DNA fragments (~450 bps), only ~20% of calls can be assigned to a haplotype using nearby heterozygous SNP; other calls are indicated as unphased. D) The combined mutation spectrum is dominated by C>T transitions in CpG motifs and matches the spectrum of developmental mutations (Fig. S2). E) When excluding hypermutable samples, the mutation burden does not correlate with age. However, there are more hypermutable brains (those are above the dotted line) in older brains (blue histogram with error bars in the background). Each data point is an individual brain with colors representing phenotypes: TS (blue), ASD (orange), SCZ (red), and normal (grey). Circles are males, crosses are females. F) Distributions of mutation counts and allele frequencies across cohorts. Numbers in parenthesis list the mutation count. Brain LIBD82 is an exception and its mutations had higher frequencies than in other brains. The VAF value of the vertical dashed line corresponds to ~15% of aneuploid cells in the hippocampus of that brain (which corresponds to displayed average VAF of ~4% between cortex and hippocampus). G) Brain LIBD82 had apparent aneuploidies in ~15% of cells in the hippocampus, which may be a signature of incipient glioma or glioblastoma.

Fig. 2.. Uneven cell lineage distribution of somatic mutations in brain.

A) Mutation allele frequencies in samples from brain NC7. Almost all mutations discovered in NC7 are present in the striatal interneuron fraction (STR-INT) at high VAFs. B) Distribution of mutation allele frequencies across samples in NC7. C) Genotyping of mutations in 16 single nuclei originated from the NC7 STR-INT fraction by sequencing at 5X depth. Black bars represent genotyped mutations. D) Examples of brains where frequencies of mutations are significantly biased toward cortex as compared to striatum (NC12) or as compared to hippocampus (LIBD80).

Fig. 3.. Relevance of somatic mutations to genome function.

A) In each cohort mutations from non-hypermutable brains have roughly the same distribution across genomes and similar predicted functional impacts by VEP. B) Counts of putative TF motifs disrupted by somatic mutations within enhancer-like elements from organoids (upper plots) or fetal brain (lower plots), showing that mutations significantly disrupt more putative TF binding sites in ASD vs controls. Each circle represents the number of TFs with x (count of mutations in ASD) and y coordinates (count of mutations in controls) of the circle. C) Somatic mutations in the ASD cohort predicted to lead to gain of binding sites for MEIS1, MEIS2, and MEIS3 in enhancer-like elements. The consensus motif is on top. Mutations on positive (+) or negative (−) strand are enumerated on the right with coordinates listed in Table S3. For each mutation, a reference base is show in small letter and the mutation base is noted by a capital letter above.

Fig. 4.. Detection and validation of somatic large structural mutations.

A) Example of a duplication on chromosome 3 (chr3:113,067,261-113,233,476) identified in brain TS1 in eight cell fractions and two bulks: cortex and striatum. Red vertical lines outline the region of the duplication. The two merged plots combine data for bulk and fractions for cortex (left) and striatum (right) and present an increase in the read depth corresponding to the cell frequency of ~32% (top), split in BAF likelihood function (brighter color corresponds to higher values) matching to the same cell frequency (middle) and BAF for individual SNPs (bottom). SNPs in the accessible genome (P-bases) are in blue, other SNPs are in green. B) Amplicon-seq validation of the duplication. Bar plot shows the number of reads mapping to the junction of the duplication in the cortical region BA17 and caudate putamen of TS1 and in control sample. C) Summary of size and cell frequency of all structural mutations detected from all analyzed individuals.