Stepwise requirements for Polymerases δ and θ in Theta-mediated end joining

Summary Paragraph

Timely repair of chromosomal double strand breaks is required for genome integrity and cellular viability. The Polymerase Theta-mediated End Joining pathway has an important role in resolving these breaks and is essential in cancers defective in other DNA repair pathways, thus is an emerging therapeutic target¹. It requires annealing of 2–6 nucleotides of complementary sequence – microhomologies – that are adjacent to the broken ends, followed by initiation of end-bridging DNA synthesis by Polymerase theta. However, the other pathway steps remain inadequately defined, and the enzymes required for them are unknown. Here we demonstrate additional requirements for exonucleolytic digestion of unpaired 3’ tails before Polymerase theta can initiate synthesis, then a switch to a more accurate, processive, and strand-displacing polymerase to complete repair. We show the replicative polymerase, Polymerase delta, is required for both steps; its 3’ to 5’ exonuclease activity for flap trimming, then its polymerase activity for extension and completion of repair. The enzymatic steps that are essential and specific to this pathway are mediated by two separate, sequential engagements of the two polymerases. The requisite coupling of these steps together is likely facilitated by physical association of the two polymerases. This pairing of Polymerase Delta with a polymerase capable of end-bridging synthesis, Polymerase theta, may help explain why the normally high-fidelity Polymerase delta participates in genome de-stabilizing processes like mitotic DNA synthesis² and microhomology-mediated break induced replication³.

Steps required for repair by TMEJ

Chromosome double strand breaks (DSBs) are repaired by Homologous recombination (HR), Non-homologous end joining (NHEJ), or a poorly understood pathway dependent on Polymerase theta (Polθ, gene name POLQ) appropriately termed theta-mediated end joining (TMEJ)¹. In mammals, TMEJ is largely equivalent to microhomology-mediated end joining and alternative end joining. Initial pathway choice is determined in part by 5’ to 3’ nucleolytic resection of DSB ends, as the resulting 3’ ssDNA tails are required for TMEJ and HR but impair repair by NHEJ. Genetic and biochemical studies argue TMEJ initiates by a Polθ-dependent search to identify and anneal 2–6 nucleotides of complementary sequence on either side of the resected ends (Fig. 1a, Step 1)^4–6. Polθ is then essential for synthesis initiated from the annealed microhomology (MH). However, MHs of sufficient size for Polθ to act are predicted to be embedded in 3’ ssDNA tails for over 95% of DSBs⁷, thus the resulting 3’ flaps must first be trimmed by a previously uncharacterized nuclease before Polθ can initiate synthesis. The steps following initiation of synthesis by Polθ are also not well understood (Fig. 1a, after Step 3).

Fig. 1. TMEJ requires a flap-trimming exonuclease and a secondary DNA polymerase.

a, TMEJ extrachromosomal reporter system and required repair steps. TMEJ is measured by qPCR, and is initiated by annealing of microhomologies (MH, red) between the head of one DNA molecule and the tail of another. Putative Polθ independent steps are highlighted in blue and yellow. b, Quantification of extrachromosomal TMEJ in RPE1 cells with a 4 bp MH (red bars), varying locations of phosphorothioates in the DNA substrate as noted (stop signs). Data are from 3 biological replicates analyzed with a one-way ANOVA and Dunnet’s method. Data are mean ± standard deviation (SD), nd; below limit of detection. c, Polθ synthesis reporter substrate. Triplicate Thymines are spaced every 8 bp along the ssDNA tract to be synthesized. d, Frequencies of mutations generated by cellular TMEJ-associated synthesis (squares) vs. control synthesis (Q5 polymerase, circles) are plotted as a function of distance from the microhomology. Data are from 3 biological replicates analyzed with a two-way ANOVA and Šidák’s test; data are mean ± SD. e, Schematic of the TMEJ strand displacement reporter. Substrates have ends with partly complementary 3’ overhangs that are either 4 (NHEJ) or 45 nts long (TMEJ) and possess a 5’ terminal nt (dN) or abasic site (Ab). A mismatched BamHI site is 20 or 50 bp downstream of the 5’ terminus, such that repair products become sensitive to BamHI if strand displacement synthesis occurs, and remain BamHI resistant in the absence of strand displacement synthesis. f, The BamHI substrates were introduced into mouse embryonic fibroblasts and digested with BamHI where indicated prior to amplification.

We investigate here the steps integral to the TMEJ pathway, as well as the enzymes required for each step. We initially employ a series of extrachromosomal substrates (Fig. 1a), which when introduced into mammalian cells require Polθ for efficient repair (Fig. 1b, Extended data Fig. 1a; repair measured by qPCR is at least 100-fold lower in cells deficient in Polθ, relative to the wildtype control).

We explored first the type of nuclease (i.e., endonuclease or exonuclease) responsible for removing 3’ flaps (Fig. 1a, step 2). We used substrates wherein a 4 nucleotide MH is annealed to generate 5 nucleotide 3’ flaps, then introduced nuclease-blocking phosphorothioate (PT) substitutions at varied phosphodiester bonds in the flaps. We assessed first the impact of PTs in the 4 bonds located closest to the 3’ terminus of both ends, leaving only the bond that must be cleaved to activate synthesis (the 5^th bond) unmodified. No significant TMEJ is observed when both ends were modified in this fashion (Fig. 1b). The nuclease required must thus cleave bonds downstream of the critical 5^th phosphodiester, progressing to this bond in steps. Notably, there was no significant effect on repair of blocking only one end, arguing TMEJ does not require synthesis to be bi-directional. We then assessed effects of a single PT substitution on one end while blocking the other with 4 PT substitutions. We observed equivalent, 2-fold inhibition when comparing PT substitution of the most 3’ terminal bond vs. the critical 5^th bond (41% vs. 47%) (Fig. 1b). That repair is inhibited approximately 2-fold is consistent with the presence of two stereoisomers in PT substituted bonds, only one of which blocks nuclease activity⁸. In accord with this interpretation, substitution of only the 3’ terminal nucleotides with locked nucleic acids also blocks TMEJ (Extended data Fig. 1b). We conclude flap trimming during TMEJ requires a ssDNA specific, 3’ to 5’ exonuclease (i.e., a nuclease that obligatorily cleaves in mononucleotide steps, starting from the 3’ terminus).

We next investigated whether Polθ-initiated synthesis from the trimmed end is sufficient to complete repair. We tracked whether synthesis in repair products was consistent with Polθ activity using its mutational signature - a tendency to insert or delete an adenine opposite 3 successive template thymidines that is much higher than other DNA polymerases⁹. We altered the extrachromosomal substrate described above to possess 3 thymidines every 5 nts in the template, then sequenced products of cellular repair, as well as a control reaction assessing error due to sample processing (Fig. 1c). The Polθ signature was evident at the first triple thymidine site only, suggesting Polθ typically performs 6–14 nucleotides of DNA synthesis during TMEJ before there is a switch to a more accurate polymerase (Fig. 1d).

Synthesis in TMEJ may then arrest after gap filling and ligation, as in NHEJ, or it may continue and displace the downstream strand. We sought to distinguish between these resolving mechanisms by embedding a mispaired BamHI site in double stranded DNA, 20 or 50 bp downstream of the 5’ end in our extrachromosomal substrate (Fig. 1e). This BamHI site remains mispaired if synthesis arrests after gap filling and ligation, as is apparent from the resistance of cellular NHEJ repair products to BamHI digestion (Fig. 1f). By comparison, TMEJ repair products are >90% sensitive to BamHI digestion when located 20 bp downstream, and 80% sensitive to BamHI digestion when located 50 bp downstream (Fig. 1f, Extended data Fig. 1c–e). Repair by NHEJ, but not TMEJ, is also impaired by a ligation-blocking abasic site at the 5’ end of the downstream strand. We conclude TMEJ is resolved with strand-displacing synthesis.

Polδ is required for two steps in TMEJ

Our results indicate a 3’ to 5’ exonuclease trims flaps before Polθ initiates synthesis, and suggest there is a subsequent switch to synthesis mediated by a more accurate, processive, and strand-displacing polymerase. We considered Polymerase delta (Polδ) as a plausible candidate for performing both steps. TMEJ is negligible in cells made Polδ deficient by shRNA mediated depletion of the catalytic POLD1 subunit (Fig. 2a,b, Extended data Fig. 2a), and is reduced in cells depleted of the non-catalytic subunits POLD2 and POLD3 (Extended data Fig. 2b–d). Importantly, non-specific effects of Polδ depletion are excluded by use of a spike-in control “minimal TMEJ” substrate that requires neither candidate Polδ dependent step (Extended data Fig. 2e, grey boxed substrate in Figures), but which remains dependent on Polθ (Extended data Fig. 1a). We conclude Polδ is as essential to repairing biologically relevant TMEJ substrates as is Polθ.

Fig. 2. Polymerase Delta is required for both flap-trimming and processive synthesis.

a, Quantification of repair of a 70 nt synthesis, 5 nt flapped TMEJ substrate relative to the minimal TMEJ substrate, normalized to repair in untreated cells. Polθ is inhibited (Polθi) by treatment of cells with 2 μM ART558. POLD1 in RPE1 cells is endogenously expressed (+) or depleted (−) by shPOLD1 treatment. Data are from 3 biological replicates and analyzed with a one-way ANOVA and Dunnet’s method. Data are mean± SD, nd; below limit of detection. b, Western blot showing shPOLD1 depletion, then complementation after expression of FLAG-tagged POLD1 in RPE1 cells. Actin was used as a loading control. c, Structural representation of the Polδ holoenzyme bound to DNA with POLD1 in cyan, remaining subunits in blue, and PCNA in pink, using 6TNY13 from the protein data base and rendered with PyMOL54. Amino acids in POLD1 that were mutated are indicated and highlighted by magenta sphere atoms. d, Quantification of repair of the 5 nt flapped substrate relative to the minimal substrate, normalized to flapped TMEJ in untreated cells. Data are from 3 biological replicates and analyzed with a one-way ANOVA and Dunnet’s method. Data are mean ± SD. e, In vitro flap cleavage experiment. 25nM double stranded, forked DNA substrates with 2, 5, and 10 nt 3’ ssDNA overhangs or 5 nt 3’ overhangs with 3 terminal phosphorothioates (5PT) were incubated with 25nM human Polδ for 10 minutes excepting 5PT, which was incubated for 30 minutes. f, Quantification of repair of the 70 nt processive synthesis substrate relative to the minimal substrate, normalized to processive TMEJ in untreated cells. Data are from 3 biological replicates analyzed with a one-way ANOVA and Dunnet’s method. Data are mean ±SD. The vertical grey line separates data from independent experiments.

We investigated next whether Polδ was required for both TMEJ steps: we independently varied catalytic competency of Polδ using separation-of-function mutations, as well as TMEJ requirements using different substrates. With respect to catalytic competency, we complemented cells depleted of the POLD1 catalytic subunit with wild type POLD1, or POLD1 mutants defective in 3’ to 5’ exonuclease activity (D402A¹⁰ or Exo^M), polymerase activity (S605del¹¹ or Pol^M, and D755A+D757A¹² or Pol^M2), or interaction with proliferating cell nuclear antigen (PCNA) (L1002A+F1005A¹³ or PIP^M) (Fig. 2b,c, Extended data Fig. 2f–2i). We then assessed activity of the different Polδ variants using TMEJ substrates dependent on only one of the two steps; a flapped substrate dependent on trimming but requiring only minimal synthesis (flap-trimming TMEJ, blue box in Fig. 2d), as well as a substrate that doesn’t require trimming (unflapped) but dependent on processive synthesis (processive synthesis TMEJ, yellow box in Fig 2f).

We assessed flap-trimming TMEJ first, and determined it was negligible in cells deficient in endogenous Polδ or Exo^M expressing cells, but unaffected in cells expressing either Pol^M or Pol^M2 (Fig. 2d, Extended data Fig. 3a). Repair requiring trimming of 2 and 10 nt flaps has similar dependencies (Extended data Fig. 3b,c). By comparison, APEX2 and ERCC1-XPF can both be implicated generally in TMEJ (Extended data Fig 3g)^14,15, but unlike Polδ, neither of these candidate flap nucleases are required for this pathway’s flap trimming step (Extended data Fig 3d–g). We additionally employed purified human Polδ and confirmed Polδ exonuclease activity in vitro effectively trims the range of TMEJ intermediates expected in cells (2, 5 and 10 nucleotide flaps)⁷ (Fig. 2e). Moreover, sequential PT substitutions in flap phosphodiester bonds inhibit trimming activity of Polδ by approximately 2-fold for each nucleotide step, which is consistent with our cellular results (Fig. 1b). Polδ exonuclease activity is thus required for flap trimming during TMEJ.

We assessed processive TMEJ next, and determined it was similarly ablated in Polδ deficient cells, but here we observed reciprocal dependencies on Polδ variants: processive TMEJ was negligible in cells expressing Pol^M or Pol^M2, and unaffected in Exo^M (Fig. 2f, Extended data Fig. 4a). An intermediate level of dependency on Polδ is observed when TMEJ requires 45 nt of synthesis (reduced 2-fold), and this residual repair was no longer associated with significant strand displacement (Extended data Fig. 4b–d). TMEJ requirements were not observed cells with null mutations in Polθ-related Pol Nu, or the Polδ-related Pol zeta (Extended data Fig. 4f,g). Notably, the intrinsic processivity of human Polδ is low^12,13,16, thus we investigated a possible role for association of the PCNA processivity clamp with Polδ. Processive TMEJ was nominal in cells expressing the Polδ PIP^M mutant (defective in PCNA association), and also reduced when cells are depleted of PCNA (Fig. 2f, Extended data Figs. 2f, 4e). We conclude PCNA-assisted Polδ polymerase activity is required when TMEJ requires filling a gap of 70 nucleotides or more. In sum, we show processive Polδ – at least POLD1, POLD2, POLD3 and PCNA - is required for efficient cellular TMEJ, and that the two different catalytic activities intrinsic to the POLD1 subunit are essential for two separate steps.

We next sought to address how the different Polδ catalytic activities contributed to chromosomal TMEJ. We introduced chromosomal DSBs at the LBR locus using Cas9, then developed quantitative PCRs specific to different pathway-dependent repair products (Fig. 3a). We define as TMEJ two MH-associated deletion products and a family of related template dependent insertion products (TINS) by showing that they are depleted in Polθ deficient cells (Fig. 3b–e, Extended data fig. 5a,b), consistent with past work^7,17,18. These TMEJ products were equivalently reduced in cells deficient in Polθ, Polδ, and both polymerases (Figs. 3b–3e), indicating Polδ is just as important for chromosomal TMEJ as is Polθ. With respect to different Polδ functions, chromosomal TMEJ is similarly dependent on Polδ exonuclease activity, polymerase activity, POLD2 and POLD3 subunits, and the ability of Polδ to interact with PCNA (Fig. 3b–d, Extended data Fig. 5c,d), with one notable exception. Levels of the terminal MH product are partly rescued in cells expressing Polδ Exo^M, relative to cells entirely deficient in Polδ. This is consistent with a reduced requirement for exonuclease activity on the predicted intermediate, which would not have a flap on one end (Fig. 3c). These effects of Polδ depletion are specific to TMEJ, as it does not impact the abundance of an NHEJ-mediated 1 bp insertion product^17,19 (Fig. 3f, Extended data Fig. 5e,f). We additionally do not observe similar effects on TMEJ upon depletion of Polymerase Epsilon (Pol ε), the leading strand replicative polymerase (Extended data fig. 5g,h). TMEJ thus requires Polδ as much as it does Polθ, engages resected intermediates that require at least 45 nt of synthesis for repair, and can dispense with the requirement for Polδ exonuclease activity on rare occasions when MHs are present at the exact termini of resected ends.

Fig. 3. Polymerases Delta and Theta are equally required for chromosomal TMEJ.

a, Diagram of Cas9 chromosomal repair reporter system at the LBR locus. Differences in repair are measured for NHEJ by quantification of a single nucleotide insertion, and for TMEJ by quantification of two different products mediated by microhomologies (MH), or products with templated insertions (TINS). b, Quantification of TMEJ at an embedded MH by qPCR. RPE1 cells express Polθ (+) or are genetically deficient (−). POLD1 in RPE1 cells is endogenously expressed (+) or depleted (−) by shPOLD1 treatment. Data are from 3 biological replicates analyzed with a one-way ANOVA and Dunnet’s method. Data are mean ± SD. c, Quantification of TMEJ at a terminal MH performed as in b. d, Quantification of TMEJ at the terminal MH in c performed as in b. e, Frequency of TINS repair products as measured by digital droplet PCR. Data are from 3 biological replicates analyzed with a one-way ANOVA and Dunnet’s method. Data are mean ± SD. f, NHEJ repair quantification performed as in b, with the exception of 2 replicates for the shPOLD1 sample.

Polδ physically interacts with Polθ

TMEJ requires in sequence the alignment of MHs by Polθ, flap trimming by Polδ, initiation of synthesis by Polθ, and finally switching to more processive synthesis mediated by Polδ. This need for alternating engagements of the two polymerases on a common substrate is best served by a physical association. We therefore introduced a tagged Polθ (Halo-Polq) into RPE1 cells and immunoprecipitated this protein. We recovered Polδ but not the related and similarly abundant Pol ε, confirming specific physical interaction between Polδ and Polθ (Fig. 4a; Extended data Fig. 6a). We also assessed if the two conserved domains of Polθ – a Helicase-like domain and its Polymerase domain¹ – independently interact with Polδ, by separately introducing flag tagged versions of each domain into Polθ deficient U2OS cells. Both Polθ domains could be recovered after immunoprecipitating Polδ (Fig. 4b), and Polδ was equivalently recovered after immunoprecipitating each Polθ domain (Extended data Fig. 6b). We again failed to recover the related Pol ε after precipitation of either Polδ domain (Extended data Fig. 6c). The two conserved domains of Polθ thus specifically and independently associate with Polδ. We further assessed the deletion of the POLD1 75 aa C-terminal CysB motif (POLD1 ΔCysB) on the two interactions. The POLD1 ΔCysB mutation ablated interaction with the Polθ helicase-like domain, but had no impact on interaction with the Polθ polymerase domain, thus identifying distinct molecular requirements for Polδ interaction with the two Polθ domains (Fig. 4c; Extended data Fig. 6f–h).

Fig. 4. Polymerases Delta and Theta physically associate.

a, Co-IP of RPE1 cells untreated (−) or treated with neocarzinostatin (+NCS) and expressing Halo-Polθ was pulled down with a Halo antibody (αHalo) and recovery probed with a POLD1 antibody (αPOLD1). b, Co-IP of FLAG-tagged domains of Polθ in POLQ−/− U2OS cells. Polδ was pulled down with αPOLD1, and recovery probed with αFLAG. c, Co-IP of FLAG-tagged domains of Polθ in POLQ−/− U2OS cells expressing endogenous POLD1, Myc-tagged POLD1, or Myc-tagged POLD1 ΔCysB. The ΔCysB mutant is truncated at amino acid 1033 (a loss of 75 amino acids). Polθ domains were pulled down with a FLAG antibody, and recovery probed with αMyc. d, Microscale thermophoresis of labeled Polθ ΔCen with varied amounts of Polδ holoenzyme. Kd ± the 95% confidence interval of 3 technical replicates. Data are mean ±SEM. e, Representative image of 53BP1, Halo-Polθ, and POLD1 in a single nucleus treated with NCS. Boxes denote triple association events. Scale bars: whole nuclei = 1500 nm, insets = 150 nm. f, Quantification of STORM in e. Relative density of POLD1 at sites of Polθ−53BP1 localization was plotted ± NCS pre-treatment. Data with median and quartiles from 3 biological replicates of 57 and 55 nuclei, respectively; analyzed with an unpaired Mann Whitney test. g, Requirement for stepwise actions of Pols θ and δ during TMEJ. DSBs containing a MH are paired by Polθ (Step 1). After alignment of non-terminal MHs, Polδ’s exonuclease domain is required for trimming of 3’ssDNA flaps (Step 2) before initiation of synthesis by Polθ (Step 3). Polδ and PCNA perform processive and strand displacing synthesis (Step 4), and may recruit Fen1 and LIG1/3 for resolution steps. The cartoon is not meant to imply stoichiometry.

We further characterized the Polδ-Polθ interaction using purified proteins. We generated a fusion of the two domains (Polθ ΔCen), confirmed this construct supports cellular TMEJ, then assessed whether purified Polθ ΔCen could interact with purified Polδ using microscale thermophoresis (Extended data Fig. 7a–c). We determined purified Polδ directly interacts with purified Polθ ΔCen with a K_d of 210 nM (Fig. 4d). No significant interaction was observed with the same Polθ ΔCen and another polymerase (Pol λ) (Extended data Fig. 7c,d).

We sought to also address whether the two polymerases interact in intact cells using super resolution microscopy of Polδ, Polθ, and the DSB marker 53BP1 (Fig. 4e, Extended data Fig. 8a). We observed a significant damage-dependent increase in Polδ density at Polθ associated specifically with DSBs, indicating both polymerases engage the same DSB in cells (Fig. 4f, Extended data Fig. 8b,c).

Discussion

We clarify here the mechanism behind two steps essential for TMEJ (Fig. 1) and the enzymes required for them. We demonstrate there is an exonucleolytic trimming step required before Polθ can initiate synthesis, and a switch to synthesis mediated by a more processive and strand displacing polymerase (Fig. 4g). Polδ is required for both steps – with no evidence for redundancy (Fig. 2–4). TMEJ is also equally impaired by deficiencies in either polymerase alone, as well as combined deficiency, indicating Polδ is just as essential to pathway function as is Polθ (Figs. 2, 3). By comparison, Polδ is important for alternative end joining in the fungus S. cerevisiae^20–22, but fungi have no counterpart to Polθ²³. The increased flexibility provided to TMEJ by Polθ helps explain the more central role this pathway plays in DSB repair in all other eukaryotes.

Polθ confers this flexibility to TMEJ in part through its ability to pair DSB ends and perform end bridging synthesis, using as little as 2 bp of complementary sequence (Steps 1 and Step 3; Fig.4g). The intermediate step – the trimming of flaps typically generated after end alignment – requires both the engagement of the exonuclease domain of Polδ, as well as sustained Polθ-mediated end pairing. The requisite tight coupling of these three steps and sequential engagement of different polymerases is likely dependent on their physical interaction (Fig. 4, Extended data Fig. 6), and further facilitated by damage dependent association of the two polymerases at sites of chromosome breaks (Fig. 4e,f) and replication stress ²⁴.

We describe here an unanticipated additional requirement during TMEJ for a more processive polymerase. We show in Polδ deficient cells that TMEJ fails when ≥70 nts of synthesis is required, and is significantly reduced when requiring ≥45 nts of synthesis (Fig. 2d, Extended data Fig. 4b). Interestingly, mutation spectra suggested switching from Polθ to Polδ occurs prior to 14 nt of synthesis (Fig. 1d), arguing Polθ disengages much earlier than 45 nucleotides when both polymerases are present. Polδ may simply have higher affinity for the primer once Polθ sufficiently extends the 2–6 bp MH, though an active switching mechanism is also possible. The switch to processive synthesis is required for biologically relevant TMEJ, since TMEJ of chromosomal DSBs is dependent both on Polδ polymerase activity, as well as the ability of Polδ to interact with the PCNA processivity clamp.

Our observation that there is a switch during TMEJ to strand displacing synthesis mediated by PCNA-bound Polδ provides a satisfying explanation for the previously described roles of Fen1²⁵ and Ligase III (or Ligase I)^26–28in the final pathway steps. PCNA also interacts with Fen1²⁹, Ligase I³⁰, and XRCC1-Ligase III³¹, with potentially both Fen1 and a Ligase recruited to the same Polδ-bound PCNA trimer.

The pairing of Polθ with the replicative polymerase Polδ is consistent with emerging evidence arguing for an important role for Polθ in response to replication stress^32–36, and especially repair of gaps due to incomplete replication in BRCA-deficient cells^24,37–39. The pivotal roles TMEJ has – both in response to replication stress and during conventional DSB repair – rely on the flexibility provided by coupling Polθ-mediated end-bridging synthesis to Polδ. However, this plasticity may come at a cost. Pairing of a promiscuous Polθ with the normally high-fidelity Polδ may help explain the role of Polδ in a variety of processes that generate large-scale genome rearrangements, including microhomology mediated break-induced replication (MMBIR), mitotic DNA synthesis (MiDAS), and translocation^2,3,37–41.

Materials and Methods

Cell lines

All cells were cultured in 5% CO₂ at 37°C and regularly tested and shown to be mycoplasma negative by PCR (detection limit less than 10 genomes/mL). Human embryonic kidney cells (HEK-293Ts), mouse embryonic fibroblasts transformed by SV40 T-antigen (MEFs), Chinese hamster ovary cells (CHO), and POLQ^−/− human bone osteosarcoma epithelial cells (U2OS) were cultured in DMEM media (Corning). Human colon carcinoma cells (HCT116s) were cultured in McCoy’s media (Corning). P53^−/− retinal pigment epithelial cells (RPE1s) immortalized by human telomerase reverse transcriptase were cultured in DMEM-F12 (Invitrogen). All media was supplemented with 10% Fetal Bovine Serum (VWR, Seradigm) and penicillin (5U/mL, Sigma). RPE1 P53^−/− POLQ^−/− and Halo-tagged POLQ cell lines were generous gifts from the lab of Dr. Gaorav Gupta, and complementation validated in Extended Fig. 7b. U2OS POLQ^−/− cells have been previously described. CHO-9 and CHO-43–3B were previously described⁴². POLN^−/−23 and POLZ⁴³ deficient (REVL3^−/−) MEF cell lines were previously described.

Generation of recombinant cell lines

For POLD1 and POLE knockdown, lentiviral constructs (Addgene 160792, 160762) were transfected with lentiviral packaging constructs (Addgene 12260, 12259) into HEK-293T cells using Transporter 5 (Polysciences). For POLD1 cDNA expression, retroviral constructs were transfected (Addgene 160805) with retrovirus packaging constructs (Addgene 35616, 14887) in the same manner described above. Mutant retroviral cDNA constructs were generated using Q5 mutagenesis (NEB) of the WT cDNA plasmid and validated by sanger sequencing. Media was changed 18–24 hours post-transfection, and virus was collected at 48 and 72 hours post transfection. Cells to be transduced were plated 1 day prior to the first viral harvest. Viral-containing media from the HEK-293Ts was filtered through a 0.45 um filter and supplemented with 1 ug/ml polybrene prior to transduction via media change of the target cells. Cells were serially transduced with both the 48 and 72 hour viral media collections from the HEK-293Ts. The day after the second transduction, cells were plated into media containing either blasticidin or puromyocin selection for 2 days. The media was changed and cells were allowed to recover for 1 day prior to experimental use or freezing down at −80 C°. Cells transduced with shPOLD1 and shPOLE were confirmed to have no significant cell cycle defects at the time of harvest (Extended data Fig. 5i,j). All plasmids used to generate recombinant lines were validated and sequenced with a combination of Oxford nanopore and Sanger sequencing. When shPOLD1 or shPOLE were used in RPE1 cells, RPE1 PAC^−/− cells were used and untreated RPE1 PAC^−/− cells were used as parental controls (wt).

A mNeonGreen-FLAG expressing line was generated by transfecting a relevant plasmid (Addgene 172864) into HEK-293T cells with protein collected 48 hours post transfection. APEX2^−/− MEFs were generated by introduction of Cas9 and a sgRNA targeting exon 5. Candidate clones with frameshift mutations were identified by TOPO cloning and Sanger sequencing, and confirmed by western blot.

siRNA knockdown

Cells were reverse transfected in antibiotic-free media using RNAiMAX (Invitrogen) in 6-well plates. 60 picomoles of the relevant SMARTpool (Horizon discovery) siRNA was used per transfection. Cells were either transfected a single time and assayed 48 hours later, or transfected twice with 24 hours between transfections, and assayed 48 hours after the initial transfection. siRNAs used include PCNA (L-003289–00), POLD1 (L-062714–01), POLD2 (L-020131–01), and POLD3 (L-026692). Knockdown efficiency was assessed by western blot. All siRNA sequences are detailed in Supplementary Table 1.

Extrachromosomal assays

All extra chromosomal substrates except for those detailed in Fig. 1f and Extended data Fig. 1 were annealed from Ultramer DNA (IDT) in a thermocycler with a 5 minute 95 °C denaturation, 1 hour at 70 °C, and finally cooled to 4 °C with a 0.5% cooldown rate between steps. DNA was annealed in 10 mM Tris pH 7.5, 100 mM NaCl, and 0.1 mM EDTA buffer. 500 ng of the TMEJ substrates and 20 ng of the NHEJ substrate were electroporated into 250,000 cells with a dual 1,350 volt, 20 ms pulse with the Neon system (Invitrogen). Where indicated, cells were pre-treated for 2 hours with 2 uM ART558 (Artios Pharma)⁴⁴ prior to electroporation and were recovered for 30 minutes post-electroporation in media or drug-supplemented media. Cells were then washed in PBS and incubated in Hank’s balanced saline solution containing 25U of Benzonase (Sigma) for 15 minutes. DNA extraction was then performed using the QIAamp DNA mini kit (Qiagen) and samples were subsequently analyzed via PCR using the TaqMan Fast Advanced Master Mix (Applied Biosystems; relevant primers and probes described in Supplemental Table 3). PCR efficiency, limit of detection (LOD), and independence of multiplex PCRs for all qPCR amplicons was determined by serially diluting a synthetically produced model amplicon product into genomic DNA containing a constant amount of the relevant reference amplicon for that target (Extended data Fig. 1). The inverse of this quality check was also performed on each target/reference pair. All model amplicon products are detailed in Supplementary Table 7. In extrachromosomal substrate experiments, TMEJ activities were normalized to repair measured with spike-in control substrates (Supplementary Table 2), either NHEJ (Fig. 1b) or minimal TMEJ substrates (Fig. 2, Extended data Fig. 2), in a multiplexed reaction, as indicated in respective figures. All experiments consisted of 3 replicates of each electroporation. Ultramer DNA oligos and qPCR primer and probe pairs are described in Supplementary Tables 2 and 3, respectively.

Strand displacement substrates employed in experiments described in Fig. 1f were assembled by golden gate ligation of left and right annealed oligonucleotide dsDNA ends to a 600 bp central DNA fragment as described using the oligonucleotides described below⁷. These substrates were introduced into mouse embryo fibroblasts as described⁷. Recovered DNA was mock digested or digested with BamHI when indicated, amplified, and electrophoresed on native 6% polyacrylamide gels to identify overhang-containing products. Next generation substrates (2.0) for strand displacement were used in Extended data Fig. 4c. These substrates were Ultramer-based and assembled as described previously.

Immunoblotting

Whole cell lysates were prepared with radioimmunoprecipitation assay (RIPA) buffer supplemented with a freshly prepared protease inhibitor cocktail (Sigma, P8340). Lysates were denatured in Laemmli sample buffer (Biorad, 1610737) and loaded onto 5–15% tris-glycine SDS polyacrylamide gels. Protein was transferred to nitrocellulose membranes in a 20% methanol supplemented tris-glycine transfer buffer. Membranes were blocked in TBST containing 3% BSA for 1 hour at room temperature. Primary antibodies were diluted in blocking buffer and incubated with membranes overnight with agitation at the following dilutions (POLD1 (Abcam, 186406) 1:2,000; Actin (Novus, NB600–535) 1:10,000; HALO (Promega, G9211) 1:1,000; POLE (GeneTex, GTX132100) 1:2,000; FLAG (Sigma, F3165) 1:2,000); APEX2 (Novus, NB100–56625) 1:1,000; POLD2 (Thermo Scientific, PA5–55401) .4 μg/mL; POLD3 (Abcam, ab182564) 1:2,000; PCNA (Santa Cruz, SC-56) 1:1,000; Vinculin (Santa Cruz, SC-25336) 1:500; Myc (Cell Signaling Technology, 2278) 1:1,000. Membranes were washed with TBST and incubated with appropriate secondary antibodies (Licor) at a dilution of 1:7,000 in blocking buffer for 1.5 hours at room temperature. Membranes were imaged and analyzed on a Licor Odessey machine. All uncropped blots are available in Supplementary Fig. 1.

Co-IP

Where indicated, cells were pre-treated with Neocarzinostatin (Sigma) at 100 ng/mL for 2 hours prior to lysate collection. Lysates were prepared in a non-denaturing buffer (25 mM Tris, 150 mM NaCl, 1 mM EDTA, 1% NP-40) supplemented with freshly prepared protease inhibitor cocktail. Magnetic beads (BioRad, 1614013) were incubated with 7 μg of the appropriate antibody for 20 minutes at room temperature. Beads were then washed 3 times with PBST and incubated at 4°C with prepared lysates for 6 hours with gentle agitation. Beads were washed 3 times and proteins were boiled and eluted into Laemmli sample buffer. Lysates were treated with benzonase to eliminate DNA (validated in Extended data Fig. 6d), and we confirmed the antibody to Polδ did not non-specifically interact with FLAG-tagged proteins (Extended data Fig. 6e). 5% input and IP elutions were subsequently processed with the previously described immunoblotting protocol. Boundaries for the POLQ helicase-like domain and polymerase domain used in these experiments were aa1–892 and aa1792–2590, respectively. The POLD1 ΔCysB deletion was made by the introduction of a stop codon in the POLD1 cDNA at aa1033.

In vitro assays

Oligonucleotides for exonuclease and polymerase substrates were purchased from IDT and annealed. We generated a Polθ construct substituting the non-conserved central domain (899–1789) with the linker GSAGSAAGSGEF (Polθ ΔCen) and confirmed expression of Polθ ΔCen in Polθ deficient RPE1s is able to complement deficiency in chromosomal TMEJ (Extended data Fig. 7b). A variant of this construct with two N-terminal maltose binding domain proteins (2xMPB) was expressed in Expi293F cells, and purified by sequential chromatography over amylose resin (NEB) and cation exchange (Hi-Trap SP, Cytiva), before removal of the MBP tag with PreScission protease and removal of the cleaved 2xMPB tag with amylose³³ (Extended data Fig. 7c). Purification of Pol λ (Extended data Fig. 7c) was performed as previously described⁴⁵. Constructs to express human POLD1, POLD2, POLD2, and POLD4 (the Polδ holoenzyme) were the gift of Dr. Kenji Kamiya. The Polδ preparation used for microscale thermophoresis was additionally cleared of DNA by streptomycin precipitation prior to chromatography. All Polδ preparations were purified by sequential affinity to Histrap (Cytiva) resin (native p66 interacts with Nickel) and heparin columns, followed by enrichment for the fully assembled 4 subunit complex using a size exclusion column⁴⁶ (Extended data Fig. 7a). Spectrophotometry indicated neither of the purified Polδ or Polθ ΔCen used for microscale thermophoresis had significant amounts of nucleic acid. Mutants were generated in the POLD1 open reading frame of the plasmid that expresses POLD1 and POLD4, validated by sequencing, and purified in the same way. Polymerase and exonuclease reactions were performed in a buffer containing 25mM potassium phosphate pH 7.5, 5 mM DTT, 5mM MgCl₂, 100 ug/mL BSA, and 10% glycerol. For exonuclease assays we employed 5’ Cy5 labeled double stranded DNA forked exonuclease substrates with 2, 5, or 10 nt 3’ ssDNA overhangs, and incubated 25 nM substrate with 25 nM Polδ for times indicated in figure legends at 37^oC in the presence of 100 μM each of dTTP, dGTP, and dCTP. For polymerase assays we employed 5’ Cy5 labeled double stranded DNA substrates with 5 nt 3’ recessed ends, and incubated 25 nM substrate with 25 nM of Polδ for 15 minutes with 100 μM of all four dNTPs at 37^oC. Reactions were stopped after the indicated time by addition of an equal volume of formamide and 10mM EDTA, heated for 5 minutes at 95^oC, separated on a 5% polyacrylamide gel under denaturing conditions, and imaged using a Typhoon FLA9500. All uncropped gels are available in Supplementary Fig. 1. Oligo sequences are described in Supplementary Table 4.

Next-generation sequencing of polymerase theta products

Cellular transfection and DNA extraction were carried out as described in the extrachromosomal assays. Desalted DNA primers (IDT) were used to amplify the repair product of interest for 25 PCR cycles. PCR products were purified via gel extraction from a 2% agarose (Lonza) gel and the QIAquick Gel Extraction Kit (Qiagen). DNA Ultramers (IDT) containing a substrate-specific primer sequence, 6 bp barcode, spacer sequence of varying length, and Illumina’s adapter sequences were used to perform a secondary amplification for 7 PCR cycles. These PCR products were further purified with AMPure Magnetic beads (Beckman). Final DNA libraries were sequenced with an Illumina Iseq 100 i2 kit (300 cycles) with a 15% PhiX Control DNA spike-in (Illumina). Ultramer oligos and primer pairs are described in Supplementary Table 5.

Sequencing data was trimmed, and reads were merged using CLC Genomic Workbench 8 (Qiagen). Triplet thymidine location was identified according to 5 nt unique barcodes both upstream and downstream of the triplet thymidines (5’-nnnnnTTTnnnnn-3’). We counted the frequency of indels (+1 or −1 nt) at each triplet. Analysis of samples was performed in Microsoft Excel.

Cas9 chromosomal reporter assay

The CRISPR RNA (crRNA) specific to a site in the human LBR gene is described in Supplementary Table 6. To generate a DSB at the LBR locus, 7 pmols of Cas9 was incubated with 8.4 pmols of crRNA annealed to tracRNA (IDT, Alt-R) for 30 minutes at room temperature. This complex was electroporated into 250,000 cells as described above. Two electroporations were pooled together to comprise a single biological replicate. Cells were then re-plated into the previously indicated media for 48 hours, and DNA harvested from cells using the QIAamp DNA mini kit (Qiagen). Repair products were quantified with qPCR using 50 ng of input DNA and the TaqMan Fast Advanced Master Mix (Applied Biosystems). Relevant primers and probes are described in Supplementary Table 3. All signature PCRs (TMEJ/NHEJ) were normalized to a reference amplicon 10 kb upstream of the Cas9-cut site multiplexed in the same reaction. PCR efficiencies and LODs for this assay were determined by diluting a model repair product into unbroken genomic DNA (Extended data fig. 1). All model product sequences are available in Supplementary Table 7.

The mouse chromosomal TINS assay was previously described and carried out as described above⁴⁷.

A custom Python script was developed to predict outcomes of theta-mediated end joining at unique genomic loci (PyCharm Community Edition 2021, JetBrains). In brief, we predict resolutions, both microhomology-mediated deletions and locally templated insertions (TINS), such that microhomologies are within 15 nucleotides and template for TINs is within 25 nucleotides of the DSB⁷.

Droplet digital PCR

Chromosomal DSBs were introduced via the Cas9 system described above at the LBR locus. Droplet digital PCR was performed with 100 ng of genomic DNA and ddPCR Supermix for Probes (no dUTP)(BioRad). The TINS signature amplicon information is described in Supplementary Table 3 and the reference amplicon was the same as previously described in the qPCR assay. Droplets were generated and read using a QX200 AutoDG Droplet Digital PCR system. QuantaSoft software was used to analyze resulting data.

Super resolution imaging and analysis

Cells were seeded onto glass coverslips (Fisher Scientific, 12–548-B) 1 day prior to experimentation. Cells were incubated with Neocarzinostain (Sigma) at either 40 (U2OS) or 80 (RPE1) ng/mL for 2 hours prior to harvest, 10 uM EdU for 30 minutes, and the Janelia Fluor 646 halo-tag ligand (Promega, GA1120) at 1 ng/uL for 30 minutes. Cells were then permeabilized with 0.5% Triton X-100 in ice-cold CSK buffer (10 mM Hepes, 300 mM Sucrose, 100 mM NaCl, 3 mM MgCl₂) for 3 minutes followed by 3 PBS washes. Cells were then fixed with 4% paraformaldehyde (EMS, 15714) for 15 minutes. Coverslips were subsequently washed twice with PBS and blocking buffer (2% glycine, 2% BSA, 0.2% gelatin, 50 mM NH₄Cl) 3 times. Cells were incubated in blocking buffer overnight at 4 °C. Click reactions were then performed on coverslips to label EdU and coverslips were subsequently washed 3 times. Coverslips were then incubated with primary antibody for 1 hour at room temperature (POLD1 sc-17776, 1:250 dilution; 53BP1 Novus NB100–304, 1:10,000). After blocking buffer washes, coverslips were incubated with secondary antibodies in blocking buffer (Invitrogen, AF488 and AF568 both at 1:10,000). Coverslips were then washed thrice with blocking buffer and mounted onto glass slides with freshly prepared imaging buffer (1 mg/ml glucose oxidase (Sigma, G2133), 0.02 mg/ml catalase (Sigma, C3155), 10% glucose (Sigma G8270), 100 mM mercaptoethylamine (Fisher Scientific, BP2664100)) flowed through prior to imaging.

For single molecule localization microscopy imaging, image stacks with at least 2000 frames per channel, acquired at 33 Hz, were taken on a custom-built optical imaging platform based on a Leica DMI 300 inverted microscope possessing three laser lines 561 nm (Coherent, Sapphire 561 LPX-500), 639 nm (Ultralaser, MRL-FN-639–1.2), and 750 nm (UltraLaser, MDL-III-750–500). Lasers were combined and aligned using dichroic mirrors and were focused on the back aperature of an oil immersion objective (Olympus, IApo N, 100x, NA=1.49, TIRF) with a multiband dichroic mirror (Semrock, 408/504/581/667/762-Di01). Fluorophores were individually excited with a Highly Inclined and Laminated Optical (HILO) illumination configuration. Emissions were expanded with a 2X lens tube and filtered using single-band pass filters in a filter wheel (ThorLabs, FW102C) and collected on a sCOMS cameras (Photometrics, Prime 95B). A 405 nm laser line (MDL-III-405–150, CNI) was used with AF647 to drive it back to its ground state. Images were acquired using Micro-Manager (v2.0) software.

Localization of each single molecule was performed as previously described in^48–51 on a minimum of 55 nuclei for each condition. Briefly, patterned noise of sCMOS cameras was accounted for based on methods described in ⁴⁹. Each pixel of the single molecule images was weighted by the inverse of its variance and were box-filtered with 4 times of the full width at half maximum (FWHM) of a 2D Gaussian point spread function (PSF). The local maximums from all the frames of the image stack were then submitted for 2D-Gaussian multi-PSF fitting (DAOSTORM)⁴⁸. The 2D-Gaussian multi-PSF fitting was achieved through the Maximum Likelihood Estimation (MLE) with up to 4 PSFs. Fitting accuracy was assessed by Cramér-Rao lower bound (CRLB) as described in ^49,52. In addition, the localizations that appeared in consecutive frames within 2.5 times of the localization precision were considered as one blinking event. Representative images were generated by rendering raw coordinates onto a 10 nm pixel canvas, convolved with a 2D-Gaussian (σ = 10 nm) kernel, and adjustment of individual channel brightness for display purposes. Result tables with the localization coordinates of each individual fluorophore blinking within a region of interest (ROI) outlining whole nuclei underwent Auto, Cross, Pair, and Triple correlation analyses as described in^53,54.

A correlation profile was generated as a function of the pair-wise distances and fit to a Gaussian model of a target species with itself (Auto-Pair correlation) or another target (Cross-Pair correlation) within the bounds of the selected ROIs. Based on these, auto-correlation analyses provided estimates of the nuclear density of POLQ, 53BP1, and POLD1 fluorophores (Extended data Fig. 8d). Triple-correlation analysis was performed to determine the correlation amplitude of three targets species as described in⁵¹and validated previously ^52,55. With this analysis, we estimated the average local density of POLD1 at sites of POLQ-53BP1 association at an inter-target association distance of ≤80 nm^51,56. This approach is specifically tailored for reliably identifying molecular configurations (or association events) that are “hidden” within dense images that consist of multi-color SMLM coordinates and distinguishes real triplet associations from the random associations prevalent in crowded images. The probability density is calculated over a continuous range of molecular distances, and as such the obtained configuration, which incorporates the variability between all the complexes, is unbiased.

To rule out false positive association events as a byproduct of the high nuclear densities of targets imaged, correlation amplitude was computed from triple correlation performed on randomized data sets as described^51,57. In short, the randomization was achieved by setting 8 × 8 μm² ROIs, then comparing triple correlation amplitude from these ROIs to a randomly assembled ROI (a random scrambling of the channels between different ROIs, so that that channels of a scrambled ROI are randomly obtained from different images).

To understand POLQ and POLD1 association independent of its association at 53BP1 labeled sites of DNA damage (Extended data Figure 8c), correlation amplitude of POLQ and POLD1 association was computed from cross-pair correlation analyses performed on experimental sample versus randomized data sets (randomized data set generated as described above) ^51,57.

Selective staining for the Halo ligand, as well as the POLD1 and 53BP1 antibodies, was appropriately restricted to cells expressing the relevant proteins (Extended data Fig. 8e–h).

Flow cytometry

Cells were cultured as described above and incubated with 10 μM EdU 30 minutes prior to harvest. Cells were collected and fixed in 70% cold ethanol for 3 hours. After fixation, EdU-positive cells were labeled with a click chemistry reaction containing 100 mM ascorbic acid, 1 μM of AF647 azide, and 1 mM copper sulfate for 30 minutes. After the click reaction, cells were stained with DAPI at 1:10,000 and incubated with RNAse at 100 μg/mL overnight at 4 °C. The next day cells were filtered through a cell strainer and run on a Attune NxT flow cytometer. Voltages were adjusted based off of forward and side scatter, doublets excluded, and the relevant lasers for DAPI and AF647 were utilized in data collection. All data was analyzed and gated with the FlowJo software.

Microscale thermophoresis

All microscale thermophoresis (MST) were carried out in MST optimized buffer (50 mM Tris-HCl, 150 mM NaCl, 10 mM MgCl₂, 0.05% Tween-20) in standard capillaries (Nanotemper). Polθ ΔCen was labeled and free label removed using the Red-tris-NHS second generation labeling kit according to supplier guidelines (Nanotemper). 50 nM of this labeled Polθ ΔCen was incubated with equal volumes of serial dilutions of other polymerases and incubated at room temperature for 5 minutes prior to data acquisition. All MST measurements were collected on a Monolith NT.115 Pico instrument set to medium MST power and 20% LED power. Analysis of MST traces was completed using MO.Affinity Analysis v.2.3 (Nanotemper). Kd calculations were performed in MO.Affinity Analysis v.2.3 (Nanotemper). All measurements were performed in triplicate.

Statistical analysis

All replicate numbers and statistical tests performed are listed with their corresponding figures. All statistical analysis was carried out using GraphPad Prism 9. Statistical significance is displayed in figures as *,**,***,*** represents p ≤ .05, .01, .001, .0001, respectively. Statistical tests for qPCR experiments were run on cycle thresholds prior to transformation of data for the linear scale representations shown in display figures. p values were adjusted using Dunnet’s method to correct for multi comparisons when more than two groups were compared.

Software

Model figures were made using BioRender and Adobe Illustrator. Graphs were generated in GraphPad Prism.

Extended Data

Extended data Fig. 1. Fundamental mechanisms and validation of TMEJ.

a, Quantification of minimal and flapped TMEJ repair normalized to NHEJ with and without 2 μM Polθi (ART558). Data are from 3 biological replicates. Data are mean ± SD. Values represent the maximum fraction of detected repair, comparing Polθi to untreated cells. b, Quantification of TMEJ with locked nucleic acids at the indicated positions. Data are from 3 biological replicates. Data are mean ± SD. c, Schematic of TMEJ mismatched BamHI substrate to measure strand displacement of 20 and 50 bp. d, Gel of DNA substrates in a. e, Quantification of strand displacement (BamHI sensitivity) of 20 bp or 50 bp. Data is from 3 independent experiments. Data are mean ± SEM. f, Standard curve of qPCR CT values of a unflapped and flapped TMEJ product where the amount of flapped product is constant and unflapped is varied. g, Identical to f, but flapped is varied and unflapped is constant. h, Standard curve of qPCR CT values of a 45 bp and 25 bp TMEJ synthesis products where the 25 bp product is constant and the 45 bp product is varied. I, Identical to h, but the 45 bp product is varied and the 25 bp product is constant. j, Standard curve of qPCR CT values of a 70 bp and 25 bp TMEJ synthesis products where the 25 bp product is constant and the 70 bp product is varied. k, Identical to j, but the 25 bp product is varied and the 70 bp product is constant. l, Standard curve of qPCR CT values of the LBR repair signature and reference where the signature product is varied and the reference product is constant. m, Identical to l, but the reference is varied and the signature is constant.

Extended data Fig. 2. Polymerase Delta is essential for TMEJ.

a, Schematic of viral timeline to generate cell lines. b, Validation of POLD2 siRNA in RPE1 cells. Actin was used as a loading control. c, Validation of POLD3 siRNA in RPE1 cells. Actin was used as a loading control. d, Quantification of repair of a 70 bp, 5 nt flapped substrate relative to the minimal TMEJ substrate in cells depleted of POLD2 or POLD3, relative to WT. Data are from 3 biological replicates analyzed with a one-way ANOVA and Dunnet’s method. Data are mean ± SD. e, Quantification of repair of the minimal TMEJ substrate in POLD1 depletion and mutant backgrounds. Data is from 5 (WT) or 6 biological replicates. Data are mean ± SD. f, Western blot validation of PolM2 expression, siScramble and shPOLD1 were included as controls. g, Western blot validation of PIPM expression, siScramble and shPOLD1 were included as controls. h, 25 nM WT human Polδ, PolM2, and ExoM was incubated with 25 nM of a 10 nt 3’ recessed 5’ labeled substrates for 10 minutes. I, 25 nM WT human Polδ, PolM2, and ExoM were incubated with 25 nM of 5’ labeled forked substrate with a 5 nt 3’ ssDNA overhang for 10 minutes. j, Polθ is inhibited (Polθi) with 2 μM ART558. POLD1 in RPE1 cells is endogenously expressed (+) or depleted (−) by shPOLD1, or depleted by shPOLD1 and complemented by expression of ExoM, PolM, or WT POLD1. Quantification of repair of a 70 bp, 5 nt flapped substrate relative to the minimal TMEJ substrate, normalized to parental cells. Data are from 3 biological replicates analyzed with a one-way ANOVA and Dunnet’s method. Data are mean ± SD. Part of these data are also displayed in Fig. 2a.

Extended data Fig. 3. TMEJ flap cleavage depends on Polδ.

a, Polθ is inhibited (Polθi) with 2 μM ART558. POLD1 in RPE1 cells is endogenously expressed (+) or depleted (−) by shPOLD1 treatment, or depleted by shPOLD1 treatment and complemented by expression of PolM2. Quantification of repair of a 5 nt flapped substrate relative to the minimal TMEJ substrate, normalized to WT. Data are from 3 biological replicates analyzed with a one-way ANOVA and Dunnet’s method. Data are mean ± SD. b, Quantification of repair of a 2 nt flapped substrate performed as in a. c, Quantification of repair of a 10 nt flapped substrate performed as in a. d, Western blot validation of APEX2−/− cells, Vinculin was used as a loading control. e, Quantification of repair of a 5 nt flapped substrate performed as in a, with WT and two independently generated APEX2−/− clones. f, Quantification of repair of a 5 nt flapped substrate performed as in a, with WT and ERCC1−/− cells. g, Quantification of a templated insertion product at the chromosomal rosa26a locus in MEFs from a Cas9-induced DSB. Experiments were performed with Polθ inhibited (Polθi), APEX2−/−, and WT MEF cells. Data are from 3 biological replicates and normalized to a chromosomal reference. Data are mean ± SD. The dashed line indicates an independent experiment.

Extended data Fig. 4. Polδ performs processive synthesis in TMEJ.

a, Polθ is inhibited (Polθi) with 2 μM ART558. POLD1 in RPE1 cells is endogenously expressed (+) or depleted (−) by shPOLD1, or depleted by shPOLD1 and complemented by expression of PolM2. Quantification of repair of a 70 nt synthesis substrate relative to the minimal TMEJ substrate, normalized to parental cells, with PolM2 complemented cells. Data are from 3 biological replicates analyzed with a one-way ANOVA and Dunnet’s method. Data are mean ± SD. b, Quantification of repair of a 45 bp synthesis substrate relative to the minimal TMEJ substrate, normalized to WT. Data are from 3 biological replicates analyzed with a one-way ANOVA and Dunnet’s method. Data are mean ± SD. c, Quantification of strand displacement in TMEJ repair substrates requiring 45 bp of synthesis. Relative joining efficiency for each background assessed was also measured. d, Western blot of POLD1 knockdown in WT MEF cells. siScramble was used as a technical control, Actin was used as a loading control. e, Western blot validation of siPCNA in RPE1 cells, siScramble was used as a technical control and Actin as a loading control. f, Quantification of repair of a 70 bp, 5 nt flapped substrate relative to the minimal TMEJ substrate, normalized to WT, with PolN−/− cells. Data are from 3 biological replicates analyzed with a one-way ANOVA and Dunnet’s method. Data are mean ± SD. g, Quantification of repair of a 70 nt, 5 bp flapped substrate relative to the minimal TMEJ substrate, normalized to REV3L−/+, with REV3L−/− cells. REV3L is the gene name for Pol zeta. Data are from 3 biological replicates analyzed with a one-way ANOVA and Dunnet’s method. Data are mean ± SD.

Extended data Fig. 5. Chromosomal LBR reporter characterization and controls.

a, Predicted microhomology-mediated deletion repair products at the LBR locus. b, Sequence alignments of predicted microhomology-mediated deletion repair intermediates. c, Quantification of TMEJ at a terminal, chromosomal MH by qPCR. RPE1 cells were untreated (−) or treated with Polθi (+). POLD1 in RPE1 cells is endogenously expressed (+) or depleted (−) by shPOLD1 treatment. Data are from 3 biological replicates analyzed with a one-way ANOVA and Dunnet’s method. Data are mean ± SD. d, Quantification of TMEJ at a terminal, chromosomal MH by qPCR. RPE1 cells were either depleted of Polθ (Polθi) or POLD2/3 by siRNA. e, NHEJ repair quantification performed as in c. f, NHEJ repair quantification performed as in d. g, Western blot of shPOLE treated RPE1 cells and an untreated control. Actin was used as a loading control. h, Quantification of the terminal TMEJ repair product at LBR relative to the signature NHEJ product for WT, shPOLD1, and shPOLE treated RPE1 cells. Bars represent data mean and SD. i, Representative flow cytometry cell cycle plot of shPOLD1 transduced RPE1 cells with EdU and DAPI staining. j, Cell cycle profiles of shPOLD1 and shPOLE transduced cells, relative to RPE1 WT.

Extended data Fig. 6. Polδ physically associates with Polθ by Co-IP.

a, Extracts of RPE1 cells expressing Halo-Polθ were immunoprecipitated with an antibody to Halo (αHalo) or without antibody (−), and recovery probed with an antibody to POLE (αPOLE). b, Co-IP of FLAG-tagged domains of Polθ in POLQ−/− U2OS cells. Polθ domains were pulled down with an antibody to FLAG, and recovery probed with a POLD1 antibody. Negative controls include parallel experiments using cells not expressing FLAG tagged constructs or cells expressing FLAG-tagged constructs but with αFLAG omitted. c, Co-IP of FLAG-tagged domains of Polθ in POLQ−/− U2OS cells. Polθ domains were pulled down with an antibody to FLAG, and recovery probed with a POLE antibody. Cells expressing FLAG-tagged constructs but with αFLAG omitted were used as negative controls. d, Protein lysates and DNA ladder incubated with and without benzonase. e, Extracts of RPE1 cells expressing FLAG-mNeonGreen or no construct were immunoprecipitated with a FLAG antibody and recovery was probed with a POLD1 antibody (αPOLD1). f, Western blot of FLAG-tagged Polθ domains in RPE1 cells expressing Myc-tagged POLD1 constructs. Actin was used as a loading control. g, Polθ is inhibited (Polθi) with 2 μM ART558. POLD1 in RPE1 cells is endogenously expressed (+) or depleted (−) by shPOLD1 treatment, or depleted by shPOLD1 treatment and complemented by expression of POLD1 cDNA. Quantification of repair of a 5 nt flapped, 70 nt ssDNA substrate relative to the minimal TMEJ substrate, normalized to WT. Data are from 3 biological replicates analyzed with a one-way ANOVA and Dunnet’s method. Data are mean ± SD. h, Western blot of Myc-tagged POLD1 constructs used in RPE1 cells for functional assays in g. Actin was used as a loading control.

Extended data Fig. 7. Polδ physically associates with Polθ by Microscale thermophoresis.

a, Protein gel of Polδ holoenzyme purification showing every second fraction which were all collectively pooled for downstream applications. Dark lines indicate cropped lanes. b, Extrachromosomal TMEJ repair frequency relative to NHEJ repair in Polθ-null cells complemented with either full length Halo-tagged or Polθ ΔCen. Data represents 3 biological replicates. Bars represent data means and SDs. c, Protein gel of Polθ ΔCen and Pol λ purifications used for downstream applications. Dark lines indicate cropped lanes. d, Microscale thermophoresis of labeled Polθ ΔCen with varied amounts of Pol λ. Fnorm was plotted as binding was not concluded. Data are mean ± SEM.

Extended data fig. 8. Polδ associates with Polθ at sites of damage.

a, Representative STORM images of 53BP1, POLD1, and Polθ in cells untreated (-NCS) or treated with Neocarzinostatin (+NCS). Scale bar = 1500 nm. b, Images from a, analyzed after ROIs were randomized or experimentally acquired. Data are from 3 biological replicates analyzed with an unpaired Mann Whitney test. All violin plots display data median (solid line) and quartiles (dashed lined). N = 55 nuclei. c, Images from a, analyzed after ROIs were randomized or experimentally acquired. Data are from 3 biological replicates analyzed with an unpaired Mann Whitney test. N =191 (-NCS) and 210 (+NCS) ROIs. (d) Nuclear density of POLD1, Polθ, and 53BP1 with and without damage (+/− NCS). Data are from 3 biological replicates and analyzed with an unpaired Mann Whitney test. N = 96 (-NCS) and 105 (+NCS) nuclei. e, Uncropped western blot of POLD1 from Fig. 2b showing antibody specificity. Endogenous RPE1 POLD1 expression (Lane 1–2), shPOLD1 in RPE1 (Lane 3), and exogenous expression of shRNA-resistant POLD1 constructs (Lane 4–6). f, Representative STORM images of POLD1 staining in an S phase (Edu positive) and non-S phase cell (EdU negative). g, Representative STORM images of HALO ligand staining in RPE1 Polθ and RPE1 HALO-Polθ both −/+ NCS treatment in an S phase nuclei. h, Quantification of 53BP1 nuclear density detected by antibody with STORM in U2OS WT and 53BP1−/− cells with and without NCS damage induction. Data was analyzed with an unpaired Mann Whitney test. N = 52 (WT), 57, (WT +NCS), 47 (53BP1−/−), 47 (53BP1−/− +NCS) ROIs.

Data Availability

MH finder code used to identify candidate TMEJ products is available on GitHub (https://github.com/aluthman/Ramsden-Lab/tree/main/Stroik_et_al). NGS data is available at (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA881334) along with relevant code for analysis at (https://github.com/aluthman/Ramsden-Lab/tree/main/Stroik_et_al). STORM localization coordinates and relevant ROIs analyzed are available on Mendeley Data (treated - https://data.mendeley.com/datasets/mg6n5r87cp/1) (untreated -https://data.mendeley.com/datasets/ymrs93b9c9/1). Code used to analyze STORM images is available on GitHub (https://github.com/d-in-crtl/SMLM).