We first investigated DNA synthesis across a short RNA tract during the repair of a chromosomal DSB, using the capacity of ssDNA oligonucleotides to serve as a template for efficient DSB repair^16,17. A site-specific DSB was induced within the LEU2 gene by overexpression of HO endonuclease¹⁸. Following DSB induction, cells were transformed with ssDNA oligonucleotides that were designed (see Fig. 1) to join LEU2 ends and introduce a unique, in-frame 12-base insert containing 0, 4, 6 or 12 RNA bases (a, b, c and d, respectively), or a 6-base insert with 0 or 6 RNA bases (e and f). To accomplish DSB repair and restore a functional LEU2 gene, the insert sequence must be used as a template. Remarkably, repair by b (containing 4 ribonucleotides) was only a factor of 3 lower than repair by the DNA-only control (a). The frequencies decreased with increasing RNA-tract length (b, c, d, Fig. 1). The appearance of the oligonucleotide sequence in the Leu⁺ transformants (see ‘Verification’ in Supplementary Table 2a) suggests that the RNA-containing molecules participate directly in the repair. This contrasts with the observation that two ribonucleotides at the mating type locus of Schizosaccharomyces pombe seem to block DNA replication and lead to a site-specific DSB¹⁹. However, the reduced transformation frequency with increased size of RNA might be due to a higher likelihood of replication arrest. If so, it is unlikely that cDNA generated by Ty reverse transcriptase of the RNA-containing oligonucleotides would be the source of DSB repair.

To exclude this possibility, that RNA-containing oligonucleotides were copied into cDNA before interacting with the DSB ends, we used our recent finding that a DSB can activate strand-biased targeting by ssDNA oligonucleotides with homology to a distant site¹⁷. Because many kilobases (kb) of the 5′ strand can be degraded before repair³, there is a bias for the oligonucleotide complementary to the 3′ strand. If a cDNA intermediate were formed from ssRNA-containing oligonucleotides, the observed bias should be opposite to that with the corresponding ssDNA oligonucleotides, and there would be no bias if dsDNA were formed. As shown in Fig. 2, targeting with RNA-containing oligonucleotides was biased in favour of the oligonucleotide complementary to the 3′ end of the break, similar to DNA-only oligonucleotides. No bias was detected without DSB induction (Supplementary Table 3). Transfer of the BamHI site contained in R.w and R.c (Fig. 2a) was confirmed in 28/30 transformants. We conclude that yeast cells have the ability to use RNA embedded in DNA as a template within the chromosome.

We then examined repair by RNA-only molecules that were homologous to both sides of a DSB. As shown in Fig. 1, the Leu⁺ transformation frequency with five nmoles of oligonucleotides h or i reached 5 × 10⁻⁷, and restoration of LEU2 sequence was precise (in 26/27 clones tested). In the absence of oligonucleotides, the frequency of Leu⁺ colonies was ~1 × 10⁻⁸ and in all tested isolates (16/16) the Leu⁺ phenotype was due to imprecise non-homologous end joining causing small insertions or deletions (Supplementary Table 2a). Thus, ssRNA oligonucleotides are estimated to increase the precise repair of a DSB in LEU2 by over 500-fold. Similar RNA oligonucleotides containing 2-base substitutions in the centre (k and l) gave comparable Leu⁺ transformation frequencies (Fig. 1). The mutations were precisely transferred (Supplementary Table 2a), indicating DNA synthesis across the RNA templates.

Several oligonucleotides were used to better understand DSB repair by RNA. The minimum size for repair by RNA-only molecules was greater than 45 bases (k.45 and l.45 in Supplementary Table 2a). The repair was enhanced by the presence of DNA at one end of the RNA oligonucleotides, regardless of whether the DNA was homologous (m, n, o and p) or non-homologous (r and s) to the DSB ends. Repair frequencies by oligonucleotides that required DNA synthesis through long RNA tracts were much lower than those of corresponding oligonucleotides requiring less synthesis on RNA (Fig. 1, compare n with p; also compare b, c, d and f). RNA oligonucleotides containing DNA homologous to a DSB end at their 5′ (m) or 3′ end (n) were over 100 and 1,000-fold more efficient, respectively, than the RNA-only molecules (Fig. 1). One possible explanation is that rather than simply enhancing annealing, the homologous DNA can provide a DNA duplex region close to the 3′ end of the break that could facilitate polymerase binding (compare n with m and with o). It is also possible that the DNA end could prevent end-degradation, especially from the 3′ end (compare m with n, and o with r and s) (Fig. 1). In agreement with the protection idea, the presence of a 20-base non-homologous DNA tail at the 3′ end of an oligonucleotide containing 60 nucleotides of homologous RNA (r and s), but not a 3-nucleotide tail (r3), increased repair by a factor of 100 when compared with the RNA-only oligonucleotides (k and l) containing an even longer RNA stretch of homology (Fig. 1 and Supplementary Table 2a). The transformation differences associated with the longer DNA tail could be due to protection from 3′ → 5′ exoRNases, which have a major role in RNA surveillance²⁰. Deletion of the non-essential 3′ → 5′ exoRNase RRP6 did not increase transformation with an RNA-only oligonucleotide (k) (Supplementary Table 4). However, because most 3′ → 5′ exoRNases are coded by essential genes in yeast, further genetic studies are required to examine the potential involvement of these enzymes. Because RNA-templated repair must involve RNA–DNA duplex intermediates, we also examined the possible role of RNases H1 (RNH1) and H2 (RNH35), which can efficiently degrade messenger RNA paired with DNA¹¹. There was no increase in RNA oligonucleotide-mediated repair for the single or the double mutants (Supplementary Table 4).

Targeting of RNA oligonucleotides was stimulated by the DSB by at least 100-fold (compare n or r added to cells in galactose medium to induce the HO-endonculease versus no galactose, Supplementary Tables 2a and b) and occurred independently from the strand invasion function of Rad51 (Supplementary Table 4). The targeting was also independent of chromosomal locus because a DSB induced within TRP5 was also precisely repaired by RNA oligonucleotides (T2 and T4 in Supplementary Table 2c).

Except for telomerase genes, the only genes in yeast known to code for reverse transcriptases are those contained in Ty elements. Deletion of the SPT3 gene, which is essential for transposition and transcription of Ty1 and Ty2 elements^4,21, did not affect RNA-templated repair (Supplementary Table 5). We conclude that Ty reverse transcription has at most a minor role in DSB repair events mediated by a homologous RNA template. Deletion of telomerase genes EST2 or EST1 (ref. ²²) also did not affect transformation by the RNA-containing oligonucleotides (Supplementary Table 5). This contrasts with the strong stimulation of Ty cDNA synthesis and transposition in an est2-null mutant⁸.

It is also possible that DNA synthesis during RNA-templated repair is accomplished by DNA polymerases. However, deletion of individual nonessential yeast DNA polymerase genes (POL4, REV1, REV3, RAD30 or MIP1)²³, as well as double and triple deletion mutants of translesion DNA polymerase REV1, REV3 and RAD30 genes, did not alter transformation by RNA-containing oligonucleotides (Supplementary Table 5). We conclude that the ability to synthesize DNA on an RNA tract during repair could be a redundant function among polymerases and/or a function of one or more essential replicative DNA polymerases.

We therefore examined the ability of yeast DNA polymerases α and δ to copy templates with sequences corresponding to oligonucleotides a and b containing 0 or 4 RNA bases (Fig. 3a and Supplementary Table 1). Both polymerases copied all 4 ribonucleotides and generated full-length products (Fig. 3b; compare lanes 8 and 9, and 14 and 15). Synthesis by Pol α, but not by Pol δ (not shown), was stimulated by Mn²⁺ (Fig. 3b, lanes 2–3 versus 8–9). These polymerases also partially copied a template (IV) in which the RNA tract starts at the first single-strand template position (Fig. 3b, lanes 11 and 17). Greater extension occurred when synthesis started from a DNA tract (for example, substrate III, lanes 10 and 16, and substrate II containing a 4 ribonucleotide tract (R 4), lanes 9 and 15). Although not tested, the DNA polymerase co-factor proliferating cell nuclear antigen might enhance synthesis further, because PCNA enhances Pol δ processivity when copying DNA. On an RNA-only template (V), Pol α incorporated up to 12 bases in the presence of Mn²⁺ (lane 6), and both polymerases added a nucleotide with or without the presence of a downstream oligonucleotide (Dw), which created a gap, if Mg²⁺ was present (Fig. 3b, lanes 12 and 18, and Fig. 3c). When the efficiency of copying the 4-nucleotide template tract was determined (as described in Section 3.1 in ref. ²⁴), Pol δ extended 80% of the products beyond R 4 and 76% of the products beyond the equivalent D 4 tract. Thus, the RNA tract was copied by Pol δ as efficiently as the corresponding DNA tract. Interestingly, once the R4 tract was copied by Pol δ, further elongation was impeded by the presence of the RNA–DNA duplex upstream of the Pol δ active site (for example, see band highlighted with a black asterisk in Fig. 3b, lane 15). Pol α also copied the R4 tract, but less efficiently (27% of products beyond R4 compared with 92% of products beyond D4). Moreover, Pol α required 20-fold more binding–synthesis-dissociation cycles (determined as described²⁴) than Pol δ to completely bypass R4. Thus, Pol δ may be the more likely candidate for reverse transcriptional repair of a DSB in vivo. It will be interesting to assess the effects of specific replicative DNA polymerase mutants on DSB repair by RNA-containing molecules.

The finding of in vivo and in vitro DNA synthesis on RNA-containing oligonucleotides is relevant to situations where RNA might appear within DNA in vivo, as shown for the mammalian mitochondrial genome²⁵. The inclusion of RNA bases could occur during normal DNA metabolic reactions, as indicated by the ability of several DNA polymerases to incorporate ribonucleotides in vitro^25,26 and the ability of DNA ligase I to ligate RNA bases into DNA during Okazaki fragment maturation in vitro²⁷.

The RNA-templated DSB repair presented here is clearly distinct from previously described RNA/cDNA-mediated DSB repair processes^4–7,12. We show that there is no barrier to the direct transfer of information from RNA templates to chromosomal DNA. On the basis of this newly discovered RNA capability, we suggest that endogenous RNA could have a direct role in repairing lesions during or after transcription, especially given its high local concentration. Our results set the stage for understanding how direct, homology-driven transfer of endogenous RNA information to DNA may occur.

The ability of RNA to transfer genetic information to homologous chromosomal DNA could lead to new directions in gene targeting, given that RNA can be amplified at will within cells. Moreover, RNA as a homologous template in DNA repair may contribute to both genome integrity and evolution.

supplement