Increased global transcription activity as a mechanism of replication stress in cancer

Cancer is a disease associated with genomic instability that often results from oncogene activation. This in turn leads to hyperproliferation and replication stress. However, the molecular mechanisms that underlie oncogene-induced replication stress are still poorly understood. Oncogenes such as HRASV12 promote proliferation by upregulating general transcription factors to stimulate RNA synthesis. Here we investigate whether this increase in transcription underlies oncogene-induced replication stress. We show that in cells overexpressing HRASV12, elevated expression of the general transcription factor TATA-box binding protein (TBP) leads to increased RNA synthesis, which together with R-loop accumulation results in replication fork slowing and DNA damage. Furthermore, overexpression of TBP alone causes the hallmarks of oncogene-induced replication stress, including replication fork slowing, DNA damage and senescence. Consequently, we reveal that increased transcription can be a mechanism of oncogene-induced DNA damage, providing a molecular link between upregulation of the transcription machinery and genomic instability in cancer.

C ancer is a disease of genomic instability, characterized by high mutation rates and genomic rearrangements that ultimately drive aggressiveness and resistance to therapy 1 . One of the mechanisms proposed to cause genomic instability in cancer is replication stress, which occurs when DNA replication fork progression in S phase slows or stalls. This leads to collapse of forks into DNA double-strand breaks (DSBs), as well as incomplete sister chromatid separation in the following mitosis 2 . Markers of spontaneous replication stress are found in tumour samples and cells expressing active oncogenes, and replication stress promotes chromosomal instability, the most common form of genomic instability in sporadic cancers [3][4][5][6] . Spontaneous replication stress is therefore increasingly regarded as a central feature of cancer cells and there is much interest in specifically targeting this phenotype for cancer therapy 7 . However, progress in this field is hindered, because the molecular mechanisms underlying spontaneous replication stress in cells are still largely unknown. This impairs our ability to investigate replication stress in vitro and in vivo, and to identify potential biomarkers or therapeutic targets.
How can mechanisms of spontaneous replication stress be identified? The overexpression of oncogenes such as RAS, MOS, MYC, CDC25A or CYCLIN E is sufficient to induce replication stress in cultured cells 4,5,[8][9][10][11] . These oncogenes all act in the growth factor signalling pathways that stimulate proliferation by promoting cell growth and division. Molecular changes associated with increased proliferation are therefore prime candidates for causing replication stress. Indeed, we and others have reported that CYCLIN E-induced replication stress results from accelerated S-phase entry and increased replication initiation during S phase 9,12 . So far, however, little attention has been paid to the fact that oncogenes such as RAS and MYC do not only activate the cell cycle machinery but also promote cell growth through the activation of transcription and protein translation [13][14][15][16] . Overexpressed c-MYC acts as a 'universal amplifier', stimulating transcription by all three RNA polymerases 14,17,18 . Oncogenic RAS promotes transcription through the mitogen-activated protein kinase extracellular signal-regulated kinase (ERK), which activates transcription factors such as TIFIA (RRN3), UBTF, TIFIIIB (BRF1) and TATA-box binding protein (TBP) 14,15,19 , and can also promote transcription through other factors such as TERT 20 . Moreover, components of the general transcription machinery itself are found mutated, differentially expressed and deregulated across a variety of cancers 14,21 .
Upregulation of transcription in cancer cells has the potential to be a direct cause of replication stress as interference between transcription and replication leads to replication fork slowing and genomic instability 22,23 . This can result from direct collisions or torsional stress between the two active protein complexes 24 . Another important source of replication stress is the collision of the replication machinery with RNA-DNA hybrids (R-loops) where the nascent RNA has re-annealed with the template 25 . Deregulated transcription is a common feature of cancers, but its importance for replication stress and consequent genomic instability has not been examined.
Here we use HRAS V12 overexpression 5,26 to investigate whether the oncogene-induced increase in transcription is a mechanism of endogenous replication stress. We report that increased transcription activity in cells expressing HRAS V12 causes replication stress, as high levels of RNA synthesis and transcription intermediates interfere with replication fork progression. Replication stress in these cells depends on R-loop accumulation and on increased expression of the general transcription factor TBP. Importantly, overexpression of TBP alone induces replication stress and genomic instability. Our data suggest that increased transcription activity is a mechanism contributing to replication stress in cancer.

HRAS V12 increases nascent RNA synthesis and R-loop formation.
To investigate whether increased transcription activity contributes to oncogene-induced replication stress, we used immortalized human fibroblasts that have been stably transfected with pBabe-HRAS V12 -ER TAM , to express tamoxifen-inducible HRAS V12 (BJ-hTert HRASV12 ER-TAM ) 27 . HRAS V12 -ER TAM has been well characterized as a system for inducing oncogenic RAS 26,28,29 . Addition of 4-hydroxytamoxifen (4OHT) led to HRAS V12 accumulation, which activated mitogen-activated protein kinase signalling as evidenced by ERK1/2 phosphorylation (Fig. 1a). Cells proliferated slightly faster for up to 6 days followed by growth arrest, which was previously shown to result from DNA damage-induced apoptosis and senescence 4,5 ( Supplementary Fig. 1a-e). When using immortalized BJ fibroblasts that were not expressing 4OHTinducible HRAS V12 , 4OHT treatment did not affect any of the phenotypes investigated in this study (Supplementary Fig. 1f-i).
First, we investigated the effect of oncogenic HRAS on global transcription activity. For this, we quantified nascent RNA synthesis using nuclear incorporation of the modified RNA precursor 5-ethynyluridine (EU) for 1 h (Fig. 1b,c). RNA synthesis increased rapidly after HRAS V12 induction and was elevated more than twofold after 48 and 72 h HRAS V12 induction (Fig. 1d). RNA synthesis after HRAS V12 induction was higher than the highest activity observed in control cells (Fig. 1d), arguing against changes in cell cycle distribution as the sole explanation for the increase. Seventy-two hours of HRAS V12 induction was used for most subsequent experiments.
We next used the S9.6 antibody that detects RNA/DNA hybrids 30,31 , to test whether R-loop formation was increased in HRAS V12 -overexpressing cells. First, we performed slot blot analysis of isolated genomic DNA 32 , which revealed a threefold increase in R-loops after 72 h HRAS V12 induction (Fig. 1e,f). The S9.6 signal could be removed by treatment with recombinant RNase H, supporting that it was specific to R-loops. We then used S9.6 immunoprecipitation of RNA/DNA hybrids from cells (DNA immunoprecipitation (DIP)) 33 after 72 h HRAS V12 induction, to investigate the distribution of R-loops across RAS target versus control genes ( Fig. 2a-f). We used quantitative PCR (qPCR) to quantify R-loop distribution on DUSP6, SPRY2 and C-FOS, genes that are upregulated by activated RAS (Fig. 2a-c and Supplementary Fig. 2a). We observed an increased R-loop formation on the promoter-proximal and selected intron regions of all three genes. DIP analysis of C-FOS showed that in line with previously described R-loop accumulation in actively transcribed genes 33 , R-loops were significantly increased over the transcribed regions of the gene ( Fig. 2c and also see Supplementary Table 1 for PCR primer sequences). We also quantified R-loop formation on non-RAS target control genes GAPDH1, ACTB (b-ACTIN) and ACTG (g-ACTIN). We observed no increase in R-loops across any of these genes ( Fig. 2d-f). RNase H treatment confirmed that DIP specifically detected R-loops ( Fig. 2a-f). RNase A treatment confirmed that DIP signal was not due to annealing of free RNA species to DNA during sample preparation or to S9.6 antibody recognizing double-stranded RNA ( Supplementary Fig. 2b-e). These data support that stimulation of transcription by HRAS V12 results in increased R-loop formation.
HRAS V12 causes replication stress and G1 53BP1 foci. HRAS V12 overexpression caused replication fork slowing after ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms13087 48 h, indicative of replication stress (Fig. 3a-c) and induced phosphorylation of replication stress response factors RPA32 (serine 33) and CHK1 (serine 345) (Fig. 3d,e). These modifications were already evident at 24 h HRAS V12 induction, which may also reflect the proposed role for RPA and ATR in the transcription stress response 34 . As expected, fork slowing was associated with nuclear foci formation of DNA damage markers gH2AX and 53BP1 (Fig. 3f,g). One consequence of replication fork slowing is mitotic entry with under-replicated DNA, which leads to micronuclei formation and appearance of 53BP1 bodies in the following G1 phase 35,36 . Accordingly, 53BP1 foci increased after 4 days HRAS V12 induction and were mostly found in G1 cells (Fig. 3h). Similarly, micronuclei formation peaked at 4 days HRAS V12 induction (Fig. 3i). This suggests that HRAS V12induced DNA damage requires mitotic progression, as was previously reported for CYCLIN E 10 . Four days of HRAS V12 induction was used for subsequent investigation of 53BP1 foci formation.
Transcription promotes HRAS V12 -induced replication stress. To test whether increased transcription causes replication stress in cells harbouring HRAS V12 , we transiently inhibited RNA synthesis using small molecule inhibitors before measuring replication fork progression. To minimize effects on gene expression, incubations were kept short at 100 min for triptolide and 5,6-dichloro-1-b-D-ribofuranosyl-1H-benzimidazole (DRB) and 4 h for a-amanitin (Fig. 4a). These treatments inhibited ongoing RNA synthesis ( Fig. 4b) but did not affect protein levels of HRAS V12 or levels of proteins with short half lives such as CYCLIN B1 and p53 ( Supplementary Fig. 3a,b). All three transcription inhibitors increased replication fork speeds specifically in the presence of HRAS V12 (Fig. 4c-f). Although such short incubations with transcription inhibitors may not be sufficient to reverse all effects of transcription, these data suggest that RAS-induced replication stress is promoted by active RNA synthesis. DRB and triptolide rescued replication more effectively than a-amanitin, suggesting that effective inhibition of early stages of transcription removes more obstacles to replication forks than does inhibiting transcription during elongation (Fig. 4f). We previously observed that increased CDK activity and new origin firing underlies replication stress in cells overexpressing CYCLIN E 12 . In contrast, inhibiting new origin firing using CDK inhibitor roscovitine could not rescue HRAS V12 -induced replication fork slowing ( Supplementary Fig. 3c-e). Together, our data suggest that replication stress in cells expressing HRAS V12 is promoted by transcription but not by CDK activity.
To test whether HRAS V12 -induced DNA damage was transcription dependent, we first incubated cells with DRB for 100 min and stained for 53BP1 24 h later. DRB reduced 53BP1 foci formation; however, its impact was limited by the short incubation time ( Supplementary Fig. 3f,g). We therefore used gH2AX chromatin immunoprecipitation (ChIP) to test whether transcription and R-loop formation was associated with DNA damage in cells harbouring HRAS V12 . Indeed, we observed an increase in gH2AX signal over the transcribed promoterproximal region of the C-FOS gene, correlating with strong induction of R-loops, 72 h after HRAS V12 induction (Figs 2c and 4g). This gH2AX induction was replication dependent, as it could be prevented by blocking replication with Aphidicolin ( Fig. 4g and Supplementary Fig. 1d). In contrast, we did not detect an increase in replication-dependent gH2AX levels across the intron 1 region of the b-ACTIN gene (Fig. 4h). This suggests that HRAS V12 triggers R-loop-associated DNA damage that also depends on replication.
R-loops promote HRAS V12 -induced replication stress. We next decided to further investigate the role of R-loops in HRAS V12induced replication stress. We used transient transfection to express green fluorescent protein (GFP)-tagged human RNaseH1, an enzyme that degrades RNA/DNA hybrids on overexpression 37 ( Fig. 5a,b). Interestingly, we observed that protein and messenger RNA levels of endogenous RNaseH1 were elevated in cells overexpressing HRAS V12 , suggesting an increased requirement for R-loop processing activities (Fig. 5b,c). The specificity of RNaseH1 antibody was verified using small interfering RNA (siRNA) depletion of RNaseH1 ( Supplementary Fig. 4a). Overexpression of GFP-RNaseH1 reduced R-loop levels in the nucleus, as indicated by S9.6 immunostaining (Fig. 5d,e and also see Supplementary Fig. 5 for validation of immunostaining method). As the expression construct contains the RNaseH1 mitochondrial targeting sequence, mitochondrial R-loops were also reduced (Fig. 5d). RNaseH1 overexpression effectively improved replication fork progression in cells harbouring HRAS V12 (Fig. 5f,g). GFP-RNaseH1 also decreased HRAS V12 induction of 53BP1 foci (Fig. 5h,i) and DSB induction measured by pulse-field gel electrophoresis ( Supplementary Fig. 4b,c). GFP-RNaseH1 overexpression did not affect hydroxyureainduced 53BP1 foci formation or the cell cycle profile ( Supplementary Fig. 4d,e). These data support that HRAS V12induced replication stress is promoted by the presence of R-loops.
We also tested whether some of the replication stress caused by HRAS V12 might be due to ribonucleotide (NTP) depletion as a result of increased RNA synthesis. Supplementing growth medium with ribonucleosides improved fork speeds and reduced 53BP1 foci formation in the presence of HRAS V12 (Supplementary Fig. 6a-c). However, HPLC quantification showed no reduction in NTP levels or dNTP levels as detectable ( Supplementary Fig. 6d,e). This is in line with previous reports demonstrating that nucleoside supplementation can rescue fork progression even in the absence of detectable nucleotide depletion 38 . Our data thus suggest that transcription contributes to HRAS V12 -induced replication stress predominantly via direct conflicts with transcription machinery and R-loops (Fig. 5j).  HRAS V12 -induced replication stress depends on TBP. We next turned our attention to transcription factors that may promote RNA synthesis and therefore replication stress downstream of HRAS V12 . TBP expression is induced by RAS signalling in a number of human, murine and Drosophila cell types 19,39,40 , and TBP mRNA and protein levels were accordingly increased after HRAS V12 induction ( Fig. 6a-c). We used siRNA depletion to test whether TBP promotes HRAS V12 -induced replication stress (Fig. 6d,e). TBP depletion decreased nascent RNA synthesis in HRAS V12 -expressing cells (Fig. 6f). Importantly, TBP depletion also prevented HRAS V12 -induced fork slowing (Fig. 6g,h) and TBP-depleted cells displayed fewer HRAS V12 -induced 53BP1 foci (Fig. 6i). Similar results were obtained using a different siRNA sequence targeting TBP ( Supplementary Fig. 7a-e). We noticed that the alternative TBP siRNA also reduced RNA synthesis and replication fork speeds in control cells, suggesting that generally low RNA synthesis may affect replication ( Supplementary  Fig. 7c,d). TBP depletion did not prevent hydroxyurea-induced 53BP1 foci (Supplementary Fig. 6f) and did not affect the cell cycle profile of 53BP1 foci-positive cells, suggesting that the reduction in foci was not due to a G2 arrest ( Supplementary  Fig. 7g).
To test whether TBP acts in the same pathway as ongoing RNA synthesis, we combined TBP siRNA with DRB treatment. Combining both treatments rescued replication fork progression and 53BP1 foci formation to a similar extent as TBP depletion alone (Fig. 6j,k and Supplementary Fig. 8a). To further test whether ribonucleoside addition was affecting replication stress via a different pathway to transcription 41 , we combined TBP siRNA and DRB with nucleoside supplementation. Compared with transcription inhibition alone, exogenous nucleosides had no additional effect on either the rescue of replication fork progression or the reduction in 53BP1 foci ( Supplementary  Fig. 8b-e).
These data suggest that the main pathway of HRAS V12induced replication stress is via upregulation of transcription, and that TBP is involved in this upregulation of transcription and the replication fork slowing and genomic instability that result from it (Fig. 6l).
TBP overexpression alone causes replication stress. We reasoned that if TBP was a downstream effector in HRAS V12induced replication stress, then overexpression of TBP alone should cause replication stress. We therefore generated BJ-hTert fibroblasts for doxycycline-inducible overexpression of TBP (BJ-TBPind; Fig. 7a). Inducing TBP overexpression over several days led to a steady increase in RNA synthesis activity as measured by EU incorporation (Fig. 7b,c). Importantly, the increase in transcription activity during TBP overexpression was accompanied by replication fork slowing, consistent with replication stress (Fig. 7d and Supplementary Fig. 9a). Similar results were observed in human MRC5 fibroblasts overexpressing TBP (Supplementary Fig. 9b-d). To test whether TBP-induced fork slowing depended on ongoing transcription, we treated TBP-overexpressing cells with DRB for 100 min, to inhibit nascent RNA synthesis (Fig. 7e). DRB treatment rescued the TBP-induced fork slowing ( Fig. 7f and Supplementary Fig. 9e), supporting that TBP overexpression causes replication stress through nascent RNA synthesis.
We next characterized the effect of TBP overexpression on DNA damage, genomic instability and proliferation. Similar to   HRAS V12 -expressing cells, cells overexpressing TBP displayed increased formation of micronuclei and 53BP1 foci (Fig. 7g), and 53BP1 foci were mostly found in G1 cells (Fig. 7h). Finally, TBP overexpression eventually led to growth arrest and senescence, similar to that caused by HRAS V12 (Fig. 7i and Supplementary  Fig. 9f,g). In agreement with this, p53 levels were increased after TBP induction ( Fig. 7a and also see Supplementary Fig. 10 for original images of western blottings).
Thus, our data show that TBP alone is able to increase transcription activity and cause replication stress with features that resemble oncogene-induced replication stress.
TBP expression and replication stress in cancer. Finally, we investigated the relationship of TBP and RNASEH1 mRNA expression with oncogenes and replication stress markers in tumour samples using The Cancer Genome Atlas (TCGA) data  (Table 1). In a number of cancers, TBP or RNASEH1 expression correlated positively with expression of RAS or MYC oncogenes and the CHEK1 or CHEK2 checkpoint kinases, which are activated by oncogene-induced replication stress 4,5 . These data support that the relationship between oncogenes, TBP, R-loops and replication stress could be present in cancer tissues.

Discussion
We report that increased transcription is a new mechanism of oncogene-induced replication stress. Overexpression of oncogenic HRAS V12 increases RNA synthesis and R-loop formation, and this directly contributes to replication stress induced by HRAS V12 . HRAS V12 -induced replication stress is mediated by the general transcription factor TBP, which is a downstream target of RAS signalling. Accordingly, overexpression of TBP itself causes replication stress and genomic instability. The transcription machinery is frequently deregulated in cancer cells 21 and our data suggest that transcription-associated replication stress could be an important mechanism promoting genomic instability in cancer (Fig. 7j).
Markers of replication stress are observed in early tumours and cancers and other conditions of high proliferation, such as stem cell reprogramming or viral infection 51,52 . Recent studies have provided first candidates for the mechanisms causing such endogenous replication stress. So far, all of these have involved deregulation of the cell cycle, including re-replication 53 and premature or increased replication initiation, resulting in depletion of nucleotides or replication enzymes 9,10,12 . In addition, reactive oxygen species play important roles in DNA damage in cancer, but whether they can cause replication stress is still uncertain 54,55 . Our data support that increased transcription is another pathway causing endogenous replication stress, which may act in parallel or independently of cell cycle deregulation. HRAS V12 overexpression did increase the density of active replication origins, which could be reversed by CDK inhibition.   However, CDK inhibition was unable to relieve HRAS V12induced replication fork slowing, suggesting that increased replication initiation is not a cause but a consequence of HRAS V12 -induced replication stress ( Supplementary Fig. 3d,e). Our data suggest that HRAS V12 causes replication stress by a mechanism that is different from oncogenes such as CYCLIN E and CDC25A 10,12 . Our data support that HRAS V12 overexpression promotes accumulation of R-loops, which can be a major cause for replication fork slowing and DNA breakage 23 . An elegant study recently reported that loss of tumour suppressors BRCA1 or BRCA2 increases R-loop levels, because these proteins act to prevent R-loop formation 56 . Our findings add an important new angle to this observation, showing that oncogenes can induce R-loops, which suggests that increased R-loop levels might also be common in cancer cells that are proficient in BRCA1 or BRCA2. Interestingly, we observed increased protein levels of endogenous RNaseH1 in cells overexpressing HRAS V12 (Fig. 5b,c). RNaseH1 and RNaseH2 are the main RNase activities counteracting R-loop formation and upregulation of these enzymes may be a response of cancer cells to increased R-loop levels 23 . Although the biology of increased RNaseH1 expression in response to HRAS V12 requires further investigation, it supports the idea that oncogenic replication stress is strongly connected with R-loop metabolism.
One exciting implication of our findings is the potential to discover new factors involved in promoting replication stress. Components of the transcription machineries are found overexpressed in cancer 21,57 . TBP is one of several general transcription factors that have been implicated in oncogeneinduced cell transformation 39,58 . The TBP promoter contains binding sites for oncogenic transcription factors and TBP is differentially expressed in cancers (Table 1) 59 . RAS has been reported to upregulate TBP expression through RAF-MEK and RALGDS signalling 19 , and TBP levels are also increased by other growth factor signalling pathways such as epidermal growth factor/epidermal growth factor receptor 59,60 (Fig. 7j). We report here that increased TBP levels cause replication stress and markers of genomic instability. TBP expression also correlates with the expression of MYC, RAS and checkpoint kinases in a number of cancers ( Table 1). As TBP overexpression did not increase RNA synthesis as strongly as HRAS V12 overexpression and TBP depletion did not completely rescue HRAS V12 -induced replication stress, other transcription factors such as UBTF may also be involved downstream of HRAS V12 (Fig. 7j).
A previous report showed that TBP is important for HRAS V12 transforming function, and that TBP overexpression alone promotes anchorage-independent growth and tumour growth in vivo 49 . Our data presented here support the idea that TBP promotes the oncogenic phenotype by increasing transcription activity and replication stress. There is however no evidence that TBP itself is an oncogene, as TBP overexpression promoted tumour growth but not tumourigenesis itself 49 . Some cancer types show recurrent mutations in TBP that affect the length of the amino-terminal glutamine-rich motif (for example, Q72dup in diffuse large B-cell lymphoma and adrenocortical carcinoma) 61 .
It is yet unknown whether such mutations have pathogenic relevance in cancer and whether they could be oncogenic.
Finally, increased RNA synthesis and elevated levels of R-loops could be correlated with replication stress in cancer. Identifying the molecular mechanisms that directly cause spontaneous replication stress should therefore help to predict and detect replication stress more accurately in cells and tissues. Nascent RNA synthesis can only be measured in live cells but increased expression of transcription factors and RNA polymerase subunits 21 , as well as increased R-loops and expression of RNases H1 or H2, might be promising candidates for markers of transcription-associated replication stress.
EU incorporation assay. EU incorporation assays were performed using the Click-iT RNA Alexa Fluor 594 Imaging Kit (Invitrogen) according to the manufacturer's instructions. Cells were incubated with 1 mM EU for 1 h, fixed with 4% PFA for 15 min at room temperature, permeabilized with 0.5% Triton X-100 for 15 min and Click-iT reaction was performed. DNA was counterstained with DAPI and images were acquired as above. ImageJ was used to generate nuclear masks based on DAPI staining and mean AlexaFluor 594 fluorescence intensities per pixel were quantified per nucleus.
RNA/DNA hybrid immunoprecipitation. DIP analysis was performed with RNA/DNA hybrid antibody (0.3 mg ml À 1 per IP reaction) purified from S9.6 hybridoma cell lines as previously described 33 . The immunoprecipitated (S9.6 antibody in IP reaction), control (beads only) and input DNAs were used as templates for qPCR. DIP RNase H-sensitivity analysis was carried out before IP step with the addition of 25 U RNase H (NEB, M0297S). One hundred microlitres of nuclease digestion reaction contained 1 Â reaction buffer and it was performed for 3 h at 37°C.
Slot-blot experiments. Slot-blot experiments were carried out as described 32 . Genomic DNA (1.2 mg) were treated with 2 U of RNase H per mg of DNA (NEB, M0297S) for 2 h at 37°C before loading on the slot blot. Half of the DNA sample was probed with S9.6 antibody (1:1,000) and the other half with anti-ssDNA antibody (MAB3031, Millipore, 1:25,000) as described 32 . Secondary antibody was goat anti-mouse horseradish peroxidase (1:10,000). Images were acquired with LAS-4000 (Fujifilm) and quantified using Image Studio Lite software (Li-COR Biosciences).
Cell proliferation and b-galactosidase assays. For proliferation assays, 1 Â 10 5 or 2 Â 10 5 cells were seeded in 12-well plates and incubated with 10 mg ml À 1 resazurin for 2 h at the indicated time points. Resorufin fluorescence at 590 nm was measured using a BMG Labtech PHERAstar FS microplate reader. Cell numbers were determined by multiplying the number of initially seeded cells with (FR n /FR 0 ) (FR n : fluorescence reading at time point, FR 0 : initial fluorescence reading). b-Galactosidase staining was performed using the Senescence b-Galactosidase Staining Kit (Cell Signaling) according to the manufacturer's instructions.
Nucleotide quantification. Cells were harvested and nucleotides extracted with 70% ice-cold methanol. Precipitated proteins were removed by centrifugation and supernatants stored at À 80°C. Supernatants were dried in a heated vacuum centrifuge and reconstituted in HPLC starting eluent. Samples were analysed by HPLC (Waters 2695, Watford, UK) with a photodiode array detector (Waters 2996). Separation was achieved using an Ace C18 (3 mm, 3 Â 125 mm, Hichrom, UK) column maintained at 35°C with eluent A: 10 mM potassium dihydrogen phosphate and 10 mM tetrabutylammonium hydrogen sulfate, 10% methanol pH 6.9; eluent B: 50 mM potassium dihydrogen phosphate, 6 mM tetrabutylammonium hydrogen sulfate and 30% methanol pH 7, using a flow rate of 0.6 ml min À 1 and a gradient of 25-80% B over 25 min, with a run time of 30 min. Nucleotides were identified by comparing with absorbance spectra and retention times of commercially available standards.
Quantitative real-time PCR. Total RNA was harvested using TRIZOL reagent (Invitrogen) followed by DNase I treatment (Roche), 1.5 mg of total RNA was reverse transcribed using SuperScript Reverse Transcriptase III (Invitrogen) with random hexamers (Invitrogen), following manufacturer's instructions. The qPCR primers for amplification are listed in Supplementary Table 1. For quantitative real-time PCR, 2 ml of cDNA was analysed using a Rotor-Gene RG-3000 real-time PCR machine (Corbett Research) with QuantiTect SYBR green (Qiagen). Cycling parameters were 95°C for 15 min, followed by 45 cycles of 94°C for 20 s, 58°C-62°C for 20 s and 72°C for 20 s. Fluorescence intensities were plotted against the number of cycles by using an algorithm provided by the manufacturer. fragments 42 Mb) were quantified by densitometry using ImageJ. Intensity of DNA entering the gel was normalized to total DNA and untreated control to obtain final values.
Statistical analysis. Unless stated otherwise, all values are means ± 1 s.e.m. of results from independent biological repeats. Scatter plots show pooled data, but numerical values displayed on plots represent the means ± 1 s.e.m. of the results from independent repeats. Numbers of repeats N are indicated in the figure legends. Statistical tests were performed using the one-tailed Student's t-test. Coexpression data were obtained using CBioPortal 61,64 and statistical analyses were performed using the Student's t-test based on t ¼ (r*On À 2)/(O À r 2 ) where r is Pearson's coefficient and n is number of samples in analysis.
Data availability. The authors declare that all the data supporting the findings of this study are available within the article and its Supplementary Information files and from the corresponding authors upon reasonable request.