Elevated pausing of RNA Polymerase II underlies acquired resistance to ionizing radiation

As the mainstay modality for many malignancies, ionizing radiation (IR) induces a variety of lesions in genomic DNA, evoking a multipronged DNA damage response to interrupt many cellular processes including transcription. How the global transcription cycle is altered by IR and whether it is contributing to the development of IR-resistance remain unaddressed. Here we report a genome-wide accumulation of paused RNA Polymerase II (RNAPII) after IR exposure. This increased pausing is partially maintained in cells acquired IR-resistance, notably on genes involved in radiation response and cell cycle, often leading to their downregulation. Individual knockdown some of these genes such as TP53 and NEK7 endows IR-sensitive cells with varying degrees of resistance, highlighting a novel link between elevated RNAPII pausing and the acquisition of IR-resistance. Accordingly, tuning-down the RNAPII pausing level by inhibiting CDK7 reverses IR-resistance both in cell culture and xenograft models. Our results suggest that modulation of the transcription cycle is a promising strategy to increase IR-sensitivity and thwart resistance.


Introduction
Radiotherapy is the mainstay treatment of non-metastatic solid tumors such as nasopharyngeal carcinoma (NPC), a geographically unbalanced malignancy prevalent mostly in southeast Asia including southern provinces of China 1 . High-energy ionizing radiation (IR) causes tumor regression by generating a variety of lesions to living cells. In addition to oxidative damages, every Gray (Gy) of IR induces roughly 1000 single-strand breaks (SSBs) and 40 double-strand breaks (DSBs) to the nuclear DNA, eliciting most of the cytotoxicity after IR exposure 2,3 . Comprehensive characterization of cellular responses to IRinduced DNA damage has been the central theme in the eld to improve treatment e cacy and limit the development of resistance.
As the rate-limiting step of gene expression, the dynamic process of transcription underlies many cellular responses towards intrinsic and extrinsic stimuli. A full transcription cycle consists of four sequential phases-initiation, pausing, elongation, and termination-which are regulated by post-translational modi cations on the C-terminal domain (CTD) of the Rpb1 subunit of RNA Polymerase II (RNAPII) 4 . After initial recruitment to promoter regions by transcription factors (TFs) and general transcription factors (GTFs), RNAPII CTD is phosphorylated at position 5 (pS5) of the Y 1 S 2 P 3 T 4 S 5 P 6 S 7 repeat by the TFIIHassociated kinase CDK7, escaping from the promoter and then quickly becoming paused after transcribing 20-60 nucleotides downstream of the transcription start site (TSS) 5,6 . Several proteins have been identi ed to govern the release of the paused RNAPII, which include the negative elongation factor (NELF) complex and the positive transcription elongation factor b (P-TEFb). The CDK9 subunit of P-TEFb along with other CDKs can phosphorylate NELF and the position 2 (pS2) of the RNAPII CTD, thereby driving the paused RNAPII to enter the elongation phase 4,6,7 . The RNAPII then continues traveling along the DNA template till it encounters the termination signal to complete a transcription cycle. The promoterproximal pausing of RNAPII has been an emerging nexus of regulation, best exempli ed by its role in synchronizing the spatiotemporal patterns of gene activities during Drosophila early embryonic development 6,8,9 . Its functional importance in other biological processes, however, is waiting to be uncovered.
Since unscheduled encounters with damages on the DNA template can lead to transcriptional failure and genome instability, the process of transcription is tightly connected with DNA damage responses toward different DNA lesions. During the repair of bulky single-strand DNA damages induced by ultraviolet light (UV), RNAPII functions as a damage sensor. UV damage triggers an immediate release of paused RNAPII into gene bodies to detect local lesions and initiate transcription-coupled nucleotide excision repair (TC-NER) 10 . Following this initial phase, the RNAPII is then cleared from the damaged DNA template and degraded by ubiquitination which leads to a global shutdown of transcription 7 . In response to doublestrand DNA damages, a repressive chromatin state mediated by histone H2AK119ub is formed around the damaged sites to halt local RNAPII activity and prevent transcription errors. The global impact of DNA DSBs on the transcription cycle is less clear. IR induces both SSBs and DSBs in the target cells 3 . How IR exposure modulates the global transcriptional dynamics and whether the altered kinetics in transcription in uences the treatment e cacy are outstanding questions need to be investigated.
Using a panel of NPC cells as well as lung cancer cell lines, here we show increased phosphorylation at serine 5 (pS5) of the RNAPII after IR, indicating an accumulation of paused RNAPII. Remarkably, a similar increase of pS5 is seen in IR-resistant cells. Chromatin immunoprecipitation followed by sequencing (ChIP-seq) analysis of RNAPII distribution con rms this increased pausing both in IR-treated and IRresistant NPC cells, notably on genes involved in radiation response and cell cycle, many of which show downregulated transcripts abundance in IR-resistant cells. Individual knockdown of these genes such as TP53 and NEK7 by small hairpin RNA (shRNA) increases the survival of IR-sensitive cells after irradiation, implying their involvement in the development of IR-resistance. Moreover, decreasing pS5 of RNAPII and hence tuning down the transcriptional pausing by inhibiting CDK7 reverses IR-resistance both in cell culture and xenograft models. Of note, the recurrent NPC tissues display a steady increase in pS5 compared to the paired primary tissues, suggesting that the CDK7 inhibitors can be used in combination with radiotherapy to increase sensitivity and prevent resistance.

Increased RNAPII serine 5 phosphorylation both in IRtreated and IR-resistant cells
The paused RNAPII is enriched in pS5 phosphorylation and the elongating RNAPII pS2 (Fig. 1A). To compare the global transcriptional kinetics in IR-sensitive and -resistant cells, we assessed the phosphorylation status of RNAPII in NPC cell line CNE2 and its derivative radiation resistant cell line CNE2-IR (Fig. S1A). Results from western blot and immuno uorescence showed a consistent increase of pS5 as well as a slight decrease of pS2 in CNE2-IR cells compared with its parental CNE2 cells (Fig. 1B-1C). We expanded the analysis to two other NPC cell lines, two lung cancer cell lines, and their derivative radio-resistant cells (Fig. S1B). Except one lung cancer cell line H460, the radio-resistant cells all showed different levels of increase in pS5 RNAPII (Fig. 1D).
Radio-resistant cells are derived from parental cell lines by repeated exposure to IR. To test if IR treatment could increase the amount of pS5 RNAPII, we irradiated the cells with different radiation doses and detected the serine 5 phosphorylation. Most of the radio-sensitive and radio-resistant cell lines analyzed displayed increased pS5 after irradiation ( Fig. 1E-1G, S1C-S1D). The H460 cells again showed a different response for reasons unknown (Fig. S1E).
These results showed that irradiation in many cells could elicit an increase of pS5 in the CTD of RNAPII, and this increase seemed to be persistent in the cells acquired IR-resistance, indicating that the kinetics of transcription was sensitive to IR and the pool of paused RNAPII was increased in IR-treated as well as IRresistant cells. The DSBs caused by IR are central in IR-induced cellular effects 3 . We inhibited the core kinases mediating the DSBs repair, ATM and DNA-PK 11 , and no longer observed the increase of pS5 RNAPII after irradiation (Fig. S1F-S1G), suggesting that signaling pathways triggered by DSBs were responsible for the transcriptional change.
Elevated transcriptional pausing is associated with IRresistance To further elaborate the altered distribution of the transcribing RNAPII, we performed ChIP-seq analysis with CNE2, CNE2 cells after 4 Gy irradiation, as well as CNE2-IR cells ( Fig. 2A). Consistent with western blot and immuno uorescence results, we detected a global increase of pS5 around the TSS both in the irradiated CNE2 cells and the radio-resistant CNE2-IR cells even without irradiation ( Fig. S2A-S2B). The pS2 of RNAPII was accordingly decreased at the TSS as well as in the gene bodies ( Fig. S2C-S2D).
We further calculated the pausing index (PI) for each gene using ChIP-seq data generated with the total RNAPII antibody (Fig. S2E) 12 . The overall PI in CNE2 cells received 4 Gy irradiation was signi cantly higher than that in control cells (Fig. 2B), and a similar increase in PI was observed in CNE2-IR cells  Table S2). We identi ed 2895 genes in the irradiated CNE2 cells and 1133 genes in CNE2-IR cells that had increased PI compared with control CNE2 cells, with a total of 331 genes shared by these two lists ( Fig. 2D and Supplementary Table S3). Gene ontology (GO) analysis of the 331 genes showed that these genes were enriched in biological processes including response to DNA damage, response to radiation, and cell cycle ( Fig. 2E and Supplementary Table S4).
Change in transcriptional dynamics could affect the gene expression level. We performed RNA-seq analysis on CNE2 and CNE2-IR cells, and identi ed the differentially expressed genes. Of particular interest, in the 331 PI increased genes, 83 showed reduced expression in CNE2-IR cells (Fig. 2F and   Supplementary Table S5). Many of these genes were involved in cell cycle regulation, such as TP53, NEK7, JUNB, etc. We selected the TP53 and NEK7 loci, designed primer sets targeting TSS and gene body regions according to the sequencing results (Fig. 2G), and validated by ChIP-qPCR that the pausing of RNAPII was indeed elevated on these genes ( Fig. S2F-S2G).
Based on these results, we concluded that a signi cant fraction of RNAPII underwent increased promoterproximal pausing after irradiation, and this increased pausing was maintained on a subset of genes involved in response to radiation as well as cell cycle, downregulating their mRNA levels in cells acquired radio-resistance.
Many pausing regulated genes are contributing to IRresistance To assess the contribution of these downregulated genes to the development of resistance to radiation, we individually knocked down the expression of several of them including TP53, NEK7, JUNB, PKN2, and POLA2 in the radio-sensitive CNE2 cells using shRNA (Fig. S3A). While the control CNE2 cells demonstrated a severe decline in viability in response to the increasing irradiation dose, the CNE2 cells treated with shRNA targeting different candidate genes all gained varying degrees of resistance to irradiation, with TP53 and NEK7 knockdown (KD) showing the highest resistance close to that of the CNE2-IR cells (Fig. 3A). Flow cytometry analysis revealed that the CNE2-IR cells had a slight G2/M arrest compared to CNE2 cells, and NEK7 KD in CNE2 cells caused a similar effect on the cell cycle (Fig. 3B).
We further characterized the cell cycle changes seen in the CNE2-IR cells and CNE2 cells treated with NEK7 shRNA. Immuno uorescence showed that both CNE2-IR and the NEK7 KD CNE2 cells displayed defects in the formation of bipolar mitotic spindle ( Fig. 3C and S3C), with a ratio of multipolar spindle greatly exceeding that seen in the control CNE2 cells (Fig. 3E, 54% in CNE2-IR, 40% in CNE2 NEK7 KD, versus 8% in CNE2). To evaluate the in uence of this spindle defect on the progression of mitosis, we performed live-cell imaging in the indicated cells using ectopically expressed GFP-tubulin and mCherry-Histone 2B to visualize mitotic spindle and chromosomes respectively (Fig. 3D). The duration of mitosis was about 2 hours in CNE2 control cells, whereas in CNE2-IR or NEK7 KD CNE2 cells, abnormal spindle formation was often observed and the mitotic phase was signi cantly extended (Fig. 3F, 113 ± 8 minutes in CNE2 cells versus 255 ± 18 minutes in CNE2-IR and 270 ± 19 minutes in CNE-2 NEK7 KD cells). In the most extreme cases, the cells were arrested in prometaphase for more than 300 minutes and could not nish mitosis (Fig. S3D).
The expression of NEK7 was reduced in the CNE2-IR cells. To test if the downregulation of NEK7 was important for maintaining the IR-resistance, we ectopically overexpressed NEK7 in the CNE2-IR cells (Fig.  S3B), and we found that increase the expression of NEK7 attenuated the viability of the radio-resistant CNE2-IR cells after irradiation (Fig. 3G).
These results indicated that many of the genes regulated by IR-induced transcriptional pausing were contributing to the development of radio-resistance, and identi ed NEK7 as a new IR-resistance factor, likely via affecting mitotic centrosomes as previously reported 13,14 .
Inhibiting transcriptional pausing reverses IR-resistance Since many of the pausing regulated genes were able to modulate IR-resistance, we reasoned that targeting transcriptional pausing could be an effective strategy to restrain resistance to radiotherapy. CDK7 in the TFIIH is required for RNAPII pausing and CDK9 in pTEFb promotes transcriptional elongation.
We rst tested if CDK7 inhibition could in uence sensitivity to radiation. We treated the CNE2 and CNE2-IR cells with two different CDK7 inhibitors, THZ1 and BS-181 15,16 . The CNE2-IR cells were more sensitive to THZ1 compared to CNE2 cells, with IC50 of 209.6 nM versus 1027.0 nM in CNE2 cells (Fig. 4A). We observed a reduced but consistent effect with BS-181 as well (Fig. S4A). We combined the chemical inhibition with radiation, and found that the joint treatment reversed the ratio-resistance of the CNE2-IR cells ( Fig. 4B and S4B). The treatment with THZ1 or BS-181 also inhibited the increase of RNAPII pS5 induced by radiation ( Fig. S4C and S4D). We next investigated if enhancing the activity of CDK9 could reduce resistance. Hexim1 is a key component of the ribonucleoprotein (RNP) complex inhibiting pTEFb 17 . We knocked down the expression of Hexim1 by shRNA (Fig. S4E), and observed a signi cant reversal of radio-resistance in the irradiated CNE2-IR cells (Fig. S4F). These results showed that inhibition of transcription pausing was an effective way to reverse IR-resistance in cell culture.
To test if inhibition of transcription pausing could be equally effective in vivo, we seeded the radioresistant CNE2-IR cells subcutaneously into nude mice and generated the xenograft model (Fig. 4C). After the tumor volume reached 200 mm 3 , mice in the experimental groups were injected with vehicle or THZ1 and received 4 Gy of irradiation treatment. The THZ1 was then administrated daily and the tumor volume was measured thrice a week. As shown in Fig. 4D, radiation alone only mildly inhibited the tumor growth, whereas radiation combined with THZ1 treatment completely blocked the tumor growth. We dissected the tumors at the end of the experiment, and prepared frozen sections of the tumor tissues for immunostainings with the RNAPII pS5 antibody (Fig. 4E). As expected, treatment with THZ1 inhibited the radiation-induced elevation of pS5 levels in the tumor tissues (Fig. S4G).
We wanted to know if the observed change in transcription pausing has clinical relevance. We collected 5 paired primary and recurrent nasopharyngeal carcinoma biopsy samples, and performed immunohistochemical analysis to detect the pS5 RNAPII levels (Fig. 4F). By carefully calculating the IHC scores 18 , we found the recurrent tumors had a small but steady increase in the pS5 level compared with the primary tumors (Fig. 4G), indicating that the observation made here was of clinical importance.

Discussion
In this study, we reported an IR-induced increase of transcription pausing indispensable for the development of resistance (Fig. 4H). The IR exposure evoked an accumulation in pS5 and paused RNAPII. This change of transcriptional kinetics was maintained on a fraction of genes, downregulating their expression and hence facilitating the cells to gain resistance to irradiation. We identi ed NEK7 as a new resistance gene subjected to such regulation. Knockdown of NEK7 disrupted the bipolar spindle formation, resulting in the prolongation of mitosis and genomic instability which might promote tumor evolution and the development of IR-resistance 3 .
The IR-induced increase in pS5 of RNAPII was sensitive to ATM and DNA-PK inhibition, indicating that the observed transcriptional change is an integral part of the cellular responses toward DNA DSBs 11 . The relationship between transcription and DSBs seems to be bilateral. On one side, transcriptional elongation at speci c loci requires local DNA breaks 19,20 , and the release of paused RNAPII promotes cancerassociated translocations, especially around the boundaries of the topologically-associating domains (TADs) 20 . On the other side, DNA DSBs induced by restriction enzymes cause a local accumulation of inhibitory H2AK119ub and transcriptional repression in proximity to the DSBs 20-23 . Our nding that IRinduced DSBs elicit a global increase in RNAPII pausing reveals another layer of regulation, adding to our understanding of transcriptional plasticity in DNA damage response.
How DNA DSBs signaling pathways modulate global transcriptional dynamics is the key question awaits future investigation. The answer is likely hidden in the complex post translational modi cations of the RNAPII CTD and its regulators. Components in the NELF complex are recruited to DSBs in a RNAPII and PARP-1 dependent manner to repress transcription and promote repair 24 . More interestingly Cyclin T, which is required for the activation of CDK9 kinase in P-TEFb, is able to undergo phase separation, forming liquid droplets in vitro that preferentially attract CTD pre-phosphorylated by CDK7 25 . It is tempting to speculate that DNA DSBs and poly(ADP-ribosyl)ation, by regulating the Cyclin T phase separation, modify the transcription cycle and increase RNAPII pausing. The increase in paused RNAPII may reduce transcription elongation and inhibit new initiation of transcription 26 .
Lastly, our discovery showed that inhibition of CDK7 reversed the IR-resistance of NPC cells, and that the recurrent NPC patient samples after radiotherapy displayed elevated pS5 of RNAPII, pointing to the possibility of enlisting CDK7 inhibitors as a new class of adjuvant chemo drugs to enhance the e cacy of radiotherapy and thwart treatment resistance. More preclinical and clinical studies are needed to further advance this emerging paradigm.

Construction and transfection of shRNA plasmids
The shRNA oligos were synthesized by Tsingke (Tsingke Biotechnology) and cloned into pLKO.1-TRC vector. The sequences of shRNA were listed in Supplementary Table S1.
Tumor cell viability analysis after irradiation
ChIP-seq raw reads were ltered using trim_galore v0.6.0, aligned to the hg38 genome assembly using Bowtie2 v2.3.5.1 with default parameters. Duplicate reads were removed using MarkDuplicates from the gatk package v.4.1.4.1. The pausing index (PI) was calculated as previously described 12 . The RefSeq gene model was downloaded from UCSC. ChIP-seq and input reads were calculated using bedtools coverage v2.26.0, mapped to the TSS regions (TSSR, -50 to +300 bp relative to TSS) and the gene bodies (TSS +300 bp to +3 kb past the TES) for each annotated RefSeq isoform. The reads density was normalized by the region length and by the mapped ltered read numbers multiplied by 1 million (rpm/bp). The input was then subtracted and PI was calculated as the ratio between RNAPII density in the TSSR and the gene body. For multiple RefSeq isoforms of the same gene, the one with the strongest RNAPII ChIP-seq signal in the TSSR, at least 0.001 rpm/bp, was selected. All those genes with PI >2 were de ned as paused, and the rest were non-paused. The bigwig tracks were generated using bamCompare from deeptools. Negative values were set to zero. IGV v.2.4.13 was used to visualize the bigwig tracks. ChIP-seq pro les were created by computeMatrix and plotPro le in deeptools.
RNA-seq and RT-qPCR RNA was extracted using TRIzol (Life Technologies, 87804). Libraries were prepared using mRNA-Seq Sample Preparation Kit (Illumina) and sequenced on an Illumina NovaSeq platform (Novogene). Raw reads were ltered using trim_galore, then mapped to hg38 genome assembly using STAR v2.7.1a. Differential expression analysis was performed using the Bioconductor package DESeq2. For RT-qPCR, RNA was converted to cDNA using the PrimeScript RT reagent Kit (Takara, RR037A).

Immunohistochemistry (IHC)
All the human tissue related experiments were approved by the Medical Ethics Committee of Central South University, and the informed consent was obtained from the patients. IHC was performed to detect the level of RNAPII pS5 in5 pairs of nasopharynx cancer samples using RNAPII phosphor S5 antibody (1:10000, Abcam, ab5408). The IHC score was evaluated by two independent investigators blinded to the histopathologic features and clinical characteristics using the intensity and proportion of positively stained tumor cells as previously described 18 .

Statistical analyses
The experiments were carried out at least three times. Data were presented as the mean ± standard deviation (SD). Statistical analysis and survival fraction analysis was performed using GraphPad Prism 9. Flow cytometry data were analyzed using ModFit LT 4.1. The statistical details of each experiment were indicated in the respective gure legends. The two-tailed unpaired Student's t-test or chi-squared test was performed to evaluate signi cant differences between the two groups. Kolmogorov-Smirnov test was used to compare signi cant differences between two groups of pausing index. P values are presented as star marks in gures: *p < 0.05, **p < 0.01, ***p < 0.001.

Data availability
ChIP-seq and RNA-seq datasets were deposited to the NCBI GEO database under the accession number GSE185423. All custom scripts are available from the authors upon request. Tables S1-s5 Tables S1-S5 are not available with this version. Figure 1 The