Hypoxia leads to reduced mito-nuclear gene expression and increased mtDNA transcriptional pausing in human cells

Abstract

Mitochondria respond to various stresses. Nevertheless, the regulation of this response while considering coordination between mitochondrial (mtDNA)- and nuclear DNA (nDNA)-encoded gene expression has been overlooked. Our RNA-seq analysis of 18 human cell lines grown in hypoxia (0.2-2% oxygen, 16-24 h) reveals a significant and coordinated reduction of mito-nuclear oxidative phosphorylation (OXPHOS) genes’ expression in most (N = 11) cell lines. mtDNA copy number assessment in U87, HCT-116, MCF-7, and HeLa cells reveals non-significant changes, suggesting that the overall reduced mito-nuclear gene expression (MNGE) in hypoxia occurs at the RNA level. Analysis of HIF1α ChIP-seq experiments from cells exposed to hypoxia reveals increased binding to upstream regulatory elements of certain regulators of mitochondrial gene expression. Furthermore, RNA-seq analysis of HIF1α knockout HCT-116 cells grown in hypoxia reveals reduced mtDNA gene expression, yet no change in nDNA OXPHOS genes, suggesting that HIF1α knockout led to departure from coordination of MNGE. Finally, nascent RNA transcripts analysis (PRO-seq) in HeLa, U87, and D407 cells grown in hypoxia shows increased intensity of pausing sites throughout the mtDNA. This finding suggests an important role for transcriptional pausing in the regulation of mtDNA gene expression. Taken together, coordinated reduction of MNGE in hypoxia underlines MNGE as a pivotal player in general mitochondrial function, and particularly in response to stress.

Hypoxia induces a common coordinated reduction of mitochondrial and nuclear OXPHOS gene expression across diverse human cell lines, accompanied by an increased level of mtDNA transcriptional pausing.

Introduction

Mitochondria are cellular hubs of metabolism and energy needs. As such, the mitochondrion readily responds to the dynamic physiological changes in the organism¹, as well as to environmental stresses². Unlike all other cellular pathways in animal cells, mitochondrial activities are operated by a unique and conserved bi-genomic system that demands tight co-evolution and regulatory coordination between elements encoded by the mitochondrial genome (mtDNA) and the nuclear genome (nDNA)^3,4. For example, the energy-producing oxidative phosphorylation system (OXPHOS) is operated by five membrane-bound multi-subunit protein complexes, four of which (i.e., complexes I, III, IV, and V) are encoded by genes that are divided between the two genomes⁵. This bi-genomic system presents a major challenge when considering gene regulation. Firstly, all mitochondrial regulatory factors are transcribed in the nucleus, translated in the cytoplasm, and subsequently imported into the mitochondria⁶. Secondly, unlike the gene-specific transcriptional regulation in the nucleus, mtDNA gene expression retained bacterial-like regulation, including polycistronic transcription⁷. Third, it is currently thought that regulation of mtDNA gene expression is governed by a set of transcription factors that are dedicated to regulate the organelle, suggesting that mtDNA gene expression regulation is separated from the regulation of nDNA gene expression⁸. Therefore, it is not trivial to consider co-expression of mtDNA- and nDNA-encoded mitochondrial genes at the RNA level. Nevertheless, coordinated expression of mtDNA and nDNA-encoded OXPHOS subunits at the transcript level has been discovered across many human tissues^9–11 and in single cells¹², supporting pervasive coordination of mito-nuclear gene expression (MNGE).

Coordination of MNGE is frequently altered in human disorders, and perturbed in several diseases such as Alzheimer’s disease⁵ and in the immune system cells of COVID-19 patients¹³ (reviewed in ref. ⁴). This raises the possibility that environmental challenges to mitochondrial activities will likely impact the coregulation of MNGE. As mitochondrial activities mostly occur in an oxidative environment, it is not surprising that hypoxia interferes with mitochondrial activities^14–16. We thus hypothesized that coordinated MNGE may be affected in cells grown in hypoxia.

Hypoxia affects almost every aspect the mitochondrial function. It induces changes in the composition and concentration of OXPHOS complexes¹⁷, leads to elevated mitophagy and mitochondrial fission¹⁸, and promotes significant changes in mitochondrial morphology¹⁹. While much is known about the cellular and physiological response to hypoxia^16,20,21, its impact on mtDNA gene expression and transcription, as well as the coordination of MNGE across human cell lines and tissues, remains unknown.

Functional genomics techniques pave the path towards highly quantitative assessment of both gene expression (using RNA-seq) and transcription (using precision run-on transcription sequencing—PRO-seq) in a variety of samples. Specifically, we previously analyzed mtDNA transcription in different human cell lines, which allowed precise identification and quantification of strand-specific mtDNA transcriptional initiation, pausing, elongation and termination in all samples analyzed²². As both RNA-seq and PRO-seq are highly sensitive and enable genome-wide analysis of gene expression and transcription, respectively, we set forth to analyze the coordination of gene expression and transcription between the mtDNA and corresponding nDNA-encoded OXPHOS genes and the mitochondrial ribosome genes in response to hypoxia.

To study the effect of hypoxia on MNGE and transcription, we exposed five different human cell lines to hypoxia (1% oxygen, 24 h), followed by genome-wide analysis of nascent RNA transcripts (PRO-seq), steady-state RNA (RNA-seq) and qPCR-based assessment of mtDNA copy number. These experiments were combined with analyses of publicly available RNA-seq experiments from 18 different human cell lines exposed to hypoxia (0.2–2% O₂ for 16–24 h). The impact of hypoxia on MNGE and on mtDNA transcription, as well as on the underlying mechanisms of the response are discussed.

Results

Hypoxia induces a coordinated down-regulation of MNGE across human cell lines

To determine the response of both nDNA- and mtDNA-encoded OXPHOS and mito-ribosome genes to hypoxia, we first generated a database of publicly available RNA-seq experiments from human cell lines exposed to hypoxia (0.2–2% O₂ for 16–24 h). To this end, we screened ENA and GEO for RNA-seq experiments from cells exposed to hypoxia and corresponding normoxic controls (Supplementary Data S1). Only datasets that are based on published results, with well-annotated metadata, and with a minimum of two controls and two hypoxia experiments performed in biological replicates were selected for further analyses. Overall, we analyzed 275 different samples assigned to 25 different RNA-seq datasets representing 18 different human cell lines. Thirteen of the eighteen cell lines had a single dataset available, one cell line had two independently published datasets (HeLa cells), and three cell lines had three RNA-seq datasets (MCF7, HCT-116, and A549 cells). Analysis of the distribution of log₂ fold-change values of hypoxia markers²³ (Supplementary Data S2) revealed elevated expression, thus confirming the response to hypoxia treatment in the analyzed cell lines (Supplementary Fig. S1). Since our goal was to determine the response of structural OXPHOS and mito-ribosome genes to hypoxia, we focused our analysis on datasets that displayed at least 25% differentially expressed genes (DEGs) out of the listed OXPHOS genes per genome (i.e., 22 out of 87 nDNA genes and 3 out of 13 protein-coding mtDNA genes, Supplementary Data S3), which significantly responded to hypoxia as compared to controls (FDR-adjusted p-value threshold below 0.1, as previously performed²⁴). Such datasets were termed hereon as “responders”. Notably, since there are only two mtDNA-encoded mito-ribosome genes (e.g., 12S and 16S rRNA genes), we defined a given dataset as a responder for mito-ribosome analysis only if the expression of both genes was significantly altered. This consideration was applied only to cells in which over 25% of the nDNA-encoded subunits (20 out of 81, Supplementary Data S4) were DEGs. Our results indicated that out of the 25 datasets analyzed, 14 and 6 were considered responders for OXPHOS and mito-ribosome genes, respectively. We observed a coordinated down-regulation of OXPHOS genes in both genomes in 78.5% (N = 11) of the responder datasets (Fig. 1a). Therefore, coordinated MNGE reduction may be regarded as a prevalent response to hypoxia. While considering mito-ribosomal genes, 3 out of 6 responder datasets showed a coordinated response to hypoxia (Fig. 1b). Interestingly, in 10 out of 18 datasets labeled as non-responders while considering the expression of mito-ribosomal genes (see above), over 25% of the nDNA-encoded mito-ribosome genes were significantly down-regulated (Supplementary Fig. S2). Taken together, our results indicate that MNGE downregulation is common to most tested cell lines. Furthermore, down-regulation of nDNA-encoded mito-ribosomal genes is likely a general response to hypoxia.

Fig. 1. Hypoxia induces a coordinated downregulation of OXPHOS MNGE in correlation with hypoxia signal.

Box-plot distributions of the log₂ fold-change per responsive dataset analyzed for OXPHOS (a) and mito-ribosome (b) DEGs. Green and orange fill—the distribution of log₂ fold-change values for nDNA-encoded and mtDNA-encoded genes, respectively. Right panels—percentage of genes that are significantly down-regulated (red), significantly up-regulated (green) or nonsignificant (white). The dot plots on the left in both (a, b) correspond to the oxygen concentrations used during cell growth. RNA-seq data generated by us is marked with an asterisk. c, d Scatter plots demonstrating the correlation between the hypoxia index, calculated according to Lombardi et al.²³ (X-axes). Y axes- log₂ fold-change in expression of c OXPHOS genes and d mito-ribosome genes in response to hypoxia. The data used to generate this figure is available in FigShare as Supplementary Data S9 (https://figshare.com/projects/Shtolz_2025_Mitochondria_Hypoxia_Project/237023).

In addition to the analysis of publicly available RNA-seq experiments, we generated RNA-seq experiments from HCT-116, MCF7, and U87 cells grown in 1% oxygen for 24 h. This data was compared to publicly available datasets generated from the same cell lines and conditions. Similar to the publicly available datasets, analysis of the distribution of log₂ fold-change values of hypoxia markers revealed upregulation, thus confirming the response to hypoxia treatment in both our and publicly available experiments (Supplementary Fig. S3a, b). Secondly, OXPHOS and mito-ribosome genes revealed reduced MNGE. These findings were significantly correlated between our newly generated HCT-116 and MCF7 data, and their corresponding publicly available datasets (Supplementary Fig. S3c, d). However, this was not the case for U87 cells (Supplementary Fig. S3e). It is worth noting that analysis of two publicly available U87 datasets²⁵ (Supplementary Data S5) showed exceedingly high correlations (R = 1, p < 10⁻¹⁰⁰), suggesting that these public datasets are likely technical rather than biological replicates. We thus excluded the latter datasets from further analysis.

We noticed that, although our analyzed RNA-seq datasets were exposed to similar hypoxia conditions, the sensitivity and intensity of the response to hypoxia varied between the tested cell lines^21,26. To measure the relationship between the magnitude of the response to hypoxia and the MNGE, we utilized a hypoxia index metric that was previously defined based on a metagene expression signature²³. Specifically, out of the multiple available hypoxia meta-analyses, which considered a variety of different experiments and cell types^27–30, we chose to utilize a recent metagene list published by Lombardi et al.²³, as it was generated based on a comprehensive re-analysis and verification of both RNA-seq and HIF1α ChIP-seq data. We observed a significantly negative Spearman correlation between the position-weighted mean log₂ fold-change values of the 48 hypoxia markers (herein termed “hypoxia index”, Supplementary Data S2) and the mean values of both OXPHOS (ρ = −0.63, p = 0.0008, Fig. 1c, Spearman’s correlation) and mito-ribosomal genes (ρ = −0.69, p = 0.0001, Fig. 1d, Spearman’s correlation). Taken together, these results indicate that an increasingly stronger hypoxia response is associated with a stronger downregulation of MNGE.

The coordinated downregulation of OXPHOS genes’ expression is part of a broader response to hypoxia

The observed significant association between the hypoxia index and the magnitude of MNGE down-regulation suggests that mitochondrial response to hypoxia may not be limited to OXPHOS and the mito-ribosome^14,16,31. Therefore, we assessed the response of mitochondrial metabolism-related pathways to hypoxia, and hence their association with the observed coordinated MNGE downregulation. Indeed, most of the DEGs related to the tri-carboxylic acid cycle (TCA, GO:0006099) and the fatty acid oxidation (GO:0019395) pathways displayed downregulation across all tested cell lines (Fig. 2a, b). In consistency with this finding, previous reports showed that lower oxygen concentrations trigger a catabolic shift from aerobic to anaerobic energy production, through glycolysis and fermentation¹⁶. Indeed, most DEGs related to glycolysis (GO:0006110) were upregulated (Fig. 2c), thus supporting the catabolic response to hypoxia. Furthermore, the expression of LDHA, a lactate dehydrogenase isoform that thermodynamically favors the conversion of pyruvate to lactate, and is more abundant in glycolytic conditions, was significantly upregulated in hypoxia compared to the OXPHOS-associated LDHB isoform³¹ (P < 0.001, Fig. 2d). We interpret these results to mean that hypoxia leads to a general metabolic response, as reflected by the similarly significant response of all tested mitochondrial pathways.

Fig. 2. The coordinated down-regulation of OXPHOS MNGE is associated with a general mitochondrial metabolic shift.

Box-plots of the distributions of log₂ fold-change in gene expression between hypoxia and normoxia per responsive dataset for all DEGs in the TCA cycle (a), the fatty acid oxidation pathway (b), and glycolysis (c). The bars next to each box-plot (same row) show the number of genes that are significantly down-regulated (red), significantly up-regulated (green), or non-significant (white). The dot plots on the left panels a–c show the oxygen concentration used while growing each of the tested cell lines. RNA-seq data generated by us is marked with an asterisk. d The distributions of the log₂ fold-change values of LDHA and LDHB lactate dehydrogenase isoforms across all tested cell lines. e Scatterplots showing the correlation between the log₂ fold-change of gene expression per pathway (x-axis), and the magnitude of mito-nuclear score for OXPHOS genes (y-axis). Spearman’s rho values and p-values are shown above the plots. The p-values were calculated using a two-sided Mann–Whitney test. P-value ranges are marked by asterisks: *0.01 < p ≤ 0.05, **0.001 ≤ p < 0.01, ***0.0001 < p ≤ 0.001, ****p ≤ 0.0001. The data used to generate this figure is available in FigShare as Supplementary Data S9 (https://figshare.com/projects/Shtolz_2025_Mitochondria_Hypoxia_Project/237023).

To test whether cell lines that exhibit a stronger metabolic response to hypoxia also display a tighter coordination of MNGE, we quantified the magnitude of coordinated MNGE in OXPHOS genes as a score (hereon termed “mito-nuclear score”). The mito-nuclear score was designed to capture the combined proportions of OXPHOS genes in both genomes, which displayed a significantly altered expression (for additional information—see “Materials and Methods”). Interestingly, the median log₂ fold-change of DEGs related to fatty acid oxidation is significantly negatively correlated with the mito-nuclear score (ρ = −0.58, p= 0.029). Genes related to the TCA cycle show a strong, but non-significant correlation to the mito-nuclear score (ρ = −0.43, p = 0.126). Hence, stronger coordination of MNGE is associated with lower expression of DEGs within these two pathways (Fig. 2e). Overall, these results indicate that cell lines with higher levels of coordinated MNGE are more prone to lower expression of fatty acid oxidation and TCA genes.

Reduction in OXPHOS MNGE in MCF7, HCT-116, HeLa, and U87 cells does not clearly associate with changes in mtDNA copy number

Unlike the nDNA, the mtDNA resides in thousands of copies per cell, and, according to previous reports, variable copies per-mitochondrion^32–34, and is transcribed in a polycistronic fashion⁷. Therefore, expression of mtDNA-encoded genes may be caused by a reduction in transcription, post-transcriptional processing rates, mtDNA copy number, or any combination of the above³⁵. To differentiate between these options, we measured the expression response of genes known to be involved in mitochondrial transcription (GO:0006390), RNA processing (GO:0000963) and replication (GO:0032042, Fig. 3a–c). We found that genes belonging to all three pathways are consistently downregulated in response to hypoxia in all cell lines. Notably, genes associated with mtDNA replication showed the least consistent response (Fig. 3c). To further investigate this phenomenon, we assessed mtDNA copy number in hypoxia versus control cells in four different cell lines (U87, MCF7, HCT-116, and HeLa) that were incubated for 24 h in 1% oxygen. Following this treatment, qPCR was employed for DNA extracts to quantify the relative mtDNA copy number normalized by several nDNA loci (JUN and ANKRD37, 18S rRNA or ACTB). While considering all four cell lines, mtDNA copy numbers were not consistently or significantly altered following exposure to hypoxia (Fig. 3d–g). This indicates that, at least in the tested cell lines, the observed reduction in mtDNA-encoded genes’ expression may not be the result of an overall reduction in the number of mtDNA per cell. TFAM is a major player in mtDNA packaging, as in high cellular concentrations it associates with tightly packaged mtDNA molecules, which attenuates mtDNA transcription^36,37. Therefore, we analyzed TFAM mRNA expression relative to mtDNA copy number in MCF-7 and HCT-116 cells. TFAM expression did not show any significant correlation with mtDNA copy number (ρ = −0.02, p = 0.93). In addition, the ratio between TFAM mRNA and relative mtDNA copy number did not significantly change in hypoxia (Supplementary Fig. S4a). These results imply that mtDNA packaging does not significantly change in hypoxia. Nevertheless, we acknowledge that future protein-level analysis would be of interest to assess TFAM-mtDNA stoichiometry. Our observed reduction in mitochondrial gene expression could be explained by loss of mitochondrial mass through enhanced expression of the mitophagy pathway genes. To test for this possibility, we measured the expression of all 60 mitophagy genes (GO:0000423) across our RNA-seq database. Our analysis indicates that mitophagy genes display only modest and inconsistent expression change in hypoxia (Supplementary Fig. S4b), with no evidence for coordinated pathway activation across the tested datasets. This indicates that enhanced mitochondrial removal is unlikely to account for the reduced mtDNA gene expression in hypoxia.

Fig. 3. Down-regulation of mtDNA-encoded genes in hypoxia is likely caused by a reduction in transcript levels rather than by altered mtDNA copy number.

Boxplots of the distributions of log₂ fold-change in gene expression (x axis) between hypoxia and normoxia per cell line (y axis) for all DEGs involved in mitochondrial transcription (a), mitochondrial RNA processing (b), and mitochondrial replication (c). In a–c, right panel—percentage of genes that are significantly down-regulated (red), significantly up-regulated (green) or nonsignificant (white). The dot plots (a–c) on the left corresponds to the oxygen concentrations used during cell growth. RNA-seq data generated by us is marked with an asterisk. Mean relative copy number in HeLa (d), U87 (e), MCF-7 (f) and HCT-116 (g) cells grown in hypoxia compared to controls. The relative copy number values were calculated as previously performed⁶⁶. The p-values were calculated using a two-sided Mann–Whitney test. The data used to generate a–c are available in FigShare as Supplementary Data S9. The data used to generate d–g are available as Supplementary Data S11 (https://figshare.com/projects/Shtolz_2025_Mitochondria_Hypoxia_Project/237023).

Recent studies have identified KLF2 and ETV1 as transcriptional repressors of mitochondrial biogenesis, with KLF2 acting upstream to regulate ETV1 expression³⁸. To test whether these factors contribute to OXPHOS downregulation in hypoxia, we analyzed their expression and activity across our hypoxia database. While ETV1 showed variable response to hypoxia (mean LFC = 0.01 ± 0.34), KLF2 displayed a more consistent upregulation (mean LFC = 0.4 ± 0.89, Supplementary Fig. S5a), being significantly upregulated in 2 of 14 responsive datasets. Furthermore, analysis of either all or mitochondria-related KLF2 target genes revealed predominantly downregulated expression patterns across cell lines, consistent with KLF2’s repressive function (Supplementary Fig. S5b, c). Additionally, we observed significant negative Spearman correlations between both KLF2 and ETV1 expression levels and mtDNA-encoded gene downregulation (ρ = −0.60 and −0.63, p = 0.023 and 0.039, respectively, Supplementary Fig. S5d). This provides an interesting mechanistic insight into the contribution of these repressive factors to coordinated mitochondrial gene suppression in hypoxia. While KLF2/ETV1 upregulation cannot fully explain the variable penetrance of MNGE coordination across cell types, these findings suggest that cell-type-specific activation of this repressive axis may partially account for the heterogeneous OXPHOS responses observed in different cellular contexts during hypoxia. Overall, our results show that the coordinated reduction of MNGE is more likely to be caused by a regulatory response at the RNA level.

Identification of candidate regulatory factors involved in the altered MNGE in hypoxia

The coordinated down-regulation of MNGE in response to hypoxia argues for an underlying regulatory mechanism. Therefore, we considered the possible involvement of the hypoxia inducible factor (HIF) family^39,40 in the observed altered MNGE in hypoxia. To this end, we analyzed publicly available HIF1α ChIP-seq experiments in four cell lines (H460, A549, HCT-116, and RKO) exposed to hypoxia (24 h of 1% oxygen, PRJNA606242)⁴¹, as well as RNA-seq datasets from HIF1α knockout HCT-116 cells exposed to hypoxia (24 h of 1% oxygen, PRJNA607557)⁴¹. If HIF1α was to directly regulate the MNGE, we would expect it to (A) preferentially bind upstream regulatory elements of nDNA-encoded OXPHOS and mito-ribosome genes; and that (B) these putative binding signals would significantly intensify in response to hypoxia. To test this hypothesis, following normalization, the above-mentioned HIF1α binding peaks were compared between hypoxia and control samples (for more information—see “Materials and Methods”). Firstly, while considering the top 1000 genes ranked based on their log₂ fold-change in differential binding upon exposure to hypoxia, we found significant enrichment for known HIF1α target genes and hypoxia markers (Supplementary Fig. S6). This indicated that our analysis captured the expected HIF1α binding response to hypoxia³⁹. However, only 3 of 88 nDNA-encoded OXPHOS genes and 1 of 78 nDNA-encoded mito-ribosome genes contained upstream HIF1α binding sites whose binding strength significantly increased in response to hypoxia (Supplementary Data S6). This suggests that HIF1α does not likely regulate nDNA-encoded OXPHOS and mito-ribosome genes directly. Thus, to identify candidate transcription factors (TFs) and/or RNA-binding proteins (RBPs) that could explain the observed reduced MNGE in response to hypoxia, we used our RNA-seq experimental dataset assembly and tested for the expression pattern of a list of all 1640 annotated human TFs⁴² and 1678 RBPs (Supplementary Data S7 and GO:0003723, respectively). Out of these genes, 27% (N = 901) displayed significantly altered expression in over 50% of the responder datasets (Supplementary Fig. S7a), and 75 of these genes contain upstream HIF1α differential binding sites in at least one of the four tested cell lines (Supplementary Fig. S7b). Four of these factors (MTERF3, HSPA9, LONP1, and IBA57) are also independently predicted to localize to the mitochondria by TargetP⁴³ and MitoCarta⁴⁴, and are hence candidates to participate in regulating mtDNA gene expression. Specifically, mitochondrial transcription termination factor 3 (MTERF3) and mitochondrial LON peptidase 1 (LONP1) are the only two genes that showed a significant change in the magnitude of the hypoxia response between datasets defined above as responders (i.e., datasets with over 25% DEG of the OXPHOS genes) and non-responders (Fig. 4a, b). Hence, MTERF3 and LONP1 constitute the best candidates to participate in the HIF1α-dependent response of mtDNA gene expression to hypoxia. To investigate whether HIF1α is crucial for maintenance of coordinated MNGE, we analyzed RNA-seq data from HCT-116 cells with a homozygous deletion of HIF1α, which were either exposed to 1% oxygen for 24 h or remained in normoxic conditions (GSE68297)⁴⁵. As expected, known hypoxia markers have significantly reduced expression response to hypoxia in HIF1α^−/− cells as compared to W.T (Supplementary Fig. S8a). Unexpectedly, HIF1α knockout by itself (i.e., without the effect of hypoxia) caused a significant reduction in the expression of 10 out of 13 mtDNA-encoded OXPHOS genes, but elevated expression of most nDNA-encoded OXPHOS and mito-ribosome DEGs (Supplementary Fig. S8b). Additionally, the deletion significantly altered the hypoxia response of nDNA-encoded OXPHOS and mito-ribosome genes but not that of mtDNA-encoded genes (Fig. 4c, d), suggesting that HIF1α is essential for the coordinated MNGE response to hypoxia in most responsive cell lines (Fig. 1a). However, the finding that mtDNA-encoded genes were still down-regulated in response to hypoxia in HIF1α^−/− cells hints that mtDNA gene expression regulation occurs via a HIF1α-independent pathway.

Fig. 4. HIF1α may indirectly regulate the MNGE response of OXPHOS and mito-ribosome genes to hypoxia.

Boxplots showing the log₂ fold-change distribution of LONP1 (a) and MTERF3 (b) in responder vs. non-responder datasets (per-dataset). Boxplots showing the log₂ fold-change distributions of nDNA-encoded (c) and mtDNA-encoded (d) OXPHOS genes’ expression between hypoxia and control for HIF1α^−/− HCT-116 cells exposed to 1% oxygen for 24 h, and W.T. HCT-116 cells exposed to either 1% or 0.5% oxygen for 24 h. e Heatmap showing the log₂ fold-change of candidate regulatory factors’ expression that are involved in the regulation of mitochondrial genes, across all tested datasets. The top bar shows the cell line studied per dataset (columns). The left column indicates mitochondrial localization source (i.e., MitoCarta experiments and TargetP prediction) for each gene (rows) (red – true, gray – false). The p-values were calculated using a two-sided Mann–Whitney test. P-value ranges are marked by asterisks: *0.01 < p ≤ 0.05, **0.001 ≤ p < 0.01, ***0.0001 < p ≤ 0.001, ****p ≤ 0.0001. The data used to generate (a, b, e) is available in FigShare as Supplementary Data S9. The data used to generate c, d are available as Supplementary Data S13 (https://figshare.com/projects/Shtolz_2025_Mitochondria_Hypoxia_Project/237023).

As a complementary approach to discover regulators, that may affect MNGE without necessarily localizing exclusively to the mitochondria, we assembled a list of genes (N-121) that were reported to be essential for OXPHOS function by a genome-wide CRISPR screen⁴⁶, along with TFs that have been previously found to be localized both to the mitochondria and to the nucleus (Supplementary Data S8)^47–49. Out of these 121 genes, 49% (N = 59) displayed significantly altered expression in over 50% of the datasets (Supplementary Fig. S9), of which only 20 genes had significantly altered expression between cell lines that we defined as responders and non-responders (Fig. 4e). Two of these genes (JUN and JUND, Supplementary Fig. S10a, b) are TFs that primarily act within the nucleus, but have been previously found to localize to the mitochondria and bind the mtDNA in human cells⁴⁹. Additionally, only the expression of JUN (i.e., c-Jun) does not significantly change in HIF1α^−/− HCT-116 cells (log₂ fold-change = 0.34, adjusted p-value—N.S.), thus making it the best candidate to mediate the hypoxia response of OXPHOS genes in the two genomes in a HIF1α-independent manner. Overall, we show that although HIF1α is essential for the coordinated MNGE response to hypoxia, mtDNA-encoded genes respond to hypoxia even without HIF1α, possibly through bi-localized transcription factors such as c-Jun.

A unique mtDNA transcriptional signature in hypoxia

The reduced expression of OXPHOS genes in the nucleus and in the mitochondria reflects changes in the steady-state RNA levels of these transcripts. This alludes to the possible impact of hypoxia on three different regulatory aspects: transcription, RNA decay, and mitochondrial biogenesis. Since we did not observe any significant changes in mtDNA copy number, and because genes related to all three aforementioned pathways were down-regulated in response to hypoxia, we have yet to decipher the relative contribution of each of these gene expression regulatory levels. Hence, we measured transcriptional response to hypoxia using PRO-seq⁵⁰ in three different cell lines (HeLa, U87, and D407) grown in 1% oxygen for 24 h. The quality control and processing pipeline PEPPRO⁵¹ was used to quantify the transcriptional response of nDNA-encoded genes to hypoxia while focusing on gene bodies. While considering 48 hypoxia markers (Supplementary Data S2), the transcription of 62%, 36% and 57% of the genes was significantly increased in HeLa, U87, and D407 cells, respectively (Supplementary Fig. S11a). This shows, as expected, that the molecular signature of hypoxia can also be detected at the transcriptional level. However, the transcription of nDNA-encoded OXPHOS and mito-ribosome genes did not significantly alter in hypoxia treatment (Supplementary Fig. S11b, c). That is in contrast to our observed significant down-regulation at the steady-state RNA level in corresponding RNA-seq data from the same cell lines (Fig. 1a). This demonstrates that, while considering nDNA-encoded OXPHOS genes, downregulation likely occurs post-transcriptionally.

While considering mtDNA transcription, U87 and HeLa cells displayed a very similar pattern in cells exposed to hypoxia: elevated levels of mtDNA transcriptional initiation, in both strands, followed by higher read coverage in hypoxia until mtDNA position ~6000 and ~9000 in the heavy and light strands, respectively. Transcription in these mtDNA regions revealed a marked shift in read coverage, which was lower than control until termination (Fig. 5a, b). Interestingly, D407 cells displayed a different pattern (Fig. 5c): heavy strand transcription was only slightly elevated around the 16S rRNA gene and light strand transcription was higher in hypoxia from initiation until termination around position 3200, suggesting that hypoxia-specific transcriptional patterns in the mtDNA may differ among cell types and origins. Taken together, since the unique mtDNA transcriptional pattern was not shared by nDNA-encoded genes, it likely reflects a separate phenomenon. The latter could be explained by two alternative mechanisms: (A) Since mtDNA transcription is regulated by nDNA-encoded factors, the response of the latter may explain the unique mtDNA transcriptional pattern in hypoxia. (B) Hypoxia may induce a reduction in the rate of mtDNA transcription due to reduced rates of the mtDNA transcriptional machinery rather than reduced expression of its protein components.

Fig. 5. Hypoxia leads to an altered mtDNA transcriptional pattern.

PRO-seq read coverage patterns in HeLa (a), U87 (b), and D407 (c) cells exposed to 24 h of 1% oxygen (hypoxia) as compared to control. The control and hypoxia samples are in red and gray for the heavy strand, respectively, and in blue and light-green for the light strand, respectively. The direction of transcription for the heavy strand is marked by a red arrow and for the light strand by a blue arrow. The data used to generate this figure is available in FigShare as Supplementary Data S14 (https://figshare.com/projects/Shtolz_2025_Mitochondria_Hypoxia_Project/237023).

To test for the above-mentioned second possibility, we further utilized our PRO-seq results, as it essentially captures transcriptionally engaged RNA polymerase at a base-pair resolution. Accumulation of RNA polymerase within certain nDNA loci (such as near transcription start sites) is generally interpreted as transcriptional pausing⁵². Within nuclear loci, pausing sites are identified based on the ratio of read-end coverage depth between transcription start sites and the downstream gene bodies. However, since 93% of the mtDNA is coding, a different approach is required. We previously used PRO-seq to identify mtDNA pausing sites and discovered a particularly strong and consistent site occurring shortly downstream of the light strand promoter²². Recently, it has been suggested that pausing sites are found throughout the human mtDNA, the rates of which are increased upon stabilization of non-B DNA structures^35,53. MtDNA PRO-seq read coverage is increased in hypoxia, even though RNA-seq coverage is reduced in hypoxia (Figs. 1a and 5). We thus hypothesized that the increased PRO-seq read coverage may not indicate elevated mtDNA transcription but rather reflects a reduced rate of mtDNA transcription due to increased rates of transcriptional pausing. To test this hypothesis, we first defined pausing sites as mtDNA positions where the 3′ read coverage is larger by three standard deviations than the surrounding 201 bp window in over 50% of samples per cell line (see “Materials and Methods”). Although we did not identify a significant difference in the number of pausing sites in the heavy strand, analysis of the light strand revealed significantly more pausing sites in samples exposed to hypoxia, with an average of 217 ± 11 compared to 162 ± 16 pausing sites in hypoxia vs. control experiments, respectively, in all cell lines tested (P = 0.00827, t-test, Fig. 6a). Additionally, the magnitude of pausing, defined as the normalized 3′ read-ends coverage, was significantly higher in hypoxia in both strands of U87 and HeLa cells and in the light strand of D407 cells as compared to corresponding controls (Fig. 6b–d). Hence, hypoxia likely leads to an increase in mtDNA transcriptional pausing, which may slow the general rate of mtDNA transcription and play a part in the lower gene expression observed for mtDNA genes (Fig. 1a, b).

Fig. 6. Hypoxia leads to a significant increase in pausing sites prevalence and intensity.

a Bar plots showing the distribution of the number of pausing sites for all tested cell lines in the heavy and the light strands, separately. SEM is indicated. b–d Box-plots showing the distribution of the reads per-million (RPM, Y axes) normalized coverage of 3′-ends of nascent RNA transcripts for the heavy strand (left panel) and light strand (right panel) in HeLa (b), U87 (c), and D407 (d) cells. The p-values were calculated using a two-sided Mann–Whitney test. P-value ranges are marked by asterisks: *0.01 < p ≤ 0.05, **0.001 ≤ p < 0.01, ***0.0001 < p ≤ 0.001, ****p ≤ 0.0001. The data used to generate (a) are available in FigShare as Supplementary Data S15. The data used to generate (b–d) are available as Supplementary Data S16 (https://figshare.com/projects/Shtolz_2025_Mitochondria_Hypoxia_Project/237023).

Discussion

In this study, we demonstrate that hypoxia induces a coordinated reduction in MNGE of OXPHOS genes across a diverse compendium of human cell lines, including both cancerous and primary cells. This finding reveals a common, and regulated, mito-nuclear regulatory response to low oxygen levels. This pattern was strengthened in correlation with the level of response of hypoxia markers, implying that the stronger the effect of hypoxia on cells, the stronger the response of MNGE. Nevertheless, although we observed reduced expression of nDNA-encoded mito-ribosome genes in response to hypoxia, this phenomenon was less apparent when considering the expression of the two mtDNA-encoded rRNA genes. We believe that any conclusion regarding coordination of MNGE in the mitochondrial ribosome should await future trustworthy gene expression analysis of the 22 mtDNA-encoded tRNA genes in response to hypoxia, once ample microRNA-seq libraries are generated from hypoxia-treated cells⁵⁴. Furthermore, analysis of the TCA cycle and fatty acid oxidation pathways revealed reduced gene expression, which was correlated with the coordinated reduction of MNGE, suggesting that increased severity or length of exposure to hypoxia would elicit a sharper coordination of MNGE response. Moreover, genes involved in mtDNA regulation, including transcription, RNA processing, and DNA replication, were generally down-regulated in hypoxia. The observed reduction in mtDNA gene expression was not linked to consistent changes in mtDNA copy number or to the expression of mitophagy pathway genes, suggesting that changes in MNGE were modulated at the RNA level rather than by general mitochondrial biogenesis. Further supporting the role of RNA regulation, the levels of mtDNA copy number did not associate with changes in the expression of TFAM, which, when elevated, may indicate mtDNA hyper packaging³⁷.

Due to its central role in the cellular response to hypoxia, it is tempting to hypothesize that our observed coordinated MNGE response is mediated by HIF1α. This notion is partially supported by a recent report, which claimed that HIF1α binds the mtDNA and represses mtDNA gene expression⁵⁵, and by the well-documented effects of HIF1α on glycolysis and OXPHOS genes in the nucleus^39,40,56. Our analysis of HIF1α knockout in HCT-116 cells revealed loss of response of nDNA-encoded OXPHOS genes expression, yet retained reduction of mtDNA genes expression in hypoxia. These findings suggest that loss of HIF1α in these cells led to the departure from coordination of MNGE in hypoxia. Hence, HIF1α is essential for the coordinated response of MNGE to hypoxia. Since analysis of ChIP-seq data did not reveal altered HIF1α binding to upstream regulatory elements of nDNA-encoded OXPHOS genes, we suspected that the involvement of HIF1α in coordinated MNGE is likely indirect. In support of this interpretation, we found two known mitochondrial regulators, MTERF3 and LONP1, whose upstream regulatory elements are differentially bound by HIF1α in response to hypoxia, which is also associated with significantly altered expression. Both genes code for well-known regulators of mitochondrial function. Indeed, the elevated expression of LONP1 in hypoxia may explain the down-regulation of mtDNA-encoded genes via its protease-associated regulation of mtDNA transcription activators such as TFAM⁵⁷. However, the response of MTERF3, which is down-regulated in association with the reduced OXPHOS gene expression, is less intuitive since it is a mitochondrial transcription termination factor, and is considered a negative regulator of mtDNA transcription^58,59. Therefore, the involvement of MTERF3 in the response of mtDNA-encoded genes to hypoxia deserves further attention in the future.

Since mtDNA genes’ expression remained downregulated in response to hypoxia even in HIF1α knockout cells, it is plausible that such a response is HIF1α independent. With this in mind, we screened for additional candidates of MNGE regulation in hypoxia, and our attention was drawn to c-Jun. The expression of JUN (c-Jun) was not only up-regulated in hypoxia across most tested cell lines, but it was previously identified both in the mitochondria and in the nucleus, and was found to directly bind the mtDNA⁴⁹. Furthermore, out of all candidate regulators, c-Jun was the only one whose expression was not affected by the knockout of HIF1α, at least in HCT-116 cells. These findings underline c-Jun as a promising candidate regulator of MNGE coordination, especially in hypoxia. This interpretation requires future experimental manipulations of c-Jun in hypoxia. Additionally, we found that expression changes of the mitochondrial biogenesis repressors KLF2 and ETV1 were significantly correlated with MNGE responses across cell lines, suggesting that these biogenesis repressors may also contribute to coordinated mitochondrial gene downregulation in hypoxia. Notably, regulation of TF that may affect OXPHOS and/or mito-ribosome gene expression does not occur only through changes in their own expression, but also post-transcriptionally. This should be assessed in the future.

Using nascent RNA sequencing, we observed a unique mtDNA transcriptional signature in hypoxia, characterized by increased transcriptional initiation and elongation immediately downstream of both heavy-strand and light-strand promoters, followed by reduced transcription (relative to control) downstream. However, since nascent RNA sequencing methods essentially capture the location of the RNA polymerase at the time of the experiment^50,60, peaks in read coverage, especially of the 3′-ends of the RNA, are frequently interpreted as transcriptional pausing⁵². Indeed, a prominent pausing site has been previously observed approximately 50 bases downstream of the human mtDNA light strand promoter (LSP)²². Recently, using a similar computational method, pausing sites have been identified throughout the human mtDNA and in association with G-quadruplex motifs⁵³. Accordingly, we identified pausing sites throughout the mtDNA and found that the number of pausing sites and their intensity increased in hypoxia. This interpretation provides an interesting explanation for the down-regulation of mtDNA gene expression that we observed in RNA-seq experiments. With this in mind, it is conceivable that this mechanism selectively leads to an overall reduction in mtDNA precursor transcripts and subsequent reduction in the levels of steady-state mitochondrial mRNA. Based on the association of G-quadruplex motifs with mtDNA transcriptional pausing sites, we speculate that direct manipulation of POLRMT pausing, perhaps via stabilization of G-quadruplex structures, may alter mtDNA-encoded gene expression in response to hypoxia. In summary, our study revealed a reduced MNGE in response to hypoxia, which is shared across a variety of cell types. This enabled us to identify candidate regulators of MNGE. Secondly, we found that hypoxia leads to a uniquely altered mtDNA transcriptional pattern, most probably by increasing pausing events. The latter suggests that hypoxia entails a reduced rate of mtDNA transcription, which calls for further investigation. These results underline MNGE as a common mechanism, especially in the frame of mitochondrial response to stresses, thus raising interest in the profound contribution of mito-nuclear coregulation to evolutionary events and to the molecular basis of mitochondrial diseases.

Materials and methods

Cell growth

HeLa, U87, MCF7, HCT-116, and D407 cells were grown in cell culture flasks (50-809-259, Greiner Bio-One, Kremsmünster, Austria) in a Forma Steri-Cycle CO₂ incubator (Thermo Fisher Scientific, Waltham, MA) at 37 °C, 5% CO₂. HeLa, U87 and MCF7 cells were cultured in Dulbecco’s Modified Eagle Medium (01-052-1 A DMEM, Sartorius, Göttingen, Germany) supplemented with 10% v/v fetal bovine serum (FBS, F7524, Merck, Darmstadt, Germany), 2% v/v L-glutamine 29.2 mg/ml (03-020-1B, Sartorius, Göttingen, Germany), 1% v/v pen-strep (penicillin 10,000 units/ml, streptomycin 10 mg/ml, L0022 Biological Industries, Beit HaEmek, Israel) and 1% v/v MEM-Eagle non-essential amino acid solution (01-340-1B, Sartorius). HCT-116 cells were grown in McCoy’s 5A (Modified) Medium (16600082, manufactured by Thermo Fisher Scientific, imported by Rhenium, Modi’in-Macabim-Re’ut, Israel). Human D-407 retinal pigment epithelium cells, a kind gift from R.C. Hunt⁶¹, were grown using DMEM medium (01-052-1 A DMEM, Sartorius) supplemented with 3% FBS (F7524, Merck), 2 mM L-glutamine (03-020-1B, Sartorius), 1% v/v pen-strep (L0022, Biological Industries). The cells were split at around 90% confluency, and the hypoxia experiment (see below) started when the cell concentration reached 10⁶–10⁷ per ml and the total cell count exceeded 10⁹ cells for PRO-seq procedures and 10⁸ for RNA-seq and mtDNA copy number measurements.

Hypoxia treatment

After the cells reached the aforementioned concentrations, and below 90% confluency, half of the cells were transferred to a CO₂ incubator (Wiggens WCI-180, Straubenhardt, Germany), designed to maintain a humidified environment in 1% O₂, 5% CO₂, and 94% N₂ for all experiments. All treated cells were incubated for 24 h. Following exposure, the cells were immediately harvested and pooled for subsequent applications. To minimize oxygen exposure of the hypoxic cells during extraction, extra care was taken to keep the flask lids closed when removed from the incubator, followed by prompt cell harvesting.

RNA and DNA extraction

To study gene expression and mtDNA copy number, RNA and DNA were extracted from both hypoxia-treated and control cells. Following harvesting, the cell pellets were split so that a quarter of the cells (~1–2.5 × 10⁶ cells) were kept in liquid nitrogen for subsequent DNA extraction, and the rest of the cells (~5–7.5 × 10⁶ cells) were used for total RNA extraction. Total RNA was extracted using the Direct-zol™ RNA Miniprep kit (R2050, Zymo Research, Irvine, CA) with DNase I treatment. Genomic DNA was extracted from the above-mentioned frozen cell pellet, using the genomic DNA extraction kit by Real Genomics (YGT50, RBC Bioscience, New Taipei City, Taiwan) with RNase treatment for HeLa and U87 cells and using Exgene™ Cell SV Mini (106-101, GeneAll Biotechnology, Songpau-gu, Seoul, Korea) with RNase treatment for HCT-116 and MCF7 cells.

Processing of RNA-seq data

Publicly available RNA-seq data analyzed in this study were retrieved from the ENA and GEO databases (Supplementary Data S1). Overall, we processed 275 different human samples belonging to 25 different datasets representing 18 different human cell lines. The samples were manually curated to remove samples that contained any manipulation other than hypoxia. Next, the trimmed reads (trimming was identical to the PRO-seq analysis described above) were uniquely mapped using STAR aligner⁶² (v2.7.11) against the primary assembly of the Gencode release of the human reference genome sequence (GRCh38)⁶³. Gene-level counts were obtained using HTseq⁶⁴ (v0.6.1) with the appropriate GRCh38 basic gene annotation GFF3 (Gencode) annotation file provided as input. Lastly, differential gene expression was evaluated using the R package DESeq2 (v1.46)⁶⁵. The per-gene log₂ fold-change and p-value table, and the DESeq2-normalized expression data are available as Supplementary Data S9 and S10, respectively, in FigShare (https://figshare.com/projects/Shtolz_2025_Mitochondria_Hypoxia_Project/237023).

Quantification of relative mtDNA copy number using qPCR

To quantify the relative mtDNA copy number, qPCR (Bio-Rad CFX96) was performed using SYBR Green (4309155, Applied Biosystems, Thermo Fisher Scientific) on genomic DNA samples with two primer pairs targeted to amplify two nDNA genes (JUN and either ANKRD37 or ACTB) and four primer pairs targeted to amplify four mtDNA fragments selected to represent four mtDNA regions (Table 1). Each sample and amplicon combination was analyzed in triplicate for four biological replicates of hypoxia experiments per cell line. The relative mtDNA copy number was quantified based on previously published formulae⁶⁶:

$$\Delta C_{\mathrm{T}}=\left(\mathrm{nDNA}{C}_{\mathrm{T}}-\mathrm{mtDNA}{C}_{\mathrm{T}}\right)\mathrm{Relative}\mathrm{mtDNA}\mathrm{copy}\mathrm{number}=2\times 2^{\Delta C_{\mathrm{T}}}$$

Where nDNA C_T and mtDNA C_T values constitute the mean C_T values, averaged across triplicates.

Table 1.

Primers for qPCR measurement of mtDNA copy number

Genome	Locus	Sequence (Forward)	Sequence (Reverse)
nDNA	ANKRD37	CACTGGCTACAAGCAGGCT	TTAGGAGAAGCTCCACTACACAA
nDNA	JUN	AGGGTCATGCTCTGTTTCAG	GACCTTCTATGACGATGCCC
nDNA	ACTB	CGCGAGAAGATGACCCAGAT	TCACCGGAGTCCATCACGAT
mtDNA	RNR2	GGGACCTGTATGAATGGCTC	GGGTTTGTTAGGTACTGTTTGC
mtDNA	CO3	GTTACATCGCGCCACCATTG	CAGCCCATGACCCCTAACAG
mtDNA	CO1	ACAGATGCAATTCCCGGACGTCTA	TAGGCCGAGAAAGTGTTGTGGGAA
mtDNA	ND5	TCGAATAATTCTTCTCACCC	TAGTAATGAGAAATCCTGCG

The raw C_T values used to quantify mtDNA copy number are available as Supplementary Data S11 in FigShare (https://figshare.com/projects/Shtolz_2025_Mitochondria_Hypoxia_Project/237023).

Cell permeabilization procedure for PRO-seq

To enable a run-on transcription experiment while maintaining all cellular mitochondria, the cells were permeabilized as previously described⁵⁰. The cells were washed three times using a custom wash buffer: 10 mM Tris-HCl, pH 7.5, 10 mM KCl, 150 mM Sucrose, 5 mM MgCl₂, 0.5 mM CaCl₂, 0.5 mM DTT, 1× Protease Inhibitor Cocktail (PIC, 78437, Thermo Fisher Scientific), and 40 U/μL RNase Inhibitor (037-25, A&A Technologies, Gdansk, Poland). The first wash cycle was followed by incubation in an ice-cold permeabilization buffer (10 mM Tris-HCl, pH 7.5, 10 mM KCl, 250 mM Sucrose, 5 mM MgCl₂, 1 mM EGTA, 0.05% Tween-20, 0.5 mM DTT, 1× PIC, and 40 U/μL RNase Inhibitor). After the final wash cycle, the cells were resuspended in a freezing buffer (50 mM Tris-HCl, pH 8.3, 40% Glycerol, 5 mM MgCl₂, 0.1 mM EDTA, pH 8.0, and 0.5 mM DTT), and the effectiveness of the permeabilization was measured by assessing the quantity of living cells penetrated by Trypan Blue (#1450021, Bio-Rad, Hercules, CA). If the permeabilization was successful, the cells were dispensed into 100 μL aliquots containing 10⁷ cells. The number of aliquots depended on the final cell count. The cells were snap-frozen in liquid nitrogen for subsequent run-on transcription experiments and library preparation.

Run-on transcription and library preparation for PRO-seq

Run-on transcription was performed as previously described⁵⁰ with minor modifications. Briefly, an aliquot containing 10⁷ cells was incubated with 100 μl of 2 ×1-Biotin run-on reaction mix: 10 mM Tris-HCl, pH 8.0, 5 mM MgCl₂, 1 mM DTT, 150 mM KCl, 50 μM Biotin-11-CTP (NEL542001EA, Revvity, Walham, MA), 50 μM rCTP, 250 μM rATP, 250 μM rGTP, 250 μM rUTP (all from NEB), 40 U/μL RNase Inhibitor and 1% v/v Sarkosyl (61747, Sigma-Aldrich, St. Louis, MO). The incubation was for 5 min at 37 °C. Reactions were terminated by the addition of TRIzol LS (10296010, Invitrogen, Carlsbad, CA), which is recommended for use on permeabilized cells⁵⁰, and RNA was extracted according to the manufacturer’s protocol. Next, the RNA samples were heat denatured at 65 °C for 40 s, placed on ice, and base-hydrolyzed for 8 min using ice-cold NaOH (0.2 N final concentration). Hydrolysis was stopped using Tris-HCl pH 6.8 (0.5 M final concentration), and the buffer was exchanged by running the samples through a P-30 column (Bio-Spin® columns with Bio-Gel®, 7326231, Bio-Rad, Hercules, CA). The samples were enriched for Biotin-11-CTP-labeled RNA by a 20-min incubation with magnetic streptavidin beads (Dynabeads M-280, 11205D, Thermo-Fisher Scientific) followed by two washes with high salt buffer (50 mM Tris-HCl pH 7.4, 2 M NaCl, 0.5% v/v Triton X-100), two washes with binding buffer (10 mM Tris-HCl pH 7.4, 300 mM NaCl, v/v 0.1% Triton X-100), and one wash with low salt buffer (5 mM Tris-HCl pH 74, v/v 0.1% Triton X-100). Additionally, 40 U/μL RNase Inhibitor was added to all buffers in a ratio of 2 μl/10 ml. After a second TRIzol (15596026, Thermo Fisher Scientific) RNA extraction, the Rev3 RNA adapter (Table 1) was ligated to the 3′ end of the RNA fragments by incubation with T4 RNA Ligase (#M0437M, NEB, Ipswitch, MA) at 20 °C overnight (~16 h). Samples were enriched for biotin-labeled RNA for a second time. Following the second enrichment cycle and RNA extraction, 5′ ends of the precipitated RNA were repaired with RNA 5′ pyrophosphohydrolase, followed by phosphorylation with T4 polynucleotide kinase (M0356S and M0201S, both from NEB). Next, the 5′ ends of the RNA molecules were ligated to the Rev5 RNA adapter (Table 2) by incubation overnight at 20 °C. With both ends of the RNA ligated, a third enrichment cycle for biotin-labeled RNA was performed. The enriched RNA was subjected to a reverse transcription reaction with the RP1 reverse transcription primer (Table 2) and SuperScript IV Reverse Transcriptase (18090010, Thermo Fisher Scientific). Next, the cDNA was PCR amplified using the Q5 High-Fidelity DNA Polymerase (M0491S, NEB), along with the RP1 primer and a reverse primer (RPI-n) containing a unique, 6-nucleotide index sequence that differs between samples (Table 2). The amplified library was size-selected using size exclusion polyacrylamide electrophoresis (200–1000 bp). Paired-end 150 bp sequencing was performed using the Illumina NextSeq 500 or 2000 platforms to generate approximately 50 M read pairs per sample. All sequencing was performed at the Genomic Technologies Facility of The Hebrew University of Jerusalem, Israel.

Table 2.

Primer and adapter sequences for PRO-seq

Oligo name	Oligo type	Sequence
Rev3	Adapter	GAUCGUCGGACUGUAGAACUCUGAAC-/inverted dT/
Rev5	Adapter	CCUUGGCACCCGAGAAUUCCA
RP1	Primer	AATGATACGGCGACCACCGAGATCTACACGTTCAGAGTTCTACAGTCCGA
RPI-n	Primer	CAAGCAGAAGACGGCATACGAGATNNNNNNGTGACTGGAGTT CCTTGGCACCCGAGAATTCCA

Processing of PRO-seq data

To accurately quantify mtDNA transcription at base-pair resolution, a custom bioinformatic pipeline was developed in-house⁴⁹, following previously described parameters⁶⁷. This pipeline was applied to both newly generated data from the current study and publicly available data (see Availability of data and materials section). The raw fastq files underwent adapter trimming using Trim Galore⁶⁸ (v0.6.10) and were subjected to quality control using a Phred cutoff score of 20. Firstly, the trimmed reads were non-uniquely mapped against the human mtDNA obtained from the RefSeq Organelle database (NC_012920.1)⁶⁹, using bwa⁷⁰ (v0.7.17). After the initial mapping, to reduce the effect of SNPs on the read depth, a new, sample-specific consensus reference sequence was generated using SAMtools⁷¹ (v1.9.0), while using the mpileup command with the “-u” parameter to create a genotype likelihood file, followed by BCFtools⁷² (v1.9.0) call command with the “-c” parameter to generate a consensus variant call format (VCF) file. The resultant file was converted into FASTA format and used as a reference for a second alignment run using bwa. Next, reads were quantified per-position using the SAMtools mpileup command. Additionally, to account for the circularity of the mtDNA, reads were mapped against an altered version of the sample-specific reference with the last 500 bases prepended to position 1. Mapping was performed, and read coverage from the former circle junction was calculated as explained above and added to the previous quantitation. The analysis of nDNA transcription, while focusing on OXPHOS genes, was conducted separately using PEPPRO (v0.10.1), a publicly available nascent RNA sequencing analysis pipeline⁵¹. PEPPRO was run in paired-end mode, with the PRO-seq protocol, hg38 as reference, and with an additional flag “—keep-mito=True” that prevents PEPPRO from filtering out mitochondrial reads as noise during the serial alignment stage. Data generated using PEPPRO is available in FigShare as Supplementary Data S12 (https://figshare.com/projects/Shtolz_2025_Mitochondria_Hypoxia_Project/237023).

Identification of pausing sites in PRO-seq data

To identify and quantify transcriptional pausing, an approach similar to the one used previously by Snyder, Ryan J., et al.⁵³, was employed. First, the per-position coverage depth at read-ends (3′ for positive strand and 5′ for negative strand) was transformed to Z-score relative to the mean and standard deviation of a distribution of read-end depth centered at a 201-base window around each position, using custom Python scripts. Positions with a Z-score equal to or greater than 3, and with at least 1 reads per million (RPM), were considered pausing peaks. Locations with consistent peaks in over 50% of the tested samples in a given cell line were considered pausing sites.

Processing and analysis of ChIP-seq data

All ChIP-seq data analyzed in this study were publicly available in the ENA database (PRJNA606242). The raw fastq read files were trimmed to remove sequencing adapters using Trim Galore⁶⁸ (v0.6.10) with a quality cutoff of 20. The trimmed reads were then aligned against the entire human genome (GRCh38) using bwa⁷⁰ (v0.7.17). After mapping, to reduce PCR biases, the Picard MarkDuplicates (v2.27.5) function was used to remove duplicate reads. Then, to analyze the differential binding of HIF1α to upstream regulatory regions in response to hypoxia, peak calling and differential binding analysis were performed using the Python-based PePr package⁷³ (v1.1.24) in sharp peaks mode “—peaktype sharp” which is designed for accurate detection of transcription factor binding peaks. The input samples (i.e., genomic DNA that has not been immunoprecipitated) were provided to PePr and considered as background signals to be subtracted for the identification of the final peaks. The output constitutes a list of peaks, the signals of which were significantly increased in normalized coverage in hypoxia vs. control samples. Finally, the peaks were annotated using the AnnotatePeaks.pl script, part of the HOMER suite⁷⁴; in brief, this script is designed to identify the nearest gene annotation to each peak in either sequence direction. To validate our output, we generated a ranked list of 1000 genes that had the lowest FDR-adjusted p-values out of all TSS-proximal (<1000 bp) binding sites and performed an enrichment analysis using gseapy⁷⁵.

Statistics and reproducibility

All statistical analyses were performed either in R version 4.40, or in Python 3.10. Spearman correlation tests, Mann–Whitney tests and DESeq2 version 1.46 were used when relevant. For publicly available datasets, only datasets with a minimum of two biological replicates per condition (control and hypoxia) were included. For newly generated RNA-seq experiments (HCT-116, MCF7, and U87 cells), samples were prepared from independent cell culture passages and processed as biological replicates. PRO-seq experiments were performed in three cell lines (HeLa, U87, and D407), with each cell line having control and hypoxia-treated samples. For mtDNA copy number quantification by qPCR, four independent biological replicates were performed for each cell line (HeLa, U87, MCF7, HCT-116), with each biological replicate representing an independent hypoxia experiment. Each biological replicate was analysed in technical triplicate for all amplicons (two nuclear reference genes and four mtDNA regions), and the mean C_T value across technical replicates was used for calculating relative copy number. Reproducibility was assessed by comparing our newly generated RNA-seq data (HCT-116, MCF7) with corresponding public datasets from the same cell lines grown in identical conditions.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

N.S. performed the analyses, created the figures and tables, and participated in all stages of writing the manuscript; S.D. performed the PRO-seq experiments; D.M. conceived the study, supervised the analysis, and wrote the manuscript.

Peer review

Peer review information

Communications Biology thanks the anonymous reviewers for their contribution to the peer review of this work. Primary handling editor: Dario Ummarino. A peer review file is available.

Data availability

Publicly available RNA-seq data are listed in Supplementary Data S1. Both raw and processed PRO-seq and RNA-seq data generated in the current study have been deposited in ArrayExpress (E-MTAB-14981). The DESeq2 output table of the per-gene log₂ fold-change and statistical test results, along with a DESeq2 normalized expression of all samples analyzed are available as Supplementary Data S9, S10 in FigShare. Source data underlying the graphs and charts in all main and Supplementary Figs. are provided as Supplementary Data S9–S17 in the FigShare repository (https://figshare.com/projects/Shtolz_2025_Mitochondria_Hypoxia_Project/237023). All other data supporting the findings of this study are available from the corresponding author upon reasonable request.

Code availability

All scripts written and used in the current study were uploaded to GitHub: https://github.com/Noam-St/2025_Mitochondria_Hypoxia_Project and released to Zenodo (10.5281/zenodo.17845989)⁷⁶.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s42003-025-09457-y.