1. Cancer Biology
  2. Computational and Systems Biology
Download icon

Mitochondrial respiratory gene expression is suppressed in many cancers

  1. Ed Reznik  Is a corresponding author
  2. Qingguo Wang
  3. Konnor La
  4. Nikolaus Schultz  Is a corresponding author
  5. Chris Sander  Is a corresponding author
  1. Memorial Sloan Kettering Cancer Center, United States
  2. Harvard Medical School, United States
  3. Dana-Farber Cancer Institute, United States
Research Advance
  • Cited 42
  • Views 4,196
  • Annotations
Cite this article as: eLife 2017;6:e21592 doi: 10.7554/eLife.21592


The fundamental metabolic decision of a cell, the balance between respiration and fermentation, rests in part on expression of the mitochondrial genome (mtDNA) and coordination with expression of the nuclear genome (nuDNA). Previously we described mtDNA copy number depletion across many solid tumor types (Reznik et al., 2016). Here, we use orthogonal RNA-sequencing data to quantify mtDNA expression (mtRNA), and report analogously lower expression of mtRNA in tumors (relative to normal tissue) across a majority of cancer types. Several cancers exhibit a trio of mutually consistent evidence suggesting a drop in respiratory activity: depletion of mtDNA copy number, decreases in mtRNA levels, and decreases in expression of nuDNA-encoded respiratory proteins. Intriguingly, a minority of cancer types exhibit a drop in mtDNA expression but an increase in nuDNA expression of respiratory proteins, with unknown implications for respiratory activity. Our results indicate suppression of respiratory gene expression across many cancer types.



Are the mitochondria of tumors characteristically different from those in normal human tissues? Several recent reports have described somatic changes to the mitochondrial genome (mtDNA) in tumors, which have expanded our knowledge of cancer genetics as well as basic mitochondrial biology (Davis et al., 2014; Joshi et al., 2015; Ju et al., 2014; Stewart et al., 2015). In our own work (Reznik et al., 2016), we estimated mtDNA copy number for thousands of tumor samples sequenced by The Cancer Genome Atlas (TCGA) consortium and found that several cancer types appeared depleted of mtDNA relative to adjacent-normal tissue.

A drop in mtDNA copy number can decrease the expression of critical (and necessary) proteins of the electron transport chain (ETC)/ATP synthase. Respirometry experiments indicate that cells typically harbor an excess of mitochondrial respiratory capacity, implying that a modest decrease in ETC protein levels will not impact the basal respiration rate of the cell (Brand and Nicholls, 2011). However, a sufficiently large reduction in the levels of proteins encoded by mtDNA (e.g. large or complete mtDNA depletion), will precipitate a drop in the rate of ATP generation via respiration. In these cases, cancer cells can partially compensate for lower respiratory ATP generation by increasing flux through glycolysis (i.e. the Warburg effect).

Thus, there remains a gap between the observation of mtDNA copy number changes and their propagation to a drop in respiration. Linking measurements of mtDNA copy number directly with respiratory flux remains difficult: flux studies of comparable sample size to contemporary genomic investigations are not available. However, one can lay a partial bridge between respiration and mtDNA ploidy by examination of mRNA expression of mtDNA (mtRNA). Mitochondrial DNA serves as the template for the transcription of 13 protein-coding genes, each of which is a critical integral membrane subunit in the electron transport chain or ATP synthase. A synchronous change (increase or decrease) in both mtDNA copy number and mtRNA expression in tumors compared to normal tissue can therefore be used as complementary evidence for a change in respiratory flux, which can then be evaluated experimentally.

Here, we use RNA-sequencing data of TCGA tumors to expand upon our prior study of mtDNA copy number variation in cancer (Figure 1). In part, our aim is to determine if tumor-associated changes in the levels of mtDNA-derived transcripts mirror changes in mtDNA copy number. More broadly, we propose using trio measurements of mtDNA copy number, mRNA transcribed from mtDNA, and mRNA transcribed from nuclear DNA (nuDNA)-encoded respiratory genes, to gauge the likelihood that a cancer type down-regulates respiration compared to normal tissue. In parallel, these data can be used to dissect the coordination of respiratory transcription, i.e. to compute the correlation between mtDNA and mtRNA levels, and to assess coordination of expression of respiratory genes from the mitochondrial and nuclear genome. As reported in detail below, our results using RNA-Seq data largely corroborate our earlier results using mtDNA copy number (Reznik et al., 2016) (with some important exceptions), and additionally offer a lens onto patterns of transcriptional coordination between the mitochondrial and nuclear genomes.

Figure 1 with 3 supplements see all
Summary of analysis.

(A) RNA-sequencing reads from the TCGA are aligned, and reads mapping to the mitochondrial genome are retained. (B) Changes in the expression of mtRNAs in tumors compared to adjacent-normal tissue are compared to analogous changes in mtDNA copy number. (C) Quantitative estimates of the correlation between mtDNA copy number and mtRNA are determined. (D) A comparison is made between the tumor vs. normal differential expression of OXPHOS subunits encoded in mtDNA and nuDNA.



Quantifying expression of mtRNAs across cancers

Many reports have examined changes in metabolic gene expression in cancer (e.g. Gaude and Frezza, 2016), but no study to date has examined the expression of the 13 mRNAs encoded by the mitochondrial genome. While estimates of gene expression in samples profiled by the TCGA are publically available (e.g. through the Firehose data pipeline at the Broad Institute, http://firebrowse.org), the abundance of the 13 proteins encoded in mtDNA are not reported there. Therefore, to quantify expression from mtDNA, we re-aligned TCGA RNA sequencing data from thousands of the same tumors analyzed in our prior study of mtDNA copy number, and retained for analysis sequencing reads aligning to the mitochondrial chromosome. Estimates of mtRNA expression (in transcripts-per-million, TPM Wagner et al., 2012) across 6614 samples profiled is available in Supplementary file 2.

As in our prior study, we tested whether mitochondrial pseudogenes in the nuclear genome, also known as nuclear integrations of mitochondrial DNA (NUMTs), confounded our estimates of mitochondrial transcription. Previous work has shown that NUMTs are, in the vast majority of cases, integrated into intergenic/intronic regions of the nuclear genome, and are unlikely to be transcribed (Collura et al., 1996; Dayama et al., 2014). Nevertheless, to directly assess the contribution of NUMT RNA to our estimates of mtDNA expression, we implemented two measures: (1) examination of expression of mitochondrial pseudogenes annotated in Gencode (Harrow et al., 2012), and (2) comparison of two methods (featureCounts [Liao et al., 2014] and RSEM [Li and Dewey, 2011]) for quantifying expression levels, which treat differently those reads mapping to more than one genomic region. Results from both analyses support the notion that NUMTs do not confound our estimates of mtDNA expression (Materials and methods, Figure 1—figure supplement 1 and Figure 1—figure supplement 2).

Across all tissues, mtRNAs were highly transcribed. However, we observed substantial differences in the abundance of any given mitochondrial (MT) transcript from one tissue to the next (as evaluated using TPM) (Figure 1—figure supplement 3). For example, compared to other tissues, samples derived from the kidney (TCGA studies KICH, KIRC, and KIRP, see Materials and methods for TCGA abbreviations) had exceptionally high levels of mtRNAs, while samples from lung adenocarcinoma (LUAD) expressed comparatively low levels of mtRNAs. When comparing our results to a prior study of MT protein abundances from the mouse (Pagliarini et al., 2008), we found good agreement between protein and mtRNA levels with respect to their ordering across tissues, (i.e. mtRNA expression and protein levels of MT-ND6 were highest in kidney, at moderate levels in liver, colon, and stomach, and lowest in lung).

Differential expression of mtRNA in tumors compared to normal tissue

In (Reznik et al., 2016), we observed widespread mtDNA copy number depletion in tumors compared to matched adjacent normal tissue in a number of solid tumor types. Here, we first set out to test if expression of mtRNAs in these tumor types also was correspondingly lower. To do so, we estimated the differential expression of mtRNA transcripts in tumor vs. normal tissues for each of 13 cancer types with adequate numbers of tumor and normal samples (Figure 2). The majority of cancer types have a tendency for lower levels of mtRNAs, and five cancer types (breast, esophageal, head and neck, kidney clear cell, and liver) have lower levels of all 13 protein-coding mtRNAs. Among the seven cancer types with statistically significant depletion of mtDNA content in our earlier report (Reznik et al., 2016), all seven had lower mtRNA levels in at least 4/13 genes, with no mtRNAs showing over-expression in these tumor types. We also observed that LUAD (lung adenocarcinoma), which we found to be the only cancer type with increased mtDNA copy number in Reznik et al. (2016), had lower expression of 6/13 mtRNAs, suggesting that any increase in mtDNA copy number in LUAD was compensated for at the transcriptional level. For one study, prostate cancer (PRAD), results were removed from analysis because differential expression using counts from featureCounts and RSEM produced differing results.

Differential expression of mitochondrial genes across cancer types.

Magnitude and statistical significance of differential expression evaluated by limma voom (see Materials and methods). The majority of mtRNAs are strongly down-regulated in several cancer types, including esophageal, breast, head and neck squamous, kidney clear cell, and liver cancers. One cancer type (kidney chromophobe), shows increases in the abundance of mtRNAs. All tumor types showing mtDNA copy number depletion in tumors relative to adjacent normal tissue (bottom annotation) show analogous depletion of mtRNAs. In contrast, mtDNA copy number changes in lung adenocarcinomas and kidney chromophobes are not reflected in differential expression of mtRNAs.


Breaking the general trend, one cancer type, KICH (chromophobe renal cell carcinomas), had higher levels of mtRNAs in tumors compared to normal tissues. The over-expression phenotype was notably large: each of the 13 genes had greater than 2-fold over-expression in tumors compared to normal tissue (Figure 2). Importantly, KICH did not have appreciable accumulation of mtDNA copies in our prior study.

The role of mitochondrial dysfunction, and of a mitochondrial accumulation phenotype, in chromophobe renal cell carcinomas has been appreciated for some time (Nagy et al., 2002). Two recent publications have highlighted the role that mtDNA mutations have in the development of one substype of chromophobes, eosinophilic chromophobe renal cell carcinomas (Davis et al., 2014; Joshi et al., 2015). Eosinophilic KICH tumors have been proposed to arise from oncocytomas, a cancer type characterized by cytoplasms swollen with respiration-deficient mitochondria. This phenotype arises from two critical dysfunctions: somatic mtDNA mutations that render mitochondrial OXPHOS non-functional, and defective mitophagy that prevents clearance of dysfunctional mitochondria. Thus, it has been proposed that oncocytoma cells experience an energy crisis due to defective mitochondria/OXPHOS (perhaps sensed through AMPK), and respond by further upregulating the biogenesis of defective mitochondria (Zong et al., 2016). Thus, our observation of increased mtRNA levels in KICH tumors is likely to counterintuitively reflect a drop in respiration. As noted before, experimental respirometry measurements must be made to confirm this hypothesis.

Interestingly, we also found differences in the tendency for any one mtDNA-encoded gene to be differentially expressed across cancer types, which likely arises from the molecular details of mitochondrial transcription. The mitochondrial genome is transcribed in a polycistronic fashion, with all mRNAs and tRNAs on a strand transcribed simultaneously. Following transcription, tRNAs are excised from the transcript, and the majority of the remaining mRNAs are polyadenylated (Pagliarini et al., 2008; Mercer et al., 2011). Polycistronic transcription ensures that mtRNAs are highly co-expressed, although it is clear from many studies that mtRNAs undergo a large degree of post-transcriptional regulation, resulting in uneven steady-state abundances (Rorbach and Minczuk, 2012). With regard to differential expression, most of the genes encoding subunits of Complex I were downregulated in the majority of cancer types. In contrast, genes encoding subunits of Complex V (ATP6 and ATP8) and to a lesser extent Complex IV (MT-CO1,MT-CO2,MT-CO3) were generally under-expressed in only the strongly mtDNA- and mtRNA-depleted cancer types.

Association of mtRNA with clinical parameters

We also evaluated the extent to which mtRNA levels were associated with clinical features (e.g. the age, pathological stage, and overall survival of patients), using available clinical data from the TCGA consortium (Figure 3, full results available in Supplementary file 3). Among these, papillary renal cell carcinoma (KIRP), esophageal carcinoma (ESCA), and thyroid cancer (THCA) showed an association between high mtRNA expression and increased age. It is not clear whether this statistical association is a secondary result of a correlation between age and other clinical/genomic features, (e.g. in THCA, age is positively associated with increased mutational density), and merits further investigation.

Association of mtRNA expression levels with overall survival across cancer types.

Multiple-hypothesis-adjusted univariate p-values from Cox regression for each mtRNA are combined using Fisher’s method for each cancer type. Several cancer types show an association between high mtRNA expression and improved outcome (negative Cox regression coefficient). Dashed line indicates threshold for statistical significance. Representative Kaplan-Meier plots are shown for overall survival in ACC (B) and KICH (C), partitioning patients into high expression vs. low expression groups based on median expression of MT-ND4.


More interestingly, we identified five cancer types (ACC,KICH,LGG,PAAD, and LIHC) in which higher mtRNA expression levels were associated with increased overall survival. Three of these cancer types (ACC,KICH, and LGG), showed similar associations in our prior analysis using mtDNA copy number (i.e. high mtDNA copy number was associated with better overall survival (Reznik et al., 2016). KIRP tumors also showed an association between higher mtRNA expression and less aggressive disease, as assessed by pathological stage (Supplementary file 3). Interestingly, these results echo similar findings by reported by Gaude and Frezza, who reported an association between down-regulation of nuclear-DNA-encoded mitochondrial transcripts and poor clinical outcome across many cancer types (Gaude and Frezza, 2016).

Correlation of mtDNA copy number and mRNA levels

Our simultaneous quantification of mtDNA copy number and mtRNA expression enabled us to address a more basic biological question: what is the relationship between the number of copies of mtDNA in a cell and the expression of mtDNA-encoded genes? A number of factors, including but not limited to mtDNA copy number, ultimately determine the steady-state abundance of mtRNAs and derived proteins in a cell. At low mtDNA copy number, transcript expression of mitochondrial genes may be limited by the number of DNA templates available for active transcription. Alternatively, other proteins (e.g. mitochondrial transcription termination factors) can control the rate of transcription, while yet others can modulate mtRNA stability and degradation (Clemente et al., 2015). Thus, it remains unclear whether mtDNA copy number is correlated to, and may be used as a surrogate for, the abundance of mtRNA.

To evaluate the association between mtDNA copy number and mtRNA abundance, we calculated (separately for each mtRNA) the non-parametric Spearman correlation between mtDNA copy number and mtRNA levels (in RSEM counts), for all samples with available mtDNA copy number and mtRNA expression estimates (18 tumor types analyzed in total, Figure 4 and Figure 4—figure supplement 1). We found that the correlation between mtDNA copy number and mtRNA expression was highly tissue-type-specific. Seven of eighteen tumor types (adrenocortical, breast, glioma, kidney clear cell, kidney papillary, liver, and thyroid) had strong (BH-corrected p-value <10−5) positive correlation across all or nearly all mtDNA-encoded genes (Figure 4—figure supplement 1). Seven cancer types (bladder, colon, glioblastoma, head and neck, kidney chromophobe, lung adenocarcinoma, and stomach) had weak but still statistically significant correlation. The remaining four cancer types (endometrial, esophagus, pancreatic, prostate) showed no statistically significant correlation. Of the seven cancer types with mtDNA depletion in tumors relative to normal tissue, five (all but esophageal and head and neck) showed positive correlation between mtDNA copy number and expression. Some genes (e.g. MT-ND5 and MT-ND6) had recurrently weaker correlation with mtDNA copy number, which may result from their lower (and thus potentially noiser) abundance compared to other MT transcripts. Thus, the data support a plausible correlation between mtDNA copy number and ‘steady-state’ mtRNA abundance across many but not all tissues.

For one cancer type (kidney clear cell), the TCGA consortium (Creighton et al., 2013) used reverse-phase protein arrays (RPPA) to measure the protein abundance of one mtDNA-encoded protein (MT-CO2) in tumors only (not normal tissue). This data enabled us to assess directly, in this one case, whether variation in mtDNA copy number and mtRNA levels translates to changes in protein levels. We found that MT-CO2 protein levels were strongly correlated with MT-CO2 mRNA (Spearman ρ 0.38, p-value <10−14), and to a lesser extent with mtDNA copy number (Spearman ρ 0.18, p-value 0.003) (Figure 4—figure supplement 2).

Interestingly, correlations between mtDNA and mtRNA were not necessarily homogeneous between tumor and normal samples from the same tissue. Upon closer examination of samples from KICH (Figure 4—figure supplement 3), we found that tumor and normal samples had strong but distinct patterns of correlation between mtDNA copy number and mtRNA levels. In particular, because KICH tumors only had increased mtRNA levels compared to normals (and not an increase in mtDNA copy number), the best-fit trend line between mtDNA and mtRNA appears shifted vertically when comparing tumor to normal tissue samples. Put another way, it appears as if KICH tumors had higher values of the steady-state mtRNA/mtDNA ratio. Given that many KICH tumors harbor mtDNA mutations, these observations suggest that compensation for such mitochondrial dysfunction in KICH may be via transcriptional mechanisms, rather than changes to mtDNA ploidy.

Patterns of nuOXPHOS and mtOXPHOS gene expression are not redundant

The mitochondrial genome encodes 13 polypeptides which serve as integral membrane subunits for 3 components of the electron transport chain and ATP synthase (herein referred to as mtOXPHOS proteins). The remaining 80 OXPHOS subunits are encoded in the nuclear genome (herein referred to as nuOXPHOS proteins) (Calvo and Mootha, 2010). The protein levels of these subunits must be coordinated in order to maintain proper stoichoimetry of OXPHOS complexes (Cancer Genome Atlas Research Network et al., 2013). Intriguingly, a study of the coordination of mtOXPHOS and nuOXPHOS protein levels in S. cerevisae found that the primary mode of regulation was through rapid translational, rather than transcriptional, synchronization (Couvillion et al., 2016). This leaves open the possibility that patterns of transcriptional changes in mtOXPHOS and nuOXPHOS may be asynchronous or even opposite.

To see if the nuclear and mitochondrial genome experience analogous changes in respiratory gene expression in the course of tumorigenesis, we examined the differential expression of mtOXPHOS and nuOXPHOS subunits for all 13 cancer types with available expression data for both tumor and normal tissues (Figure 5). We calculated a differential expression (DE) score for each cancer type, which captured the tendency for a set of genes to be down-regulated (corresponding to a score of −1) or up-regulated (a score of +1) in tumors compared to normal tissue. In 9/13 cancer types, we observed agreement in DE scores between mtOXPHOS and nuOXPHOS subunits (e.g. see volcano plot of KIRC in (Figure 5). More interestingly, we found that four cancer types (BLCA, BRCA, LUAD, and UCEC) have opposite patterns of differential expression in mtOXPHOS and nuOXPHOS genes. For example, in BRCA, while nearly all mtOXPHOS genes are down-regulated in tumors compared to normal tissue (mirroring the mtDNA copy number depletion in BRCA), nearly all nuOXPHOS genes actually had increased mRNA levels.

For a functional electron transport chain, subunits of each OXPHOS complex must assemble with compatible stoichiometries in a process that requires coordination of individual subunit protein levels (Mouchiroud et al., 2013). Taken together, the above findings speak to the potential shortcomings of transcriptional data as surrogates for OXPHOS protein abundance and respiratory activity. As mentioned above, several recent studies have reported on the critical role of translational processes in the coordination of respiration. One report by Wagner, Kitami, and colleagues found that treatment of differentiated C2C12 murine myotubes with certain drugs (including eukaryotic translation inhibitors) induced anti-correlated changes in nuOXPHOS and mtOXPHOS genes, respectively (Wagner et al., 2008). Separately, Couvillion and colleagues reported that the sudden shift of S. cerevisae to a non-fermentable carbon source led to rapid induction of transcriptional changes in nuOXPHOS, but not mtOXPHOS. Instead, rapid redistribution of mitochondrial ribosomes supported fast adapation to respiration (e.g. via an increase in translation efficiency of Complexes III and IV). Together, these two studies suggest that the examination of translational activity (e.g. by ribosome profiling) may be useful tools for examining an imbalance in mtOXPHOS and nuOXPHOS transcription.

Questions and future directions

The purpose of this study was to compare cancer-associated changes in mitochondrial transcript levels with changes in mtDNA copy number as reported in Reznik et al. (2016). An equally important goal, for all scientific studies but perhaps more so for ‘replication’ studies of the sort here, is to lay out the loose ends and unanswered questions, and to make explicit those findings which did not agree with prior results. Below we offer a reflection on our findings, which may steer future investigations towards interesting territory for future discovery.

In tandem with our prior study (Reznik et al., 2016), this work finds that several tumor types suppress the expression of genes involved in oxidative phosphorylation/respiration. Five cancer types (kidney clear cell, kidney papillary, head and neck squamous cell, liver, and esophageal) were mtDNA-depleted, under-expressed mtRNAs, and under-expressed nuOXPHOS genes, in comparison to normal tissue. Importantly, because cells harbor excess respiratory capacity, a drop in the protein expression of mtDNA-encoded OXPHOS subunits will not necessarily precipitate a drop in respiration unless it is sufficiently large. Experimental measurements, e.g. respirometry or measurements of flux will be necessary to determine if drops in mtRNA expression translate to a drop in oxygen consumption/respiration (Brand and Nicholls, 2011). A reasonable subset of cancer types to initially investigate for such experiments would be the five listed above to suppress OXPHOS gene expression.

While changes in mtDNA copy number and mtRNA expression cannot be used as surrogates for changes in respiratory flux, several of the cancer types examined here are driven by genetic changes affecting mitochondrial respiration, suggesting that changes in mtRNA levels may reflect bona fide changes in respiration. For example, clear-cell renal cell carcinomas are driven by homozygous loss of VHL, which leads to activation of hypoxia inducible factor (HIF), and subsequently increased transcription of glycolytic enzymes (Creighton et al., 2013). The signature of HIF activation is evident not only in gene expression data, but also in metabolomic measurements of KIRC tumors (Hakimi et al., 2016), and suggests that KIRC tumors experience a drop in respiration. In particular, KIRC tumors exhibit increased levels (compared to normal tissue) of metabolites upstream of SDH (succinate dehydrogenase, Complex II), i.e. citrate, cis-aconitate, and succinate, and decreased levels of metabolites downstream of SDH (fumarate and malate). This partitioning of changes in metabolite levels could arise by a decrease in the activity of the electron transport chain and SDH, which would induce a metabolic bottleneck and the accumulation (depletion) of metabolites upstream (downstream) of SDH.

Furthermore, while our data suggests that several cancer types may decrease mitochondrial respiration, mitochondrial metabolism provides several other important services to the cell. In some cases, respiration may be critical to these processes; for example, aspartate biosynthesis is heavily reliant on electron acceptors produced by mitochondrial respiration, and respiration-deficient cells become limited by the availability of aspartate (Birsoy et al., 2015). Beyond the ETC, cytosolic acetyl-CoA utilized for lipogenesis is produced from the action of ATP-citrate lyase on mitochondrial citrate, and substantial evidence supports the critical role of mitochondrial metabolism in producing adequate one-carbon precursors for nucleotide biosynthesis (Ahn and Metallo, 2015). Finally, it should be noted that xenografted ρ0 (depleted of mtDNA) breast cancer cells form tumors only after re-acquiring mtDNA from the host, emphasizing that some basal level of mitochondrial respiration is likely required for tumor viability and malignancy (Tan et al., 2015).

When comparing the differential expression of mtRNA (tumor vs. adjacent-normal tissue) to changes in mtDNA copy number (tumor vs. adjacent-normal tissue), we found two cancer types (LUAD and KICH) had opposite changes in the two measures. To highlight a single example, lung adenocarcinomas were the only cancer type to accumulate mtDNA copies compared to normal tissue. Using mtRNA data, however, these tumors had lower mtRNA levels of 6/13 MT genes. We can speculate as to the source of this inconsistency: it may be that our inference from sequencing data of mtDNA copy number or mtRNA expression was erroneous for these tumor types. While this is a possibility, we would remind readers that the same analytical pipeline was applied to all cancer types, the majority of which had features consistent with our expectations, e.g. correlation of whole-exome and whole-genome estimates of mtDNA copy number in Reznik et al. (2016), correlation of mtDNA copy number with expression of mitochondrially-localized genes in Reznik et al. (2016), consistency of changes in mtDNA copy number and mtRNA expression across most cancer types in Figure 2 of this study. Alternatively, if one takes our calculations to be reasonably accurate estimates of mtDNA and mtRNA abundance, then the unexpected results become more intriguing. Regarding the inconsistency between mtDNA ploidy and mtRNA abundance described above, it is known that a number of proteins regulate mtDNA transcription, including transcription factors (TFAM,TFB1M), transcriptional termination factors (MTERFs), as well as mitochondrial biogenesis in general (NRF-1 and NRF-2, PGC1α,ERRα) (Scarpulla, 2008; Rorbach and Minczuk, 2012). A more detailed investigation of these proteins may reveal the factors driving inconsistency between changes in mtDNA copy number and mtRNA.

The coordinated expression of mtDNA- and nuDNA-encoded subunits of the respiratory chain is critical to efficient respiration. Prior studies of mitochondrial-nuclear (mitonuclear) protein imbalance in C. elegans claimed that it promotes longevity via activation of the mitochondrial unfolded protein response (UPRMT), a transcriptional response preserving mitochondrial function in the face of stress (Cancer Genome Atlas Research Network et al., 2013). Intriguingly, two recent studies reported activation of the UPRMT counterintuitively promoted tolerance of mutant mtDNA levels by activating mitochondrial biogenesis in an effort to recover OXPHOS activity (Lin et al., 2016; Gitschlag et al., 2016). If cancer cells indeed have substantial changes in mtDNA copy number and transcription (and mtDNA mutations), then activation of the mammalian UPRMT may be critical in supporting their proliferation in the face of increased mitochondrial stress.

Several of our findings discussed above allude to intriguing tissue-specific patterns of variation in mtDNA and mtRNA. Both the accumulation/depletion of mitochondrial transcripts (Figure 2,4,5) and their association with mtDNA ploidy were highly dependent on the cancer type. It is well-appreciated that the physiology and molecular constitution (Fernández-Vizarra et al., 2011; Calvo and Mootha, 2010) of mitochondria varies substantially from one tissue to the next. Can some of the unexpected phenomena described here (e.g. inconsistent changes in mtDNA copy number and mtRNA expression, decoupling of nuOXPHOS and mtOXPHOS) be explained by the tissue-specific physiology of mitochondria? We expect that, in the pursuit of such questions, tools for profiling the morphologic and proteomic composition of mitochondria, in parallel with organelle-targeted transcriptomics and genomics, will yield valuable insights.

Figure 4 with 3 supplements see all
Correlation between mtDNA copy number and mtRNA expression.

Relative mtDNA copy number was correlated against mtRNA expression (log2 normalized counts from limma voom). The expression of one gene, MT-ATP6, is depicted, although other mtDNA protein coding genes are similar. Lines indicate best fit linear trend between mtDNA copy number and log2 MT-ATP6 expression. Cancer types titled with an asterisk indicate a statistically significant correlation (Spearman, BH-adjusted p-value <0.05). Double asterisk indicates an especially strong correlation (BH-adjusted p-value <10−5).

Comparison of differential expression (tumor vs. adjacent-normal tissue) of mtDNA-encoded OXPHOS subunits (mtOXPHOS) and nuclear-DNA-encoded OXPHOS subunits (nuOXPHOS).

(A) Differential expression scores for mtOXPHOS and nuOXPHOS across cancers. Red dashed boxes highlight cancer types with opposite trends in mtOXPHOS and nuOXPHOS differential expression. (B) Volcano plots highlighting differential expression of mtOXPHOS (red) and nuOXPHOS (blue) genes in BRCA and KIRC.


Materials and methods

RNA sequencing alignment, quantification, and quality control

Request a detailed protocol

The input of our analysis pipeline is raw sequencing reads (in FASTQ format). The raw reads of the RNA-seq samples were retrieved from the Cancer Genomics Hub (CGHub, https://cghub.ucsc.edu, Research Resource Identifier RRID:SCR_003193). When FASTQ files were not available, e.g. for stomach adenocarcinoma, we downloaded aligned sequence reads (in BAM format) and extracted reads from BAM files with the Java program ubu.jar (https://github.com/mozack/ubu) before processing samples using our pipeline. All acronyms for cancer types conform to the TCGA nomenclature (ACC: adrenocortical; BRCA:breast; BLCA: bladder; COAD: colon; ESCA: esophageal; GBM: glioblastoma; HNSC: head and neck squamous cell; KICH: kidney chromophobe; KIRC: kidney clear cell; KIRP: kidney papillary; LGG: low grade glioma; LIHC: liver, LUAD:lung adenocarcinoma; PAAD: pancreatic; PRAD: prostate; STAD: stomach; THCA: thyroid; UCEC: endometrial). Sample sizes for each cancer type are reported in Supplementary file 1.

We employed STAR aligner (Dobin et al., 2013), a fast accurate alignment software used widely in the NGS community, to map reads to UCSC human reference genome hg19 and reference transcriptome GENCODE (v19), using recommended parameters, e.g. ’–outFilterType BySJout’ and ’–outFilterMultimapNmax 20’, etc., which are also standard options of the ENCODE project for long RNA-seq pipeline. Samples with alignment rates less than 40% were excluded from further analysis.

The software tools RseQC (Wang et al., 2012) and mRIN (Feng et al., 2015) were used to evaluate sample quality. RNA degradation, as detected by mRIN (Feng et al., 2015), was present in some TCGA samples. Since degradation can bias expression level measurements and cause data misinterpretation, we excluded samples with evidence for degradation. Specifically, we used prostate cancer samples from the TCGA project as test data to decide on a degradation cutoff for mRIN. TCGA prostate samples had undergone extensive pathological, analytical, and quality control review and had been shown to include a significant portion of degraded samples (Clemente et al., 2015). We compared mRIN scores with RNA Integrity Numbers (RIN) calculated by TCGA for prostate samples and found they are highly negatively correlated (Pearson correlation <−0.93). To filter degraded samples, TCGA used a cutoff 7.0 for RIN, which corresponds roughly to −0.11 for mRIN. We also manually examined other tumor types including bladder urothelial carcinoma and breast invasive carcinoma and a mRIN cutoff −0.11 worked reasonably well for these studies as well. Hence, we used −0.11 as the degradation threshold for mRIN throughout our study. Samples with mRIN <−0.11 were regarded as degraded and, thus, excluded from further analysis.

When running STAR, we specified an option ’–quantMode TranscriptomeSAM’ to make STAR output a file, Aligned.toTranscriptome.out.bam, which contains alignments translated into transcript coordinates. This file was then used with RSEM (Li and Dewey, 2011) to quantify mitochondrial gene expression. The program ’rsem-calculate-expression’ in the RSEM package requires strand specificity of the data, which is estimated using software RseQC (Wang et al., 2012).

To ensure that reads aligning to the mtDNA are not derived from NUMTs/mitochondrial pseudogenes in the nuclear DNA, we implemented two quality control measures. First, we examined the expression (using RSEM, which handles multi-mapping reads, see next paragraph) of 175 regions annotated as mtDNA pseudogenes in GENCODE v19. While these genes are a subsample of the total number of NUMTs in the genome (on the order of 1000 [Dayama et al., 2014]), they are nevertheless useful for roughly estimating the genome-wide expression of NUMTs. Across all studies, we failed to see substantial expression of NUMT transcripts. In general across all cancer types, the vast majority of samples showed 100- to 1000-folder greater expression of bona fide mtDNA-encoded genes than expression of mitochondrial pseudogenes (Figure 1—figure supplement 1).

As a second quality control measure, we compared estimates of expression counts produced from two computational approaches, RSEM and featureCounts. These methods differ fundamentally in the way that they count reads mapping to multiple regions in the genome (multi-mapping reads). RSEM handles multi-mapping reads using an expectation-maximization procedure, while featureCounts by default simply discards multi-mapping reads. Thus, if a read maps ambiguously to both mtDNA and a NUMT, then featureCounts will ignore this read while RSEM will not. For all 13 mtDNA-encoded mRNAs, we compared (1) the total expression (in counts) as reported by RSEM and featureCounts (Figure 1—figure supplement 2A) and (2) the tumor vs. normal differential expression estimate (Figure 1—figure supplement 2B). With the exception of a single study, we found excellent agreement between RSEM and featureCounts by both measures. In the PRAD study, we noted that the two approaches produced different results for differential expression of MT genes, and we therefore removed PRAD from our differential expression analysis. These data support the hypothesis, as reported by in prior work by others (Dayama et al., 2014; Collura et al., 1996), that NUMT expression is negligibly low and does not confound estimates of true mtDNA expression.

Differential expression analysis

Request a detailed protocol

Differential expression analysis was completed using the limma voom package (Law et al., 2014). TMM normalization was applied to counts before running differential expression analysis. Estimates of differential expression were compared against analogous estimates using publicly available data from the Broad Institute’s TCGA Firehose using the RTCGAToolbox R package (Samur, 2014), with excellent agreement between the two (data not shown). Cancer types with fewer than five adjacent-normal tissue samples profiled by RNA-Seq were discarded. As mentioned in the above section, the TCGA cancer type PRAD was removed from differential expression analysis due to differences when comparing results for mtDNA differential expression from RSEM and featureCounts.

Differential expression score is defined as

(1) DE=# of Genes Overexpressed in Gene Set# of Genes Underexpressed in Gene Set# of Genes in Gene Set

A score of 1 indicates that all genes in the gene set are overexpressed in tumor compared to normal tissue, while a score of −1 indicates that all genes in the gene set are underexpressed.

Genes included in the nuDNA OXPHOS geneset were derived from the MSigDB Reactome geneset file (Liberzon et al., 2015).

mtDNA copy number analysis

Request a detailed protocol

All data for mtDNA copy number analysis was derived from Supplementary file 1 of (Reznik et al., 2016).

Clinical association analysis

Request a detailed protocol

All data for clinical analysis was downloaded from the Broad Firehose. For survival calculations, Cox regression was used to associate log2 TPM expression of each mtRNA with overall survival. For association with pathological stage or age, non-parametric Spearman correlations were calculated. For each clinical variable, association was calculated with each of the 13 mtDNA-encoded mRNAs. Then, p-values for the associations were corrected using the Benjamini-Hochberg procedure, and then combined using Fisher’s method. Because Fisher’s method assumes independent p-values (which is not the case in this analysis, as mtRNAs are transcribed polycistronically), we additionally report in Figure 3 the number of mtRNAs significantly associated (BH-adjusted p-value<0.05) with the clinical variable under examination.


  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
    The molecular taxonomy of primary prostate cancer
    1. Cancer Genome Atlas Research Network
    2. A Abeshouse
    3. J Ahn
    4. R Akbani
    5. A Ally
    6. S Amin
    7. CD Andry
    8. M Annala
    9. A Aprikian
    10. J Armenia
    11. A Arora
    12. JT Auman
    13. M Balasundaram
    14. S Balu
    15. CE Barbieri
    16. T Bauer
    17. CC Benz
    18. A Bergeron
    19. R Beroukhim
    20. M Berrios
    21. A Bivol
    22. T Bodenheimer
    23. L Boice
    24. MS Bootwalla
    25. R Borges dos Reis
    26. PC Boutros
    27. J Bowen
    28. R Bowlby
    29. J Boyd
    30. RK Bradley
    31. A Breggia
    32. F Brimo
    33. CA Bristow
    34. D Brooks
    35. BM Broom
    36. AH Bryce
    37. G Bubley
    38. E Burks
    39. YS Butterfield
    40. M Button
    41. D Canes
    42. CG Carlotti
    43. R Carlsen
    44. M Carmel
    45. PR Carroll
    46. SL Carter
    47. R Cartun
    48. BS Carver
    49. JM Chan
    50. MT Chang
    51. Y Chen
    52. AD Cherniack
    53. S Chevalier
    54. L Chin
    55. J Cho
    56. A Chu
    57. E Chuah
    58. S Chudamani
    59. K Cibulskis
    60. G Ciriello
    61. A Clarke
    62. MR Cooperberg
    63. NM Corcoran
    64. AJ Costello
    65. J Cowan
    66. D Crain
    67. E Curley
    68. K David
    69. JA Demchok
    70. F Demichelis
    71. N Dhalla
    72. R Dhir
    73. A Doueik
    74. B Drake
    75. H Dvinge
    76. N Dyakova
    77. I Felau
    78. ML Ferguson
    79. S Frazer
    80. S Freedland
    81. Y Fu
    82. SB Gabriel
    83. J Gao
    84. J Gardner
    85. JM Gastier-Foster
    86. N Gehlenborg
    87. M Gerken
    88. MB Gerstein
    89. G Getz
    90. AK Godwin
    91. A Gopalan
    92. M Graefen
    93. K Graim
    94. T Gribbin
    95. R Guin
    96. M Gupta
    97. A Hadjipanayis
    98. S Haider
    99. L Hamel
    100. DN Hayes
    101. DI Heiman
    102. J Hess
    103. KA Hoadley
    104. AH Holbrook
    105. RA Holt
    106. A Holway
    107. CM Hovens
    108. AP Hoyle
    109. M Huang
    110. CM Hutter
    111. M Ittmann
    112. L Iype
    113. SR Jefferys
    114. CD Jones
    115. SJ Jones
    116. H Juhl
    117. A Kahles
    118. CJ Kane
    119. K Kasaian
    120. M Kerger
    121. E Khurana
    122. J Kim
    123. RJ Klein
    124. R Kucherlapati
    125. L Lacombe
    126. M Ladanyi
    127. PH Lai
    128. PW Laird
    129. ES Lander
    130. M Latour
    131. MS Lawrence
    132. K Lau
    133. T LeBien
    134. D Lee
    135. S Lee
    136. KV Lehmann
    137. KM Leraas
    138. I Leshchiner
    139. R Leung
    140. JA Libertino
    141. TM Lichtenberg
    142. P Lin
    143. WM Linehan
    144. S Ling
    145. SM Lippman
    146. J Liu
    147. W Liu
    148. L Lochovsky
    149. M Loda
    150. C Logothetis
    151. L Lolla
    152. T Longacre
    153. Y Lu
    154. J Luo
    155. Y Ma
    156. HS Mahadeshwar
    157. D Mallery
    158. A Mariamidze
    159. MA Marra
    160. M Mayo
    161. S McCall
    162. G McKercher
    163. S Meng
    164. AM Mes-Masson
    165. MJ Merino
    166. M Meyerson
    167. PA Mieczkowski
    168. GB Mills
    169. KR Mills Shaw
    170. S Minner
    171. A Moinzadeh
    172. RA Moore
    173. S Morris
    174. C Morrison
    175. LE Mose
    176. AJ Mungall
    177. BA Murray
    178. JB Myers
    179. R Naresh
    180. J Nelson
    181. MA Nelson
    182. PS Nelson
    183. Y Newton
    184. MS Noble
    185. H Noushmehr
    186. M Nykter
    187. A Pantazi
    188. M Parfenov
    189. PJ Park
    190. JS Parker
    191. J Paulauskis
    192. R Penny
    193. CM Perou
    194. A Piché
    195. T Pihl
    196. PA Pinto
    197. D Prandi
    198. A Protopopov
    199. NC Ramirez
    200. A Rao
    201. WK Rathmell
    202. G Rätsch
    203. X Ren
    204. VE Reuter
    205. SM Reynolds
    206. SK Rhie
    207. K Rieger-Christ
    208. J Roach
    209. AG Robertson
    210. B Robinson
    211. MA Rubin
    212. F Saad
    213. S Sadeghi
    214. G Saksena
    215. C Saller
    216. A Salner
    217. F Sanchez-Vega
    218. C Sander
    219. G Sandusky
    220. G Sauter
    221. A Sboner
    222. PT Scardino
    223. E Scarlata
    224. JE Schein
    225. T Schlomm
    226. LS Schmidt
    227. N Schultz
    228. SE Schumacher
    229. J Seidman
    230. L Neder
    231. S Seth
    232. A Sharp
    233. C Shelton
    234. T Shelton
    235. H Shen
    236. R Shen
    237. M Sherman
    238. M Sheth
    239. Y Shi
    240. J Shih
    241. I Shmulevich
    242. J Simko
    243. R Simon
    244. JV Simons
    245. P Sipahimalani
    246. T Skelly
    247. HJ Sofia
    248. MG Soloway
    249. X Song
    250. A Sorcini
    251. C Sougnez
    252. S Stepa
    253. C Stewart
    254. J Stewart
    255. JM Stuart
    256. TB Sullivan
    257. C Sun
    258. H Sun
    259. A Tam
    260. D Tan
    261. J Tang
    262. R Tarnuzzer
    263. K Tarvin
    264. BS Taylor
    265. P Teebagy
    266. I Tenggara
    267. B Têtu
    268. A Tewari
    269. N Thiessen
    270. T Thompson
    271. LB Thorne
    272. DP Tirapelli
    273. SA Tomlins
    274. FA Trevisan
    275. P Troncoso
    276. LD True
    277. MC Tsourlakis
    278. S Tyekucheva
    279. E Van Allen
    280. DJ Van Den Berg
    281. U Veluvolu
    282. R Verhaak
    283. CD Vocke
    284. D Voet
    285. Y Wan
    286. Q Wang
    287. W Wang
    288. Z Wang
    289. N Weinhold
    290. JN Weinstein
    291. DJ Weisenberger
    292. MD Wilkerson
    293. L Wise
    294. J Witte
    295. J Zhang
    296. J Zhang
    297. W Zhang
    298. L Yang
    299. J Zhang
    Cell 163:1011–1025.
  6. 6
  7. 7
  8. 8
  9. 9
    Comprehensive molecular characterization of clear cell renal cell carcinoma
    1. CJ Creighton
    2. M Morgan
    3. PH Gunaratne
    4. DA Wheeler
    5. RA Gibbs
    6. A Gordon Robertson
    7. A Chu
    8. R Beroukhim
    9. K Cibulskis
    10. S Signoretti
    11. F Vandin Hsin-Ta Wu
    12. BJ Raphael
    13. RGW Verhaak
    14. P Tamboli
    15. W Torres-Garcia
    16. R Akbani
    17. JN Weinstein
    18. V Reuter
    19. JJ Hsieh
    20. A Rose Brannon
    21. A Ari Hakimi
    22. A Jacobsen
    23. G Ciriello
    24. B Reva
    25. CJ Ricketts
    26. W Marston Linehan
    27. JM Stuart
    28. W Kimryn Rathmell
    29. H Shen
    30. PW Laird
    31. D Muzny
    32. C Davis
    33. M Morgan
    34. L Xi
    35. K Chang
    36. N Kakkar
    37. LR Treviño
    38. S Benton
    39. JG Reid
    40. D Morton
    41. H Doddapaneni
    42. Y Han
    43. L Lewis
    44. H Dinh
    45. C Kovar
    46. Y Zhu
    47. J Santibanez
    48. M Wang
    49. W Hale
    50. D Kalra
    51. CJ Creighton
    52. DA Wheeler
    53. RA Gibbs
    54. G Getz
    55. K Cibulskis
    56. MS Lawrence
    57. C Sougnez
    58. SL Carter
    59. A Sivachenko
    60. L Lichtenstein
    61. C Stewart
    62. D Voet
    63. S Fisher
    64. SB Gabriel
    65. E Lander
    66. R Beroukhim
    67. SE Schumacher
    68. B Tabak
    69. G Saksena
    70. RC Onofrio
    71. SL Carter
    72. AD Cherniack
    73. J Gentry
    74. K Ardlie
    75. C Sougnez
    76. G Getz
    77. SB Gabriel
    78. M Meyerson
    79. A Gordon Robertson
    80. A Chu
    81. HE Chun
    82. AJ Mungall
    83. P Sipahimalani
    84. D Stoll
    85. A Ally
    86. M Balasundaram
    87. YSN Butterfield
    88. R Carlsen
    89. C Carter
    90. E Chuah
    91. RJN Coope
    92. N Dhalla
    93. S Gorski
    94. R Guin
    95. C Hirst
    96. M Hirst
    97. RA Holt
    98. C Lebovitz
    99. D Lee
    100. Li
    101. M Mayo
    102. RA Moore
    103. E Pleasance
    104. P Plettner
    105. JE Schein
    106. A Shafiei
    107. JR Slobodan
    108. A Tam
    109. N Thiessen
    110. RJ Varhol
    111. N Wye
    112. Y Zhao
    113. I Birol
    114. SJM Jones
    115. MA Marra
    116. J Auman
    117. D Tan
    118. CD Jones
    119. KA Hoadley
    120. PA Mieczkowski
    121. LE Mose
    122. SR Jefferys
    123. MD Topal
    124. C Liquori
    125. YJ Turman
    126. Y Shi
    127. S Waring
    128. E Buda
    129. J Walsh
    130. J Wu
    131. T Bodenheimer
    132. AP Hoyle
    133. JV Simons
    134. MG Soloway
    135. S Balu
    136. JS Parker
    137. D Neil Hayes
    138. CM Perou
    139. R Kucherlapati
    140. P Park
    141. H Shen
    142. T Triche Jr
    143. DJ Weisenberger
    144. PH Lai
    145. MS Bootwalla
    146. DT Maglinte
    147. S Mahurkar
    148. BP Berman
    149. DJ Van Den Berg
    150. L Cope
    151. SB Baylin
    152. PW Laird
    153. CJ Creighton
    154. DA Wheeler
    155. G Getz
    156. MS Noble
    157. D DiCara
    158. H Zhang
    159. J Cho
    160. DI Heiman
    161. N Gehlenborg
    162. D Voet
    163. W Mallard
    164. P Lin
    165. S Frazer
    166. P Stojanov
    167. Y Liu
    168. L Zhou
    169. J Kim
    170. MS Lawrence
    171. L Chin
    172. F Vandin
    173. H Wu
    174. BJ Raphael
    175. C Benz
    176. C Yau
    177. SM Reynolds
    178. I Shmulevich
    179. RGW Verhaak
    180. W Torres-Garcia
    181. R Vegesna
    182. H Kim
    183. W Zhang
    184. D Cogdell
    185. E Jonasch
    186. Z Ding
    187. Y Lu
    188. R Akbani
    189. N Zhang
    190. AK Unruh
    191. TD Casasent
    192. C Wakefield
    193. D Tsavachidou
    194. L Chin
    195. GB Mills
    196. JN Weinstein
    197. A Jacobsen
    198. A Rose Brannon
    199. G Ciriello
    200. N Schultz
    201. A Ari Hakimi
    202. B Reva
    203. Y Antipin
    204. J Gao
    205. E Cerami
    206. B Gross
    207. B Arman Aksoy
    208. R Sinha
    209. N Weinhold
    210. S Onur Sumer
    211. BS Taylor
    212. R Shen
    213. I Ostrovnaya
    214. JJ Hsieh
    215. MF Berger
    216. M Ladanyi
    217. C Sander
    218. SS Fei
    219. A Stout
    220. PT Spellman
    221. DL Rubin
    222. TT Liu
    223. JM Stuart
    224. S Ng
    225. EO Paull
    226. D Carlin
    227. T Goldstein
    228. P Waltman
    229. K Ellrott
    230. J Zhu
    231. D Haussler
    232. PH Gunaratne
    233. W Xiao
    234. C Shelton
    235. J Gardner
    236. R Penny
    237. M Sherman
    238. D Mallery
    239. S Morris
    240. J Paulauskis
    241. K Burnett
    242. T Shelton
    243. S Signoretti
    244. WG Kaelin
    245. T Choueiri
    246. MB Atkins
    247. R Penny
    248. K Burnett
    249. D Mallery
    250. E Curley
    251. S Tickoo
    252. V Reuter
    253. W Kimryn Rathmell
    254. L Thorne
    255. L Boice
    256. M Huang
    257. JC Fisher
    258. W Marston Linehan
    259. CD Vocke
    260. J Peterson
    261. R Worrell
    262. MJ Merino
    263. LS Schmidt
    264. P Tamboli
    265. BA Czerniak
    266. KD Aldape
    267. CG Wood
    268. J Boyd
    269. J Weaver
    270. MV Iacocca
    271. N Petrelli
    272. G Witkin
    273. J Brown
    274. C Czerwinski
    275. L Huelsenbeck-Dill
    276. B Rabeno
    277. J Myers
    278. C Morrison
    279. J Bergsten
    280. J Eckman
    281. J Harr
    282. C Smith
    283. K Tucker
    284. L Anne Zach
    285. W Bshara
    286. C Gaudioso
    287. C Morrison
    288. R Dhir
    289. J Maranchie
    290. J Nelson
    291. A Parwani
    292. O Potapova
    293. K Fedosenko
    294. JC Cheville
    295. R Houston Thompson
    296. S Signoretti
    297. WG Kaelin
    298. MB Atkins
    299. S Tickoo
    300. V Reuter
    301. W Marston Linehan
    302. CD Vocke
    303. J Peterson
    304. MJ Merino
    305. LS Schmidt
    306. P Tamboli
    307. JM Mosquera
    308. MA Rubin
    309. ML Blute
    310. W Kimryn Rathmell
    311. T Pihl
    312. M Jensen
    313. R Sfeir
    314. A Kahn
    315. A Chu
    316. P Kothiyal
    317. E Snyder
    318. J Pontius
    319. B Ayala
    320. M Backus
    321. J Walton
    322. J Baboud
    323. D Berton
    324. M Nicholls
    325. D Srinivasan
    326. R Raman
    327. S Girshik
    328. P Kigonya
    329. S Alonso
    330. R Sanbhadti
    331. S Barletta
    332. D Pot
    333. M Sheth
    334. JA Demchok
    335. T Davidsen
    336. Z Wang
    337. L Yang
    338. RW Tarnuzzer
    339. J Zhang
    340. G Eley
    341. ML Ferguson
    342. KR Mills Shaw
    343. MS Guyer
    344. BA Ozenberger
    345. HJ Sofia
    346. Cancer Genome Atlas Research Network
    Nature 499:43–49.
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
  34. 34
  35. 35
  36. 36
  37. 37
  38. 38
  39. 39
  40. 40

Decision letter

  1. Chi Van Dang
    Reviewing Editor; University of Pennsylvania, United States

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

[Editors’ note: this article was originally rejected after discussions between the reviewers, but the authors were invited to resubmit after an appeal against the decision.]

Thank you for submitting your work entitled "Mitochondrial Respiratory Gene Expression is Suppressed in Many Cancers" for consideration by eLife. Your article has been favorably evaluated by a Senior Editor and three reviewers, one of whom, Chi Van Dang, is a member of our Board of Reviewing Editors.

Our decision has been reached after consultation between the reviewers. Based on these discussions and the individual reviews below, we regret to inform you that your work will not be considered further for publication in eLife.

The manuscript by Reznik et al. reports alteration mitochondrial gene expression amongst human cancers, for which they have previously reported in eLife the spectrum of mtDNA alterations relative to normal tissues. Understanding somatic alterations of mitochondrial genome during tumorigenesis is critical for the next generation of cancer therapeutics. Several recent studies based on the TCGA specimens have revealed that mitochondria DNAs are altered in cancers, including mtDNA mutations as well as copy number changes. The authors previously reported that mtDNA copy number depletion in human cancers. In this follow-up study, they analyzed mitochondria gene expression using the TCGA RNA-seq profiles. They presented an expressional landscape of 13 mitochondria genes across >10 cancer types which have normal control specimens from the TCGA, and found that mtRNAs were dysregulated in cancers. This comprehensive mtRNA expression profile could contribute to better understanding both mitochondria biology and tumorigenesis.

However, there is a similar recent paper in Nature Communications on this topic by Frezza and colleagues "Tissue-specific and convergent metabolic transformation of cancer correlates with metastatic potential and patient survival".

Further, the mitochondrial genomic changes documented in this manuscript do not provide any insight into what advantage decreasing respiratory genes will have on tumorigenesis.

Complete loss of the respiratory chain impairs cancer cell proliferation and tumorigenesis. The authors do not observe a complete loss but a suppression of respiratory chain. What advantage does moderate suppression of respiratory chain provide for the cancer cells in vivo?

Additionally, two statements are not correct:

"Our results provide quantitative evidence for suppression of respiration across many cancer types." Any given cell only utilizes 10-30% of maximal respiratory capacity. Thus, a cell can have large drop in mtDNA without respiration being affected. The authors don't have any direct measurements that examine respiration.

"On the other hand, mtDNA depletion will not necessarily lead to a drop in respiration". If cells have complete mtDNA depletion then they will be respiratory deficient. mtDNA encodes for critical subunits of the respiratory chain.

Other omissions that detract from the paper include the following issues:

1) Is there clinical significance of mtRNA dysregulation in cancers? E.g. are the expression levels of mtRNAs correlated with disease stage/grade, patients' age, outcome and other clinical factors?

2) A detailed information for sample numbers of each cancer type needs to be provided, especially for the corresponding control specimens. Relatively small number of normal specimens may create biases for analyses.


Author response

[Editors’ note: the author responses to the first round of peer review follow.]

[…] However, there is a similar recent paper in Nature Communications on this topic by Frezza and colleagues "Tissue-specific and convergent metabolic transformation of cancer correlates with metastatic potential and patient survival".

The reviewers are correct in noting that a recent paper by Gaude and Frezza describes an association between aggressive disease and downregulation of mitochondrial gene expression across many cancers. While the Gaude paper complements our own, the two analyses are completely distinct. Gaude and Frezza use the expression of nuclear-DNA-encoded mitochondrial genes as a surrogate for respiratory gene expression. In contrast, our manuscript is the first (and to our knowledge the only one to date) to quantify and analyze expression of the mitochondrial genome. Publically available data from the TCGA consortium (used by Gaude and Frezza and in dozens of other manuscripts) does not quantify expression of mitochondrial DNA; mitochondrial genome expression data is in fact discarded prior to expression quantification.

We realized that we had not done a sufficiently good job in emphasizing the difference between other publications examining the expression of metabolic genes in cancer samples. We have created a new subsection and included additional text to clarify the novelty of our study with respect to prior work (“Quantifying Expression of mtRNAs Across Cancers). More importantly, because data for mtRNA expression is not available through the Broad Firehose data portal for TCGA data, we have included estimates of mtRNA expression across all 6,614 samples investigated as a supplementary resource with this publication (Supplementary file 2).

Furthermore, our analysis in Figure 5 demonstrates that expression of nuclear-DNA-encoded OXPHOS genes (used by Gaude and Frezza) is not redundant with mtDNA-encoded OXPHOS gene expression. Several cancer types (e.g. breast cancer, BRCA, and bladder cancer, BLCA) exhibit discordant changes in the expression of these sets of genes. These findings, and previous work by others examining mitonuclear coordination of gene expression, suggest that coupling between expression of the mitochondrial and nuclear genome, in addition to the level of gene expression itself, may become dysfunctional in these cancer types. Our data will be essential to those seeking to examine such phenomena.

Further, the mitochondrial genomic changes documented in this manuscript do not provide any insight into what advantage decreasing respiratory genes will have on tumorigenesis.

Complete loss of the respiratory chain impairs cancer cell proliferation and tumorigenesis. The authors do not observe a complete loss but a suppression of respiratory chain. What advantage does moderate suppression of respiratory chain provide for the cancer cells in vivo?

Additionally, two statements are not correct:

"Our results provide quantitative evidence for suppression of respiration across many cancer types." Any given cell only utilizes 10-30% of maximal respiratory capacity. Thus, a cell can have large drop in mtDNA without respiration being affected. The authors don't have any direct measurements that examine respiration.

"On the other hand, mtDNA depletion will not necessarily lead to a drop in respiration". If cells have complete mtDNA depletion then they will be respiratory deficient. mtDNA encodes for critical subunits of the respiratory chain.

The reviewer raises two critical and related points.

We agree that we did not adequately explain our results in the context of physiological mitochondrial respiration. Respirometry experiments can be used to estimate the difference between the basal oxygen consumption rate (OCR) of a cell and the maximal OCR upon decoupling of oxygen consumption from ATP synthesis (e.g. using an uncoupler such as FCCP). These kinds of experiments frequently (e.g. see Brand, Nicholls, Biochemical Journal 2011) demonstrate that cells harbor excess respiratory capacity. The implication, therefore, is that a reduction in functionally respiratory mitochondria in a tumor will not necessarily reduce the basal respiration rate of the tumor. However, in cases where the reduction in respiratory capacity is sufficiently large (e.g. large or complete mtDNA depletion), the basal respiratory rate of the cell will drop, because mtDNA encodes necessary components of the respiratory chain. These changes are now noted throughout the Introduction. We further emphasize in the discussion in subsection “Questions and Future Directions” that a drop in mtRNA expression does not necessarily indicate a drop in respiratory flux, but suggest that because of a consistent decrease in mtDNA copy number, mtOXPHOS expression, and nuOXPHOS expression in 5 cancer types, future experiments may initially focus on these cancers when investigating changes in respiration.

We further agree with the reviewer that a critical question remaining from our study is the selective advantage provided by a potentially moderate suppression of respiration. We are unable to perform in vivo or in vitro experiments to measure how suppression of mtRNAs affects respiratory flux and proliferation/malignancy. Indeed, an adequate investigation into the quantitative link between mtRNA levels and respiratory flux is a project in its own right. Nevertheless, we agree it is important to understand how potential changes in respiratory capacity enable tumors to survive selective pressures. To do so, we used three approaches.

In the first (Results section), we utilized established cancer biology to discuss how the observation of changes in mtRNA expression may arise from known tumorigenic driving mechanisms in two cancer types, clear-cell renal cell carcinoma (KIRC) and chromophobe renal cell carcinoma (KICH). In both of these cancer types, we propose that changes in mtRNA are an adaptive response to “driver” genetic mutations and upstream changes in cell signaling. For example, VHL inactivation in KIRC leads to activation of HIF which upregulates glycolytic genes and suppresses OXPHOS gene expression. The signature of HIF activation is present not only in gene expression data, but also in metabolomic data published by us (Hakimi et al. 2016, Cancer Cell), and suggests that KIRC tumors may experience a bona fide drop in respiration. In particular, KIRC tumors show a distinct metabolomic signature in TCA cycle metabolites, with those metabolites upstream of SDH (succinate dehydrogenase, Complex II), i.e. citrate, cis- aconitate, and succinate, exhibiting increased levels in tumors (compared to normal tissues) and those metabolites downstream of SDH exhibiting decrease levels. One explanation (but perhaps not the only) for this partitioning of changes in metabolite levels is a decrease in respiration and SDH activity, which would induce a metabolic bottleneck and the accumulation (depletion) of metabolites upstream (downstream) of SDH.

In the second approach (subsection “Association of mtRNA with Clinical Parameters”), we leverage the reviewer’s suggestion to discuss our results in light of available clinical data (Comment 1 below). We find that down-regulation of mtRNAs is significantly associated with decreased survival in 5 cancer types, including adrenocortical carcinomas and low grade gliomas. While this analysis does not speak to the mechanism by which changes in mtRNA expression are linked to a selective advantage, it does indicate that a drop in mtRNA expression is relevant, and is associated with more aggressive disease in some cancer types.

In the final approach, we felt it was also important to emphasize that mitochondria (and their metabolic “services”) are required for tumor cells to remain viable and malignant. In subsection “Questions and Future Directions, we briefly discuss the essential metabolic roles of mitochondria beyond respiration (e.g. production of biosynthetic precursors, production of aspartate, redox balance).

Other omissions that detract from the paper include the following issues:

1) Is there clinical significance of mtRNA dysregulation in cancers? E.g. are the expression levels of mtRNAs correlated with disease stage/grade, patients' age, outcome and other clinical factors?

We agree with the reviewer that, in light of the analysis in Gaude et al. (see our first response above), it would be interesting to analyze our results alongside clinical data. We have now analyzed our results with respect to clinical data provided by the TCGA, including age, pathological stage, and overall survival. In particular, we associated the expression of each mitochondrial mtRNA with overall survival (using Cox regression) and age and pathological stage (using non-parametric Spearman correlations). These results are described in Figure 3, subsection “Association of mtRNA with Clinical Parameters”, and in Supplementary file 3, and are largely in line with clinical associations reported in our original study of mtDNA copy number across cancer. Overall, we find that several cancer types (e.g. adrenocortical carcinomas, low grade gliomas, chromophobe renal cell carcinomas) show an association between low expression of mtRNA and poor outcome. We also found several associations with age and stage, including an increase in mtRNA levels in older patients with thyroid tumors.

2) A detailed information for sample numbers of each cancer type needs to be provided, especially for the corresponding control specimens. Relatively small number of normal specimens may create biases for analyses.

This information is now available in Supplementary file 1.


Article and author information

Author details

  1. Ed Reznik

    Center for Molecular Oncology, Memorial Sloan Kettering Cancer Center, New York, United States
    ER, Conception and design, Analysis and interpretation of data, Drafting or revising the article
    For correspondence
    Competing interests
    The authors declare that no competing interests exist.
    Additional information
    Ed Reznik and Qingguo Wang are joint first authors.
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-6511-5947
  2. Qingguo Wang

    Center for Molecular Oncology, Memorial Sloan Kettering Cancer Center, New York, United States
    Present address
    College of Computing and Technology, Lipscomb University, Nashville, United States
    QW, Conception and design, Analysis and interpretation of data, Drafting or revising the article
    Competing interests
    The authors declare that no competing interests exist.
    Additional information
    Ed Reznik and Qingguo Wang are joint first authors.
  3. Konnor La

    Center for Molecular Oncology, Memorial Sloan Kettering Cancer Center, New York, United States
    KL, Analysis and interpretation of data, Drafting or revising the article
    Competing interests
    The authors declare that no competing interests exist.
  4. Nikolaus Schultz

    Center for Molecular Oncology, Memorial Sloan Kettering Cancer Center, New York, United States
    NS, Conception and design, Analysis and interpretation of data, Drafting or revising the article
    For correspondence
    Competing interests
    The authors declare that no competing interests exist.
    Additional information
    Nikolaus Schultz and Chris Sander are joint senior authors.
  5. Chris Sander

    1. Department of Cell Biology, Harvard Medical School, Boston, United States
    2. cBio Center, Dana-Farber Cancer Institute, Boston, United States
    CS, Conception and design, Analysis and interpretation of data, Drafting or revising the article
    For correspondence
    Competing interests
    The authors declare that no competing interests exist.
    Additional information
    Nikolaus Schultz and Chris Sander are joint senior authors.


National Cancer Institute (5U24CA143840)

  • Ed Reznik
  • Nikolaus Schultz
  • Chris Sander

National Institutes of Health (P41GM103504)

  • Ed Reznik
  • Nikolaus Schultz
  • Chris Sander

Marie-Josee and Henry R. Kravis Center for Molecular Oncology

  • Ed Reznik
  • Qingguo Wang
  • Konnor La
  • Nikolaus Schultz

National Cancer Institute (P30-CA008748)

  • Ed Reznik
  • Qingguo Wang
  • Konnor La
  • Nikolaus Schultz
  • Chris Sander

Roberston Foundation

  • Nikolaus Schultz

Prostate Cancer Foundation

  • Nikolaus Schultz

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Reviewing Editor

  1. Chi Van Dang, University of Pennsylvania, United States

Publication history

  1. Received: September 23, 2016
  2. Accepted: December 9, 2016
  3. Version of Record published: January 18, 2017 (version 1)
  4. Version of Record updated: January 19, 2017 (version 2)


© 2017, Reznik et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.


  • 4,196
    Page views
  • 778
  • 42

Article citation count generated by polling the highest count across the following sources: Crossref, Scopus, PubMed Central.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

Further reading

    1. Cancer Biology
    Krystal Ann Orlando et al.
    Research Article Updated

    Small cell carcinoma of the ovary, hypercalcemic type (SCCOHT) is a rare and aggressive form of ovarian cancer. SCCOHT tumors have inactivating mutations in SMARCA4 (BRG1), one of the two mutually exclusive ATPases of the SWI/SNF chromatin remodeling complex. To address the role that BRG1 loss plays in SCCOHT tumorigenesis, we performed integrative multi-omic analyses in SCCOHT cell lines +/- BRG1 reexpression. BRG1 reexpression induced a gene and protein signature similar to an epithelial cell and gained chromatin accessibility sites correlated with other epithelial originating TCGA tumors. Gained chromatin accessibility and BRG1 recruited sites were strongly enriched for transcription-factor-binding motifs of AP-1 family members. Furthermore, AP-1 motifs were enriched at the promoters of highly upregulated epithelial genes. Using a dominant-negative AP-1 cell line, we found that both AP-1 DNA-binding activity and BRG1 reexpression are necessary for the gene and protein expression of epithelial genes. Our study demonstrates that BRG1 reexpression drives an epithelial-like gene and protein signature in SCCOHT cells that depends upon by AP-1 activity.

    1. Cancer Biology
    2. Genetics and Genomics
    Iris E Glykofridis et al.
    Research Article

    Germline inactivating mutations in Folliculin (FLCN) cause Birt–Hogg–Dubé (BHD) syndrome, a rare autosomal dominant disorder predisposing to kidney tumors. FLCN is a conserved, essential gene linked to diverse cellular processes but the mechanisms by which FLCN prevents kidney cancer remain unknown. Here we show that deleting FLCN activates TFE3, upregulating its downstream E-box genes in human renal tubular epithelial cells (RPTEC/TERT1), including RRAGD and GPNMB, without modifying mTORC1 activity. Surprisingly, deletion of FLCN or its binding partners FNIP1/FNIP2 also induces interferon response genes, but independently of interferon. Mechanistically, FLCN loss promotes STAT2 recruitment to chromatin and slows cellular proliferation. Our integrated analysis identifies STAT1/2 signaling as a novel target of FLCN in renal cells and BHD tumors. STAT1/2 activation appears to counterbalance TFE3-directed hyper-proliferation and may influence the immune response. These findings shed light on unique roles of FLCN in human renal tumorigenesis and pinpoint candidate prognostic biomarkers.