The pseudo-mitochondrial genome influences mistakes in heteroplasmy interpretation

The unique maternal inheritance pattern of mitochondrial DNA (mtDNA), its small genome size, accelerated mutation rate, lack of recombination, and multiple copy number per cell, in comparison to nuclear DNA, are ideal biological traits for investigating evolution, population genetics and for forensic and medical applications. Thus, the mitochondrial genome has been used as a biosensor for the timing and movement of human populations in antiquity [1, 2]. MtDNA analysis is routinely used in forensic biology to type biological material when degradation prevents nuclear STR amplification [3]. In addition, the entire mitochondrial molecule has potential medical utility because it can serve as a repository of cancer mutations and as a biosensor indicative of genetic alterations [4–13].

Frequently, identifying legitimate mtDNA mutations is confounded by heteroplasmy, a condition in which wild-type and mutant mitochondrial genomes co-exist in a cell. The interpretation of heteroplasmy can further be confounded by the widespread integration of portions of the mitochondrial genome into the nuclear genome [14, 15]. These homologous, yet divergent nuclear and mtDNA sequences can be co-amplified in PCR reactions intended to replicate targeted mtDNA sequences only. Although this problem has previously been considered to be muted because of the high copy number of mtDNA over corresponding nuclear loci, caution is warranted [16]. For example, there are specific regions of the mitochondrial genome that have corresponding nuclear mitochondrial pseudogenes (numts) distributed across multiple chromosomes. Hence, there are regions of the mitochondrial genome that have a high nuclear copy number, which are not completely "swamped" during amplification. We report that some heteroplasmies detected in prostate cancer samples are a result of co-amplification of these multiple loci.

A large number of manuscripts addressing errors related to the interpretation of mtDNA and mtDNA heteroplasmy has been published [17–25]. Notably not all these errors are due to pseudogene co-amplification; however, mistakes from pseudogenes may increase with improved sequencing methods and highly sensitive re-sequencing microarray technologies that have a lower detection limit than traditional sequencing and which readily detect low-level heteroplasmy [11, 26]. In some cases, if the heteroplasmy is inherited, it substantially increases the power of mutation detection, which becomes an important aspect since heteroplasmy has been reported as an early indication of disease [27–31]. In addition, if the disease process invokes mitochondrial depletion, this could increase nuclear pseudogene signal in reactions as a result of reduced mitochondrial genome copy number [32]. Loss of mitochondria has been described in several human cancers [33–36]. As well, the number of mitochondria and mtDNA copy number vary for different cell types [37–39]. These important matters relating to sequence interpretation have been generally neglected, in part, due to the lack of numt reference material, which would help investigators determine the relevance of detected mtDNA sequence variations. Hence, the need to validate somatic mitochondrial mutations is a pressing one.

Heteroplasmic issues have already complicated data obtained from other species. For example, in elephant hair, low mtDNA content is the reason why numts were co-amplified and misinterpreted as authentic mtDNA. In contrast, numts were not detected in DNA derived from elephant blood due to the presence of mitochondria-rich platelets [40]. Moreover, the hominid, Gorilla, is well known for significant numt interference with mitochondrial sequences, highlighting the need for diligence when interpreting human mtDNA heteroplasmy [41]. Not surprisingly, the effort for using mitochondrial cytochrome c oxidase as a primate "barcode" is plagued by numt amplification as well [42].

Further, laser capture microscopy has improved the ability to separate and analyze cancer cells, but because of the decreased amounts of sample DNA, many primer pairs are required to obtain a robust amplification of the entire mitochondrial genome [43]. Moreover, a sufficient number of cells must be captured to avoid incorporation errors associated with low template quantity [44]. This will also be relevant to studies that use formalin-fixed paraffin embedded samples [45]. The use of many primers means that smaller amplicons will be synthesized translating into a higher risk of co-amplification of numts and the potential misinterpretation of heteroplasmic calls.

There is limited in silico and wet-lab evidence indicating that fragments of the human mitochondrial genome are embedded in the nuclear DNA archive [46–50]. These findings emphasize the critical need to minimize erroneous interpretation of heteroplasmy, a vital necessity for precise forensic discrimination, evolutionary studies, and potential diagnostics. We provide evidence of numts for the entire mitochondrial genome by the amplification, cloning, and identification of numts from rho-zero (ρ⁰) cells and clinical cancer specimens. Here we present results from overlapping primers, which co-amplified nuclear embedded, paralogous mitochondrial sequence. Surprisingly, our data shows a relatively small number (when compared to hypothetical sequence information obtained from BLAST searches) of multiple nuclear loci that co-amplify with the mitochondrial genome. These findings demonstrate that accurate interpretation of heteroplasmy not only requires careful primer design and testing, but also indicates that a compendium of the sequence information from multiple-copy number numts is an important reference tool that will facilitate correct mtDNA interpretation and support reliable mitochondrial genome sequencing data.

In this study, we recovered and assembled the entire mitochondrial genome from nuclear loci. Moreover, this "pseudo-mitochondrial genome" involves numts from over half of the human complement of chromosomes, including the X chromosome. This suggests a widespread allocation of numts in the human nuclear genome. Surprisingly, this distribution was achieved with primers routinely used to amplify mtDNA, yet designed without consideration for numts. Seventy-one percent (24/34) of the primers co-amplified numts in prostate cancer tissue samples. This validates prior suggestions that numts are a potential source of misinformation and serves to illustrate the ease of co-amplification of both mtDNA and nuclear embedded paralogous mitochondrial DNA sequences [16]. Our data demonstrate that contrary to a consensus of opinions that the copy number of the mitochondrial genome "swamps" the signal from numts loci, there are circumstances which favor PCR recovery of numts, such as multiple pseudogene copy number [18]. For instance, heteroplasmic mutations had been associated with late-onset Alzheimer's disease [17, 53]; however, these false heteroplasmies resulted from co-amplification of numts [19, 54, 55]. Indeed, human numts perplexed ancient DNA studies as well when it was reported that DNA had been recovered and amplified from a Cretaceous dinosaur bone [56]. This sequence corresponded to a human numt containing cytochrome b sequence [57], probably from reagent or sample contamination.

Direct pseudogene contribution is not always obvious and can confound suggested mtDNA biomarkers. For example, one set of primers in our data set amplifies tRNA^leu and ND1 (3230–3893). Subsequent cloning data identified co-amplification of paralogous numts on five chromosomes with this amplicon. Of specific interest are bases 3316, 3496, 3697 and 3796, which are reported as potential disease associated sites [52]. These sites are problematic since the reported base changes are consistent with pseudogene presentation in our data. Further, results from new sequencing technologies have suggested that homoplasmic signals may indeed be heteroplasmic in nature [11]. In addition, the heteroplasmic patterns seen at bps 3697 and 3796 are mirrored by the nuclear pseudogene patterns (Table 3).

Re-examination of the raw data from the above studies could address if the disease mutations are actually due to co-amplified numts. Potential markers must be thoroughly investigated to preclude the inclusion of false mutations in the interpretation of mtDNA mutations. BLAST searches of nuclear pseudogenes belie the possibility of widespread integration and/or replication of these sequences, since primers may amplify homologous numts embedded elsewhere in the genome. Thus, high copy numbers for these nuclear segments can produce potential misleading heteroplasmic signals.

Comparative marker studies require the same conditions and primers for meaningful results. For example, proposed MELAS sites T3250C and T3291C (tRNA^leu) have paralogous nuclear sites [58–60]. Follow-up work by Akanuma et al. (2000) [61], demonstrate a corollary numt associated site for T3250C, but not T3291C. A BLAST search shows that these competing sequences are on separate chromosomes (17 and 20) indicating comparison of dissimilar data. In addition, the cloning data here identifies similar regions on chromosomes 2, 4 and 17 (Table 3). Clearly there are numerous paralogous loci for mitochondrial tRNA^leu, the amplification of which depends on location and homology of primer sets.

Because of this association in our data, we compared our cloning results to the suggested mitochondrial genome disease associated sites listed on MITOMAP [52]. Results suggest that many mutations require meticulous scrutiny because of paralogous nuclear commonalities. Although many of these mutations may well be actual disease markers, the possibility of numt association may confound detection. For example, proposed prostate cancer mutations (G 5913A, G 5973A and G 6081A) are identified, by cloning data, as resident on chromosomes 1 (2 homologous copies) and 17 even though the authors exercised precautionary measures by scanning a database of known nuclear pseudogenes [49, 62]. A locus on chromosome 6 was identified as a potential co-amplification product, yet chromosomes 1 and 17 were not detected. Co-amplification of numts is primer dependent, which may explain the differences seen here; however, database limitations and the absence of extensive wet-lab numt data obscure the meaning of the marker. Particularly suspicious are those sites which demonstrate heteroplasmy in both normal and disease tissue. This may reflect a consistent pattern of numt amplification, an unintended characteristic of primer design. A subset of the marker work prior to the seminal numt work of Lopez et al. (1994) [14] may need clarification as well. For example, an ND2 lesion associated with Alzheimer's and Parkinson's diseases (G 5460T/A) has modulated T and A nucleotides. Both mutations are seen in a BLAST search of numts. Moreover, our cloning data also identifies a G at this point, but not a T, again suggesting the relevance of primer selection; nevertheless, all possible modulations are described in this early work (see Kosel 1994. [63]), yet it remains a primary reference for this lesion.

If the use of mitochondrial DNA and in particular somatic mitochondrial genome mutations has important utility and medical merit, much of the data requires critical follow-up from a pseudogene perspective. Amplification of ρ⁰ DNA template with primers to identify and eliminate those which co-amplify nuclear pseudogenes is a vital and necessary procedure [16]. For example, mitochondrial PCR protocols were simultaneously run on clinical samples and nucleic acid recovered from ρ⁰ cells to identify and exclude co-amplification of numts in work by Coskun et al. [64]. Alternatively, data may be screened by amplification and sequencing of ρ⁰ derived DNA and conflicting sites then backed out of actual data generated with identical primers; however, this approach is labor-intensive [43]. Phylogenetic analysis of the data would also help distinguish polymorphisms from authentic mutations [22]. In general, and unfortunately the advice by Parfait et al. has been largely ignored [16].

Our surprising results are not limited to short amplicons, but are also detected in much larger amplicons. For example, the overlapping amplification of chromosome 5 from bp 8816 to 15051 cautions against assuming that long amplicons are pseudogene free. These possibilities and characteristics of the nuclear genome must be considered when using mitochondrial sequence data for population, forensic or disease studies. Although designing and testing primers to avoid co-amplification of numts is a good laboratory practice, compilation of numts representative of the entire mitochondrial genome is valuable to catalog and characterize the overall nuclear burden of these sequences.

Amplification of overlapping numts paralogous to the mitochondrial genome indicates that co-amplification of nuclear mitochondrial pseudogenes is a real problem for accurate sequence interpretation. Not only is co-amplification dependent on the particular amplicon used, but the copy number of these loci is also important. Only certain positions across the mitochondrial genome are associated with multiple copies of numts. Mitochondrial DNA heteroplasmy should be interpreted with caution since they can be the result of nuclear/cytoplasmic co-amplification. Herein, we have demonstrated the robust amplification of numts. This paper is the first to fully sequence the 46 paralogous DNA fragments that represent the entire mitochondrial genome using 36 primer pairs. This is a surprisingly low number, but reveals that only a limited number of paralogous numts are relevant when considering if heteroplasmic call are authentic mutations. Compilation of a complete data set of numt sequences will help others distinguish paralogous nuclear based heteroplasmy in forensic, population and medical applications.