Clinical manifestations of mtDNA deletion syndromes range from acute, multisystem, and often lethal disorders in infants and young children, to relatively mild neuromuscular symptoms that are often of adult onset 
. The underlying molecular features of these deletions are thought to play a role in the deletion process, and may also be related to the severity of the disorder. The aim of this study was to investigate the molecular characteristics of mtDNA deletions and their correlation to the age at onset and disease clinical phenotype, with the hypothesis that specific molecular signatures are associated with mtDNA deletion mechanism and thus clinical manifestation. We maintain a large clinical and molecular database of patients with mitochondrial disorders, including a large cohort of patients with an mtDNA deletion syndrome, which provided us with a unique opportunity to test this hypothesis.
It has been shown previously that mtDNA deletions exhibit enrichment for direct repeats at the breakpoints 
. These data have been corroborated by many groups and resulted in stratifying these deletions as type I (with a direct repeat) or type II (imperfect or no direct repeat) in relation to sequences at the two breakpoints. While the presence of the repeat may play a role in molecular events leading to an mtDNA deletion, the precise mechanism is not clear. On the other hand, there are numerous reports of young children severely affected with multisystem mtDNA deletion syndrome who did not have direct repeats at the breakpoints 
. This prompted us to characterize deletion breakpoints molecularly. Our data reveal that, regardless of the presence of a direct repeat, most mtDNA deletion regions have a significant increase in sequence homology surrounding the breakpoints. About one third of the patients in this study that had type II breakpoints nevertheless showed a significant increase in sequence homology relative to what would be expected by chance. These data provide evidence that sequence homology is the primary determinant of breakpoint distribution in mtDNA deletion syndromes. Since large direct repeats have the longest continuous stretches of sequence homology, they would be expected to be the most frequent sites of mtDNA deletion breakpoints. The most common mtDNA deletion, which is flanked by the longest (13 nt) direct repeat, is observed in more than one third (25/67
37%) of our patients. The second most common deletion is flanked by the 12 nt direct repeat in six of our patient samples. The presence of sequence homology at the breakpoints suggests its role not only in the generation of the break but also in the repair of mtDNA damage. It has been suggested that repeats drive breakpoint generation when there is an error in mtDNA replication due to inappropriate alignment of direct repeats 
, or a mtDNA damage 
. Both, defects in mtDNA replication caused by inappropriate alignment of direct repeats 
and mis-annealing of a single strand mtDNA molecule following double stranded breaks 
require the presence of direct repeats or sequence homology.
Although significantly higher than random mitochondrial genome homology, the youngest age group 1 showed significantly lower breakpoint homology relative to the older age groups. Furthermore, group 1 patients harbored a significantly lower percentage of samples with type I breakpoints (), and almost 3-fold decreased incidence of the common 5 kb mtDNA deletion relative to other three age groups, as well as increased heterogeneity in breakpoint distribution. However, disease severity is not affected by the size of deletion and genes deleted. These data suggest that molecular events responsible for mtDNA deletions in young patients may differ from those found in older age groups. One possibility is that mtDNA deletions in these patients are inherited from maternal germ line mutations, or are acquired during early embryogenesis, while the mtDNA deletions in older age groups represent later somatic random events. This is consistent with the clinical data obtained from these patients, where the vast majority of young patients present with the multisystem disease, while older age groups predominantly display a KSS spectrum or myopathy phenotype. Significantly increased levels of heteroplasmy in young patients relative to the older age groups provide further evidence for this hypothesis. Consistent with an earlier embryonic occurrence of the deletions, we have previously shown that levels of heteroplasmy for mtDNA deletions in young patients are present at similar levels in blood, muscle, skin and other tissues 
, whereas mtDNA deletions are almost exclusively localized to neuromuscular tissues in older patients. Whether there is interplay between the tissue specificity and the molecular mechanism of mtDNA rearrangement is not clear. It is possible that proteins involved in DNA rearrangement are differentially expressed in rapidly dividing/actively differentiating cells and non-dividing muscle and brain cells.
There is a significant difference (p<0.0014) in the frequency of the common deletion in patients with multisystem disorders (where the deletion is detected in blood) compared to patients with neuromuscular/PEO/MM disorders (11.5 versus 53.6%), further suggesting a different molecular mechanism for mtDNA rearrangement between germ cells and somatic cells. MtDNA deletions in germ cells may be driven by DNA replication, while mtDNA deletions in postzygotic somatic cells may be due to mis-pairing of homologous regions during repair of random oxidative mtDNA damage.
A recent study of mtDNA deletions in aged human skeletal muscle fibers has reported an increased incidence of mtDNA deletions with the presence of direct repeats compared to no direct repeats at breakpoints in older (>60 years old) relative to younger individuals 
. The authors proposed that oxidative-damage induced DNA-replication errors resulting in mtDNA deletions accumulated up to a pathological threshold over time. Presumably there would be a higher frequency of deletions at the direct-repeat hotspots that would selectively accumulate and be predominant over time. Conversely, a study of mtDNA deletion in substantia nigra neurons of age-matched patients with Parkinson Disease, mtDNA multiple deletion disorder, and single deletions, showed no difference in the type of deletion breakpoints in these samples, suggesting that a similar mechanism may be involved 
. Krishnan and colleagues suggested that increased reactive oxygen species produced in these neurons could cause double strand breaks through DNA damage or replication stalling, and proposed aberrant DNA repair as a mechanism for mtDNA deletion formation 
. Further support for this mechanism comes from a study in which restriction endonuclease PstI
-induced breakpoints resulted in the formation of mtDNA deletions flanked by short sequences with or without direct repeats 
. A common theme across all these studies is the detection of increased degree of sequence homology around the breakpoints of mtDNA deletions with breakpoints containing direct repeats, imperfect repeats, or no repeats at the breakpoint. Our data support these findings, and further demonstrate that a common feature of the breakpoints, irrespective of the type, is a significantly increased sequence homology around the breakpoint. Similar to the DNA replication model of mtDNA deletions, DNA break repair models rely on miss-annealing of the short homologous sequences in the single stranded DNA molecule, which in the latter case would be generated via 3′ to 5′ exonuclease activity. Three major types of error-prone double-strand breakpoint repair in the context of genomic DNA include non-homologous endjoining (NHEJ), microhomology-mediated end joining (MMEJ), and single-strand annealing (SSA). NHEJ was shown to rely on small (1–4 nt) homologies resulting in small <5 nt deletions/insertions, while MMEJ involves larger (5–25) nucleotide homologies that may or may not be direct repeats and can result in larger deletions 
. SSA requires direct repeats >30 nt. Therefore, MMEJ-type double-strand break repair mechanism would most closely fit the criteria based on the type of deletions seen in the mtDNA. The mitochondrial proteins that may be involved include mitochondrial polymerase gamma (POLG1) and SFN which are known to have a 3′ to 5′ exonuclease activity and are targeted to mitochondria 
. Another mechanism named microhomology-mediated break-induced replication (MMBIR), involving small (2–5 nt) homologous sequences at or near the breakpoint junctions, was recently proposed to be involved in human copy number variance (CNV) formation, and could theoretically be involved in mtDNA deletions 
. This mechanism is based on the assumption that during DNA replication, single stranded DNA breaks would generate single-stranded 3′ tails which could anneal with any single-stranded homologous DNA nearby and thereby would induce DNA rearrangements.
In conclusion, breakpoint sequence homology is a consistent feature of most mtDNA deletions. These data also suggest that the underlying molecular signatures determine the mtDNA deletion mechanism and may also correlate with clinical manifestations in patients with a mtDNA deletion syndrome. While we are still far from fully understanding the molecular mechanisms underlying these disorders, certain specific molecular features such as increased percentage of heteroplasmy, relatively lower homology at breakpoints, and a decreased frequency of the common deletions, may play a role in the severity and earlier age of onset of mtDNA deletion syndromes.