Pol ε, one of the paralogous B family polymerases that are conserved in all eukaryotes, is a very large protein that typically consists of 2000 or more amino acid residues 17. The functionally characterized proofreading 3'-5' exonuclease (Exo) and polymerase (Pol) domains are located in the N-terminal half of this protein whereas the C-terminal half contains no experimentally characterized or readily detectable domains except for two Zn-finger modules at the end of the sequence 1117242526. The Pol ε holoenzyme heterotetramer 27, the 20 Å resolution structure of which has been determined by cryo-electron microscopy (cryo-EM) 28, contains, in addition to the large catalytic subunit, three smaller subunits, DPB2-4; the DPB2 subunit is essential for viability, and its proper structure is required for high fidelity of genome replication 29. Site-directed mutagenesis experiments demonstrated that the Zn-fingers of Pol ε are required for its interaction with DPB2 29. Deletion of the other two accessory subunits is not lethal but leads to elevated mutation rates 3031.

The sequences of the Zn fingers in Pol α, Pol δ and Pol ζ are adjacent to the C-terminal portion of the catalytic domain that is homologous to the sequences of the Thumb subdomain in the available crystal structures of B-family DdDps. By contrast, the Zn fingers in Pol ε are separated from the N-terminal catalytic domains by a large insert that is similar in size to the N-terminal Exo-Pol module. Examination of the Cryo-EM structure 28 indicates that this insert and the DPB2-binding subdomain (Zn fingers) are, largely, structured and bound to each other; furthermore, the presence of this insert places the DPB2-binding area spatially apart from the N-terminal, catalytic Exo-Pol module. Somewhat paradoxically, it was shown by deletion mutagenesis and site-directed mutagenesis that the N-terminal, catalytic portion of Pol ε is not required for viability whereas the uncharacterized C-terminal portion is essential 1116242526.

We employed secondary structure prediction and fold recognition in combination with different sequence similarity search strategies in an attempt to elucidate the origin and possible functions of the essential C-terminal region of Polε. Secondary structure prediction and automated three-dimensional model building for the N-terminal 1200 amino acids of human Pol ε using the Phyre server 32, as expected, revealed a typical DNA polymerase fold (pdb: 1wn7, 1d5a, 1s5j, 1q8i, 2gv9, 2p5o) with a 100% confidence. Strikingly, the search with the remaining amino acids 1201–2286 of human Pol ε also revealed a DNA polymerase fold for the sequences preceding the Zn fingers with the confidence of 95% (E. coli DNA polymerase II, PDB code 1q8i), 90% (Desulfurococcus sp. tok DNA Polymerase, 1d5a) and 85% (Thermococcus kodakaraensis family B DNA polymerase, 1wn7). Although we did expect to detect some Thumb subdomain-like fold that would stabilize the positions of Zn fingers, the discovery of the entire second polymerase and exonuclease module was highly surprising. This unexpected finding prompted us to initiate a further, in-depth sequence analysis in an attempt to elucidate the origin and possible functions of the essential C-terminal region of Pol ε.

A PSI-BLAST search 33 with the C-terminal portion of the Pol ε sequence from Saccharomyces cerevisiae (amino acid positions from 1170 to 2085 aa) used as the query (with E = 0.001 inclusion threshold and composition based statistics on) reveals similarity to the sequence of DNA polymerase II of Photobacterium profundum (GI:90410522) of the B-family at the 3^rd iteration, with E-value = 2e-05; numerous sequences of B-family polymerases were detected in subsequent iterations. The same sequence was used as a query for an HHpred search 34. This method detects the similarity with a B-family polymerase from the archaeon Thermococcus sp. (pdb: 1qht) with E-value = 4.9e-06 as the second top hit (the first one is a self-hit to pfam08490: DUF1744, Domain of unknown function) and several additional hits to different sequences and profiles of B-family polymerases with statistically significant E-values.

The results of these searches strongly support the possibility, originally brought up by the structural comparisons described above, that the C-terminal portion of Pol ε is homologous to B family polymerases. A more detailed analysis showed that, although the C-terminal region of Pol ε readily aligned with B family DdDps, the motifs that contain the catalytic amino acid residues in both the Exo and Pol domains are disrupted in the Pol ε sequence, with the only apparent exception of the 'DIE' motif of the Exo domain (Figure 1). The partial conservation of this motif might indicate that the inactivated Exo domain of Pol ε retains metal-binding capacity, although not the catalytic activity. Thus, it appears that the C-terminal portion of the eukaryotic Pol ε is a derived B-family DdDp in which both the Exo domain and the Pol domain are inactivated. Inactivation of catalytic domains or subunits in DNA polymerase has been observed previously. In particular, we recently described a family of inactivated B family polymerases that is widespread in diverse archaea 35. In addition, the small subunits of eukaryotic B-family DdDps including DPB2, the essential second subunit of Pol ε, are inactivated versions of the exonuclease subunits of archaeal PolD 3637383940. However, to our knowledge, Pol ε is the first detected case of the combination of an active and inactive polymerase within the same protein. Thus, it seems particularly remarkable that both essential subunits of Pol ε are inactivated derivatives of replicative enzymes. The fusion of active and inactivated B-family polymerases in Polε supports the prediction that active and inactive forms function in concert in archaeae and some bacteria although so far no fusions analogous to Polε were detected in prokaryotes 35. As proposed previously, inactivated polymerase subunits are likely to perform essential functions in the assembly of replicative complexes 253536.

In general, when two homologous domains follow one another within the same protein, one would be inclined to suspect that they evolved by tandem duplication. However, the inactivated C-terminal part of Polε was much more similar to a variety of B-family polymerases, in particular, bacterial ones, than to the active, N-terminal polymerase moiety of Pol ε. Moreover, the latter, active moiety of Polε differed from other B-family DdDps including the inactivated C-terminal part of Polε by the presence of multiple, unique inserts (Figure 2). These observations do not support the intuitively plausible hypothesis of a tandem duplication in Pol ε and prompted us to investigate in greater detail the domain architectures of eukaryotic DdDps and their likely origins.

Eukaryotic B-family DdDps contain two Zn-finger modules at their C-termini. Unexpectedly, when examining the results of PSI-BLAST searches with the C-terminal portion of Pol ε as a query, we detected significant similarity to the C-terminal Zn-finger modules of the archaeal PolD rather than to those of Pol α, Pol δ, or Pol ζ. Specifically, the PSI-BLAST search with C-terminal Zn-finger sequence of the yeast Pol ε (amino acids 2116 to 2199) reveals highly significant similarity to the C-terminal Zn-fingers of many archaeal Pol D sequences (E-value 4e-06 in the second search iteration). By contrast, significant similarity to the Zn-fingers of other eukaryotic B-family polymerases could not be easily demonstrated. The reverse search with the Zn-fingers of archaeal PolD yielded, essentially, the same results (data not shown). The same relationship was detected using HHpred: the Zn-finger sequence from yeast Pol ε gave the top hits to several profiles of PolD with E-value as low as 1.4e-24. The multiple alignment of the Zn-finger domains of DdDps clearly reveal the specific similarity between the distal Zn-finger of Pol ε and the sole Zn-finger of the archaeal Pol D as opposed to the limited similarity to the Zn-fingers of other eukaryotic B-family polymerases (Figure 3). These observations suggest an unexpectedly complex evolutionary scenario for the origin of eukaryotic DdDps from archaeal ancestors. After performing this analysis, we became aware of the fact that the specific similarity between the Zn-finger of the catalytic subunit of Pol ε and archaeal PolD has been noticed previously although evolutionary implications of this finding have not been examined 40.

Because of the high sequence divergence of the Pol ε C-terminal domain, we were able to construct a reliable alignment only for approximately 280 amino acid residues from the Exo and Pol domains of B-family DdDps. Nevertheless, when this alignment was employed for phylogenetic tree reconstruction, the topology of the tree was quite stable as demonstrated with a variety of tree-building methods and different parameter combinations (Figure 4; see Methods for details). The results were, mostly, compatible with those of previously published phylogenetic analyses of B-family DNA polymerases 354142. The tree has a complex structure, with the active, N-terminal region of Pol ε clustered with the "major" group of archaeal B-family polymerases (PolBI) that is represented in nearly all archaea, whereas the rest of the eukaryotic B-family polymerases including the inactivated C-terminal portion of Pol ε are affiliated with a distinct, "minor" group of polymerases (PolBII) found in a smaller subset of archaea (Figure 4). These findings are in a general agreement with the previous results of phylogenetic analysis of archaeal and eukaryotic B-family polymerases 4344.

Together with the observations on the Zn-finger domains of archaeal and eukaryotic B-family polymerases, the results of phylogenetic analysis suggest an unexpectedly complicated scenario of the evolution of eukaryotic DdDps that is not limited to duplications and diversification as central trends at the early stage of eukaryogenesis 45. Instead, the results suggest distinct archaeal pedigrees for eukaryotic polymerases and imply that the archaeal ancestor of eukaryotes possessed at least two B-family polymerases as well as PolD from which the "major" B-family form acquired the Zn-finger, either prior to or during eukaryogenesis (Figure 5). The combination of a B-family polymerase with the PolD Zn-finger is not seen in any of the sequenced archaeal genomes, in accord with the conclusions of a recent phylogenetic analysis that derives the "archaeal" subset of eukaryotic genes from a deep branch of archaea 46. Under this scenario, eukaryotic Pol ε and the rest of the eukaryotic B-family polymerases appear not to be ancient eukaryotic paralogs sensu strictu, but rather, pseudoparalogs originating from paralogous archaeal ancestors 45.

The subsequent events in the evolution of eukaryotic B-family DdDps that occurred prior to the radiation of the major lineages of eukaryotes included not only two duplications of the Pol-Exo block that led to the origin of polymerases α,δ, and ζ, but also the duplication of the Zn-finger, probably, in the ancestral Polε, with the subsequent acquisition of the two-finger module by the common ancestor of Polα, Polδ, and Polζ. The inactivated C-terminal portion of the Polε is more likely to result from a fusion of two distantly related B-family polymerases as opposed to the intragenic duplication scenario. The topology of the phylogenetic tree suggests that the source of the C-terminal portion of Pol ε could be a proteobacterial (or bacteriophage) B-family polymerase (Figure 4) although, given the long Pol ε branch, its origin cannot be determined with any confidence. In principle, a long-branch artifact could even obscure a duplication of the N-terminal portion of Pol ε; however, this seems unlikely considering that the N-terminal sequence shows a distinct pattern of indels as opposed to a common pattern in the C-terminal sequence and the rest of the eukaryotic B-family polymerases (Figure 2).

The paper by Tahirov and colleagues reports a very exciting observation: they have shown convincingly, using a combination of in silico approaches based on structural comparison and iterative Psi-BLAST analyses, that the C terminal domain of the eukaryotic DNA polymerase ε, corresponds to an inactivated DNA polymerase of the B family. The eukaryotic DNA polymerase ε thus appears to be formed by the fusion of an active DNA polymerase B (in N-terminal) and an inactive DNA polymerase B (in C-terminal). Amazingly, the inactive DNA polymerase B does not seem to have originated from a duplication of the active one, but by the fusion of a bacterial-like DNA polymerase B (such as E. coli DNA polymerase II). This is a very interesting observation that deserves publication. The authors also notice that the two Zinc fingers of the eukaryotic DNA polymerase ε are more related to Zinc fingers of archaeal DNA polymerases D than to those of other DNA polymerases B. This is in line with the fact that archaeal DNA polymerases D and eukaryotic DNA polymerases ε both interact with homologous subunits in Archaea and Eukarya. From these two observations, the authors speculate about the origin and evolution of eukaryotic DNA polymerases. I think that the authors should more clearly distinguish between their observations and evolutionary hypotheses. For instance, in the abstract, the hypotheses are described in the "result section" and even introduce this section as if they were bona fide results. The main and exciting result is only presented as an additional observation!!! "In addition, we found that.....". I think that the hypothesis favoured by the author should be mentioned only in the conclusion.

Authors response: We appreciate these constructive suggestions and have revised the Abstract accordingly. In the main text, the description of the inactivated module of Pol ε already preceded the rest of the analysis, so no change was necessary.

Ideally, the authors should have discussed their observations in the context of alternative hypotheses on the origin of eukaryotes and the eukaryotic DNA replication apparatus (see below). In my opinion, in discussing evolutionary scenarios, terms such as "archaeal ancestor" (already in the title and abstract conclusions) (see Figure also 5) should be avoided. The term archaeal ancestor is confusing since the common ancestor of Archaea and proto-eukaryotes was probably neither a proto-eukaryote nor an archaeon. Similarly, the Human does not descend from Apes, but Apes and Human have a common ancestor.

Authors response: This point is often brought up, and a reminder, we hope, will be helpful to the reader. It is true that Homo sapiens did not evolve from Pan troglodytes or any other living great ape species but rather shares a common ancestor with them. However, that common ancestor was, necessarily, an ape (distinct from any extant ape, of course), so the phrase "ape ancestor of humans" is not confusing, in our opinion. Ditto regarding "archaeal ancestor of eukaryotes".

From our own analysis of the evolution of the DNA replication apparatus (unpublished), it is indeed likely that the last common ancestor of Archaea had two DNA polymerases of the B family and one of the D family (as suggested in Figure 5A) and this was possibly also the case for the last common ancestor of Archaea and proto-eucaryotes (as suggested by the authors). The authors imagine a scenario of evolution going from this "simple" ancestor to modern eukaryotes (transformation of the two ancestral polymerases B in four polymerases B and loss of the polymerase D in the lineage of modern eukaryotes). However, one cannot exclude other scenarios, such as the presence of more than two polymerases B in the common ancestor of archaea and proto-eucaryotes (with loss of DNA polymerase D in Archaea and of some DNA polymerase B in Archaea), and/or introduction of DNA polymerases of viral origin in Archaea and/or in the lineage of proto-eucaryotes 62. Since viral DNA polymerases of the B family are intermixed with cellular DNA polymerases in phylogenetioc tree 6364, it should be in any case interesting to extend the present analysis to viral DNA polymerases as well.

Authors response: We agree that alternative scenarios are imaginable. They might somewhat less parsimonious but parsimony is at best a rough guide in the study of such complex evolutionary scenarios. Analysis of viral polymerases is interesting although it is complicated by the typical high rate of evolution of viral proteins, even essential ones.

The authors discuss the compelling evidence for the complex evolutionary history of eukaryotic Family B DNA polymerases. Observations and their analysis are technically sound, and I have only minor questions.

1. Abstract: "of archaeal, eukaryotic, and bacterial B-family DNA polymerases, the main replicative polymerases in archaea and eukaryotes" is awkward.

Authors' response: corrected to a (hopefully) less awkward phrase

2. Ibid. "eukaryotic B-family polymerases, most likely, originate from two distinct archaeal ancestors" – perhaps change to "there are two subgroups of eukaryotic B-family polymerases, each most likely originating from its own archaeal B-family ancestor". Otherwise. "two distinct archaeal ancestors" can be mistaken for the description of B+D chimera in the following sentence.

Authors' response: modified for clarity

3. "As proposed previously, inactivated polymerase subunits are likely to perform essential functions in the assembly of replicative complexes" (also Conclusions) – a bolder suggestion may be that these proteins still facilitate a subset of catalytic reactions, if the maintainance of a proper conformation of subsrates/ligands is sufficient for catalysis – processive synthesis may not work well that way, but perhaps some sort of proofreading or ejection of abortive products might – discuss?

Authors' response: a bold proposal, indeed, in our opinion, too bold to be considered justified at this time. Actually, it has been shown that the 145 kDa proteolytic fragment of Pol ε, missing the C-terminal Pol Exo module, is indistinguishable from the four-subunit complex with respect to the exonuclease and polymerase activities but less rapidly dissociates from primer-template 65It is also known that the C-terminal domain of Pol2p and/or the auxiliary subunits are specifically involved in dsDNA-binding 66. Obviously, the C-terminal domain of Pol ε is critically important for replication but the complete elucidation of its specific functions requires much more experimentation. Nevertheless, in the revised version of the manuscript we are more specific about the possible role of the conserved DIE motif of the inactivated module of Pol ε.

4. Evolutionary scenario and Fig. 5: Why proteobacterial-type PolB, the source of the C-terminal domain tandem in eukaryotic Pol epsilon, has to be the symbiogenetic/mitochondrial acquisition – can it be a phage contribution instead?

Authors' response: in principle, it could be a phage contribution but we do not see any specific indications of such an origin of the inactivated polymerase module of Pol ε.

This manuscript reports the identification of a tandem exonuclease-polymerase homology module within the C-terminal regions of DNA polymerase epsilons. The authors propose these domains arose by gene fusion rather than intra-gene duplication and that they dispensed with their enzymatic activities. Also discussed are the evolutionary implications of similarities between zinc fingers of DNA polymerase epsilon and archaeal PolD.

This is a well-written and compelling report that contributes significantly to our understanding of the evolution of cellular DNA replication and repair. The sequence similarity methods and the statistical analyses used are entirely appropriate. These findings should now re-focus attention on the molecular mechanisms of these apparently inactivated domains in Pol-epsilon.

Authors' response: We appreciate these constructive comments and cannot agree more with regard to the importance of experimental investigation of the functions and mechanisms of the inactivated module of Pol ε. Moreover, such experiments are currently underway in the laboratory of one of us (YIP) at the University of Nebraska Medical Center.