Absence of coding somatic single nucleotide variants within well-known candidate genes in late-onset sporadic Alzheimer’s Disease based on the analysis of multi-omics data

Abstract

Somatic mutations arise randomly or are induced by environmental factors, which may increase the risk of Alzheimer’s disease (AD). Identifying somatic mutations in sporadic AD (SAD) may provide new insight of the disease. To evaluate the potential contribution of somatic single nucleotide variations (SNVs), particularly that of well-known AD-candidate genes, we investigated sequencing data sets from four platforms: whole-genome sequencing (WGS), deep whole-exome sequencing (WES) on paired brain and liver samples, RNA sequencing (RNA-seq), and single-cell whole-genome sequencing (scWGS) of brain samples from 16 AD patients and 16 non-AD individuals. We found that the average number, mean variant allele fractions (VAFs) and mutational signatures of somatic SNVs have similar distributions between AD brains and non-AD brains. We did not identify any somatic SNVs within coding regions of the APP, PSEN1, PSEN2, nor in APOE. This study shows that somatic SNVs within the coding region of AD-candidate genes are unlikely to be a common causal factor for SAD.

1. Introduction

Alzheimer’s disease (AD) is the most common cause of dementia. Only the rarer form, early-onset familial AD (FAD) is known from genetic causes. A definitive understanding of the much more common sporadic AD (SAD) is lacking. SAD likely emerges from the combined workings of genetics and environmental factors. Somatic mutations are post-zygotic genetic variations and not inherited from one’s parents. In the brain, somatic mutations are known to develop and accumulate during development and aging. Researchers have begun searching for associated somatic single-nucleotide variants (SNVs) in candidate genes in the brains of patients with AD, but most of studies showed negative results. In contrast, a recent study showed positive results and brought a focus on the role of somatic mutations in the APP (amyloid precursor protein) gene in the pathogenesis of SAD (Lee, M.H. et al., 2018). However, an abundance of somatic SNVs within the coding region of the APP gene conflicts with earlier findings (Sala Frigerio et al., 2015), (Nicolas et al., 2018), (Keogh et al., 2018), (Parcerisas et al., 2014). Given these studies with inconsistent evidence, we are interested in evaluating whether somatic SNVs may be involved in the pathogenesis of SAD. The first aim of this study was to evaluate the genome-wide landscape of somatic SNVs within the brains of patients with SAD using whole-genome sequencing (WGS) data. The second aim of this study was to further investigate the somatic SNVs in the APP and other AD candidate genes, namely PSEN1 (presenilin 1), PSEN2 (presenilin 2), and APOE (apolipoprotein E) using multiple data sets (Supplementary Table 1–4).

2. Methods

We received human postmortem frontal cortex and liver tissue samples from patients with AD (n=16) and individuals with non-AD (n=16) who had earlier consented to postmortem research at the Banner Sun Health Research Institute (BSHRI) and the Stanley Medical Research Institute (SMRI). All patients were clinically confirmed definitively with a diagnosis of AD.

Here we performed four sequencing strategies to identify somatic SNVs in AD as follows: 1) WGS (average read depth 32×) in brain and matched liver samples from eight patients with SAD and seven individuals with non-AD; 2) deep whole-exome sequencing (WES, average read depth 300× in brain and 100× in liver samples) of samples from independent eight patients with SAD and eight individuals with non-AD; 3) RNA sequencing (RNA-seq) of brain samples from five patients with AD and five individuals with non-AD and 4) single-cell whole-genome sequencing (scWGS, average read depth 16.5× per neuron) of 96 single neurons with six neurons per individual from the same subjects analyzed using WES (Supplementary Methods).

3. Results

We identified somatic SNVs from the brain samples in each subject and quantified somatic SNVs, variant allele fraction (VAF), and mutation subtypes of somatic SNVs in patients with SAD and individuals with non-AD based on WGS data. After filtering out SNVs with low coverage depth (<10×) and high frequency (>0.4), we detected an average of 94.8 ± 40.0 (mean ± standard deviation (SD)) somatic SNVs per sample (range 46-176) in the AD group and an average of 84.3 ± 22.9 (mean ± SD) somatic SNVs per sample (range 58-125) in the non-AD group. No significant difference was detected in the average numbers of somatic SNVs between AD and non-AD groups (t-test, p-value=0.57). The VAFs of detected somatic SNVs ranged from ~4.2% to 39.9% (median: 17.8%) in the AD group, and ~5.7% to 40% (median: 17.65%) in the non-AD group. Again, we observed no difference in the mean frequency of somatic SNVs between the AD group and non-AD group (t-test, p-value=0.65) (Supplementary Figure 1). We found seven somatic SNVs (7/758, 0.92%) in exons and one in a splicing region in the AD group, and eight somatic SNVs (8/580, 1.36%) in exons in the non-AD group (Supplementary Table 5). No somatic SNV was detected in coding regions of the AD candidate genes. Meanwhile, no somatic SNV was detected in coding regions of genes involved in synaptic plasticity, axonal transport, and differential expression in AD samples (Supplementary Table 6).

We examined the mutation signatures of brain somatic SNVs using pooled somatic SNVs from AD samples (758 SNVs) and non-AD samples (590 SNVs). In AD brains, we found that single base substitution (SBS) signature 5 and signature 2 accounted for 94.9% and 5.1% of all somatic SNVs, respectively. Meanwhile, in non-AD brains, SBS signatures 5 and 2 explained 93.1% and 6.9% of all somatic SNVs, respectively (Supplementary Figure 2). The proportion of signature 5 and signature 2 between AD and non-AD were not statistically different (χ², p-value = 0.81), indicating similar rates of various mutation mechanisms in AD brains and non-AD brains. Gene-set enrichment analysis showed that somatic SNVs annotated genes in AD samples were not significantly enriched in any Gene Ontology or KEGG pathway under the threshold of adjusted p-value 0.05 (Supplementary Table 7, 8).

To search deeper for somatic SNVs in coding regions of the AD candidate genes, including PSEN1, PSEN2, APOE, and verify the observation from WGS, we looked for somatic SNVs in three other data sets, including the deep WES data, RNA sequencing data, and scWGS data. Despite the coding regions of the AD candidate genes in patients having a sufficient depth in both brain and liver samples to enable somatic SNVs detection (Supplementary Figure 3), no somatic SNVs within coding regions of four candidate genes were detected. Meanwhile, we did not detect any somatic SNVs in AD candidate genes within the RNA level. Furthermore, still no somatic SNVs were detected within any exons of candidate genes at the single-cell level in AD brains using scWGS data. As observed in deep WES data, while a few somatic SNVs were detected in intronic or intergenic regions of candidate genes, no somatic SNVs were detected in the exonic regions (Supplementary Table 9). Ultimately, no somatic SNVs were detected within coding regions of AD-candidate genes using multiple sequencing datasets from independent samples.

4. Discussion

Our extensive investigation yielded no SNVs within coding regions of APP, nor in other major candidates (i.e., PSEN1, PSEN2, and APOE). We analyzed two independent sample sets with four different sequencing technologies, which is to our knowledge the most comprehensive sequencing dataset for the detection of somatic SNVs in SAD to date. The detection of somatic SNVs in exonic regions of non-candidate genes (AF from 2.8% to 31.5%) from brain tissues proved that our system had adequate ability to detect SNVs if they existed (Supplementary Table 5, 10). The absence of somatic SNVs in the candidate genes from bulk WGS data, deep WES, and RNA-seq data may reflect a lower mutation rate in those genes, even if they exist. In addition, we used scWGS to analyze 96 single neurons from the eight patients and eight controls that were also analyzed using WES to reduce the possibility of missing rare mutations in a single cell. For the scWGS data, the number of cells determines the probability of detection of somatic mutations. For 48 cells of the AD group, we can achieve 60% power to detect somatic mutation at 1% frequency. But again, still, no somatic SNVs were detected within any exons of AD-candidate genes. Furthermore, our results with AD-candidate genes are in line with recent findings from previously targeted gene or whole-exon sequencing studies.

In conclusion, based on our consistent failure to detect somatic SNVs in the coding region of AD candidate genes in four sequencing datasets, our results do not support an important role for somatic SNVs in APP, PSEN1, PSEN2, and APOE genes in the pathogenesis of SAD, though the sample size in our study was relatively small and larger cohorts should be tested to confirm this finding.

Highlights.

1. Multi-omics data from 16 AD patients and 16 controls were studied for somatic SNVs.

2. A comprehensive detection of somatic SNVs in APP, PSEN1, PSEN2, and APOE in AD.

3. The distribution of number and VAFs of somatic SNVs were similar in AD and non-AD.

4. No somatic SNVs were detected in the coding region of candidate genes in AD brains.

Data availability

Clinical, and raw fastq files from our SMIB-AD study (The Somatic Mutations in Brains of Alzheimer’s Disease) have been submitted to the AMP-AD Knowledge Portal (https://adknowledgeportal.synapse.org/) under accession number https://www.synapse.org/#!Synapse:syn22278344.