Using amplicon-based UDS at unprecedented depth, we have shown that detectable very low-frequency mtDNA variants (0.2–2% heteroplasmy) are present in all tested healthy subjects.
It is highly unlikely that nuclear pseudogene (NumtS) contamination influenced our results for the following reasons: Primer-BLAST analysis (hg19 build) failed to identify any predicted non-mtDNA PCR products using our specific MT-HV2
primers; we were unable to generate a product from rho0 genomic DNA using these primers. Moreover, further evidence came from the post hoc
analysis of individual sequence reads and scrutiny of all predicted nuclear pseudogenes (Supplementary Material, Table S2
). Based on this analysis, a minimum of six changes from the mtDNA consensus sequence would be expected in a read from a pseudogene with our MT-CO3
primers and ~48 changes for the MT-HV2
primers. Read-by-read analysis never identified this number of variants, so pseudogene contamination is extremely unlikely in our dataset. We conclude, therefore, that heteroplasmy is a universal finding in humans. Although our approach does not allow us to state with absolute certainty whether any one variant is inherited or a somatic mutation, a comparison of maternal relatives shows substantial sharing between first-degree relatives, and our findings, therefore, show that inherited variants make a significant contribution to the overall mutation load in any one individual.
None of the heteroplasmic variants detected were haplogroup defining or haplogroup specific. Twelve of 40 base positions containing heteroplasmic variants are reported to be polymorphic (>1% of reported sequences with variants) at the population level (http://www.mtdb.igp.uu.se/
). None of the variants detected have been ascribed definite pathogenic potential, although a single mutation at position m.9544 has previously been associated with optic neuropathy. We also examined whether variants detected had previously been reported as somatic mutations. In our data, 21 of 31 control region (MT-HV2
) heteroplasmic variants were detected, and 2 of 9 coding region (MT-CO3
) variants have previous been reported as somatic variants (http://www.mitomap.org/MITOMAP
) (Supplementary Material, Table S4
Using this approach, we also show that a disrupted mtDNA polymerase γ due to a POLG
defect leads to increased levels of mutations, and that these increase during life, mirroring observations in POLG-
deficient mice (29
). In contrast, we did not observe a correlation between age and mutation burden in control subjects, whether in the blood or muscle. This suggests that de novo
mutagenesis throughout life is unlikely to contribute significantly to the cellular COX defects observed in healthy aged individuals. Rather, age-related COX defects are more likely to be the result of clonal expansions of mutations occurring during early life (31
). This notion has recently been demonstrated in aged colonic crypt cells, where multimethod measurement of mtDNA mutation load failed to demonstrate an age-dependent increase (32
). However, our observation of very low-frequency heteroplasmy transmission indicates that that many of these mutations will be inherited down the maternal line. As a result, all clonally expanded species need not be the result of somatic mutation events in early life, and some could have arisen from low-level inherited variants.
The changes in low-level heteroplasmy value between each mother and offspring (Supplementary Material, Figure S1 and Supplementary Material, Table S4
) are relatively small when compared with the shifts observed in patient pedigrees carrying higher levels of pathogenic mutations (33
). However, this is consistent with the neutral drift theory (34
) which predicts that the variance in the offspring heteroplasmy due to the mtDNA bottleneck decreases as the mother's mutation level decreases. These are the first data to show the inheritance of such low-level mtDNA mutations. The high rate of inheritance of the low-level mutation in muscle (71%) may be surprising, considering our expectations of the effect of the mtDNA inheritance bottleneck; however, two points should be considered here. First, that muscle data is from only two sibling pairs [subjects A-B and P-Q (Supplementary Material, Figure S1 and Supplementary Material, Table S4
)]. Second, as noted above, the neutral drift theory predicts that very low-level mutations will have low variance in the offspring, making them more likely to be preserved through the mtDNA inheritance bottleneck.
The different patterns of mutation observed in the blood and skeletal muscle are, at least in part, likely to be a consequence of different rates of cell and mtDNA turnover between dividing (blood) and post mitotic tissues. Rapid turnover of blood cells can lead to the loss of mtDNA mutations during life, either through selection against a particular mutation (35
) or simply by genetic drift (36
). On the other hand, the loss of mutations is much less likely in a post mitotic tissue such as the skeletal muscle, where the replication of mtDNA can lead to an increase in mutation load during life within individual cells and the tissue as a whole, even from very low levels of heteroplasmy (31
). Thus, it is plausible that an inherited mutation is lost from the blood, but detected in the muscle, explaining why some inherited mutations are more likely to be detected in the muscle than in the blood.
It is intriguing that the frequency of low-level variants in MT-HV2
is significantly greater than in MT-CO3
in healthy control subjects (OR 3.3). Why should this be the case, given that our UDS of cloned mtDNA showed no intrinsic sequence-specific difference between the two templates? One possible explanation is that much of the non-coding D-loop heteroplasmy is actually inherited at a very low level. Being a non-coding region, these substitutions may be tolerated during transmission, unlike coding region variants that undergo strong negative selection during transmission (37
Perhaps most importantly, we show here that next generation sequencing has the potential to reliably detect very low levels of heteroplasmy, when a very stringent analytical approach is employed. It is important to note that we did not perform a head-to-head comparison of different next generation sequencing platforms, so it would be premature to conclude that the 454 approach is superior to other platforms. However, using this method, we have gained novel insight into mtDNA within individuals and within pedigrees. Prospective studies in larger family-based cohorts will substantiate these findings. However, given that mtDNA heteroplasmy levels can change dramatically during life, and during maternal transmission, the finding of universal mtDNA heteroplasmy has significant implications for our understanding of mtDNA at the population, family and individual level. If deleterious mutations are inherited, these have the potential to accumulate within single cells during life and thus contribute to neurodegenerative disease. Or, if they segregate rapidly through the mtDNA bottleneck, they could lead to a maternally inherited mitochondrial disorder. This places greater emphasis on the importance of developing techniques to prevent the transmission of mtDNA heteroplasmy and preventing the clonal expansion of pre-existing mtDNA mutations.