Mitochondrial DNA abnormalities provide mechanistic insight and predict reactive oxygen species-stimulating drug efficacy

Hallmarks of cancer include genome instability and alterations in metabolism, for the latter of which mitochondria are key regulators [1, 2]. Mitochondria are semiautonomous eukaryotic organelles that are considered as cellular energy powerhouses. They consist of thousands of proteins, mainly encoded by the nuclear genome, but also 13 encoded by their own genome [3, 4]. Mitochondrial DNA (mtDNA) is a small circular DNA of 16,569 bp, which codes for 2 rRNAs (12S and 16S), 22 tRNAs, and protein subunits for four of the five complexes of the oxidative phosphorylation system (OXPHOS), namely seven subunits of NADH dehydrogenase (complex I), one of cytochrome c reductase (complex III), three of cytochrome c oxidase (complex IV) and two of ATP synthase (complex V). The four subunits of respiratory enzyme complex II are totally encoded by nuclear DNA (nDNA), and are transcribed and translated in the cytoplasm and then transported into the mitochondria [5]. The displacement-loop region (D-loop) is a long noncoding region of mtDNA approximately 1.1 kbp long and contains regulatory sequences for replication and transcription. A number of proteins are implicated in the replication and transcription of mtDNA, and mitochondrial transcriptional factor A (TFAM) plays dual roles in these biological processes through its binding to D-loop [5]. The number of mitochondria that exist in a human cell ranges from hundreds to thousands, and each mitochondrion contains 2–10 mtDNA copies [6]. The synthesis and degradation of mtDNA is rapid and independent of the cell cycle so the dynamic equilibrium between the two processes determines the mitochondrial DNA copy number (mtDNAcn) (reviewed in Ref. [7]). MtDNAcn is strictly regulated during differentiation, and normal healthy cells with a high demand for ATP generation via OXPHOS have higher mtDNAcn than cells with a low ATP requirement [8].

The OXPHOS system is implicated in many cellular processes such as cell communication, differentiation and apoptosis. Consequently, OXPHOS dysfunction can lead to the development of several human diseases, including cancers [9]. For example, mutations in nuclear genes that encode the succinate dehydrogenase (SDH, MRC complex II) subunits can alter corresponding subunit structure and consequently affect MRC complex II activity, substrate levels, reactive oxygen species (ROS) production and cell redox state, which in turn activates oncogenic transcription factors to regulate expression of target genes and stimulate cancer cell proliferation [10, 11].

To date, mitochondrial dysfunction and various mtDNA alterations including point mutations, deletions and copy number changes have been observed in several human cancers such as breast, colon, stomach, kidney, thyroid, head and neck and ovarian cancers [12‐14], and some linked with clinical parameters [15].

A total of 635 mutations have been reported in the D-loop region, including 510 point mutations, 56 deletions and 69 insertions. Additional mutations have been identified in the coding regions: 593 mutations in the complex I subunits, 343 mutations in the complex III, IV and V subunits, and 165 mutations in the tRNA and rRNA genes [16]. An increased mtDNAcn has been observed in endometrial [17], untreated head and neck [18], prostate [18], pancreatic [19] and colorectal [20] cancers compared to normal cells. On the contrary, a decrease of mtDNAcn has been associated with renal [21], thyroid [22], breast [23], previously treated head and neck [14], hepatic [24] and ovarian cancers compared to normal cells [25]. While mtDNA alterations such as mutations and changes in mtDNAcn are commonplace in cancers compared to normal cells, much more needs to be learnt about how they interrelate with each other and how they associate with underlying biological processes, including ROS production and ROS-mediated drug response.

Mitochondria are considered a primary intracellular site of ROS generation via OXPHOS during ATP generation [26]. ROS play important roles in cell signalling pathways such as growth, differentiation, metabolism and apoptotic cell death signalling (reviewed in Ref. [27]). ROS have been purported to have a double-edged-sword effect in cancer cells since low levels of ROS can play a critical role in promoting cell proliferation and invasion, whereas excessive levels of ROS can cause oxidative damage to intracellular bio-macromolecules and consequently induce cell death (reviewed in Ref. [28]). This makes cancer cells vulnerable to agents that further increase ROS levels, e.g. cisplatin (CDDP) that has been reported to cause excessive accumulation of ROS resulting in cytotoxic effect in various types of cancer such as ovarian cancer [29], breast cancer [30] and prostate cancer [31, 32]. The mitochondria-targeting compound, dequalinium chloride (DQA), has also been reported to demonstrate potent anticancer activity in vitro and in vivo in different malignancies due to the resulting damage to mtDNA and the inhibition of mitochondrial complex I rendering elevated ROS generation [33‐35]. DQA belongs to a group of chemical agents knows as delocalised lipophilic cations (DLCs). Positively charged DLCs are drawn into mitochondria in response to the negative electric potential across the mitochondrial membrane. More importantly, the difference in mitochondrial transmembrane potential between normal and cancer cells can cause a ten-fold increase in DLC accumulation in cancer cells compared to their normal counterparts (reviewed in Ref. [36]), making the targeted mitochondrial accumulation of DLCs an attractive approach to cancer-specific therapy. Our previous study confirmed observations by other groups that DLCs such as DQA preferentially target mitochondria in cancer cells [37]. Furthermore, DQA may afford enhanced drug efficacy with reduced side effects due to its cancer-specific targeting property compared to non-cancer-specific / nucleus-targeting compounds such as CDDP, even though both DQA and CDDP induce ROS production [37].

The main objectives of our ongoing research are to prove the therapeutic potential of targeting mitochondria via DLCs in treating cancer and to identify effective biomarkers (such as ROS and specific mtDNA variations) that can predict drug response. Our previous study revealed a positive correlation between baseline intracellular ROS levels and drug response in human cancer cell lines. Therefore, baseline ROS levels have been proposed as a novel biomarker to predict drug sensitivity [37]. However, since measuring ROS in tissues is technically challenging, a more straightforward, e.g. genetic, biomarker to indicate the response to ROS-stimulating agents was sought.

Due to the complexity of the pathways for cancer development and resistance, which involves the functional cross-talk of mtDNA genes and their related nuclear genes, it is challenging to prove direct links between mtDNA variations and phenotype, one reason being the absence of routine laboratory methods for the genetic manipulation of mammalian mtDNA. However, one way of proving a link is to demonstrate a corresponding mitochondrial/cellular effect in cells/tissues that contain the mtDNA variation. However, more than one nuclear or mitochondrial variation may be present, potentially confounding the interpretation. Recently, we have demonstrated that 3D structural analysis of mitochondrial proteins in silico can complement the former approach revealing detailed mechanistic insights into the functional role of variations within mtDNA encoded-OXPHOS genes [10].

Consequently, the present study aimed to explore the interrelationship between various mitochondrial genetic abnormalities (mtDNA variations and mtDNAcn change), baseline intracellular ROS levels and sensitivity to ROS-stimulating drugs such as CDDP and DQA within the same panel of cell lines employed in our previous study [37]. Here we show that mtDNAcn and the total number of non-synonymous variations within the mitochondrial genome have a strong correlation with both baseline ROS level and drug resistance. 3D structural modelling of these reveals a subset of variations that could be responsible for the observed mtDNA genotype-mitochondrial/cellular phenotype links. While repeating these experiments with the different mtDNAs on a standardised nuclear genetic background is warranted, we propose that this is a useful first step in the characterisation of the cancer cell lines, and reveals a subset of variations that have the potential to be used to stratify patient into subgroups that are most likely to benefit from ROS-stimulating agents in the future.

It has become broadly acknowledged that the redox status of cancer cells could be manipulated to achieve cancer-specific killing since high levels of intracellular ROS activate death signal pathways (reviewed in Ref. [27]). Given that mitochondria are a primary intracellular site of ROS production via OXPHOS during ATP generation [26], it is not surprising that mitochondria have attracted increasing research interest as a promising target for anticancer therapy. It has been reported that mtDNA mutations could influence intracellular ROS levels [45, 46]. In particular, since complexes I and III are the main ROS generation sites, variations in their mtDNA genes could have a significant impact on the overall intracellular ROS level (reviewed in Ref. [47]). However, how specific mtDNA variations influence the efficacy of mitochondria-targeting/ROS-stimulating therapy remains unclear. Our previous data have shown that elevated baseline intracellular ROS levels among different cancer cell lines correlate positively with increased resistance towards cisplatin and dequalinium, both being ROS-stimulating drugs [37]. We hypothesised that certain mitochondrial genetic abnormalities, including variations and copy number (mtDNAcn) changes, could influence intracellular ROS levels, and therefore be utilised as biomarkers to predict cancer cells’ response to ROS-stimulating agents. The present study characterised the interrelationship between mitochondrial genetic parameters (variation and mtDNAcn), ROS levels and ROS-mediated drug response with the aim of generating future novel biomarkers that have a significant association with drug response of cancer cells. Such biomarkers have the potential to be integrated into future clinical practice in order to facilitate personalised medicine.

As a first step to investigating the possible link between mitochondrial genetic abnormalities and the cells’ response to ROS-stimulating agents, a range of molecular, biochemical and bioinformatics methods were employed in the present study to analyse ROS levels, cytotoxicity of cisplatin and dequalinium, as well as mtDNA variations and relative copy numbers in 4 cancer and 1 non-cancerous cell lines (two of the cell lines, cancerous PC-3 and non-cancerous PNT-2, originated from the same tissue type, i.e. prostate).

We found that cancer cells generally carry more variations and a higher mtDNAcn in their mitochondrial genome than non-cancerous cells and have higher baseline ROS levels. This is consistent with the observations of others [47‐49] and likely reflects the cancer cells’ adaption to oxidative stress via enhanced DNA repair mechanisms, upregulated anti-apoptotic pathways and consequently increased drug resistance/cell survival (reviewed in Ref. [28]).

We also confirmed our previous observation that increasing baseline ROS level correlates positively with increasing resistance to the ROS-stimulating agents in cancer cell lines. We further demonstrated that increased ROS and drug resistance levels also correlate positively with greater mtDNAcn, higher number of non-synonymous complex I and III variations, and higher number of complex I and III functional variations predicted by the 3D structural modelling (i.e. A10398G, T11120C, C12084T, A13681G, G13708A, C13802T, A13966G and T14798C) (Table S4). The positive correlation between the number of predicted complex I/III functional variations and ROS levels appears more significant for mitochondrial ROS than for overall intracellular ROS. Since complexes I and III are the main contributors to mitochondrial ROS generation, it is not surprising that cancer cells carrying larger numbers of functional variations located in these two complexes would produce higher levels of mitochondrial ROS, as shown in this study.

Intriguingly, the positive correlation between the number of predicted complex I/III functional variations and drug IC50 seems more marked for cisplatin than for dequalinium. This suggests that cisplatin, a conventional nDNA-targeting compound, may rely more on elevated mitochondrial ROS to promote drug resistance in cancer cells compared to dequalinium, a mitochondria-targeting compound. Our observation was echoed by a recent study in which cisplatin-resistant lung cancer cells demonstrated increased mitochondrial mass through upregulation of PGC-1α (the predominant mitochondrial biogenesis promoter) and consequently enhanced intracellular ROS production upon cisplatin treatment [50]. Our data support the previously proposed theory that cancer cells may develop resistance towards cisplatin by upregulating mitochondrial ROS production in order to stimulate anti-apoptotic pathways and promote cell survival [50]. We speculate that this mechanism could be triggered by certain nucleus-mitochondria crosstalk in response to cisplatin treatment. On the contrary, unlike cisplatin, dequalinium is a mitochondria-targeting compound and therefore may not initiate such crosstalk, at least not to the same degree, upon entering the cells. This may explain the much lower IC50s of dequalinium observed in the cancer cells compared to the IC50s of cisplatin in the present study. Furthermore, our data and the previously published data support our hypothesis that mitochondria-targeting therapy can be more effective than conventional therapy since the latter may trigger nucleus-mitochondria crosstalk to promote cell survival.

Using 3D structural modelling, we predicted 8 functional OXPHOS variations: A10398G, T11120C, C12084T, A13681G, G13708A, C13802T, and A13966G are likely to destabilise complex I, impacting on corresponding enzyme activity, which in turn may affect ROS levels and drug response. In support of this view, A10398G was reported in various types of cancer including brain, breast and cervical cancer, and strongly associated with elevated levels of ROS [51‐54]; T11120C and C13802T was found in the PC3 cells (but not the PNT-2 cells) previously and shown to inhibit OXPHOS and increase ROS production, and consequently promote tumour growth in vivo when those PC-3 cells were injected into nude mice [55]; C12084T and A13966G variations found in the MDA-MB-231 cells were previously reported to be responsible for reduced complex I activity and enhanced metastatic potential via ROS overproduction and the resultant overexpression of Hif-1α [56, 57]. While G13708A in complex I has been reported in various types of cancer such as breast and colorectal cancer [58, 59], A13681G has no previously reported disease associations. There are no previous reports documenting G13708A or A13681G and ROS levels, although our predictions suggest this would be worth investigating. On the other hand, we predict that the final functional OXPHOS mutation: T14798C, despite frequently occurring in the healthy population (like some of the other variations; Table S3), is likely to affect the association/dissociation of ubiquinone at the Qi-site, impacting on the activity of complex III, and this in turn may affect the levels of ROS produced by the complex and consequently drug response. In support of this view, a study conducted by Keatley et al. demonstrated that glioblastoma cell lines that carried T14798C had elevated complex III activity, oxidative stress, and different drug sensitivity levels compared to cells that did not contain the mutation. Further, glioblastoma patients with T14798C had worse prognosis than non-carriers [10]. This is in keeping with an earlier study which demonstrated that the activity of yeast mutant complex III genetically manipulated to contain the equivalent variation of T14798C was significantly impaired and the mutant showed more sensitivity to mitochondrial-targeting compounds [60].

In contrast to increased ROS and drug resistance levels correlating positively with greater mtDNAcn, larger number of non-synonymous complex I and III variations, and larger number of predicted complex I and III functional variations, none of these parameters correlated with the number of D-loop variations. The mtDNAcn result is surprising given that the D-loop region is a major control site for mtDNA replication and transcription, and higher mtDNAcn has been reported in various types of cancer carrying D-loop variations [61, 62]. One possibility is that specific variations located within the light and heavy strand promoters in the D-loop region rather than the total number of D-loop variations are more important. Such variations could affect the binding affinities of the initiators and modulators of mtDNA transcription, thus, disturbing the rate of transcription, RNA primer synthesis as well as mtDNA replication at the H-strand origin of replication (O_H) [63]. In other words, quality is more important than quantity in terms of functional variations in the D-loop region. In the present study, only one variation appeared to be located in a key regulatory region. Indeed, 310InsC, present in Ishikawa, MDA-MB-231 and PC-3 cells, is located in CSB2, one of the three conserved sequence blocks within the H-strand origin of replication (O_H) that harbours critical functional motifs for the initiation and regulation of H-strand replication. Given that insertion variations have a real fundamental impact on DNA function, as they completely change the sequence of the region and result in the shift of the original functional motifs, it is likely that the 310InsC variation may interfere with DNA-primer interaction and have a negative impact on the replication of the H-strand in those cell lines. This observation might explain why the above cell lines had lower mtDNA copy numbers compared to Caco-2 that did not harbour this insertion.

The apparent relationship between mtDNAcn, ROS levels and drug IC50s could also represent cellular adaptation to the level of OXPHOS activity that in turn might be related to the number of functional OXPHOS variations. One example of such adaptation includes the retrograde signalling pathway, where dysfunctional mitochondria (caused by functional OXPHOS variations) can lead to differences in ROS levels, which in turn lead to the activation of various nuclear responses, consequently promoting multiple pathways that regulate energy homeostasis, oxidative stress, mitophagy, fission, fusion and other functions to facilitate cellular adaption strategies and hence regulate the transcription and translation of genes responsible for mitochondrial biogenesis (e.g. PGC-1α), mtDNA replication and OXPHOS functions [64]. Taking these observations together, we propose that cell line-specific functional OXPHOS variation profiles could mechanistically contribute to the mitochondrial activity, baseline ROS levels, mtDNAcn and response to ROS-stimulating agents observed in cancer (Fig. 8), and the best illustration of this from our study is the comparison between the cancerous PC-3 cells and non-cancerous PNT-2 cells, as they are both derived from prostate.

A recent publication by Cocetta and co-authors provides a sophisticated review on mitochondrial involvement in cisplatin resistance [65]. As pointed out by the authors, so far, existing knowledge of the retrograde signalling is still very limited. Based on our data and the findings from a previous study linking PGC-1α upregulation with increased mitochondrial mass in cisplatin-resistant lung cancer cells [50], we present a hypothesis illustrated below in Fig. 8. Our data clearly indicate that specific functional variations in the OXPHOS complexes I and III are responsible for enhanced ROS production and the subsequent retrograde communication, rendering resistance to cisplatin. Our data not only support previous findings that redox imbalance and nucleus-mitochondria crosstalk play important roles in cisplatin resistance, but also provide a potential mechanistic link to specific mtDNA variations. Furthermore, our studies were not restricted to cisplatin only, hence our findings would provide insights into the broader area of ROS-stimulating/mitochondria-targeting therapy as well as the related biomarker studies.

Measuring mitochondrial activity and the expression level of genes controlling the mitochondrial biogenesis and mtDNA replication, such as PGC-1α, DNA polymerase γ (POLγ) and mitochondrial transcriptional factor A (TFAM), in the future would obviously help corroborate this hypothesis.

Now that we know the characteristics of the cell lines in detail, there are a number of approaches that could be employed in the future to further validate our hypothesis that functional OXPHOS variation or a group of variations contribute to the intracellular ROS/drug response phenotypes observed. One way would be to create cybrids which involves the transfer of the cancer cell mitochondria/mtDNA containing the functional OXPHOS variations onto a common wild type nuclear background, and the observation whether the cellular phenotype is transferred. However, the process of generating cybrids is not without its problems and can create irreversible epigenetic changes [66], as well as nuclear genome instability and variation [67], potentially confounding the ability to identify one single mtDNA variation (or a group of mtDNA variations) as the contributing factor(s), and so the insights generated in this paper, especially the comparison between PC-3 and PNT-2, are still useful. Another possibility would be to introduce the functional OXPHOS variations into Saccharomyces cerevisiae (Baker’s yeast) cells, as unlike mammalian mtDNA, its mitochondrial genome is amenable to genetic manipulation. This has proved useful for the functional validation of T14798C [60], but the lack of respiratory complex I in S. cerevisiae precludes a similar validation of functional complex I variations identified in this study. Recent technological advances suggest that such precise manipulation of mammalian mtDNA (and thus the functional validation of complex I variations) may be possible in the future, and may in fact provide a better alternative for functionally validating mtDNA variations than both cybrids and the yeast model system [68].

It is worth mentioning that two cell lines of distinctly opposite characteristics were identified among the tested cancer cell lines in our studies that was either the most sensitive (i.e. Ishikawa) or the most resistant (i.e. Caco-2) to ROS-stimulating therapy. The profiles of Ishikawa and Caco-2 based on the ROS/mtDNAcn/functional complexes I/III mutation parameters could be further validated as benchmarks to assess cancer cells’ response towards ROS-stimulating therapy. In addition, these two cell lines would make for interesting candidates for future downstream cybrid/mtDNA gene editing studies to help validate that the differences observed in the present study are due to mtDNA pattern and not due to differences in their nuclear proteomes.

Our data showed significant correlations between mitochondrial genetic abnormalities and the baseline ROS levels and drug resistance in cancer cells. In particular, we highlight an association between elevated mtDNAcn, higher frequency of predicted functional non-synonymous variations in complexes I/III, elevated baseline ROS levels and enhanced drug resistance (Table S4). Consequently, screening patients for mtDNAcn and the key functional variations identified in this study (A10398G, T11120C, C12084T, A13681G, G13708A, C13802T, A13966G and T14798C) could provide healthcare professionals with useful information in terms of predicting a patient’s response to the treatment and designing personalised therapy in the future. It is worth mentioning that since the mtDNA mutations were identified using the Sanger sequencing method, we were unable to ascertain the level of heteroplasmy of these key functional variations. However, the fact that those variations were successfully detected by Sanger sequencing indicates that they are significantly abundant in the collective mitochondrial genome. Nevertheless, a more advanced technique such as Next Generation Sequencing would be able to provide such information in future studies and allow us to evaluate whether the extent of heteroplasmy also affects intracellular ROS and the associated drug sensitivity. Furthermore, the present study was carried out on a small number of samples, and we were unable to test more cell lines (particularly more paired cancer/non-cancerous cell lines) due to limited resources. However, our existing data clearly indicate the potential of the aforementioned mtDNA variations as biomarkers to predict cancer cells’ response to ROS-stimulating agents, and therefore warrant further investigations in this direction.