Erratum to: Mitochondrial DNA point mutations and relative copy number in 1363 disease and control human brains

Erratum In the original publication of this article [1], the author Marzena Kurzawa-Akinibi was written incorrectly. The correct spelling should have been Marzena Kurzawa-Akanbi. The original article has now been updated to reflect this change. References 1. Wei et al (2017) Mitochondrial DNA point mutations and relative copy number in 1363 disease and control human brains. Acta Neuropathol Commun 5:13. permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver


Introduction
Mitochondria are critical intracellular organelles involved in calcium signaling, lipid biosynthesis, apoptosis [31], and the generation of adenosine triphosphate (ATP) via the mitochondrial respiratory chain [28]. Over the last decade it has become clear that mitochondria play a key role in the pathogenesis of common neurodegenerative disorders. Mitochondria contain their own 16.5 kb circular mitochondria genome (mtDNA), which codes for key components of the mitochondrial proteome. MtDNA is present in 10s-1000s of copies in each cell and undergoes lifelong replication in post-mitotic cells including neurons [10]. With rudimentary mechanisms for repair, mtDNA is vulnerable to mutations, which accumulate within cells and also within the germ line.
There is emerging evidence that genetic variation of mtDNA contributes to the pathogenesis of neurodegenerative disorders [14]. Common population genetic variants divide the human population into a geographicallydefined 'haplogroups' [30], which have been associated with several neurodegenerative disorders including Alzheimer's disease (AD) and Parkinson's disease (PD), conferring a small increase in disease risk [13,22]. Furthermore, high-density genotyping arrays provide preliminary evidence that rare (minor allele frequency, MAF <5%) mtDNA polymorphisms are also associated with neurodegenerative disease, supporting previous work on AD [12].
In addition to maternally inherited polymorphisms, acquired somatic mutations of mtDNA have also been associated with neurodegenerative disorders. Unlike the maternally inherited germ line variants (which are 'homoplasmic'), the somatic mutations are usually present alongside the original wild-type molecules (heteroplasmy), and the proportion of mutated alleles determines whether a biochemical defect is manifest at the cellular level. MtDNA deletions accumulate in the ageing brain, reaching higher levels in regions vulnerable to neurodegeneration [3,26], but evidence describing the accumulation of somatic point mutations or small insertion-deletion mutations (indels) is less compelling, with conflicting reports in the literature [6,20]. Finally, several recent studies have described abnormal amounts of mtDNA both in cerebrospinal fluid or the brains of patients with neurodegenerative diseases [23,24], but only in a limited number of individuals.
To provide definitive evidence in all three areas, we studied the entire mtDNA sequence and quantity in1363 post mortem brains with Alzheimer's disease (AD), Amyotrophic-frontotemporal dementia (ALS-FTD), Creutzfeldt-Jakob Disease (CJD), Parkinson's disease (PD) and Dementia with Lewy Bodies (DLB), and compared them to healthy age-matched control brains. We found no evidence that rare inherited polymorphisms or mtDNA heteroplasmy contributes to the pathogenesis of neurodegenerative diseases, although differences in mtDNA content provide a clue to disease mechanism in AD and CJD.

Study samples
Full mitochondrial genome sequence data was extracted from Exome Sequencing data of 1363 case or control brain tissue samples from the Medical Research Council Brain Tissue Resource [15]. Cases fulfilling both antemortem and post-mortem diagnostic criteria for major neurodegenerative disease were included (Table 1, Additional file 1: Methods, Additional file 1: Table S1). DNA was extracted from the cerebellum in 87.3% of cases (n = 1190), cerebral cortex in 6.5% of cases (n = 89) with other brain regions in 6.16% (n = 84).
Known insertions or deletions (ins/dels) were defined, and all variants scored in HmtDB [27] and MITOMAP [28]. Remaining reads reconstructed the mitochondrial genome. Nucleotide mismatches and ins/del with quality scores (QS ≥25) and read depth (rd ≥5) were included.

Determining heteroplasmy and homoplasmy
We determined the proportion of variant alleles at each site of the mitochondrial genome. We then calculated the heteroplasmic fraction (HF, %) by dividing the number of variant reads by the total number of reads (for SNVs and deletions) or of the total number of 5′ flanking reads (for insertions). If the HF was <10% or >90%, we conservatively considered the variant site to be homoplasmic. If the HF was between 10 and 90%, the Key: AD Alzheimer's disease, CJD Creutzfeldt Jacob Disease, DLB-PD Dementia with Lewy Bodies or Parkinson's disease, FTD-ALS Frontotemporal Dementia or Amyotrophic Lateral Sclerosis. Information about Other disorders can be seen in Additional file 1: Table S2 site was considered to be heteroplasmic, and the HF was studied further.

Defining rare variants
Minor allele frequencies for each base of the mitochondrial genome were calculated from 30,506 full-length mitochondrial sequences in NCBI-GenBank using custom Python scripts. Rare homoplasmic variations were defined as those alleles present in less than 5% of individuals within their haplogroup using MITOMASTER [21], and novel variants those not present in the NCBI-GenBank dataset, 1000 genomes [9], MITOMAP [28] or HmDB [27].

MtDNA copy number estimation
The relative amount of mtDNA in each brain (referred to as mtDNA copy number) was calculated as the ratio between the mean mtDNA read depth and the mean exome read depth as previously described [5].

Statistical analysis
Mean variant counts or fractions both within and between groups were performed using Mann-Whitney, Fisher's exact, or Spearman's rank test tests as appropriate and as defined at the uncorrected threshold. A Poisson loglinear model was used to test the association between the number of variants and age among individuals within each main group. All statistical analyses were performed using R (v3.0.2) (http://CRAN.R-project.org/ doc/FAQ/R-FAQ.html), and plots made using Python.

Coverage and quality of assembled mitochondrial genomes
Complete assembly of the mitochondrial genome was obtained in all cases, with a mean read depth across the genome of 289 (SD = 169.8) (range 66-1328) and a mean base quality score of 36.8 (sd = 0.25) (range 36.0-37.5) (Additional file 1: Figures S1 and S2). There was no difference in mean read depth or mean base quality score for any group vs controls (Additional file 1: Figure S3).

Haplogroup associations
Haplogroups and phylogenetic relationships were determined for all 1363 samples (Fig. 1). There was no difference in major overall haplogroup frequency when compared to 2360 UK population controls (Additional file 1: Table S2), confirming the accuracy of haplogroup calling and that the cohort as representative of the UK population. We saw no association between any disease cohort and specific haplogroups in our study (Additional file 1: Table S3).

Homoplasmic variants
One thousand, nine hundred twenty-three homoplasmic variants were detected within the cohort, with a mean of 22.9 (sd = 10.8) variants per sample. Four hundred sixtyseven variants were defined as 'common' (minor population Allele Frequency, MpAF >0.05 within their haplogroup), and included known haplogroup defining variants. One thousand, four hundred fifty-six homoplasmic variants were defined as rare (MpAF <0.05 within their haplogroup). Twenty-five of the rare variants were novel, not seen in the NCBI database (n = 30,506), MITOMAP (n = 30,589) or the 1000 genomes database. Four of the novel variants were in non-coding regions, two were in rRNA genes, and 19 were synonymous (Additional file 1: Table S4).

Association with diseasecommon homoplasmic variants
Here we considered all homoplasmic variants, and a subgroup analysis of all non-synonymous homoplasmic variants. We saw no evidence of a disease association with any single variant, the burden of homoplasmic variants in any gene, nor the burden of homoplasmic variants in groups of genes forming a respiratory chain complex (Additional file 1: Figures S4-S8, Additional file 1: Table S5). However, when stratifying by age, there was a trend towards young onset AD cases (age of death <60) having a greater number of total variants in MT-TR compared to controls (6/13 vs 9/139) (p = 0.002) (Additional file 1: Figure S5).

Association with disease -rare homoplasmic variants
No single rare homoplasmic variant was present at greater frequency in any disease compared to controls (Additional file 1: Figure S8). There was a trend towards a greater number of rare homoplasmic point mutations in two genes in AD compared to controls; MT-RNR1 (AD; 30/282 (10.6%), Controls; 16/344 (4.7%)) (p = 0.005)) and again MT-TR (AD; 6/282 (2.1%), Controls; 0/ 344) (p = 0.008), although both failed to reach significance at the corrected threshold of p = 0.0014. (Additional file 1: Figure S8, Additional file 1: Table S5). When stratified by age, this suggested that the trend towards an excess burden of rare homoplasmic variants in MT-RNR1 was likely driven by variants in young onset AD cases vs controls (AD: 9/53 (16.9%), Controls (2/65), p = 0.012 (3%) (Additional file 1: Figure S9). We also saw that young onset PD-DLB cases (death aged <70) had a significantly greater number of rare homoplastic mutations in MT-CO2 (PD-DLB: 5/23 (21.7%), Controls: 5/ 213 (2.3%), p = 0.0010 (Additional file 1: Figure S9). The majority of the variants in MT-CO2 in both cohorts were in non-coding D-loop, but when combined this did not reach the corrected threshold for significance. There was no association between any rare non-synonymous variant, nor the burden of rare non-synonymous variants in any gene or respiratory chain complex in any disease group vs controls.

Heteroplasmic variants
Three hundred eleven heteroplasmic variants (>10% MAF) were detected (mean HF = 7%, sd = 1.0), in 440 cases, with 10 of these variants entirely novel. 55.7% of all heteroplasmic variants occurred within the D-loop, 33.3% in coding regions, 4.8% in rRNA genes and 6.2% in tRNA genes (Fig. 2). There was no association between any disease group and controls for any single heteroplasmic variant, the total number of heteroplasmic variants, or the mean variant pathogenicity score. There was also no association between the number of non-synonymous  Figure S12).

Heteroplasmy and age
We subsequently used a Poison loglinear model to determine the relationship between heteroplasmy and age within each group. There was no age correlation with the total number of heteroplasmic variants, mean level of heteroplasmy (HF), nor the mean variant pathogenicity score in any disease group (Fig. 3). mtDNA number mtDNA copy number was significant lower in AD and CJD compared to controls (p = 2.85 × 10 −7 (AD), p = 3.34 × 10 −7 (CJD)), and we observed a strongly positive correlation between age and mtDNA copy number in CJD (p = 2.7 × 10 −11 ). No association with age was seen in other groups (Fig. 4). The frequency of cerebellar samples in the AD cohort was no different to controls (p = 0.64) or the FTD-ALS cohort (p = 0.87). However, the CJD cohort did show a greater proportion of cases from the cerebellum compared to all other cohorts (100%, p < 0.001 vs all other groups).

Discussion
Here we report a comprehensive study of mtDNA sequence variation and abundance in brain tissue from 1363 neuropathologically characterized post mortem samples from MRC Brain Tissue Resource. After correction for multiple significance testing, we saw no difference in the frequency of mtDNA haplogroups, no difference in the frequency of common or rare homoplasmic variants, and no difference in the presence or degree of mtDNA heteroplasmy between different disease groups and control subjects. Overall, we conclude that neither rare homoplasmic variants nor heteroplasmic variants play a substantial role in the pathogenesis of these disorders. However, further work is required to clarify whether mtDNA copy number is important for the pathogenesis of AD and CJD.
Our most interesting finding was the frequent occurrence of mtDNA heteroplasmy in human brain tissue (HF >10% found in 32.3% of cases and controls), but contrary to previous reports, this did not change with age, nor was it associated with disease. This suggests that, although ongoing replication errors in mtDNA occur with age [16,25], moderate level heteroplasmic variants (>10%) are likely to have either occurred de novo in early development, or have been inherited within the germ-line, perhaps clonally expanding to these levels in early life. Resolving this issue will be difficult in human subjects, but will require the analysis of serial samples from the same tissue in the same subject. We cannot exclude the possibility that low level heteroplasmy (MAF <10%) is associated with these different disorders, or that different regions of the brain contain a significant burden of mtDNA variants, as recently described [8]. However, if low level heteroplasmic variants are important (perhaps because the low mean level masks high levels in individual neurons or glia), it is surprising that the higher levels of heteroplasmy we detected were not associated with disease, and that patient brains did not contain more pathogenic mutations than control subjects.
Finally we observed a significant reduction in mtDNA copy number in both Alzheimer's disease and CJD brains. Our data support previous work in AD [7], but to our knowledge is the first study of mtDNA copy number in CJD brain tissue. It is possible that the CJD finding reflects the higher proportion of cerebellar samples studied when compared to controls. In addition, the mtDNA copy number in the variant CJD (vCJD) cases (n = 40, mean mtDNA copy number = 4.14, sd = 2.96) was lower than all other CJD types (mean = 10.03, sd = 7.67; p = 2.8 × 10 −6 ). Therefore, it is therefore also possible that the trend observed for CJD actually reflects unusually low levels of mtDNA in the brains from patients with vCJD, who died at a younger age than other forms of CJD.  The relative mtDNA copy number in each disease cohort. a Top -Relative copy number of each cohort calculated as the ratio between the mean mtDNA read depth and the mean exome read depth as previously described [5]. ** (P < 0.01). b Bottom -The association between relative mtDNA copy number and age for all CJD cases (n = 182) with the Spearman Rank ρ and p-value shown. Key -AD -Alzheimer's disease, CJD -Creutzfeldt Jacob Disease, DLB-PD -Dementia with Lewy Bodies or Parkinson's disease, FTD-ALS -Frontotemporal Dementia or Amyotrophic Lateral Sclerosis