Mitochondrial mRNA transcripts predict overall survival, tumor recurrence and progression in serous ovarian cancer: Companion diagnostics for cancer therapy

Here, we performed a systematic analysis to discover new biomarkers of overall survival and tumor progression in ovarian cancer patients. More specifically, we determined whether nuclear-encoded mitochondrial genes related to mitochondrial biogenesis and function are effective in predicting clinical outcome in ovarian cancer. As a consequence, we are able to provide in silico validation of the prognostic value of these mitochondrial markers, in a well-defined population of ovarian cancer patients. Towards this end, we used a group of N=111 ovarian cancer patients (serous type; stage III), with optimal de-bulking. Importantly, in this group of cancer patients, CA125 and PCNA (conventional markers) were associated with poor overall survival, as would be expected. Using this approach, we identified >100 new individual mitochondrial gene probes that effectively predicted significantly reduced overall survival, with hazard-ratios (HR) of up to 3.68 (p < 9.8e-05). These mitochondrial mRNA transcripts included membrane proteins, chaperones, anti-oxidant enzymes, as well as mitochondrial ribosomal proteins (MRPs) and key members of the OXPHOS (I-V) complexes. Based on this bioinformatics analysis and in silico validation, we conclude that mitochondrial biogenesis and OXPHOS should both be considered as new therapeutic targets, for the more effective treatment of human ovarian cancers. The mitochondrial biomarkers that we have identified could also be employed as new companion diagnostics to assist oncologists in: i) more accurately predicting clinical outcomes and ii) improving the response to therapy, in ovarian cancer patients.


INTRODUCTION
Drug-resistance dramatically limits the effectiveness of most cancer therapies, and especially for ovarian cancer patients [1,2]. As such, treatment failure remains a significant barrier to successful cancer therapy and precision medicine [3,4]. As a result, new biomarkers are urgently required for the treatment stratification of ovarian cancer patients, into different risk sub-groups at diagnosis (high-risk versus low-risk) [5].
In this report, we tested the hypothesis that mitochondrial markers might have prognostic value for the identification of high-risk ovarian cancer patients, with increased progression and poor overall survival. For this purpose, we used a data-mining and informatics strategy to determine the potential effectiveness of mitochondrial gene transcripts, in predicting clinical outcome.
Our results indicate that >100 mitochondrial gene probes can be used individually or in various combinations, to predict poor overall survival in ovarian cancer patients. Based on these current findings, we speculate that mitochondrial biogenesis and/or OXPHOS could be targeted therapeutically to prevent ovarian cancer recurrence and extend overall survival.

Prognostic value of conventional markers (CA125 and PCNA) in the patient population
To identify novel biomarkers for ovarian cancers, we employed publically available transcriptional profiling data from the tumors of patients with serous ovarian cancer (stage III), with optimal de-bulking, low CA125 levels at diagnosis, and 5-years of follow-up data ( Figure  1).
First, we assessed the prognostic value of CA125 in this context. The results of this analysis are shown in Figure 2 and Table 1. Note that the hazard-ratio (HR) for CA125 was 2.29 (p = 0.005) for overall survival (OS). As proliferative markers are often used as key endpoints in Phase II clinical trials, we next assessed the prognostic value of Ki67 and PCNA. Figure 2 and Table 2 both show the prognostic value of these markers. The results with Ki67 were not significant, but PCNA showed a hazardratio of 2.85 (p = 0.00025). Similarly, we determined the utility of macrophage-specific markers of inflammation. However, Table 3 shows that that CD68 and CD163 did not show significant prognostic value.
Thus, a subset of conventional markers (CA125 and PCNA) can be used to predict overall survival in ovarian cancer patients.

Prognostic value of individual mitochondrial markers
Our hypothesis is that increased mitochondrial biogenesis drives poor overall survival in ovarian cancer patients. To directly test this hypothesis, we next determined the prognostic value of a series of mitochondrial markers.
Firstly, we interrogated the utility of the behavior of mitochondrial chaperones and mitochondrial membrane proteins. Table 4 and Figure 3A both show that SLC25A5 and TIMM10 have significant prognostic value, with hazard-ratios of 2.67 and 2.63, respectively. Other members of the SLC25A, TIMM, TOMM and VDAC families also had prognostic value. Mitochondrial-related antioxidant proteins (NQO1 and SOD2), as well as mitochondrial creatine kinase, also had significant value (summarized in Table 4 and Figure 3B).
Next, we carefully examined the prognostic value of mitochondrial ribosomal proteins (MRPs). They functionally control the biosynthesis of essential components of the OXPHOS complexes, driving mitochondrial biogenesis (Table 5). Ten members of the large subunit (MRPLs) showed significant prognostic value, with hazard-ratios between 3.56 and 1.90. Interestingly, MRPL49 had the best prognostic value. Eleven different members of the small subunit (MRPSs) showed significant prognostic value, with hazard-ratios       Note that that high transcript levels of NQO1 are associated with significantly reduced overall survival.  Similarly, we also determined the prognostic value of key components of the OXPHOS complex. These results are summarized in Table 6. Surprisingly, 52 different gene probes for the OXPHOS complexes showed hazard-ratios between 3.68 and 1.76. Complex I had the most subunits with significant prognostic value (21 in total). However, UQCRFS1 (complex III) had the best individual prognostic value (HR = 3.68; p = 9.8e-05). NDUFA3 (complex I) also showed significant prognostic value (HR = 3.55; p = 2.3e-05). Kaplan-Meier curves for members of complex I and II are shown in Figure 5A

Three new mitochondrial gene signatures for predicting overall survival, recurrence and the response to therapy
To significantly amplify the prognostic power of these unique mitochondrial markers, we then combined the most promising markers and to derive three new mitochondrial gene signatures.
As such, both of these Ov-Mito-Signature(s) were a dramatic improvement over individual mitochondrial biomarkers, as well as CA125 and PCNA (Tables 1 & 3; Figure 2).
To further improve the predictive value of Ov-Mito-Signature-2, we next added the proliferative marker PCNA, to create Ov-Mito-Signature-3. The robust nature of Ov-Mito-Signature-3 is highlighted in Table 9   and Figures 12-14, which shows a hazard-ratio of 5.63 (p = 7.6e-06). Ov-Mito-Signature-3 was also the most effective in predicting the response to therapy ( Figure 13). Importantly, Ov-Mito-Signature-3 retained its prognostic value in a larger group of serous ovarian cancer patients (N = 442), without restricting our analysis to patients with low serum CA125 levels ( Figure 14).

Understanding
CSCs, telomerase and mitochondrial activity: targeting ovarian cancer with doxycycline and/or palbociclib The exact functional role of telomerase activity in ovarian cancer stem cell (CSC) propagation remains largely unknown. Recently, to address this issue, we indirectly monitored telomerase activity, by linking the hTERT-promoter to eGFP [6,7]. Using SKOV3 ovarian cancer cells, stably-transfected with the hTERT-GFP reporter, we then used GFP-fluorescence to fractionate these cell lines into GFP-high and GFP-low populations. We functionally compared the phenotype of these GFPhigh and GFP-low cell sub-populations. Importantly, we showed that ovarian cancer cells with higher telomerase activity (GFP-high) are energetically-activated, with increased mitochondrial OXPHOS and glycolysis [6]. This was confirmed by unbiased label-free proteomics analysis. A sub-population of SKOV3 cells with high telomerase activity showed i) increased "stemness" (3D-spheroid formation) and ii) enhanced cell migration (Boyden-chamber assay). These cellular phenotypes were halted by inhibitors of energy-metabolism, targeting either OXPHOS or glycolysis, or by using doxycycline, a clinically-approved antibiotic, that inhibits mitochondrial biogenesis [6,7].
Telomerase activity also determined the ability of hTERT-high ovarian CSCs to proliferate, as determined by monitoring DNA-synthesis. Use of Palbociclib, a CDK4/6 inhibitor (an FDA-approved drug) specifically blocked ovarian CSC propagation, with an IC-50 of ~100 nM [6]. Thus, telomerase-high ovarian CSCs are the most energetically-activated, migratory and proliferative cell sub-population [6]. These findings suggest a mechanistic  interpretation for why long telomere length (a specific marker of high telomerase activity) is strictly correlated with metastasis disease progression and poor outcome in ovarian tumors and other cancer types [8,9].
As such, elevated telomerase activity may "fuel" the propagation of ovarian CSCs by activating mitochondrial biogenesis, ultimately leading to poor clinical outcome. These observations may help explain why combining mitochondrial markers, together with the proliferation marker PCNA, so significantly increased the prognostic value of this Ov-Mito-Signature.

Employing mitochondrial markers and mitosignatures, as companion diagnostics for treatment stratification: implications for drug repurposing
In support of our current hypothesis, integrating telomerase activity with increased mitochondrial function, we demonstrate that a sub-set of mitochondrial gene transcripts are able to predict survival in serous ovarian cancer patients, with optimal de-bulking. As such, these particular mitochondrial markers could ultimately be used to select high-risk ovarian cancer patients at diagnosis, up to 5 years in advance, for close monitoring. As such, our results provide an excellent justification for the therapeutic targeting of mitochondria in ovarian cancer cells, to improve patient survival.
In this new paradigm, high-risk patients would be identified at diagnosis by the over-expression of mitochondrial mRNA transcripts in their ovarian tumors ( Figure 15). As a consequence, these patients could then be treated with certain FDA-approved drugs (e.g., Doxycycline or Palbociclib; together with the standard of care), to improve overall survival. These therapeutics have been previously documented to halt the proliferation of the ovarian CSC population [6].
FDA-approved antibiotics can safely prevent mitochondrial biogenesis and/or OXPHOS as a manageable, off-target, "side-effect" [6,7,[10][11][12][13][14][15][16]. These antibiotics include the tetracyclines, the erythromycins, as well as pyrvinium pamoate, atovaquone, and bedaquiline [10,11,13,14]. For example, the new mitochondrial markers and Mito-Signatures we have discovered, could be used as companion diagnostics, for re-purposing these FDA-approved drugs as novel anti-cancer agents. More specifically, this would facilitate the ability of medical oncologists to identify the correct patient sub-population for new phase II clinical trials for drug re-purposing/repositioning in serous ovarian patients, as an add on to conventional chemo-therapy (e.g., platins and taxol).

Mitochondrial markers and mito-signatures: implications for new drug discovery
The three new Mito-Signatures that we developed may also be useful for selecting new "druggable" targets for new drug development, to prevent treatment failure and improve overall survival. As a consequence of our K-M analyses, the mitochondrial ribosome would be an attractive new target for developing novel inhibitors of mitochondrial protein translation in cancer cells; similarly, mitochondrial chaperones, the OXPHOS complexes and the mitochondrial ATP-synthase may also be suitable drug  targets. Multiple members of these multi-subunit protein complexes show significant prognostic value, suggesting that modulation of their intrinsic activity may provide therapeutic benefits. Targeting of these large complexes would be predicted to suppress tumor recurrence and prevent disease progression in these serous ovarian cancer patients.
In addition, such mitochondrial markers could also be employed as companion diagnostics for novel therapies targeting either mitochondria or telomerase (hTERT) and/ or cell proliferation, to select the high-risk sub-population of ovarian cancer patients, resulting in the necessary treatment stratification. In direct support of this assertion, we showed here that three different Mito-Signature(s) could be used to successfully identify the sub-population of high-risk ovarian cancer patients that failed "platin" or "taxol" based therapies. These results indicate that mitochondrial markers could be used to monitor and/or predict the response to therapy, specifically identifying patients at high-risk for treatment failure at diagnosis, up to 5 years in advance, even before therapy is initiated.

METHOD OF ANALYSIS
Kaplan-Meier (K-M) Analyses. To perform K-M analysis on nuclear mitochondrial gene transcripts, we used an open-access online survival analysis tool to interrogate publically available microarray data from up to 1,435 ovarian cancer patients [5]. This allowed us to determine their overall prognostic value. For this purpose, we primarily analyzed 5-year follow-up data from serous ovarian cancer patients (stage III) that had optimal debulking (N = 111). Biased array data were excluded from the analysis. This allowed us to identify >100 nuclear mitochondrial gene probes, with significant prognostic value. Hazard-ratios for overall survival (OS), progression free survival (PFS; recurrence) and post-progression survival (PPS) were calculated, at the best auto-selected cut-off, and p-values were calculated using the logrank test and plotted in R [5]. K-M curves were also generated online using the K-M-plotter (as high-resolution TIFF files), using univariate analysis: http://kmplot.com/analysis/index. php?p=service&cancer=ovar.
This allowed us to directly perform in silico validation of these mitochondrial biomarker candidates. The 2012 version of the database was originally utilized for all these analyses; however, virtually identical results were also obtained with the 2015 and 2017 versions.