Pol theta-mediated end joining: Mechanism, cellular functions, and role in cancer

Abstract

Pathways that repair chromosomal double strand breaks play a pivotal role in cell growth, development, and cancer. The characterization of these pathways and clarification of pathway mechanisms has been the target of intensive investigation, but one of these pathways – alternative end joining, or a-EJ – has long resisted these efforts. Here, we highlight recent progress that places DNA polymerase theta as the predominant mediator of a-EJ in most eukaryotes, discuss a candidate molecular mechanism by which Polymerase (Pol) theta-mediated end joining (TMEJ) occurs, and consider candidate mechanisms and biological significance of a-EJ activity that is independent of Pol theta. We consider possible cellular functions for TMEJ in helping resolve double strand breaks that are refractory to repair by nonhomologous end joining, those generated after replication fork collapse, as well as those present after stalling of repair by the homologous recombination pathway. We also review how these context-dependent cellular roles for TMEJ explains how it is both a cure and cause of genome instability, as well as its emerging potential as a therapeutic target in cancer.

Introduction.

Chromosome breaks are a rare, but potent source of genome instability. They typically arise upon replication fork dysfunction, after exposure to exogenous agents (e.g. ionizing radiation used in cancer therapy), and are also normal intermediates in meiotic recombination and the assembly of Immunoglobulins and T cell receptors. One of the pathways for repairing chromosome breaks in mammals, alternative end joining (a-EJ), has attracted considerable recent attention as both a source of cancer-causing chromosome aberrations, as well as a vulnerability in these same cancers that can be therapeutically exploited.

a-EJ is one of 4 pathways employed for DNA double-strand break (DSB) repair¹ (Figure 1). Nonhomologous end joining (NHEJ) directly ligates broken chromosome ends together, and requires a heteromeric ligase (LIG4 and XRCC4) and heteromeric DNA binding scaffold (Ku; XRCC5 and XRCC6). Homologous recombination (HR), single strand annealing (SSA), and a-EJ all require a 5’>3’ nucleolytic resection of broken ends to generate ends with 3’ ssDNA overhangs, which is initiated by CtIP (Sae2 in budding yeast) and a complex of MRE11, RAD50 and NBS1 (MRN). Resection both generates a critical intermediate for HR, SSA, and a-EJ, and impairs repair by NHEJ, identifying this as an important step in pathway choice. Repair of resected ends by HR then requires RAD51 and Breast Cancer associated genes 1 and 2 (BRCA1 and BRCA2) for subsequent steps. This includes the identification of homology to the resected end within a region of the unbroken sister chromatid or homologous chromosome, the invasion of the resected end(s) into the homologous duplex, synthesis using the sister or homologue as template, and finally resolution of the intertwined sister chromatids or homologues. SSA requires annealing of a long tract of complementary sequence (>50p) in resected tails, and additionally requires RAD52 and XPF.

Figure 1.

DSB repair pathways. Resection by CtIP, Mre11, Rad50 and Nbs1 impairs repair by Nonhomologous end joining (NHEJ) and is required for Theta mediated end joining (TMEJ), Single strand annealing (SSA), and homologous recombination (HR). Repair products for TMEJ favors deletions at microhomologies in flanking DNA (in red).

Unlike other DSB repair pathways, a-EJ has been difficult to define with clarity. Below we discuss the merits of various definitions, the evidence identifying Polymerase theta (Pol θ; gene name POL θ) as the predominant mediator of a-EJ in most eukaryotes, and the molecular mechanism of Pol theta mediated fraction of a-EJ (Theta-mediated end joining; TMEJ). We’ll discuss next cellular contexts needed to engage TMEJ, as well as its important role in genome instability, cancer progression, and cancer therapy.

Definition and mechanism of TMEJ

a-EJ is also known as alternate nonhomologous end joining (Alt-NHEJ), microhomology mediated end joining (MMEJ), backup nonhomologous end joining (B-NHEJ), synthesis dependent microhomology mediated end joining (SD-MMEJ), and theta mediated end joining (TMEJ). 5 “akas” is never a good look; for a-EJ it reflects in part an ongoing lack of clarity in how best to define this pathway. a-EJ has been defined as DSB repair that i) is not NHEJ, ii) acts on ends with resected 3’ ssDNA overhangs, and iii) favors deletions to flanking microhomologies in repair products. As summarized below, its frequently not possible to identify repair as unambiguously a-EJ using this criteria, thus we focus in this review on the recent work showing a-EJ is also iv) largely dependent on Pol θ, at least in organisms that have this gene. We then further outline our rapidly improving picture of the molecular mechanism of Pol theta mediated end joining (TMEJ).

Defining a-EJ and TMEJ

Early work made it clear that there was an end joining pathway that does not require Ku, XRCC4, or LIG4^2–4, thus was an “alternative” to NHEJ – hence, a-EJ. This definition is still useful, since it is inclusive of both TMEJ as well as end joining that is independent of both Pol θ and NHEJ factors (Box 1; Pol theta independent a-EJ). a-EJ also preferentially joins ends such that deletions extend to short sequence identities in flanking DNA^2,3 (microhomologies; Figure 1). Unfortunately, not all a-EJ is microhomology mediated, and microhomologies play an important role in other pathways as well (see Box 2, Stop calling it MMEJ?). a-EJ requires 5’>3’ nucleolytic resection of DSB ends and thus the MRN complex, CtIP, and ATM, which impairs recognition of ends by NHEJ^5–13. However, resection isn’t a binary switch that channels repair to NHEJ vs. a-EJ, as discussed later in this review (“Regulation of TMEJ”).

Box 1: Alternative end joining without Pol theta.

The yeast S. cerevsiaie^8,71,153,154 and S. Pombe¹⁵⁵ don’t have a Pol θ ortholog, thus employ a-EJ that is natively Pol θ-independent. Relative to mammalian TMEJ, longer microhomologies are required (>5 bp of perfect microhomology, or >20 bp of microhomology with internal mispairs), consistent with evidence that extended microhomologies can suppress requirement for Pol θ during a-EJ in mammalian cells¹⁵⁶. Synthesis during a-EJ in yeast instead requires at least Pol δ; a-EJ was severely reduced in strains deficient in the non-essential Pol32 subunit of Pol δ, as well as strains expressing a hypomorphic mutant in the Pol3 catalytic subunit⁷¹. Efficient a-EJ in yeast also employs POL4, and possibly Pol η (Rad30) and Pol ζ (Rev7) in some contexts^71,155,157. POL4 associates with breaks prior to Pol δ⁷¹, suggesting a unified a-EJ model where synthesis is initiated using a polymerase with a relatively low requirement for microhomology (POL4 in yeast, Pol θ in other eukaryotes) but also low processivity and fidelity. This is followed by a switch to more processive and accurate synthesis by Pol δ.

Is there a robust a-EJ in other eukaryotes in the absence of Pol θ? Biochemical evidence argues for a synthesis (and thus Pol θ) independent a-EJ mediated by bridging of a pair ends by Mre11, reversed polarity of end resection through MRE11 5’>3’ exonuclease activity, annealing of microhomologies in the resulting 5’ overhangs, and finally ligation^75,158–160. It is also possible to recover end joining products in mammalian cells deficient in both Ku and Pol θ¹⁷, and other work argues for an aphidicolin sensitive (thus synthesis dependent) and Pol θ-independent end joining that requires long microhomologies¹⁶¹. However, both proliferation and colony forming ability of Ku and Pol θ deficient cells are severely reduced relative to singly deficient cells¹⁷, suggesting Pol θ independent a-EJ is far from “robust”.

Box 2. Stop calling it microhomology mediated end joining (MMEJ)?

a-EJ is also referred to as MMEJ, reflecting a preference that deletions in repair products extend to short sequence identities – microhomologies – that are fortuitously present in DSB flanks (Figure 1)^2,3. Microhomology mediated deletions are a defining feature for HR defective cancers (see “insertion and deletion signature 6 or ID6” in “TMEJ and deletion scars”, and box3), have been leveraged for genome engineering^124–128 (potentially with significant clinical impact^129,130), and are essential for a-EJ in yeast. In other eukaryotes, however, microhomologies are neither essential^50,61,63 (Figure 2C), nor are the microhomology sizes that are preferentially employed nearly as long – thanks to Pol θ (see Box 2 Microhomology usage Table, below). The definition of what constitutes a preferred microhomology is also influenced by break context (e.g. nuclease generated DSB vs. replication-associated DSB), and species. Worse, many microhomology mediated products are not a-EJ, as a preference for microhomologies is also observed for products of repair by NHEJ¹⁶² and microhomology mediated break induced replication¹⁶³ (though TMEJ and MMBIR may be related processes^69,154,164).

The biochemical basis for microhomology preference is clear. Ligases and polymerases can better synthesize phosphodiester bonds when strand-break termini can anneal to complementary sequence. Since the complementary strand is also broken during DSB repair, the amount of complementary sequence opposite a strand break terminus is limited to that provided by a microhomology. This is a critical limitation, since the amount of microhomology available by chance is less than canonical eukaryotic ligases and polymerases require to be active (Box 2 Table, bottom two rows). Accordingly, a defining feature of both NHEJ and a-EJ is their employment of specialized ligases and polymerases with reduced requirement for complementary sequence opposite strand break termini. Specialization is usually attributable to structural motifs directly responsible for stabilizing strand break termini in the active site^31,165.

Box 2 Table.

The relative usage of different lengths of microhomology employed by the noted polymerases and ligases within NHEJ, TMEJ, and yeast a-EJ repair pathways is shown, with increasing usage of different microhomology sizes denoted by increasing darkness of blue shade. Only relative usage within pathway function is considered; these enzymes remain active with increasing size of microhomology, but larger microhomologies will not be as frequently used when shorter microhomologies are used effectively.

Species	Pathway	Context	Protein	Citation	Microhomology usage
					0	1	2	3	4	5–10	>10
Mammals	NHEJ	DSB	Pol μ	166
			Pol λ	166
			LIG4	162
	TMEJ	DSB	Pol θ	17,50,56
Worms	TMEJ	DSB	Pol θ	20
	TMEJ	Replication	Pol θ	20,24,54
Flies	TMEJ	DSB	Pol θ	14
Yeast	a-EJ	DSB	POL4+POL32	153,155
Other ligases and polymerases
By chance

Pathways are best defined through their requirement for a factor that performs a key step in the pathway, especially when this factor is uniquely specialized to perform this step (Box 2). For most a-EJ in most eukaryotes this factor is Pol θ. Evidence that Pol θ accounts for most a-EJ was first apparent from elegant experiments in Drosophila melanogaster, where end joining repair after excision of a transposon (a P-element) was shown to be independent of NHEJ, and dependent on Pol θ¹⁴. P-element excision generates ends with extended 3’ overhangs¹⁵, thus are effectively “pre-resected”, and less readily engaged by Ku and NHEJ^16,17. This possibly explains why NHEJ makes little contribution to repair of the DSB generated after P-element excision. In accord with this idea, this same study reported a more balanced engagement of Pol θ vs. NHEJ for repair of DSBs with much shorter 3’ overhangs (4 nucleotides, generated by the nuclease I-SceI).

Similar evidence points to the majority of a-EJ being Pol θ dependent in other eukaryotes. POLQ deficiency led to an over 10-fold reduction in the ability to join an extrachromosomal substrate with resected ends in mouse cells¹⁸, as well as a loss of the major class of microhomology mediated products (i.e. microhomologies 2–10 bp) generated after repair of targeted chromosome breaks in mice¹⁷, zebrafish¹⁹, and Caenorhabditis elegans²⁰. Moreover, cells deficient in both NHEJ and Pol θ had at best trace levels of end joining of breaks generated by nucleases and deprotected telomeres as well as reduced viability, relative to cells deficient in either pathway alone^17,21–23. Taken together, the available evidence argues that in most eukaryotes a clear majority of repair that satisfies the above definitions for a-EJ – end joining that is independent of NHEJ, engages resected 3’ssDNA tails, and favors microhomologies – requires Pol θ. We thus favor reference to such repair as Theta mediated end joining (TMEJ), consistent with pioneering use of this term²⁴, unless its possible to definitively demonstrate independence of the repair products from both NHEJ and Pol θ.

The molecular mechanism of TMEJ

Pol θ is a large protein (Figure 2a) – 290 kDa in mammals – with an N-terminal helicase-like domain (HelD) and a C-terminal polymerase domain (PolD)²⁵. The pairing of these domains together, as well as their sequence, is highly conserved^26,27. TMEJ activity in vitro requires both domains to be present in cis (i.e. addition of separated domains is not similarly functional)²⁸. The sequence of the central domain (CenD) is much less well conserved, is likely disordered²⁸, varies in size from 265 amino acids in worms to 920 amino acids in humans^26,27. The CenD may play a regulatory role, as a construct entirely missing this domain (the HelD instead fused to the PolD with a linker) is active for TMEJ in vitro²⁸.

Figure 2.

Molecular mechanism of Polymerase theta mediated end joining. a) Pol θ has an N-terminal superfamily 2 helicase-like domain (HelD, cyan) separated from a C-terminal A family Polymerase domain (PolD, yellow) by a poorly conserved central domain (CenD, black); domain boundaries are for the human Pol θ protein (Uniprot O75417). b) Model for steps in TMEJ, identifying at right the domains of Pol θ or other factors argued to perform these steps. Hydrolysis of ATP (ATPase) by the HelD removes Replication protein A (RPA) from resected ends to enable subsequent steps, and may drive sampling of end pairings to find complementary sequence (in red) at microhomologies before a nuclease (scissors) removes non-homology tails to activate synthesis (in blue) by the PolD. Resolution could occur without strand displacement (short patch) or with strand displacement (long patch) c) Mechanism for generation of templated insertions (TINS) through aborted intermolecular synthesis to generate direct repeat TINS, or aborted intramolecular synthesis primed by annealing of inverted repeats to generate inverted repeat TINS. Complementary sequences at microhomologies are in red, inserted sequence in junctions is bolded, and the orientation of inserted sequence relative to template is noted by blue arrows. Intramolecular synthesis cannot directly contribute to productive repair, and may be blocked by the nuclease activity of the PolD (scissors) and by unwinding activity of the HelD.

The PolD belongs to the A family [G]²⁵, though it has insertion loops relative to other A family polymerases that are likely linked to its unique biological role^29–31. Additionally, though the proofreading exonuclease normally associated with A family polymerases is inactive²⁵, a recent report indicates the polymerase domain is also a structure-specific endonuclease³². Endonuclease activity requires dNTP, Mg²⁺, and the same Mg²⁺-binding amino acids employed by synthesis activity, suggesting an ability to remodel the same active site to switch between extending and trimming 3’ termini³², which is unprecedented for polymerases. Importantly, this sharing of active site for the two activities has hindered the development of a mutation allowing separation of these functions, thus making a definitive determination of its biological role difficult. The HelD is a ssDNA activated ATPase and a member of the SF2 superfamily of helicases²⁵. However, it is missing structural motifs that were shown in its paralog HEL308 to be important for strand separation, and it has at best very low levels of DNA unwinding activity^25,33–35.

The most parsimonious model for TMEJ involves the following steps (Figure 2b): 1) end recognition, 2) end pairing and microhomology search, 3) trimming of nonhomologous tails, 4) synthesis from the microhomology, including displacement of the downstream 5’ terminus and 5) resolution. Tight coupling of these steps together would appropriately mitigate the loss of flanking sequence that is a direct consequence of the role of microhomologies in stimulating synthesis across two DSB ends. How could the diverse Pol θ activities described above contribute to these steps, and what additional factors are required?

1. DNA end recognition

Loading of Ku on DNA ends, and consequently repair of these ends by NHEJ, is progressively impaired by increasing length of resected 3’ ssDNA overhangs^16,17. TMEJ in mammals is thus favored over NHEJ when overhangs are 30–70 nucleotides long¹⁷. These observations generally correlate well with the <100 nucleotide 3’ ssDNA overhangs generated by initial activity of the slow phase of resection (i.e. through action of CtIP and MRN) in mitotic budding yeast as described in a recent preprint³⁶. It is not clear, though, if TMEJ loses efficiency on longer ssDNA overhangs, e.g the many 100s of nucleotide overhangs generated^36–38 after prolonged MRN activity or by fast resection (EXO1 or DNA2 and BLM)^10,39,40. It also isn’t yet clear if both ends must be single stranded, i.e. if Pol θ/TMEJ can directly access potential “one ended” intermediates generated after replication fork collapse, a biological role in which Pol θ activity has been implicated (see below, “Cellular functions of TMEJ”).

Recruitment of Pol θ to damage is reduced in cells deficient in Poly(ADP-ribose) Polymerase-1 (PARP-1) or if cells are treated with PARP inhibitors^21,41. This is consistent with early work arguing deficiency in PARP-1 or inhibition of PARP-1 enzymatic activity leads to a reduction in end joining that is independent of NHEJ factors, as well as radiosensitivity^42–44. PARP-1 is activated by single and double strand breaks to add Poly(ADP-ribose) chains (PARylation) to nearby protein substrates as well as itself⁴⁵. This in turn aids in recruitment of factors that repair strand breaks, though it isn’t yet clear whether PARylation has a direct role in recruiting Pol θ during TMEJ. Notably, PARP1 deficiency was dispensable for resolution of V(D)J recombination intermediates by TMEJ⁴⁶, and there were additive effects of Pol θ deficiency and PARP inhibition on viability of BRCA1 deficient cells³³, suggesting PARP-1 activity isn’t essential for all TMEJ.

Pol θ recruitment to damage and TMEJ is similarly reduced by deficiency in, or inhibition of, the Fanconi anemia complex D2 (FANCD2) ubiquitin ligase⁴¹. A role for several FANC members, including FANCD2, A, C, E, F, M, and FAAP was also identified in a systematic screen for factors required for a-EJ⁴⁰.

ssDNA in eukaryotic cells is bound by Replication protein A (RPA), which inhibits repair by a-EJ whether a Pol θ ortholog is present (mammals)⁴⁷ or not (yeast)⁴⁸. The HelD of Pol θ has been implicated in mitigating this barrier; loss of the HelD ATPase activity in Pol θ resulted in a reduced ability to displace RPA from ssDNA in vitro, reduced chromosomal TMEJ, and a reduced ability to rescue viability of BRCA1 deficient cells⁴⁷. The HelD may thus mitigate the block RPA poses to TMEJ by removing RPA from resected DNA ends, thus enabling end pairing and subsequent steps. RPA may also be displaced by HMCES (5-hydroxymethycytosine binding, ES-cell specific antigen) during TMEJ, as this widely expressed factor also binds ssDNA, but unlike RPA, HMCES promotes a-EJ⁴⁹.

2. End pairing and Microhomology search

Resected 3’ ssDNA overhangs must be paired together, after which microhomologies fortuitously present within these tails are identified and annealed (Figure 2b). In vitro and cellular studies in mammals both argue that for repair to be efficiently completed in a single cycle of synthesis microhomologies typically need to be 3 bp or more^17,50,51, with both copies of the microhomology within 15 nucleotides of the 3’ terminus⁵⁰. This is consistent with a fraction of deletions <50bp associated with Pol θ activity at targeted DSBs in somatic cells^17,50,52, as well as in presumptive replication associated breaks in genomes of C. elegans^24,53,54. Frequent longer, >50 bp Pol θ-dependent deletions can be observed in cells deficient in 53BP1 or NHEJ factors^{17,22,50,55,56} as well as in wild type embryonic stem cells⁵⁷. This may reflect engagement of Pol θ after degradation of resected 3’ssDNA tails when resection is excessive (see also “Regulation of TMEJ”.

Microhomologies 3 bp or more are expected to be present within 15 bp of either terminus for approximately 90% of breaks⁵⁰. Repair can also be mediated by TMEJ for contexts possessing only 0–2 bp microhomologies, but this more often requires multiple – typically 2–3 – rounds of synthesis. This is apparent from distinctive TMEJ repair products that possess inserted sequences at the break site that are either a direct repeat or inverted repeat, relative to sequence within 40 nucleotides of the break^{14,17,19,20,24,50,54,58–60}(templated insertions, or TINS; Figure 2c). Direct repeat vs. inverted repeat TINS reflect an aborted round of intermolecular or intramolecular synthesis, respectively, and are less frequent than canonical microhomology-associated deletions. However, TINS are a more specific footprint of pathway activity than microhomology associated deletions^50,61, and consequently have been used as a biomarker for TMEJ activity behind disease alleles⁶¹, genomes of breast cancers⁵⁰, as well as in parathyroid cancers arising after the nuclear plant accident at Chernobyl⁶².

Intramolecular annealing of a 3’ terminus (annealing of the 3’ terminus to complementary sequence on the same side of the break) is typically favored over intermolecular annealing of equivalent complementary sequence. Synthesis primed from intramolecular annealing is thus expected to be frequent, but it cannot directly contribute to successful repair (Figure 2c). This problem appears particularly relevant for TMEJ in Drosophila melanogaster, where junctions with insertions consistent with intramolecular primed synthesis (TINS that are inverted repeats, relative to flanking sequence) are typically more frequent than junctions formed after a single intermolecular round of synthesis^59,63. A striking solution to the problem of intramolecular-primed synthesis is possibly provided by way of the endonuclease activity recently identified in the mammalian Pol θ PolD. This nuclease preferentially cleaves hairpin structures generated after intramolecular annealing of the primer, thus releasing a primer terminus that will now favor intermolecular annealing, synthesis, and thus end joining (Figure 2c)³². A similar role in suppressing intramolecular synthesis was demonstrated for the Pol θ HelD in vitro²⁸, although for this latter observation it is difficult to distinguish whether it is mediated through suppression of intramolecular synthesis, or (as suggested below) promotion of the competing intermolecular synthesis reaction.

Structural studies argue the Pol θ PolD and Pol θ HelD can form a dimer³¹ and tetramer³⁴, respectively, so DNA binding by either or both domains could mediate end pairing. Indeed, the ability of both domains to pair ends may be important to sustain pairing as the reaction progresses. A model of the Pol θ HelD oligomer bound to DNA ends allows for alignment of a pair of 3’ termini in anti-parallel orientation, which would promote annealing of near terminal microhomologies³⁴. The ability of the Pol θ HelD to promote intermolecular synthesis at the expense of intramolecular primed synthesis during TMEJ in vitro is perhaps consistent with this argument²⁸. Finally, the Pol θ HelD ATPase likely drives translocation of Pol θ along ssDNA, from 3’ to 5’, which given the above structural model would drive a directed “search” for microhomologies starting from the 3’ terminus. In accord with this model, TMEJ in cells strongly favors the use of the most 3’ proximal of two equivalent nearby microhomologies⁵⁰.

3. Trimming of nonhomologous tails

Unless the microhomology is at the extreme 3’ terminus – this is rare, given a 3 bp or larger microhomology is the preferred substrate for mammalian Pol θ – annealing of microhomologies will generate nonhomologous 3’ termini (Figure 2b). Synthesis from a mispaired 3’ terminus is possible in vitro, though inefficient^51,64,65, consistent with evidence that such events contribute little to cellular TMEJ¹⁷. Nonhomologous tails present after annealing of microhomologies must thus be trimmed by either a 3’ flap endonuclease or a 3>5’ ssDNA exonuclease. The 3’ structure-specific nuclease in the Pol θ PolD is a strong candidate nuclease for this trimming function³¹, since it would allow the requisite tight coupling of nuclease activity and subsequent synthesis. Another candidate is the 3’ flap endonuclease XPF–ERCC1, since deficiency in XPF reduces recovery of an a-EJ product⁶⁶ and increases radiosensitivity of Ku deficient cells⁶⁷.

4. Synthesis

Pol θ is a template dependent A family polymerase; its ability to extend in the absence of template in vitro is at best low^18,29,51,68, and dependent on extended incubation and non-physiological conditions. In accord with this observation, 2/3^rd of insertions in cellular Pol θ dependent repair products have unambiguous local template (the remainder are plausibly templated as well, but tracts of synthesis were not sufficiently long enough to determine this with enough confidence)⁵⁰. Synthesis by Pol θ is error prone^65,68, and generates substitutions, insertions, and deletions at comparable rates (1–3×10⁻³), which is a much higher frequency than other A family polymerases.

The instability of intermolecular end pairing means processivity, the ability to sustain synthesis in a single primer/template binding cycle, is a key requirement for polymerase involvement in TMEJ. This is evident from cellular TMEJ products, as restriction of TMEJ to regions with microhomologies <3 bp and AT rich template sequence results in frequent failure of synthesis after addition of less than 10 nucleotides, followed by re-engagement at an alternate microhomology (Direct repeat TINS; Figure 2c)⁵⁰. By comparison, Pol θ in vitro and in cells is able to sustain >50 nucleotides of synthesis in a single binding cycle on stable primer/template complexes (i.e with 6 bp or more microhomology), especially when full length Pol θ is employed^28,68.

Pol θ – even full length Pol θ – is unlikely to be sufficiently processive to complete TMEJ when the length of resected 3’ssDNA overhang exceeds 100 nucleotides, or in the context of a possible role in a distinct microhomology primed synthesis repair reaction termed microhomology mediated break induced replication (MMBIR)⁶⁹. Pol δ has been implicated in a-EJ^70,71, consistent with an initial Pol θ dependent, microhomology primed synthesis step, after which complete repair requires a switch to a much more processive synthesis mediated by Pol δ.

5. Resolution

Resolution of TMEJ could involve gap filling synthesis sufficient to allow ligation to a downstream 5’ phosphate. However, the insensitivity of cellular TMEJ to downstream 5’ flapped substrates¹⁷ suggests recognition of downstream 5’ phosphate isn’t essential, with resolution instead often involving synthesis that displaces the downstream strand, followed by eventual removal of the resulting 5’ flap and ligation (Figure 2b). Candidate 5’ flap endonucleases implicated in a-EJ include DNA2⁴⁰ and FEN1⁷². This “long patch” model has the advantage of allowing bypass of 5’ adducted proteins and repair of forked substrates (e.g. collapsed replication forks; see Cellular functions of Pol θ). The PolD alone has limited strand displacement activity⁵¹. Strand displacement activity may be enabled by the HelD²⁸, or an upstream switch from Pol θ to Pol δ (suggested above, as a means to promote processive synthesis).

Ligation, whether resolution is by short or long patch mechanisms, is probably mediated primarily by DNA Ligase III (LIG3), as it can be functionally implicated in cellular a-EJ/TMEJ^21,73, and a number of biochemical studies confirmed this ligase can perform a-EJ’s final step^42,74–76. However, it’s also clear that Ligase I (LIG1) can act in place of LIG3I during cellular a-EJ in many contexts^73,77,78.

Cellular functions of TMEJ

The early limitation of defining a-EJ/TMEJ primarily by what it wasn’t (i.e. “not NHEJ”) made it difficult to characterize its cellular functions, at least beyond an ability to act as a backup for NHEJ. The identification of Pol θ as essential for a/EJTMEJ has allowed for a description of this pathway as more than just a backup, and hints at important roles for TMEJ in rescue of DSB repair in contexts where NHEJ or HR have been blocked, as well as a role in the re-establishment of replication after replication fork collapse.

TMEJ: Backup, or the third DSB repair pathway?

Deficiency in Pol θ alone typically causes mild phenotypes^79–81, but is required for survival and radiation resistance in cells with genetic deficiency in the NHEJ pathway^17,56. Cell viability is similarly severely compromised when deficiency in Pol θ/TMEJ and HR genes are combined (e.g. BRCA1 or BRCA2; Figure 3a)^{14,21,33,53,55}. More surprisingly, synthetic lethal POLQ genetic interactions extend well beyond HR and NHEJ pathways, as determined in a CRISPR screen that looked for synthetic growth defects when deficiency in POLQ was combined with deficiency in a broader panel of DNA damage response genes⁵⁵. While confirming synthetic lethality with HR and NHEJ pathway deficiencies this screen revealed a synthetic sick/lethal landscape of Pol θ deficiency that also included factors active in base/nucleotide excision repair, damage response signaling, and anti-resection pathways^55,81,82.

Figure 3.

Biological roles for TMEJ. a) Recue of blocked HR by TMEJ. TMEJ generates deletion scars, but can restore chromosome integrity when resected ends are unable to form a presynaptic filament (block 1; e.g. due to BRCA2 deficiency) or unable to perform strand exchange (block 2; e.g. due to BRCA1 deficiency). When non-cross over (NCO) resolution of HR by synthesis dependent strand annealing (SDSA) is blocked (block 3), TMEJ suppresses Holliday junction (HJ) resolvase-mediated crossovers (CO) between maternal and paternal chromosomes and associated loss of heterozygosity (LOH), or alternatively sister chromatid exchanges (SCE). c) Protein-adducted ends block NHEJ, and can be resolved by TMEJ after removal by Mre11 or strand displacement synthesis. d) Stalling of the replication fork at protein adducts and secondary structures, including G quadruplexes (G4), activates Ataxia Telangiectasia related kinase (ATR) dependent pathways including fork reversal, or can result in daughter strand gaps that break (potentially after cell division, in the next S phase), and are repaired by TMEJ. d) V(D)J recombination intermediates are retained in a complex with RAG1 and RAG2 proteins (red circle) during the G1 phase of the cell cycle, which blocks access of ends to TMEJ. Intermediates not repaired in NHEJ deficient cells are resolved by TMEJ after entry of cells into S phase after S-phase dependent destruction of RAG proteins and activation of resection.

That these genetic interactions are so diverse, as well as the evolutionary conservation of Pol θ across plant and metazoan species, suggests a more important role in cellular DNA metabolism than simply acting as a backup when cells are deficient in NHEJ or HR. In accord with this, loss of Pol θ alone in diverse backgrounds results in an increase in unrepaired DNA damage and genome instability^18,24,55,81, observations suggesting that Pol θ’s broad synthetic lethality landscape may reflect more than simply a “backup” repair function. Rather, genetic deficiency in canonical repair pathways may represent a more sensitive means of identifying specific repair contexts where TMEJ is the preferred pathway for DNA repair -- where its activity, while error-prone relative to other DSB repair, nevertheless protects against more catastrophic types of genome rearrangements. Accumulating recent evidence has begun to shed insight into the nature of these Pol θ-dependent repair settings.

TMEJ and rescue of blocked DSB repair

TMEJ is synthetic lethal with deficiencies in RAD51 and BRCA2 (which loads RAD51 on resected ends)^21,33,55, arguing it aids in repair of resected ends unable to form a presynaptic filament (Figure 3a, block 1). Additionally, a recurrent feature of Pol θ synthetic lethality with factors involves in later steps is the accumulation of abnormally large RAD51 foci, consistent with an ability of TMEJ to rescue HR that would otherwise lead to unproductive RAD51-coated pre-synaptic filaments (Figure 3a, block 2)⁵⁵. There is also evidence HR blocked at late steps can be rescued by TMEJ (Figure 3a, block 3). In somatic cells, HR preferentially employs synthesis-dependent strand annealing (SDSA) for DSB repair, which typically does not generate crossovers^83–85. However, SDSA can be blocked -for example if the sister chromatid has a polymerase-blocking lesion - which may result in engagement of error-prone HR pathways such as interhomolog or non-allelic HR (Figure 3a, block 3)^86,87. These alternative recombination pathways require the activity of structure-specific Holliday junction resolvases, including SLX4 and GEN1⁸⁸, which can promote genetic crossovers and are consequently drivers of loss of heterozygosity (LOH) of tumor suppressors in cancer^{86,87 88}. Combined loss of Pol θ, SLX4, and GEN1 in Drosophila is synthetic lethal, an observation consistent with a strong induction of DSB-induced crossovers in somatic cells when deficient in Pol θ (Figure 3a)⁸⁹. DSB-induced interhomolog recombination was similarly suppressed by Pol θ in human cells⁹⁰. Pol θ/TMEJ also promotes appearance of long terminal duplications in a background of BRCA1 deficiency in worms, an observation that is most consistent with rescue of HR after initiation of synthesis, i.e. at a post-synaptic step⁵³. In sum, TMEJ may be able to salvage repair after blocked HR irrespective of what HR step is blocked. These observations also reinforce the critical role resection plays in regulation of TMEJ engagement; once ends are resected in cells lacking HR activity, TMEJ is often the predominant alternative repair pathway.

TMEJ can also help resolve DSBs refractory to repair by NHEJ, as was first apparent from observations noted above where TMEJ alone mediates end joining repair of resected ends^7,8,10,17. Pol θ deficient cells are additionally sensitive to agents that generate NHEJ-blocking protein adducts, including topoisomerase inhibitors (Etoposide and camptothecin) and formaldehyde^18,91, consistent with an ability of TMEJ to resolve substrates with protein blocked ends both in mammalian cells and xenopus extracts^17,91. Resolution of these end structures may follow resection activity of MRN and CtIP or, with 5’ adducts, be enabled by strand displacement synthesis by Pol θ and a long patch resolution mechanism (Figure 3b).

TMEJ and collapsed replication forks

A role for Pol θ /TMEJ in rescuing replication after collapse of replication forks was made apparent in early work in C. elegans (Figure 3c). C. elegans genomes accumulate small, 100 bp microhomology associated deletions at G quadruplex (G4) DNA when deficient in FANCJ, a helicase that promotes replication through these secondary structures⁹². A similar genomic scar is associated with sites of presumptive spontaneously introduced oxidative damage in genomes of C. elegans deficient in the translesion polymerases Pol κ and Pol η²⁴. In either context, when C. elegans are additionally deficient in Pol θ there is a reduction in viability, as well as an exchange of the characteristic TMEJ and microhomology mediated small deletion scars for much larger (several 1000 bp) deletions in the genomes of survivors. Interestingly, G-quadruplexes can be acted on by TMEJ after replication⁹³. This block is transmitted without repair through a cell division as a daughter strand gap, generating a two-ended DSB in a subsequent S phase (Figure 3c). Given that the sister chromatid harbors a synthesis-blocking G quadruplex it cannot be used effectively as a template for repair by HR, but after resection is an ideal substrate for TMEJ.

In accord with this work in C. elegans, Pol θ in mammals promotes viability and antagonizes accumulation of ssDNA and DSBs in diverse conditions that induce stalled or collapsed forks, such as polymerase inhibitors, G quadruplex stabilizers, interstrand crosslinking agents, UV light, and topoisomerase inhibitors^18,55,94,95. Furthermore, inhibition of the ATR replication stress response pathway (e.g. ATR activated fork reversal) renders cells strongly dependent on Pol θ to prevent lethal accumulation of unrepaired DSBs⁹⁴ (Figure 3c). Pol θ thus has an important, though mutagenic role in helping re-initiate replication after replication fork collapse.

What is the signal that recruits Pol θ to collapsed forks? In BRCA1- or BRCA2-deficient cells, FANCD2 monoubiquitination may be required for Pol θ recruitment to sites of replication fork collapse, which promotes resistance to PARP inhibitors⁴¹. In HR proficient cells, p53 may regulate the balance between HR and TMEJ, as p53 deficiency induces reduced RAD51 and increased Pol θ recruitment to stalled replication forks⁹⁶. Pol θ may also only gain access to collapsed forks after dissolution of the replisome complex. In a Xenopus extract model, mitotic CDK activity stimulated fork collapse and replisome disassembly at sites of DNA protein cross links, resulting in Pol θ-dependent end joining events that harbored deletions and templated insertions⁹⁷. Restricting TMEJ to settings where replication forks have collapsed may ensure preferential utilization of other, less error prone mechanisms for replication fork restart (e.g. fork reversal or HR) when possible. Whether Pol θ contributes to replication stress-associated mitotic DNA synthesis [G], or to the development of clustered, highly complex rearrangements during chromothripsis [G] remains to be clarified^98,99.

TMEJ in programmed gene rearrangements

The adaptive immune response relies on the ability to generate diverse antigen specific receptors through V(D)J recombination. This recombination is initiated at immunoglobulin loci in B cells, and T cell receptor loci in T cells, by targeted breakage of these antigen receptor loci by the RAG1 and RAG2 nuclease. Repair of the resulting double strand break intermediates is almost entirely dependent on NHEJ. This near complete exclusion of TMEJ participation can be reversed using a truncation mutant of RAG2¹⁰⁰. The authors argue the mutant destabilizes a complex of RAG1 and RAG2 bound to broken intermediates that normally ensures the channeling of these ends to repair by NHEJ (Figure 3d). In accord with this interpretation, fusion of Rags to I-SceI is sufficient to block TMEJ of I-SceI generated breaks in NHEJ deficient cells¹⁰¹.

RAG1 and RAG2 are active only in the G1 phase of the cell cycle. This cell cycle restriction is also important in excluding participation of a-EJ/TMEJ; an elegant study determined that in NHEJ deficient cells, breaks induced in G1 (whether induced during V(D)J recombination or by Cas9) can be repaired by TMEJ only when cells enter S-G2⁴⁶. This cell cycle dependence could plausibly be due to S-phase specific destruction of the RAG proteins that block TMEJ, the need for S-phase specific phosphorylation of CtIP to allow resection, or both (Figure 3d). Regardless, the consequences of this interrupted repair are striking: TMEJ after S phase entry is associated with excessive deletion (100s-1000s of bp), as well as a >100 fold increase in chromosomal translocations.

B cells in lymph nodes and the spleen also employ a second programmed gene rearrangement to switch immunogloblobulin isotype, from IgM to either IgG, IgA, or IgE (immunoglobulin class switch recombination, or CSR). Like V(D)J recombination, breaks are restricted to the G1 phase of the cell cycle and are normally repaired primarily by NHEJ; Pol θ deficiency alone does not result in large reductions in CSR^18,102,103, though there is a loss of 9% of a class of CSR products with insertions >1bp that are likely templated insertions (TINS)¹⁸. Notably, in NHEJ deficient cells CSR is much more efficient than is V(D)J recombination^104–106, though is delayed¹⁰⁷, possibly again reflecting a requirement for S-phase entry for resection and repair.

Regulation of TMEJ

Maintenance of genome integrity requires precise regulation of error-prone TMEJ; yet, how this regulation is achieved remains enigmatic. Engagement of this pathway is under the control of core components of the DSB recognition machinery, particularly factors that regulate resection of DSBs (Figure 1; discussed next). PARP activity has been implicated in Pol θ recruitment^21,41, although the precise molecular mechanism for damage site recognition by Pol θ remains poorly understood. Also unknown is how the balance between HR and TMEJ is maintained, as both pathways act on resected DSBs. While some groups have suggested TMEJ competes with HR^21,33, such a role would not be consistent with a genome stabilizing function of Pol θ. Within cells, there are likely multiple layers of regulation that restrain TMEJ activity to situations where canonical DSB repair programs are unable to effectively repair the damage. A more complete understanding of TMEJ may also reveal how the pathway can be over-utilized in cancer.

Regulation of TMEJ by resection

DSB repair pathway choice is closely related to the extent of end resection (Figure 4). TMEJ has an intermediate requirement for end resection that falls between NHEJ and SSA or HR (Figure 4). An initial “slow” phase of resection, requiring MRN, CtIP, and ATM mediated phosphorylation, generates 3’ ssDNA overhangs varying from 10s to 100s of nucleotides (Figure 4); requirement for these factors is a defining characteristic of a-EJ and TMEJ in all eukaryotes. Resection is activated in part through CDK-mediated phosphorylation of CtIP in the S and G2 phases of the cell cycle. That resection is at least mostly restricted to S phase is consistent with evidence for similar cell cycle restrictions for activity of TMEJ^46,108.

Figure 4.

Regulation of DSB repair pathway choice by end resection. The 5’>3’ resection of DSB ends has emerged as a major node for regulation of DSB repair. The 4 major pathways – NHEJ, TMEJ, SSA, and HR – have distinct requirements for homology that determines the extent of resection required for successful pathway engagement. a) The first regulatory node is the recruitment of the 53BP1-Shieldin-CST/Polα anti-resection complex to DSB-associated chromatin. Recruitment of the anti-resection complex is antagonized by BRCA1. The 53BP1-Shielding-CST/Polα complex promotes NHEJ and antagonizes TMEJ, SSA, and HR. b) Initiation of end resection is mediated by the MRN complex and CtIP, the activity of which is regulated positively by ATM and CDK1/2, the latter of which restrains resection to the S/G2 phases of the cell cycle. This initial phase of end resection typically is limited to resection of <100bp, which is sufficient to inhibit Ku binding (thus impeding NHEJ) and promote TMEJ. c) Fast, processive resection involves recruitment of Exo1, DNA2, and/or BLM, nucleases that are also positively regulated by ATM and CDK1/2 phosphorylation. These factors can promote resection of 1000’s of bases, which facilitates homology search for SSA and HR. Whether extensively resected ends can be processed for TMEJ is currently unknown, but may involve degradation of the 3’ overhang.

There are also antagonists of resection that inhibit TMEJ. 53BP1 is a chromatin binding protein that nucleates the recruitment of a constellation of factors that antagonize resection, thereby promoting NHEJ and antagonizing HR and TMEJ repair (Figure 4a)^55,109. Engagement of HR—and likely TMEJ as well—requires inhibition of 53BP1 binding by BRCA1 (Figure 4b)¹¹⁰. Intriguingly, combined 53BP1 and Pol θ deficiency is synthetic lethal^17,55, while 53BP1 deficiency has the opposite effect (i.e., synthetic viability) when combined with BRCA1 loss^111,112. A plausible explanation is that while 53bp1 pathway deficiency restores HR when the defect is in end resection¹¹³, it also stimulates resection of a subset of HR-incompatible DSBs that now require TMEJ for repair. Synthetic lethality with Pol θ loss is also observed for factors recruited by 53BP1 and similarly implicated in antagonizing resection, including PTIP, RIF1, and MAD2L2/REV7⁵⁵. 53BP1, RIF1, and REV7 also recruit the Shieldin complex (SHLD1, SHLD2, and SHLD3), CST complex (CTC1, STN1, TEN1), and DNA polymerase α, which are all essential components of the anti-resection machinery (reviewed in¹¹⁴). Whether loss of these various anti-resection factors equally promotes TMEJ has not been assessed, yet collectively anti-resection mechanisms likely represent an important node for regulating TMEJ pathway engagement.

Analysis of repair in cells defective in antagonists of resection or NHEJ factors provides an example of how TMEJ guards against genome instability when restrained, but has the opposite effect if overused. TMEJ is normally rarely used, is associated with deletions of less than 50 bp, and suppresses larger deletions^17,50,115. However, in cells deficient in 53BP1 or NHEJ factors the picture is very different: Pol θ is engaged in a much larger fraction of repair, and is associated with much larger TMEJ-mediated deletions^17,22,50,56. This may be because when resection is excessive/mis-regulated, Pol θ/TMEJ can no longer engage ends before significant degradation of 3’ ssDNA termini. This may also explain why Pol θ/TMEJ can reduce the frequency of deletions >50 bp in several somatic cell types, but appears to have the opposite effect in embryonic stem cells⁵⁷.

Limited resection by MRN triggers engagement of EXO1, or DNA2 and BLM/SGS1, and a faster resection phase, resulting in 3’ssDNA tails that can now extend several 1000s of nucleotides to promote homology-directed repair^37,38 (Figure 4c). Like deficiency in antagonists of resection, deficiency in fast resection has a clear impact on the spectra of products generated by a-EJ. In yeast deficient in Sgs1 and Exo I, a-EJ using a distal microhomology was ablated, with this pathway instead stimulating employment of a more proximal microhomology ~20 fold⁴⁸. This effect of regulation of resection on product spectra can confound assessment of a-EJ engagement when using a single “signature” product, possibly explaining contradictory evidence regarding the role of factors required for fast resection (EXOI, BLM/SGS1, and DNA2) in TMEJ in mammalian cell models (Figure 4c)^10,40,116.

Emerging Regulatory Mechanisms

Chromatin context can have a major impact on DSB repair pathway choice^117,118. Systematic analysis of signature NHEJ vs. TMEJ products of Cas9-induced breaks at various sites in the genome argues locations with heterochromatic features, including regions rich in Histone 3 tri-methylated at lysine 27, as well as di-methylated at lysine 9, skew repair towards TMEJ¹¹⁹. Cellular signaling may also influence repair pathway choice. TGF-beta has recently been proposed as a DNA repair modulator that suppresses TMEJ, which may be important for maintaining genome integrity in normal tissue cells^120,121. Conversely, the Human Papillomavirus (HPV) E7 oncoprotein suppresses NHEJ and stimulates TMEJ repair, which is corroborated by significantly higher levels of Pol θ-related genomic scars [G] in HPV+ human cancers¹²². However, the molecular mechanisms underlying these diverse TMEJ regulatory effects remain poorly understood.

TMEJ: both a cause and cure for cancer?

Cancer, fundamentally, results from a loss of genome integrity. Gene alterations in the DNA damage response pathway are prevalent in human cancers, and these cancers often also exhibit Pol θ overexpression^{30,33,55,58,123}. Thus, there is accumulating evidence that TMEJ upregulation may be an adaptive property of tumors that enables them to tolerate higher levels of endogenous DNA damage. The discovery of Pol θ synthetic lethality in cells with DNA damage response pathway alterations has elevated Pol θ as an attractive target for cancer therapy^17,21,33,55. Yet, despite the allure of Pol θ as a therapeutic target, relatively little is currently known about the effects of inhibiting Pol θ in preclinical cancer models, and the settings where its inhibition will be most impactful for tumor growth. Thus, identifying the optimal therapeutic contexts for Pol θ inhibitors represents a major need for clinical translation.

TMEJ as a driver of genome instability

TMEJ and genomic scars

The mutagenic consequences of TMEJ repair can be identified in the form of characteristic genomic scars, especially deletions commonly defined as microhomology length ≥2 bp and deletion size ≥5 bp (the latter restriction helps exclude contribution from more abundant NHEJ products, which similarly favors microhomologies)(Box 2). Indeed, the microhomology-preference that is partly driven by TMEJ helps make CRISPR/Cas9 genome editing more predictable^124–129, and has been proposed as a means to correct disease-causing genetic mutations (Figure 5a)^128,130.

Figure 5.

Clinical translation of Pol theta. a) A muscular dystrophy causing insertion mutation in the TCAP gene (TCAP^ins) can be corrected (TCAP^wt) through favoring of a specific repair product generated after cleavage with Cas9 (scissors) that is microhomology mediated (in red). b) Predictive biomarkers of Pol theta hyperactivity may help identify tumors that may be most vulnerable to Pol θ inhibition. Transcriptional upregulation of Pol θ and/or TMEJ-associated factors, measurement of TMEJ genomic scars (e.g., MHD or TINS), and presence of gene mutations that induce Pol theta dependency are candidate predictive biomarkers. A composite biomarker for TMEJ activity in tumors may be required for optimal patient selection. c) Pol θ has multiple functional sub-domains and molecular activities that can potentially be inhibited. These include helicase function (i.e., microhomology search), ATPase activity, and polymerase synthesis. The optimal domain for pharmacological targeting has not been determined. d) Pol θ inhibitors may be used in a variety of settings to improve cancer therapy. Preclinical studies support its investigation as monotherapy in highly selected tumors with strong dependency on Pol theta. Broader anti-tumor activity may be observed with combination therapy, particularly with DSB inducers, S phase damaging agents, and targeted therapy, including immunotherapy.

These scars are also be helpful in assessing pathway activity in cancer (Figure 5b). The profile of Polθ/TMEJ repair is essentially equivalent to the microhomology-rich insertion and deletion (indel) signature ID6 that was empirically derived in the Pan-Cancer Analysis of Whole Genomes¹³¹. Notably, expression of the ID6 signature is also associated with single base substitution (SBS) signature 3, both of which are enriched in cancers with HR deficiency^132,58. Based on these correlative associations, the ID6 and SBS3 signatures—sometimes in combination with other genomic features—have been investigated as predictive biomarkers for HR deficiency and sensitivity to platinum-based chemotherapy or PARP inhibitors^133–136. We propose that upregulation of the ID6 indel signature is an indicator of TMEJ activity levels rather than a direct measure of HR status (see Box 3 Reinterpreting BRCAness signatures in cancer). As noted earlier, though, the relationship between repair pathway and microhomology is complicated (Box 2); indeed, ID8 (deletion size ≥5 bp, but with a microhomology preference more aligned with NHEJ) is also enriched in cancers with high levels of Pol θ⁵⁸. Thus, the utility of ID6 as a predictive biomarker of Pol θ dependence remains to be established.

Box 3: Reinterpreting BRCA-ness signatures in cancer.

BRCA1/2 deficient cancers exhibit unique patterns of genome instability that include: SBS3, ID6, large scale state transitions, telomere allelic imbalance, and loss of heterozygosity^{131,132,167–169}. Various combinations of these genomic scar signatures have been used to establish classifiers that effectively distinguish BRCA-mutant cancers genomes from cancer genomes not associated with BRCA deficiency^133,136. Invariably, a subset of cancers that have no identifiable BRCA deficiency will be classified together with BRCA-deficient cancers. These tumors are often referred to as having “BRCA-ness,” and it has been postulated that this may reflect a functional HR deficiency that would exhibit similar therapeutic susceptibilities as bona fide BRCA-deficient cancers^134,170. However, the utility of BRCA-ness signatures to predict therapeutic benefit from platinum chemotherapy and PARP inhibitors has been mixed^171–174. Furthermore, correlation between BRCA-ness genomic signatures and the gold standard assay of HR proficiency – Rad51 foci formation – has been inconsistent¹⁷⁵.

Definitive proof that TMEJ is responsible for one of the scar signatures (ID6) induced by BRCA-deficiency warrants a re-interpretation of BRCA-ness⁵³(add Feng and Humphries REFs). Undoubtedly HR deficiency is sufficient to induce TMEJ scars^33,47,55,176. However, other deficiencies in DNA repair, such as those that disrupt the NHEJ or 53BP1 pathways, also induce TMEJ signature repair while retaining (and sometimes stimulating) HR proficiency^17,55,56,58. Simply put, while HR deficiency induces ID6, the presence of ID6 does not obligately indicate HR deficiency. Therefore, BRCA-ness signatures comprised of ID6 may identify HR proficient tumors that have activated TMEJ through other mechanisms. This may explain the limitations of BRCA-ness signatures in predicting PARP inhibitor sensitivity and their imperfect correlation with functional HR assays. In contrast, ID6 (and other TMEJ genomic scars) should be investigated as a predictive biomarker of vulnerability to Pol θ inhibition.

As previously discussed, TINS (Figure 2C) are rare relative to microhomology mediated deletions, but are more specifically associated with Pol θ mediated repair^50,61. TINS frequency is also highly correlated with BRCA deficiency in breast cancer genomes^50,137, but their paucity poses challenges to employment as a clinically pragmatic biomarker. Whether one or more of these TMEJ genomic scar signatures, possibly in conjunction with Pol θ expression levels, may predict Pol θ addiction in cancer remains a clinically pertinent yet unanswered question (Figure 5b).

Translocations and telomere fusions

Translocation between DSBs generated by targeted nucleases is reduced in mouse cells deficient in the a-EJ factors CtIP and LIG3^13,73, but there is contradictory evidence for the role of Pol θ/TMEJ. It promoted translocation in one experiment using cultured cells²¹, acted as a backup to NHEJ in suppressing translocation in another¹⁷, and alone was sufficient to suppress translocation in an in vivo (mouse) model¹⁸, as well as in a human cell line⁵⁸. Most translocations and large deletions observed in human tumors lack a strong microhomology signature¹³⁸, yet a subset of tumors harbor them at disproportionately high levels, possibly reflecting TMEJ hyperactivity¹³⁹. It remains unclear what different contextual factors are behind these diverse observations.

TMEJ can also mediate fusions of de-protected telomeres, but predominantly when cells are also deficient in NHEJ²¹. a-EJ/TMEJ may nevertheless confer a survival advantage over NHEJ as cancer cells undergo telomere crisis, possibly due to a predilection for promoting intrachromosomal rather than the ultimately lethal interchromosomal telomere fusions¹⁴⁰. Thus, pathologic TMEJ activity can drive the accumulation of chromosomal rearrangements, particularly when such activity promotes the survival of incipient tumor clones.

Integration of extra-chromosomal DNA

Two parallel studies demonstrated that NHEJ and TMEJ were responsible for nearly all random integrations of plasmid DNA in mammalian cells^22,23. Remarkably, combined deficiency of Pol θ and LIG4 abolished random integration and drastically increased gene targeting efficiency, in some cases up to 20-fold, and is a strong argument that Pol q-independent a-EJ is of little significance in mammals (Box 1). A role for TMEJ in driving integration of foreign DNA is also apparent in pathogenic settings. Integration of T-DNA [G] into plant genomes exhibits hallmarks of TMEJ and was shown to be primarily Pol θ-dependent in Arabidopsis^60,141. Similarly, integration of HPV genomes into cellular host DNA exhibits frequent use of microhomology, and accumulating evidence suggests that HPV oncoproteins stimulate Polθ/TMEJ activity^122,142.

Pol θ as a therapeutic target

Pol θ is a multifunctional protein with at least three distinct activities that may be targetable (Figure 5c). The polymerase domain of Pol θ has unique structural features suggesting that specific inhibitors may be attainable, and certainly represents an attractive drug target³¹. In vitro experiments argue the Pol θ HelD is essential for TMEJ²⁸ but it lacks robust dsDNA unwinding activity activity^25,33–35; the HelD ATPase activity and translocation along ssDNA may instead be critical to removing inhibitory RPA bound to resected end substrates (Figure 2b)⁴⁷, or in blocking unproductive intramolecular primed synthesis (Figure 2c)²⁸. Evidence for cellular requirement for the HelD or ATPase activity is mixed^18,47,143. A preprint describing a small molecule screen for Pol θ ATPase inhibitors identified Novobiocin, which is a known inhibitor of bacterial DNA gyrase¹⁴⁴. This compound demonstrated selective toxicity in BRCA1-deficient cancer models, and was able to overcome resistance to PARP inhibitors induced by 53BP1 deficiency. These findings reinforce the promise of Polθ as a therapeutic target, and should motivate the development of even more specific inhibitors for preclinical evaluation.

The optimal clinical scenarios where Pol θ targeting will be most impactful are not yet clear (Figure 5d). Pol θ inhibition as monotherapy relies on true “addiction” to Pol θ (such as in BRCA1/BRCA2, TP53BP1, or PRKDC mutant cancers), and sufficiently high levels of endogenous DSBs to induce cell lethality (Figure 5d). Alternatively, Pol θ inhibitors may find their broadest clinical utility as part of a combination therapy paradigm. Genetic inhibition of Pol θ has been shown to enhance sensitivity to a broad range of DSB-inducing therapeutics, including ionizing radiation, clastogens (e.g., neocarzinostatin, bleomycin), interstrand crosslinking chemotherapy (e.g., carboplatin, mitomycin C), alkylating agents (e.g., cyclophosphamide), and topoisomerase inhibitors^{18,55,91,94,145}. Pol θ targeting also enhances the response to PARP, ATR, and DNAPK inhibitors^{33,55,91,94,108}. The relatively minor role for Pol θ in normal cells coupled with its hyper-utilization in cancer suggests that targeting Pol θ may increase the therapeutic ratio of other DNA-directed cancer therapies.

Pol θ inhibitors may also have utility as a means for altering DNA repair programs that drive adaptive evolution in tumor cells¹⁴⁶. Reversion mutations in BRCA1 and BRCA2 are a common mechanism for resistance to PARPi therapy, and often bear hallmarks of TMEJ repair^147–149. Whether Polθ inhibitors may reduce the emergence of mutation-based resistance to targeted therapy should be explored. Finally, by shifting DNA repair patterns, Polθ inhibition may influence the immunogenicity of tumors during therapy, particularly as frameshift mutations can induce expression of potent tumor neoantigens^150–152. Thus, preclinical evaluation of Polθ inhibitors in immunocompetent cancer models, and in combination with immunotherapy, will be important.

Conclusion and future perspective

The last few years have seen an explosion of new knowledge regarding the mechanisms and physiological significance of a-EJ in DSB repair, most notably, the discovery of Pol θ/TMEJ as required for most a-EJ, in most eukaryotes. This has transformed our ability to interrogate this pathway and helped us sketch an outline of pathway mechanism, but gaping holes remain. Moreover, whether there is significant Pol θ-independent a-EJ in species that have a POL θ ortholog isn’t yet clear, nor is the functional relationship between TMEJ and other genes implicated in a-EJ.

It’s long been appreciated TMEJ can act as a backup for the major DSB repair pathways (HR and NHEJ), but recent work hints at cellular contexts where TMEJ is instead the primary choice for repair. The contextual cues that make it the primary choice, as well as the regulatory mechanisms that enforce this decision, remain to be determined. Repair of genomes by TMEJ is “error-guaranteed”: what is it that determines when such repair is nevertheless protective, vs. pathogenic? Finally, cutting edge algorithms have identified manipulation of TMEJ activity as useful for genome engineering, and early work shows it has promise as a therapeutic target in cancer. Here again, though, we need a more complete understanding of molecular mechanism, as well as pathway regulation, before the promise of TMEJ as a target for genome engineering and cancer therapy can be fully realized.

Glossary

A Family Polymerases

One of seven groupings of eukaryotic DNA polymerases, consisting in mammals of Polymerase gamma, nu, and theta.

G Quadruplex

Stable secondary structures in DNA generated by sequences that are rich in guanine (sometimes abbreviated as G4), and can impede both DNA replication and gene transcription.

Mitotic DNA synthesis

A DNA replication stress-induced repair process that involves DNA synthesis during mitosis, possibly through break-induced replication.

Chromothripsis

A mutational process whereby a portion of a chromosome is shattered into many fragments, most likely during erroneous mitotic progression, and rejoined inappropriately.

Genomic Scars

A recurring pattern of mutagenesis that can be attributed to a specific etiology or DNA repair process.

T-DNA

“Transfer DNA” that is transferred from the plasmid genome of some tumor-inducing bacterial species, such as Agrobacterium tumefaciens, into a host plant’s genome.