Roles of mitochondrial genetics in cancer metastasis

Abstract

The contributions of mitochondria to cancer have been recognized for decades. However, the focus on the metabolic role of mitochondria and the diminutive size of the mitochondrial genome compared to the nuclear genome have hindered discovery of the roles of mitochondrial genetics in cancer. This review summarizes recent data demonstrating the contributions of mitochondrial DNA (mtDNA) copy-number variants (CNVs), somatic mutations, and germline polymorphisms to cancer initiation, progression, and metastasis. The goal is to summarize accumulating data to establish a framework for exploring the contributions of mtDNA to neoplasia and metastasis.

Mitochondrial contributions to disease

Cancers are polygenic diseases with genes driving oncogenesis, suppressing tumorigenicity, or modifying disease severity. Most research has focused on nuclear genome, mostly because of its relative size and greater level of characterization. However, eukaryotic cells also possess a genome within the mitochondrial organelle. Mitochondria are referred to as the ‘powerhouse of the cell’ because they are central to many metabolic and catabolic processes. Accumulating data demonstrate that mitochondria can no longer be seen simply as energy plants because they contribute to complex phenotypes in previously underappreciated ways [1–3].

To understand how mitochondrial genetics contribute to cancer, it is first important to define the origins and evolution of the mitochondrial organelle (Box 1) and the unique characteristics of the mitochondrial genome (Box 2). It is beyond the scope of this review to do so comprehensively, but a few key principles will be crucial in dissecting its contributions in cancer. (i) mtDNA is maternally inherited, and several different SNPs have been selected through evolution and can be used to distinguish ancestral groups. (ii) Copies of the circular mtDNA genome vary greatly between cells and tissues. (iii) When cells divide, mtDNA is not equally divided between progenitor cells. (iv) Mitochondria are dynamic organelles, changing shape and position according to physiologic needs (Box 3). When assessing mitochondrial contributions to disease, each of these characteristics must be considered. Consequently, SNPs and somatic mutations in mtDNA have been associated with age-related pathologies, for example neurodegenerative disorders and cancers [4–6]. Together these observations lead to the question: do mtDNA mutations or adaptations associated with human ancestry impact on disease susceptibility, progression, or severity? The goal of this brief review is to summarize recent data and establish a framework for dissecting the contributions of mitochondria (mtDNA SNPs and/or mutations) in neoplasia and metastasis (see Glossary).

Box 1. Mitochondrial evolution.

In 1905 the Russian botanist Konstantin Mereschkowski first postulated that the chloroplast was the result of a symbiotic relationship in plants [7]. In 1967, Lynn Sagan (née Margulis) proposed that the related organelle, the mitochondrion, was similarly symbiotic [8]. These concepts have been expanded to propose that mitochondria arose from incorporation of an Alphaproteobacterium into a host cell over a billion years ago [9]. From which, subsequent coevolution led to added functions to meet cellular needs, including intercellular communication [10,11], epigenetic tagging [12–14], innate immunity [15], regulation of apoptosis [16], and involvement in disease.

Current theories surrounding SNPs comprising human haplotypes support the existence of a mitochondrial ‘Eve’ from which matrilineal selection of variants based upon energy demands and thermoregulation occurred when humans migrated from Africa through Europe to Asia, the Americas, and Australasia [9,17–23]. As a result, SNPs which define haplogroups can be used as a criterion by which to recognize human population groups (i.e., ancestry or colloquially ‘race’).

Box 2. Mitochondrial DNA.

Mammalian mitochondrial genomes consist of ~16.5 kb of circular DNA that encodes 13 protein subunits of the electron transport chain (ETC), two rRNAs (16S and 12S), and 22 tRNAs. Replication of the mtDNA starts in the displacement (D)-loop, which also contains transcriptional control regions for mtDNA. Some bases within the D-loop region are conserved whereas others are highly variable. The latter have proved to be useful for studying vertebrate evolution. Mammalian mitochondrial genomes are similar but not identical. Table I compares human and murine mtDNA as an example. The strands of the mtDNA duplex can be distinguished based upon base composition. The heavy (H) strand contains more guanine and is the template from which 12 of the 13 mitochondrial protein mRNAs are transcribed, whereas the light (L) strand contains more cytosines and encodes only one protein [24].

Within cells, there can be tens to thousands of mtDNA copies [1,25,26]. Over time, the originally engulfed bacterium has ceded much of its genome to nDNA. Nuclear genes contribute >90% of the molecules responsible for mitochondrial structure, function, bioenergetics, replication, and repair [27,28]. A consequence is protection of crucial molecules from the 10- to 100-fold higher mutation rates found in mtDNA compared to nDNA [19]. Recently, it was reported that seven mitochondrial tRNAs have identical copies embedded in the human nuclear genome [29], presumably as a protection against mutations. Evidently this phenomenon is also present in animals – at least in marsupials [30], and plants (I.R., unpublished). The primary contributor to mtDNA mutations is the high level of reactive oxygen species (ROS) generated during electron transport accompanied by lower-fidelity DNA replication mediated by polymerase γ (Pol γ) [31,32] and Pol τ under conditions of oxidative stress [33]. Although Pol γ has high fidelity, an increased mutation rate is observed because of an elevated rate of replication rate, limited repair mechanisms, high ROS, and lack of protective histones [34–37]. A vast number of mtDNA mutations arise from DNA replication errors with heavy- and light-strand biases [38]. mtDNA includes regions with abundant nucleotide repeats which have been attributed to increased mtDNA mutations because of DNA polymerase slippage leading to the introduction of run-length variation at these sites [38,39]. Increased ROS levels have also been demonstrated to correlate with increased mtDNA mutation burden and tumor progression [40].

In addition, it was recently shown that mitochondrially encoded tRNAs correlate positively with nuclear mRNAs whose exons and introns contain repetitive elements [93]. This finding links repetitive elements to cancer via mtDNA. Notably, recent work has shown that mt-tRNA-derived fragments are associated with the nuclear localization, membrane localization, and secretion of specific proteins [93,94]. The details of these associations and the identity of the involved proteins differ by cancer type and are the subject of current investigation.

Box 3. Mitochondrial dynamics.

Some cells require longer mitochondrial filaments to fulfill cellular metabolic needs, whereas others have less physical space for organelles to function and are more uniform and spherical to fit into smaller spaces (e.g., hepatocytes). Although mitochondria within specific tissues have characteristic phenotypes, morphology is not static. The balance of fission and fusion changes in response to growth, mitochondrial damage, and energy levels [41,42]. These processes are crucial for the survival of all types of cells, even senescent cells [43]. For example, nonproliferating neurons require a proper balance of fission and fusion for normal physiological function because, without it, disease states such as Charcot–Marie–Tooth disease [44] and Leber’s hereditary optic neuropathy (LHON) can occur [45,46]. Embryogenesis cannot proceed and is often fatal when mechanisms connected to fission/fusion are dysfunctional. Fission is essential for growing and dividing cells to ensure organelle distribution between progeny, whereas fusion is key to maintaining mitochondrial health. Generally, when energy needs are increased, fusion promotes OXPHOS when protein synthesis is inhibited by starvation. Mitochondria fuse, mixing content in an apparent effort to correct damage or complement needs of malfunctioning mitochondria. When fission/fusion mechanisms are broken or hijacked because of acquired mutations, pathologies emerge, including cancers [47,48]. Recent data reveal that the subcellular localization of mitochondria also varies widely during cellular processes including migration and invasion that are most pertinent to this review [49–53]. It is worth interjecting that both Charcot–Marie–Tooth and LHON have also been linked to tRNA synthesis and functions as well as to mitochondrial dynamics [54–57].

Mitochondrial diseases

Mitochondrial diseases manifest largely in tissues which rely heavily on metabolism (e.g., skeletal and cardiac muscle) or in tissues that rely on glucose as their major energy source (e.g., brain and neurons) [7–10]. The latter result in LHON (Leber’s hereditary optic neuropathy), MELAS (mitochondrial encephalopathy, lactic acidosis, and stroke-like episodes), MERFF (myoclonic epilepsy with ragged red fibers), and NARP (neuropathy, ataxia, and retinitis pigmentosa) syndromes [3,11,12]. Mitochondria play many roles in other common diseases such as Alzheimer’s, Parkinson’s, glaucoma, multiple sclerosis, amyotrophic lateral sclerosis, cardiovascular disease, osteoarthritis, bipolar disorder, and type 2 diabetes [2,3], but those roles are typically secondary in nature and, at present, remain relatively ill-defined. Neoplastic cells are often metabolically unbalanced, eventually accumulating changes that snowball into the late-stage phenotypes observed clinically. Cancers are increasingly included among diseases in which mitochondrial mutations are observed [3,13–19], but whether those mutations are the cause or consequence (or both) is still under investigation.

The severity of most mitochondrial diseases is related to a ‘biochemical threshold’ which is when the number of mutant copies surpasses a given ratio within a cell or group of cells. The threshold can be widely distributed (i.e., affecting multiple tissues) or in specific tissues. This heterogeneity is, in part, why mitochondrial diseases are challenging to diagnose and understand. In some cells the normal, unmutated mtDNA can protect cells from mutated variants by rescue mechanisms such as fission/fusion [7,11,20]. Shifts in mtDNA content can alter metabolism and nuclear epigenetic marks [21,22]. It is also possible that mtDNA epigenetic modifications are also affected by mtDNA copy number, but to the best of our knowledge this has not been measured.

At birth, mtDNA is generally identical by sequence in all cells (homoplasmy), although copy numbers can vary by tissue. With aging, mtDNA can accumulate mutations such that cells contain mixtures of mtDNA (heteroplasmy). Because mtDNA is not distributed equally among progeny, heteroplasmy contributes to cellular evolution [11]. Further, heteroplasmy impacts on disease severity and disease subtype [20,23]. Interestingly, heteroplasmy levels fluctuate depending on tissue type and cellular energy needs, but revert toward homoplasmy that provides a selective advantage over another mutation. As a result, it is crucial to consider both sequence and copy number when assessing the contributions of mtDNA in cancer. From a technical perspective, the increased depth of genome sequencing has resulted in increased detection of heteroplasmy [4,5] which in turn complicates definitive attribution of driver versus ancillary mutations compared to the mostly diploid composition of nucleus-encoded genes. As a rule, persistent heteroplasmy contributes to phenotype instability and reduced cellular fitness [6]. Thus, there is selection pressure to reduce heteroplasmy even within cancer cells [24,25], which suggests that homoplasmy may be important for maintenance of mitochondrial function. Sercel and colleagues recently evaluated homo- and heteroplasmy in iPSCs and showed that, for retention of pluripotency, homoplasmy was a crucial determinant [26]. Whether the same patterns are observed in cancer stem cells has not yet been measured to the best of our knowledge.

Metabolism and cancer

The primary functions of mitochondria relate to energy homeostasis. Electron transport achieves this through a series of redox reactions that shuffle electrons and pump protons across the inner membrane, driving ATP synthesis via proton motive forces. The ‘tighter’ the complex, the more ATP is produced, the more reactive oxygen species (ROS) are produced, and less heat is generated. The loosely packed electron transport chain (ETC) (i.e., uncoupled) exhibits less efficient ATP production, reduced ROS levels, and greater heat emission. Uncoupling respiration from ATP synthesis was thought to be one the most crucial adaptations of Homo spp. living in colder climates. Several proteins responsible for uncoupling (uncoupling proteins, UCPs) have been implicated in cancer but also vary by tissue and uncoupling is influenced by the microenvironment.

In the 1920s Warburg reported a phenomenon referred to as ‘aerobic glycolysis’ in which neoplastic cells shift glucose conversion to ATP via glycolysis rather than via oxidative phosphorylation (OXPHOS) despite adequate oxygenation [27–29]. Subsequently, these observations have been replicated, but the hypothesis that mitochondrial dysfunction is a driver of cancer needed to be modified because most cancer cells maintain mitochondrial respiration [30,31]. Nevertheless, Warburg’s observations ushered in the concept that mitochondria are pivotal to cancer biology because of their regulation of metabolism. In addition, it is now appreciated that gain-of-function mutations in isodehydrogenase 1/2 (IDH1/2) can drive cancer [32]. Despite the discovery that some nuclear (n)DNA-encoded mitochondrial proteins regulate tumorigenesis [33,34], mtDNA SNPs or mutations are generally not considered to be driver mutations [35–37]. A rare benign class of tumors, termed oncocytomas, are an exception. They are characterized by abnormal accumulations of dysfunctional mitochondria [38], or accumulations of normal mitochondria that have increased respiration due to loss of a tumor-suppressor gene FLCN in the nucleus, both of which can cause malignancy.

mtDNA mutations in cancer and metastasis

It is important to emphasize that mtDNA encodes quantitative trait loci (QTLs) that work with nucleus-encoded genes to regulate many complex diseases [39–41]. Measurable phenotypic changes are influenced by the combined effects of SNPs, mutations, and environmental factors. It is therefore unlikely that mitochondrial polymorphisms will provide a complete explanation of differences in disease progression. These SNPs will, however, influence disease development and progression by gene–gene interactions and altering responses to the tumor microenvironment. This is a crucial point because mitochondrial SNPs would otherwise be predicted to exhibit strong maternal inheritance because mtDNA is maternally inherited. Instead, crosstalk between the nucleus and mitochondria conveys signals from the various microenvironments in which disseminated cells find themselves during the metastatic cascade. Depending upon the signals and the response of each cell, the efficiency of metastasis will be altered. Likewise, it must be emphasized that not all SNPs in mtDNA will be QTLs for a given phenotype.

The designation of mtDNA mutations as drivers has been difficult because of the limitations of experimental models and technology. Intriguingly, in a study evaluating breast cancers, mtDNA mutational burden was not correlated with survival [42]. Nevertheless, The Cancer Genome Atlas (TCGA) datasets identified some fascinating correlations [24,25] that will be elaborated later together with more recent findings. Definitive cause–effect relationships, however, are difficult to ascribe owing to the existence of multiple mtDNA copies per cell and the technical challenges associated with manipulating all copies of mtDNA [11,43,44]. In addition, most next-generation sequencing (NGS) studies filter out mtDNA sequences, possibly because the small mitochondrial genome was suspected to be inconsequential. Those raw data still exist and are expected to be a goldmine of information as the relevance of the mitochondrial genome, not only of metabolism, in cancer is appreciated more fully.

Mitochondrial CNVs correlate with cancer and metastasis

Recent evidence shows that CNVs exist for mtDNA in multiple cancers [2,13,37,45] (Figure 1, Key Figure). Ovarian cancers tended to have the most (>600 copies) whereas myeloid cancers had the least (~90 copies). Comparing cancer to matched non-cancerous tissues, increased mtDNA was found in patients with chronic lymphocytic leukemia, lung squamous cell carcinoma, or pancreatic adenocarcinoma, whereas the pattern was reversed in kidney clear cell carcinoma, hepatocellular carcinoma, and myeloproliferative neoplasms. CNVs correlated positively with age in prostate, colorectal, and skin cancer [37]. Interestingly, restoration of KISS1 metastasis-suppressor expression corresponded to an increase in the mitochondrial biogenesis regulator, PGC1α, with a corresponding increase in mitochondrial mass in the tumor cells [46]. Collectively, these data show that mitochondrial biogenesis regulation can result in apparently discordant results that differentially regulate oncogenesis and/or metastasis. Although the biogenesis pathways are clearly changed, definitive cause–effect relationships have not been established.

Figure 1. Key figure. Mitochondrial genetics contribute to cancer.

Mitochondrial (mt)DNA copy-number variants (CNVs) or germline SNPs are associated with the development and progression to metastasis of multiple cancer histotypes. In addition, somatic mutations emerge and are associated with progression of most cancers, but their roles as drivers are not established. Note: it is possible for SNPs and CNVs to coexist, but those data have not yet been widely studied nor reported. Abbreviation: osteoSa, osteosarcoma.

Mitochondrial transfer between cells

In recent years the mitochondrial content of tumor and neighboring host cells has been shown to change through exchange of mtDNA or whole mitochondria between cells via tunneling nanotubes (TNTs) [47–54]. Transfer has been associated with proliferation, aggressiveness, and glycolysis, as well as with pentose phosphate and lipid metabolism, in a variety of neoplastic cells [48–54]. These changes often occur concomitantly with chemoresistance [48,50,55]. It is currently only speculation what triggers TNTs, but cellular attempts to repair mtDNA damage is among the possible reasons. In a recent study, transfer of normal mitochondria into cisplatin-resistant MDA-MB-231 breast carcinoma cells restored chemosensitivity [56], leading the authors to speculate that mitochondrial transfer should be considered as a therapeutic option. Although an intriguing possibility, technical hurdles related to efficiency of transfer and instability associated with introduced heteroplasmy will need to be more thoroughly considered beforehand. Relatedly, recent data suggest that elevated levels of uncoupled protein 2 (UCP2) may contribute to chemoresistance [57] but this requires further investigation.

Germline mtDNA polymorphisms may explain cancer disparities

In 2020 Yuan et al. [37] performed a meta-analysis aggregating whole-genome sequencing data from 2658 cancers across 38 tumor types to identify germline and somatically acquired mutations coupled with assessment of transcriptional activities of mitochondrial genes. Collectively, they confirmed many of the patterns described earlier. They found that MT-ND5 was the most frequently mutated ETC gene in most cancers, whereas MT-ND4 was most frequently mutated in prostate and lung cancers, and MT-COX1 was most frequently mutated in breast, cervical, and bladder cancers. Most mutations in tumor types were similar, with C:G>T:A in >50% of cases. mtDNA somatic mutations were acquired at an early age and shifted toward homoplasmy throughout life in the cellular lineage of the neoplastic cells. Whether the shift to homoplasmy was due to asymmetric segregations during cell division or by selection of the mutations remains unknown. Interestingly, in the subset of kidney and thyroid cancers with no identifiable oncogenic driver, they found non-silent mtDNA mutations, suggesting a potential contribution of these mutations in the absence of nuclear drivers. They also observed negative selection for truncating mutations of mtDNA-encoded proteins, suggesting the importance of an intact ETC for most cancer cells. Exceptions were found in kidney, colorectal, and thyroid cancers, possibly implicating inactivation of mitochondrial genes in tumorigenesis for some cell types.

Although the aforementioned analyses are informative, a key conclusion is that mitochondrial contributions to cancer vary by tissue of origin, cancer subtype, age, microenvironmental stress, and other so far unidentified variables. Therefore, it is instructive to examine several of the individual studies that contributed to the analysis by Yuan et al. as well as some newer data obtained since their publication.

NADH-ubiquinone oxidoreductase (complex I) is the largest complex of the ETC. It is essential for biosynthesis and redox control during proliferation, resistance to cell death, and metastasis [58–61]. Several inhibitors of complex I act as selective anticancer agents. Paradoxically, other studies show that complex I inhibition shows protumor effects [59]. Similarly, mutations in MT-ND1, that encodes a component of complex I, can enhance non-small cell lung cancer (NSCLC) metastasis [62]. Mito-mice carrying the G13997A mutation, which induces complex I, exhibited high lactate production but no observable phenotypes [63], suggesting that additional genetic or epigenetic cofactors are necessary to influence tumorigenicity and metastasis, consistent with our assertions that mtDNA SNPs are QTLs.

Oncogenes contribute to altered metabolism [64–66]. For example, mutant p53 with the P72R gain-of-function mutation regulates PGC1α owing to changes in binding and leads to poorer prognosis [67]. In addition, c-Myc can cause mtDNA fragmentation [68]. Knowing this, it is reasonable to ask: do oncogenes alter tumorigenesis in the context of mtDNA mutations or SNPs? Do the combinatorial effects of mtDNA QTLs and nDNA QTLs explain differential susceptibilities to cancer and/or metastasis? mtDNA mutations do not occur with equal frequency or location in all cancer types – prostate, liver, stomach, and colorectal cancers have the highest, whereas heme malignancies generally have the fewest [25]. We previously posed several questions that arose from these observations [69]. Are there selective pressures for specific mitochondrial mutations? Are there tissue-specific variations for mtDNA mutations in oncogenesis or metastasis? Are there ‘hotspots’ for mutation in cancer (progression)? Are there germline differences in mtDNA (i.e., SNPs) which could (partially) explain racial disparities in cancer and/or metastasis development? (see Outstanding questions). Tables 1 and 2 show changes in relative risk associated with mtDNA haplotypes.

Outstanding questions.

What selective pressures drive the acquisition of, or selection for, mtDNA mutations?

Are there ‘hotspots’ in mtDNA that are mutated in most cancers, in select cancers, or in embryologically related cancers?

Are somatic mtDNA mutations drivers, passengers, or hitchhikers?

Are there germline differences in mtDNA (i.e., SNPs) that predispose people to cancer and/or metastasis, or that can predict responses to therapy?

Do mutations or SNPs in mtDNA regulate subcellular distribution or the transfer of mitochondria or mtDNA between cells?

Do mutations in nDNA disrupt the production of mtRNA?

What are the signals to and from the mitochondria that mediate the phenotypic changes in cancer?

What is the molecular basis for interactions between the mitochondrial genome and the nuclear genome?

Table 1.

mtDNA mutations and polymorphisms in cancer^a

Cancer type	Gene	Posion	Cancer type	Gene	Posion
Bladder	MT-ND2	T4823C	NSCLC	MT-ND5	C12813A
Breast	MT-CYB	A15836C	NSCLC	MT-ND4	C11409T
Breast	MT-CYB	T14819insTTCTATA	NSCLC	MT-ND3	T10363C
Breast	MT-ND5	G13333A	NSCLC	MT-ND1	T3394C
Breast	MT-ATP6	T8821C	NSCLC	MT-ND1	C3497T
Breast	MT-CO3	A9664G	NSCLC	MT-ND1	C3689G
Colorectal	MT-ND6	T14470C	Ovary	D-loop	C16296T
Colorectal	MT-ND1	C3497T	Ovary	D-loop	C16294T
Colorectal	MT-ND1	T3394C	Ovary	D-loop	C16261T
Colorectal	NMT-D5	G12630A	Ovary	D-loop	A16163G
Colorectal	MT-TT	G15928A	Ovary	D-loop	A16162G
Colorectal	MT-CO1	C6371T	Ovary	MT-TW	T5567C
Colorectal	MT-ND5	T14138C	Ovary	D-loop	C16527T
Colorectal	MT-ND1	C3689G	Pancreas	MT-RNR2	G2946A
Colorectal	MT-TR	T10463C	Pancreas	MT-RNR2	G2145A
Colorectal	MT-ND1	G3955A	Pancreas	D-loop	G316A
Endometrial	MT-CYB	T15831C	Prostate	MT-ND3	A10398G
Endometrial	MT-TW	T5567C	Prostate	MT-TT	G15928A
Endometrial	MT-ND3	G10290A	Prostate	MT-TR	T10463C
Endometrial	MT-TR	T10463C	Prostate	MT-CO3	G9820A
Endometrial	MT-TP	G15995A	Prostate	MT-ND5	A13966G
Glioblastoma	D-loop	C16134T	Prostate	D-loop	G207A
Head & neck	MT-ND2	T4823C	Thyroid	Intergenic	A5581G
Leukemia	MT-ND1	T3394C	Thyroid	D-loop	G207A
Lung	MT-ND2	T4823C	Thyroid	MT-TR	T10463C
Lung	MT-ND1	T3394C	Thyroid	MT-CO1	T7389C
Melanoma	D-loop	T310C
Melanoma	D-loop	A302CC
Melanoma	D-loop	T16519C
Nasopharyngeal	D-loop	C16261T

^aShaded boxes indicate polymorphisms.

Table 2.

mtDNA haplotype predictions in cancer risk

Haplotype	Ancestry/race	Cancer tissue of origin	Risk change
B2	American	Cervix	Increase
D	Asian	Endometrium	Increase
D	South American	Cervix	Increase
D	Asian	NSCLC	Reduced survival
D	Asian	Lung	Decrease
D	Asian	Esophagus	Increase
D4	Asian	Thyroid	Increase
D5	Asian	Thyroid	Increase
D5	Asian	Breast	Increase
H	European	Breast	Increase
I	European	Breast	Increase
J	Middle East	Uveal	Increase
K	European	Pancreas	Decrease
K	European	Breast	Increase
K	European	Thyroid	Decrease
L0	African	Prostate	Increased severity
M	Asian	Oral	Increase
M	Asian	Cervix	Increase
M	Asian	Stomach	Increase
M	Asian	Breast	Decrease
M7	Asian	Lung	Increase
M7	Asian	Liver	Decrease
N	European	Prostate	Decrease
N	Asian	Thyroid	Increase
N	Asian	Stomach	Longer survival
N	Asian	Breast	Increase
N	Asian	Prostate	Decrease
N	Asian	Breast	Increase
T	Multiple	Colorectal	Increase
U	European	Prostate	Increase
U	European	Breast	Decrease
U	European	Kidney	Increase
X	European	Breast	Increase

At the outset it is important to emphasize that findings regarding linkage of mtDNA SNPs and disparities are mostly anecdotal. Note, however, that we have shown that mtDNA-derived short RNA transcripts containing no SNPs are associated with ancestry or sex disparities in prostate [70], bladder [71], lung [71], kidney [71], and triple-negative breast cancer [72], and with survival in uveal melanoma patients [73]. A point worth emphasizing is that not all SNPs are within protein-coding mtDNA genes. Unfortunately, although examples exist, the literature exploring the other regions of the mitochondrial genome has not been as extensively studied so far.

Definitive cause–effect relationships are incomplete. Nonetheless, there are some intriguing observations and patterns that certainly warrant further investigation. An example includes SNPs found in the J haplotype which alter uveal cancer metastasis as well as mitochondrial biogenesis [13]. In addition, haplogroup N contributes to good survival in gastric carcinoma patients compared to haplogroup M, suggesting that SNPs in the displacement loop (D-loop) can be used as predictors of disease outcome [74]. Mutations in the D-loop are associated with age of onset, relapse rate, and metastasis for fibrous histiocytoma [75]. Polymorphisms in head and neck squamous cell carcinoma (HNSCC) are correlated with the mitochondrial unfolded protein response [76]. In a cohort of hepatocellular carcinomas, Li and colleagues [77] reported a large number of sequence variants in 140 cases, most of which were SNPs. The presence of these variants identified independent predictive factors in the D-loop for time to recurrence, tumor-free survival time, and overall survival. In another study, mitochondrial genomic variation in prostate cancers was greater at metastatic sites than in the primary tumor [78]. Relatedly, polymorphisms in the control region have been associated with melanoma etiology, pathogenesis, and progression [79].

Using cybrids, transfer of mtDNA SNPs found in breast cancer cells from African-American women slowed proliferation, increased complex I activity and ROS production, depolarized the mitochondria corresponding to apoptosis resistance, increased anchorage-independent growth, and increased metastases [80]. Likewise, mitochondria from an aggressive osteosarcoma cell line, a moderately aggressive breast carcinoma cell line, or an immortalized, but otherwise normal, breast epithelial cell demonstrated that mitochondria from the benign cells could reverse the cell proliferation, responses to hypoxia, apoptotic sensitivity, invasiveness, anchorage-independent growth, and tumorigenicity when injected into athymic mice [81]. In addition to the insights regarding tumor cell behavior, these findings suggest that mitochondrial gene therapy may be a viable option once deeper understanding is gained. Collectively, the results show that the mtDNA genome is not mutated randomly in metastases, suggesting that there were selective pressures for specific mutations associated with bone metastasis. Equally importantly, many of the SNPs are not in the protein-coding genes of the mtDNA. Both points will impinge upon future attempts to repair or restore normal phenotypes in cancer cells.

Because the presence of ROS is often associated with DNA damage, Blein et al. [82] and Ju et al. [25] measured mtDNA mutations in the presence or absence of DNA repair machinery. They reported that BRCA1/2 mutation carriers from the T haplogroup had an increased risk of developing breast cancer. By contrast, Ju et al. found no correlations between oncogenic drivers that impact on nDNA mutation rates and mtDNA mutation rates. Of 2453 colorectal cancers and 11 930 controls, the most significant mtDNA SNP was A4917G in the T haplogroup of American men and women of Asian, African, European, Latino, or Native Hawaiian ancestry [83]. This SNP was associated with an increased risk of developing colorectal cancer. Interestingly, European-Americans with G4655A were associated with higher colorectal cancer risk, but the effect was not observed throughout the population. Perhaps this SNP adds risk depending upon the nuclear factors with which it interacts.

Somatic mitochondrial mutations in cancer progression

Accumulating evidence suggests that mtDNA is correlated with cancer. Although many mitochondrial transcriptomic differences have been linked to disparities and survival, few driver mutations have been found. Some somatic mutations have been associated with specific cancer types (and subtypes) in both protein-coding and noncoding regions of the mitochondrial genome (reviewed in [84–87]). Of note, some studies report a higher mutation frequency in the D-loop ([24,88–90] and Table 1), but the underlying role(s) have not yet been determined. Another common deletion occurs in a stretch of cytosines (D310) that is variable. Similarly, correlations of mtDNA mutations or SNPs with metastasis have been observed, but ascribing a mechanism to those changes has not been possible to date. This situation is also true in the context of metastasis, but some mtDNA changes associated with oncogenesis will also likely provide insights into the ontogeny and/or the severity of metastasis.

In clear cell renal carcinoma, Zhang et al. observed a generalized decrease of complex I despite relatively few mutations in complex I proteins [58]. The expression patterns appeared to correlate with development of metastasis and immune response. Zhan and colleagues observed a novel tRNA fragment in the serum of hepatocellular carcinoma patients that had predictive value [91]. In animal models of liver cancer, c-Myc-driven mtDNA fragmentation was measured as a possible driver of tumor formation. Knockout of the mitochondrial fusion molecule MFN1 or overexpression of mitochondrial fission molecule DRP1 promoted liver cancer development [68]. The opposite effect was observed when MFN1 was overexpressed.

Mitochondria are closely associated with the actin cytoskeleton and are highly mobile within cells. Recent elegant studies by Caino and colleagues demonstrate that mitochondria mobilization in cancer cells co-opts a neuronal mitochondrial traffic machinery involving syntaphilin (SNPH), kinesin KIF5B, and GTPase Miro1/2 to localize mitochondria to the cortical cytoskeleton [92,93]. Overexpression of SNPH suppressed the kinetics and distance traveled by mitochondria, with a corresponding reduction in chemotaxis and metastasis in vivo. More extensive analyses reveal that mitochondrial subcellular localization is dysregulated in multiple cancer cells [94,95]. However, another question arises: are there differences in mtDNA subcellular distribution in different primary and metastatic lesions of different cancers, or between different sites of metastasis?

Mitochondrial genetics and metastasis

Metastasis is a highly inefficient process that is responsible for the majority of cancer-related morbidity and mortality, and is a culmination of cellular evolution from transformation to the neoplastic state [96,97]. Both pro- and antimetastatic genes have been discovered, but there are many pathways to achieve metastatic colonization [97]. Reviewing the literature reveals several suggestive links between mitochondria and metastasis, but definitive roles still require experimental testing. A common conclusion is that mitochondrial genetic changes are phenotypic ‘modifiers’ rather than drivers per se.

Mitochondrial haplogroups have been implicated in multiple hallmarks of metastasis [44]. For example, native American haplogroups in Mexican women with breast cancer showed no statistical difference between haplogroups A–D, and L according to tumor subtype [42] whereas in-depth analysis uncovered many important results. Single-nucleotide variants comprised 98% of somatic mutations; 59% of total mutations were in coding regions and 41% were in noncoding regions. The D-loop had the most mutations, followed by CO1, 16S, 12S, and ND5. Somatic mutations were homoplasmic (21.4%), but most were heteroplasmic.

Invasive breast cancers in African-American women participating in the University of North Carolina Breast Cancer Study had increased risk with G10398A, which had previously been linked to neurodegenerative diseases [98]. Curiously, Caucasian women harboring the same allele showed no difference in relative risk. These findings affirm the cooperativity between genomes as QTLs. It is interesting to speculate, based upon the higher frequency of developing triple-negative breast cancer in African-American women, that part of these differences could be due to disruption of estrogen-based regulation of mitochondrial function [99,100]. However, there are no direct data that establish such a link.

Most data aforementioned describe mutations in mtDNA coding regions. Webb et al. reported no correlation between overall colorectal cancer risk and mitochondrial haplogroup [101]. They did, however, find that A5657G in the non-coding region between mt-tRNA^Ala and mt-tRNA^Asp correlated with colorectal (compared to rectal) cancers. A synonymous change in MT-ND2 (T4562C) was highly associated with microsatellite instability of colorectal cancers [101].

Hunter and colleagues studied genetic background as a metastasis efficiency-modifier loci by crossing FVB/NJ-TgN (MMTVPyMT)^634nul mice (‘PyMT’) to dams from different mouse strains. F₁ progeny had varied metastatic efficacy [102–105]. They have since mapped several metastasis efficiency-modifiers in breast and prostate cancers [102,104,106,107]. However, their experimental design allowed us to posit an alternative interpretation based upon maternal inheritance [108]. We developed a genetically engineered mouse model by transferring pronuclei from one mouse strain into the enucleated embryonic cytoplast of another strain [109,110]. Following transfer into pseudopregnant nuclear-matched mice, pups were designated mitochondrial-nuclear exchange (MNX) mice. We replicated a scaled-down version of the Hunter experiment. Changes in tumor incidence, latency, and metastasis were nearly indistinguishable from those using wild-type mice [108,111,112]. Comparable changes were observed using HER2/neu as an oncogenic driver [111]. MNX mice also exhibited differences in spontaneous tumor formation even in the absence of an oncogenic driver [113] in addition to differences in atherosclerosis, adiposity, and diabetes [110,114–117].

Recognizing that genetic crosses also change stromal mtDNA, we asked whether syngeneic (i.e., histocompatible) mammary and melanoma tumor cells behaved differently when injected into MNX mice. Whenever host stroma contained C57BL/6J mtDNA, metastasis was inhibited; whenever stroma had C3H/HeN mtDNA, metastasis was promoted [118]. In efforts to understand the underlying mechanisms, we determined that MNX mice selectively alter nuclear DNA methylation and histone marks ([112]; X. McGuire and D.R.W., unpublished). C3H/HeN mtDNA consistently produces higher levels of ROS than C57BL/6J mtDNA. Metastasis increased with higher ROS whereas scavenging ROS decreased metastases [111]. Baseline immune profiles differed in MNX mice compared to nDNA-matched counterparts, and the polarization states of myeloid populations infiltrating lung metastases were also different [119]. Principal component analyses of >5000 metabolites in wild-type and MNX mice clustered slightly differently but were largely similar (D.R.W., unpublished). Specific metabolites corresponding to changing metastatic potential have not yet been identified, consistent with previous reports [110,111,120–123]. Because mitochondria evolved from ancient bacteria [124], and because bacterial ecosystems form when microbes communicate with each other [125–127], we reasoned that mitochondria could communicate with some bacteria, thereby selectively promoting/inhibiting bacterial growth and fecal microbiome change. Deep sequencing of 16S RNA identified selective changes (only 8–17 bacterial species) (D.R.W. et al. unpublished).

MNX mouse findings mice support pioneering work from Ishikawa, who utilized cybrids to demonstrate that mitochondrial transfer could change metastasis [61,128,129]. Mutations disrupting complex I (e.g., an insertion into the MT-ND6 gene, 13885insC) elevated ROS and metastatic propensity [61,62]. mtDNA mutation also increased the transcription of glycolysis- and metastasis-related genes [62]. The functional role of complex I mutations in MT-ND6 (C12084T) and MT-ND5 (A13966G) was later linked to metastasis [129]. Mutant MT-ND6 increased invasion in A549 lung cancer cells [130]. Multiple mutations in NADH dehydrogenase genes (T3398C, T12338C, C3689G, G3709A, G3955A, T10363C, C11409T, G13103A, and T14138CC in MT-ND1; G12813A, G13366A, and 14504delA premature truncations in MT-ND5 or MT-ND6) were associated with distant metastasis [62,131,132]. Two SNPs were found in MT-ND1 (C3497T and T3394C), consistent with the notion of ancestral disparities in cancer aggressiveness. Paralleling our observations that mtDNA SNPs in stroma affect metastasis in MNX mice, T3394C mutations in adjacent mucosa in NSCLC and colon tumors [62] support the concept that there are inherited susceptibilities to metastasis. Comparison of non-invasive versus invasive breast cancer cell lines observed shifts from OXPHOS to glycolysis together with increased heteroplasmy [133]. However, such patterns are not universally reported in all cancer types [46,134]. Other reports identified antioxidants (e.g., mitochondrial catalase, mtCAT) that affect metastasis [135,136]. mtCAT reduced macrophage infiltration and CD34⁺ endothelial cells, the latter suggesting reduced angiogenesis.

The strongest evidence for germline metastasis efficiency loci in mtDNA comes from experimental models, most commonly mice. Although powerful, mouse models have weaknesses (reviewed by Bussard and Siracusa [137]). Indeed, extrapolating murine data to humans is limited by incomplete tools and the limited number of haplogroups, making it difficult to separate nuclear and mitochondrial contributions. However, recent publications implicate mitochondrial haplotypes as modifiers of hepatocellular carcinoma [138] and neuroblastoma [139].

Most examples focus on mtDNA changes in the cancer cells themselves. Understanding the involvement of mtDNA in the stroma is complicated by variances in tissue, gender, age, and stress. The experiments cited below are individually well controlled, but current technology and experimental costs make it impractical to address all these variables.

Immune activation and mitochondria functions are inextricably linked. Mitochondria regulate innate immune system activation through recognition of cellular damage and activation of the NLRP3 inflammasome [140,141], and formylated peptides in bacteria can also activate specific receptors [142]. In addition, ROS production can indirectly or directly regulate the immune system [143,144]. Cancer-associated fibroblasts (CAFs) can produce pyruvate, lactate, ketone bodies, and fatty acids using glycolysis that can be utilized by cancer cells [145–151]. These same metabolites also alter immune cell functions [152] and polarization [153]. As the primary consumers of oxygen in most tissues, mitochondrial differences in oxygenation regulate metabolism [154]. Mitochondria-derived ROS can induce hypoxia-inducible factor α (HIF-1α) which in turn alters cellular responses [155]. Thus, hypoxia will have overarching effects on many cells within the tumor microenvironment, which may be further impacted by differences in mtDNA. Because hypoxia increases metastasis [156–160], the capacity of mitochondria to respond may be intrinsic to connecting the two phenotypes.

Concluding remarks and future perspectives

Despite limited models that isolate mitochondrial SNPs and mitochondrial genetics as experimental variables, abundant data support the notion that mtDNA QTLs exist for multiple diseases including cancer. It is unfortunate that most investigators default to ignoring the mitochondrial genome because of its relatively small size, and others focus solely on the protein-coding regions of the mitochondrial genome. We remind readers that small size does not always mean little impact and that ‘junk’ DNA is now well recognized as a crucial regulator of cellular function. The influence of mitochondria on cancer development, progression, and disease severity occurs because of both intrinsic and extrinsic effects. As conveyers of signals from the microenvironment to the nucleus and vice versa, mitochondrial SNPs can alter signaling in context-dependent ways. What are the signals? Are the same molecules used to signal multiple locations? With what are those molecules interacting? How do nuclear genes and the proteins (and non-coding RNA) derived from them modify the mtDNA?

Where should we go from here? Hopefully, this review overcomes the misconception that the mitochondrial genome is (relatively) unimportant. Throughout this review we have attempted to introduce questions and topics for future research. Some of the bigger issues are summarized in the Outstanding questions. The most important recommendation is to increase the incorporation of mtDNA analyses in cancer genomic analyses. When doing so, pairing normal, primary tumor, and metastatic tissues will be crucial, especially from different metastatic sites.

Because germline differences in mtDNA perhaps predict tumor behavior, and because ostensibly normal tissues adjacent to diseased tissues are not truly normal [161], there is immense potential that mitochondrial information within serum, plasma, or non-tumor tissues will be predictive or prognostic, even before oncogenic mutations appear or before tumors arise [73]. Similar findings have been described for other human conditions [162]. The relatively small size of the mitochondrial genome could, in fact, be a benefit because deep sequencing would be comparatively fast and inexpensive compared to whole-genome or even whole-exome sequencing.

Another crucial step will be to develop databases that capture mitochondrial genomic properties fully and effectively allow comparisons of non-human with human mtDNA. Furthermore, the development of technologies that allow systematic manipulation of the mitochondrial genome without introducing heteroplasmy [163] will accelerate the study of mitochondrial genomics. The field of mitochondrial cancer genetics is still in its relative infancy. Nonetheless, the foundations have been laid. New and emerging models hold great promise for addressing these questions and initiating the pathway to translating the foundational questions into clinical practice.

Table I.

Organization of human and murine mitochondrial genomes

Human						Murine
Genea	ETCb	Start	End	Length (bp)	Strand (H or L)	Start	End	Length (bp)	Strand (H or L)
MT-TF	–	577	647	70	H	1	68	67	H
MT-RNR1	–	648	1601	953	H	70	1024	954	H
MT-TV	–	1602	1670	68	H	1025	1093	68	H
MT-RNR2	–	1671	3229	1558	H	1094	2675	1581	H
MT-TL1	–	3230	3304	74	H	2676	2750	74	H
MT-ND1	I	3307	4262	955	H	2760	3704	944	H
MT-TI	–	4263	4331	68	H	3706	3774	68	H
MT-TQ	–	4329	4400	71	L	3772	3842	70	L
MT-TM	–	4402	4469	67	H	3845	3913	68	H
MT-ND2	I	4470	5511	1041	H	3914	4948	1034	H
MT-TW	–	5512	5579	67	H	4950	5016	66	H
MT-TA	–	5587	5655	68	L	5018	5086	68	L
MT-TN	–	5657	5729	72	L	5089	5159	70	L
MT-TC	–	5761	5826	65	L	5193	5257	64	L
MT-TY	–	5826	5891	65	L	5260	5326	66	L
MT-CO1	III/IV	5904	7445	1541	H	5328	6872	1544	H
MT-TS1	–	7446	7514	68	L	6869	6939	70	L
MT-TD	–	7518	7585	67	H	6942	7011	69	H
MT-CO2	IV	7586	8269	683	H	7013	7696	683	H
MT-TK	-	8295	8364	69	H	7700	7764	64	H
MT-ATP8	V	8366	8572	206	H	7766	7969	203	H
MT-ATP6	V	8527	9207	680	H	7927	8607	680	H
MT-CO3	IV	9207	9990	783	H	8606	9389	783	H
MT-TG	-	9991	10058	67	H	9391	9458	67	H
MT-ND3	I	10059	10404	345	H	9457	9803	346	H
MT-TR	–	10405	10469	64	H	9805	9872	67	H
MT-ND4L	I	10470	10766	296	H	9874	10167	293	H
MT-ND4	I	10760	12137	1377	H	10161	11537	1376	H
MT-TH	–	12138	12206	68	H	11539	11606	67	H
MT-TS2	–	12207	12265	58	H	11607	11665	58	H
MT-TL2	–	12266	12336	70	H	11665	11735	70	H
MT-ND5	I	12337	14148	1811	H	11736	13559	1823	H
MT-ND6	I	14149	14673	524	L	13546	14064	518	L
MT-TE	–	14674	14742	68	L	14065	14133	68	L
MT-CYB	III	14747	15887	1140	H	14139	15281	1142	H
MT-TT	–	15888	15953	65	H	15283	15449	166	H
MT-TP	–	15956	16023	67	L	15350	15416	66	L
D-loop	–	16024	576	1121		15417	16295	878

^aGene names are preceded by ‘MT’ in accordance with Human Genome Organization (HUGO) Gene Nomenclature Committee recommendations.

^bIndicates whether the designated gene encodes a component of electron transport chain (ETC) complexes I–IV; –, does not encode an ETC component.

Highlights.

Mitochondrial DNA (mtDNA) harbors cancer/metastasis quantitative trait loci.

Both somatic mutations and germline mtDNA polymorphisms are associated with cancer development and metastasis in tissue-specific and cancer subtype-specific manners.

Mitochondria alter nuclear epigenomes.

Not all mitochondrial effects on cancer/metastasis are metabolism (i.e., electron transport)-related.

Changes in stromal mitochondria influence neoplastic behavior.

Glossary

Cybrid

a fused cell containing a parental nuclear genome and the mitochondrial genome of another parent (or a combination of mitochondrial genomes)

Haplotype

alleles inherited from a single parent

Haplogroup

a clade in which a unique polymorphism is represented

Heteroplasmy/heteroplasmic

a eukaryotic cell in which mitochondrial DNA (mtDNA) is non-identical

Homoplasmy/homoplasmic

a eukaryotic cell in which all copies of mtDNA are identical

Quantitative trait loci (QTLs)

genes with alleles that influence the expression of a phenotypic trait

Metastasis

the spread of neoplastic cells to discontiguous sites where neoplastic cells proliferate to become macroscopic lesions

Metastatic cascade

the series of steps used by cancer cells undergoing the process of metastasis

Tunneling nanotubes (TNTs)

cellular protrusions that enable cells to touch over long distances