Ribonucleotide incorporation enables repair of chromosome breaks by nonhomologous end joining

See allHide authors and affiliations

Science  14 Sep 2018:
Vol. 361, Issue 6407, pp. 1126-1129
DOI: 10.1126/science.aat2477

RNA takes over DNA repair

Damage to DNA genomes is normally thought to be repaired with DNA. Pryor et al. now describe a clear exception (see the Perspective by Modesti). They found that RNA is routinely incorporated during the repair of DNA double-strand breaks through the mammalian nonhomologous end–joining (NHEJ) pathway. In a variety of contexts, including V(D)J recombination and Cas9-induced genome engineering, two “DNA” polymerases specific to NHEJ preferentially added RNA in cells. These RNA additions facilitated the critical step of ligation and were later replaced by DNA to complete the NHEJ repair process.

Science, this issue p. 1126; see also p. 1069


The nonhomologous end–joining (NHEJ) pathway preserves genome stability by ligating the ends of broken chromosomes together. It employs end-processing enzymes, including polymerases, to prepare ends for ligation. We show that two such polymerases incorporate primarily ribonucleotides during NHEJ—an exception to the central dogma of molecular biology—both during repair of chromosome breaks made by Cas9 and during V(D)J recombination. Moreover, additions of ribonucleotides but not deoxynucleotides effectively promote ligation. Repair kinetics suggest that ribonucleotide-dependent first-strand ligation is followed by complementary strand repair with deoxynucleotides, then by replacement of ribonucleotides embedded in the first strand with deoxynucleotides. Our results indicate that as much as 65% of cellular NHEJ products have transiently embedded ribonucleotides, which promote flexibility in repair at the cost of more fragile intermediates.

Nonhomologous end joining (NHEJ) is the primary pathway for repairing chromosomal double-strand breaks (DSBs) in mammals and is required for genome stability in all cell types, as well as for the assembly of antigen-specific receptors by V(D)J recombination in lymphocytes (1). NHEJ employs specialized nucleases and polymerases, including the widely expressed polymerase μ (Pol μ) (encoded by Polm) and lymphocyte-specific terminal deoxynucleotidyl transferase (TdT), to modify broken end structures in preparation for ligation (2). Accordingly, the loss of Pol μ or TdT results in impaired immune responses (36). Loss of the more widely expressed Pol μ additionally interferes with cell growth (7, 8), hematopoiesis (7), and resistance to DNA damage (79). Pol μ and TdT notably favor deoxynucleotides over ribonucleotides by a factor of 1.4 to 11 (depending on the nucleotide base) (10). By comparison, other polymerases that maintain DNA genomes (including the closely related Pol λ and Pol β) typically incorporate deoxynucleotides several thousand times more efficiently than ribonucleotides (1113). However, it is unknown whether ribonucleotide incorporation occurs during cellular NHEJ and, if ribonucleotide incorporation occurs, whether it substantially affects NHEJ function.

We initially investigated whether ribonucleotides are incorporated during NHEJ after introducing linear DNA substrates into transformed mouse embryonic fibroblasts (MEFs). We optimized this assay to allow for rapid harvesting of repair products, in anticipation that ribonucleotides were only transiently present. Ribonucleotides embedded in NHEJ products were quantified by assessing the template lost in samples upon cleavage of ribonucleotide-containing strands (Fig. 1A and fig. S1, B and C) with validated quantitative polymerase chain reactions (qPCRs) (figs. S1A, S3A, and S4B). We determined that embedded ribonucleotides were present in 60% (SD, 4.2%) of NHEJ products (Fig. 1B) when products were assessed within the first minute after electroporation, and these ribonucleotides were dependent on either Pol μ or TdT (Fig. 1C and fig. S1, E and F).

Fig. 1 Ribonucleotide incorporation during repair of extrachromosomal substrates.

(A) DNA fragments with 3′G overhangs at ends were introduced into MEFs, and the percentages of cellular NHEJ products with embedded ribonucleotides (% ribo.) were determined by comparing amplification efficiencies with and without prior cleavage at sites of ribonucleotide incorporation (see also fig. S1C). 5′P, 5′-phosphate. (B) Percentage of products with embedded ribonucleotides after the introduction of the substrate into Rnaseh2a+/+ or Rnaseh2a−/− MEFs. Data points represent the mean for three transfections, and error bars represent SD. (C) Percentage of products with embedded ribonucleotides among products recovered after 1 min. Data points represent the mean for three transfections, and error bars represent SD. Means were compared in pairs with values for Polm−/− by analysis of variance (ANOVA) (***P < 0.001). (D and E) Digestion of amplified products with NsiI and electrophoresis distinguishes Pol μ–dependent +C products (products with the addition of a single complementary C) from products with deletions of flanking sequence (Δ). (E) The mean percentages ± SD of products with embedded ribonucleotides among NsiI-susceptible (NsiIs) products recovered after 1 min from three independent transfections are shown. NsiIr, NsiI resistant.

Embedded ribonucleotides in NHEJ products decreased in frequency until they were almost undetectable after 20 min (Fig. 1B, gray line). To determine whether this reduction was due to the replacement of incorporated ribonucleotides with deoxynucleotides [ribonucleotide excision repair (RER)], we employed CRISPR-Cas9 to generate a MEF variant deficient in Rnaseh2a (fig. S1G), which initiates RER (14). Levels of embedded ribonucleotides in Rnaseh2a-deficient cells were initially equivalent to those in wild-type cells; unlike those in wild-type cells, embedded ribonucleotides in Rnaseh2a-deficient cells were not completely removed and stabilized at levels approximately half those initially observed (Fig. 1B, orange line). Reexpression of RNaseH2A in the Rnaseh2a-deficient variant was sufficient to reduce embedded RNA in NHEJ products to the low levels observed in wild-type cells (fig. S1H).

The substrate used in the experiments described above had a single-nucleotide 3′ overhang (3′G). Approximately half of repair products require ligation after the addition of a single complementary C, are dependent on both Ku and Pol μ (15), and can be identified by sensitivity to a restriction enzyme (NsiI) (+C product; Fig. 1D). Sequencing indicates that the remaining NsiI-resistant products have 1– to 5–base pair deletions of flanking sequence that are at best modestly affected by Polm deficiency (15). Embedded ribonucleotides were present in 91% (SD, 8%) of NsiI-sensitive products after 1 min (Fig. 1E). Similar results were observed when we used a different method to detect ribonucleotide-containing products (fig. S2A) and when we used a substrate with a different overhang template (C3′) and a different Pol μ–dependent added nucleotide (G) (fig. S2B). We conclude that most Pol μ– and TdT-dependent NHEJ products contain embedded ribonucleotides and that the modest preference of Pol μ and TdT for the addition of deoxynucleotides in vitro (10) is overwhelmed by higher concentrations of ribonucleotides in cells (15).

As also informed by data in subsequent figures, we suggest that early products involve one ligated strand only. Subsequent repair of the complementary strand with deoxynucleotides accounts for the twofold dilution of products with embedded ribonucleotides that is independent of Rnaseh2a (Fig. 1B, orange line), whereas complete removal of ribonucleotides requires Rnaseh2a-dependent RER.

We determined whether ribonucleotides are similarly incorporated during repair by NHEJ of chromosomal breaks. We used a pre–B cell line that can be induced to arrest in G1 phase and undergo V(D)J recombination at the immunoglobulin kappa locus (Igk) (Fig. 2A and fig. S3E) (15), because Pol μ is efficiently engaged by the 3′ overhang intermediates in this process (5, 1517). Embedded ribonucleotides were undetectable 24 hours after induction when cells were proficient in RER. By comparison, 35% of Igk recombination products had embedded ribonucleotides in an Rnaseh2a-deficient variant (Fig. 2B and fig. S3, A and B). This frequency is approximately half of the frequency for Igk products where Pol μ is active (5, 17), consistent with the model proposed above, where only the first strand of a chromosome DSB is repaired with ribonucleotides. Embedded ribonucleotides were again largely dependent on either Pol μ or TdT (Fig. 2B and fig. S3, C and D).

Fig. 2 Ribonucleotide incorporation during repair of chromosomal breaks.

(A and B) SP9 pre–B cells were induced for 24 hours, resulting in the expression of RAG1-RAG2 nuclease and the introduction of chromosome breaks adjacent to VK (variable region) and JK (joining region) coding segments (boxes). The percentage of products with embedded ribonucleotides in VJK coding junctions was measured as for Fig. 1A. (B) Data points represent the mean for five independent inductions, and error bars represent SD. Means were compared in pairs by ANOVA as noted (***P < 0.001). (C and D) Rosa26 locus–targeting Cas9 ribonucleoprotein was introduced into MEFs deficient in Rnaseh2a and expressing TdT. The percentage of products with embedded ribonucleotides was detected as for Fig. 1A by using a qPCR specific for the TdT-dependent +GG product (see also fig. S4). Ch. 6, chromosome 6. (D) Data points represent the mean for three independent transfections, and error bars represent SD.

We sought to track polymerase-dependent ribonucleotide incorporation during chromosomal NHEJ earlier than was possible using the V(D)J recombination model and also to extend analysis to nonlymphoid cells. We directly introduced Rosa26 locus–targeted Cas9 nuclease into Rnaseh2a-deficient MEFs, which allowed for rapid accumulation of repair products (fig. S4D) and thus assessment of ribonucleotides in these products immediately after they were generated. Sequencing of repair products from wild-type versus Polm−/− MEFs confirms that Pol μ promotes repair accuracy (fig. S4A). However, the blunt ends generated by Cas9 engage Pol μ less frequently (in 16% of all repair) than V(D)J recombination intermediates, and the contribution of Pol μ is distributed over several mostly template-dependent products that cannot be easily distinguished from polymerase-independent repair products. We therefore expressed TdT in these cells (fig. S1F), as this generates a class of repair products—with the addition of two or more G’s and no loss of flanking DNA (+GG products)—that are abundant (18% of NHEJ products) (table S1), unambiguously polymerase dependent, and that can be detected by a sensitive product-specific qPCR (Fig. 2, C and D, and fig. S4, B to D). Ribonucleotide incorporation by TdT does not differ significantly from that by Pol μ in cells (Figs. 1C and 2B) and in vitro (10), supporting the characterization of ribonucleotide incorporation by TdT during NHEJ as directly comparable to Pol μ–dependent repair. Embedded ribonucleotides were present in 84% and 77% of +GG NHEJ products 1 and 4 hours after the introduction of Cas9, respectively (Fig. 2D), and were reduced by a factor of two in these RER-deficient cells over the next 20 hours. This is consistent with strong favoring of ribonucleotide incorporation by these polymerases for first-strand repair, followed by repair of complementary strands with deoxynucleotides. As expected, levels of embedded ribonucleotides in repair products were much lower when cells were proficient in RER (23% after 1 hour) (fig. S4F) and undetectable at a nearby locus where breaks were not induced (whether cells were RER deficient or not) (fig. S4E).

We investigated the consequences of ribonucleotide addition for ligation, the next step of cellular NHEJ. We initially focused on a substrate with 3′GA overhangs, where 66% of NHEJ occurs by ligation after Pol μ–dependent addition of a single complementary C (fig. S5A). We made two variants of this substrate with a C already added: one where the added C was a ribonucleotide (+rC) and one where it was a deoxynucleotide (+dC) (Fig. 3A). We then introduced these two substrate variants into cells that express neither Pol μ nor TdT to isolate the effects of the different pre-added nucleotides on ligation and confirmed that repair under these conditions relies on the NHEJ-specific ligase (LIG4) (fig. S5B). Notably, only the +rC variant was able to efficiently promote direct ligation in cells, whereas the +dC variant was largely ineffective (the efficiency of direct ligation was reduced by a factor of >20) (Fig. 3B and Table 1). Direct ligation was also stimulated when the ribonucleotide 2′-OH was replaced with fluorine and blocked when the terminal nucleotide was replaced with a ribonucleotide stereoisomer, arabinofuranosylcytidine (Ara-C) (Fig. 3B and Table 1). Ara-C differs from rC only by the orientation of the 2′-OH and the favored sugar pucker (rC favors C3′-endo, and Ara-C favors C2′-endo), suggesting that a terminal C3′-endo nucleotide is required for the stimulation of ligation (fig. S5C). A terminal ribonucleotide also stimulated LIG4-dependent ligation in vitro, confirming that the effect on cellular NHEJ is specific to the ligation step. In contrast, T4 DNA ligase gained no benefit from a terminal ribonucleotide (fig. S5D). LIG4 may be alone among mammalian ligases in the ability to take advantage of added ribonucleotide termini, analogous to the in vitro activity of bacterial Pseudomonas LigD compared with that of other prokaryotic ligases (18).

Fig. 3 Effect of ribonucleotide termini on the NHEJ ligation step.

(A to C) Termini of NHEJ substrates were varied to be consistent with the polymerase-dependent addition of a ribonucleotide versus a deoxynucleotide and introduced into MEFs expressing neither Pol μ nor TdT. The sensitivity of amplified products to a diagnostic restriction enzyme (RE) was used to identify examples of direct head-to-tail ligation. (B and C) Substrates with the indicated terminal nucleotides were introduced into MEFs expressing neither Pol μ nor TdT. The mean percentage ± SD of directly ligated products for three independent transfections is noted below. (D) dGTP, rGTP, or an equivalent amount of the relevant salt (“none”) was added to Rosa26 Cas9–single guide RNA ribonucleoprotein (RNP) transfections performed as for Fig. 2, C and D, and genomic DNA was harvested after 1 hour. Data are the mean from four transfections, and error bars represent SD. Means were compared in pairs by ANOVA with values for no nucleotide triphosphate (NTP) addition (***P < 0.001). (E) Triple strand break repair model. Pol μ– or TdT-dependent ribonucleotide addition is noted in red.

Table 1 Stimulation of NHEJ repair pathway by a terminal ribonucleotide.

To represent head-to-tail alignment of substrate ends, 3′ overhang sequences for head ends are listed N3′, and tail ends listed 3′N (for example, GAC3′ and 3′AG align as in Fig. 3B). Underlined sequence letters correspond to the indicated terminal C. The relative joining efficiency is the joining efficiency for rC divided by the joining efficiency for dC, as measured by qPCR that amplifies all NHEJ products. Ligation stimulation = relative joining efficiency × (% direct ligation of rC)/(% direct ligation of dC).

View this table:

We addressed whether end structure context affected whether ribonucleotide additions stimulated ligation. An added ribonucleotide was required for direct ligation in cells whenever the opposite strand was mostly mispaired or gapped (Fig. 3C and Table 1); in the context of mispaired or gapped end structures, Pol μ and TdT are distinctively active during NHEJ (15, 17). By comparison, direct ligation is similarly efficient for ribonucleotide and deoxynucleotide additions in contexts where other polymerases are more active—mostly complementary overhangs—both in cells (Fig. 3C and Table 1) and in vitro (10, 19). The class of end structures where Pol μ and TdT distinctively contribute to cellular NHEJ thus correlates well with the class of end structures where ribonucleotides are required for direct ligation. Moreover, deoxynucleotide additions in these contexts were associated with frequent deletion of both the added nucleotide and flanking DNA (fig. S5, E and F), and repair was less efficient (Table 1). Our results imply that ribonucleotide addition is required for the biological activity of Pol μ and TdT.

We sought to more directly address whether polymerase function during cellular NHEJ relies on ribonucleotide addition. We introduced breaks in the chromosome with Cas9 and assessed whether the introduction of high concentrations of deoxynucleotide triphosphate affected the accumulation of the TdT-dependent repair products (+GG products) characterized in Fig. 2, C and D. The introduction of excess deoxyguanine triphosphate (dGTP) impaired the accumulation of these products by a factor of four relative to the accumulation in parallel experiments with unperturbed nucleotide pools. In contrast, the introduction of excess riboguanine triphosphate (rGTP) modestly stimulated +GG product recovery, consistent with the already high cellular rGTP pools in unperturbed cells (Fig. 3D). Similar results were also obtained by using two methods that more generally measure NHEJ-dependent short insertions and deletions (fig. S6, A and B). NHEJ was also impaired upon the introduction of Ara-GTP (fig. S6), in accordance with the inability of additions of this ribonucleotide stereoisomer to promote repair (Fig. 3B).

We show that Pol μ and TdT preferentially add ribonucleotides (Figs. 1 and 2) and contribute to the repair of a specific subset of end structures (15), that the same subset of end structures requires ribonucleotide additions for efficient LIG4-mediated repair (Fig. 3), and that only LIG4 may be able to take advantage of added nucleotides. Within all mammalian DNA metabolism, only the synthetic enzymes specific to NHEJ appear to cooperate in this unusual manner. Our results suggest that this reflects a coevolution of these enzymes to better repair damaged or mispaired ends, a central problem of this pathway.

Our results also have more general relevance. Ribonucleotides destabilize DNA genomes unless removed by RER (20, 21), and when incorporated during NHEJ, they pose special problems for the RER pathway. We show that safe accommodation of ribonucleotide-containing intermediates during NHEJ is most likely enabled by three sequential coupled strand break repair reactions (Fig. 3E), with each reaction coupled to the next: repair of the first strand with ribonucleotides, repair of the second strand with deoxynucleotides, and RNaseH2A-dependent excision of the ribonucleotides embedded during first-strand repair. This model explains the twofold dilution of embedded ribonucleotides that is independent of RER and RNaseH2A (Fig. 1B). Alternative models—where ribonucleotides are incorporated into both strands or where RNaseH2 incises the first strand before second-strand repair is complete—risk rebreakage of the chromosome. Additionally, the transient nature of intermediates with embedded ribonucleotides (for which the half-life is likely less than 5 min) (Fig. 1B) suggests that all three strand break repair reactions are joined together, possibly by physical interactions between pathway components.

Pol μ is widely expressed and participates in as little as 16% (e.g., fig. S4A) to as much as 66% (e.g., fig. S5A) of repair, depending on the end structure (15). In lymphocytes, either TdT or Pol μ is active in 65% of NHEJ events required for V(D)J recombination (5, 17). The triple stand break repair model proposed here (Fig. 3E) is thus relevant to a large fraction of mammalian NHEJ and is a fundamental departure from the previously accepted model. It is probably relevant to NHEJ in other species (e.g., yeast and bacteria) (18, 22, 23) as well. Our work further indicates that ribonucleotide incorporation is required if mammalian Pol μ and TdT are to be effective in promoting long-term cellular proliferative capacity, the development of adaptive immunity, and radioresistance (39).

Supplementary Materials

Materials and Methods

Figs. S1 to S6

Tables S1 and S2

References (2427)

References and Notes

Acknowledgments: We thank S. N. McElhinny, J. Havener, T. Kunkel, R. S. Williams, Q. Zhang, and Ramsden lab members for help guiding the experiments described in this work; L. Blanco for providing MEFs; Y. Chang for providing SP9 cells; and E. Hendrickson for providing HCT116 cells. Funding: This work was supported by NCI R01CA097096 (D.A.R.), ACS PF-14-0438-01-DMC (J.M.P.), NCI F31CA203156 (M.P.C.), T32GM007092 (M.P.C. and M.E.L.), and T32GM119999 (A.J.L.). Author contributions: J.M.P., M.P.C., and D.A.R. authored the manuscript, designed experiments, and analyzed data. J.M.P., M.P.C., J.C.-G., M.E.L., A.J.L., and G.W.S. performed experiments. Competing interests: J.M.P. is currently an employee at New England Biolabs, a manufacturer and vendor of some of the molecular biology reagents used in this work. Data and materials availability: All materials and raw data are available upon request.
View Abstract

Stay Connected to Science

Navigate This Article