Functional Evolution in Orthologous Cell-encoded RNA-dependent RNA Polymerases

Many eukaryotic organisms encode more than one RNA-dependent RNA polymerase (RdRP) that probably emerged as a result of gene duplication. Such RdRP paralogs often participate in distinct RNA silencing pathways and show characteristic repertoires of enzymatic activities in vitro. However, to what extent members of individual paralogous groups can undergo functional changes during speciation remains an open question. We show that orthologs of QDE-1, an RdRP component of the quelling pathway in Neurospora crassa, have rapidly diverged in evolution at the amino acid sequence level. Analyses of purified QDE-1 polymerases from N. crassa (QDE-1Ncr) and related fungi, Thielavia terrestris (QDE-1Tte) and Myceliophthora thermophila (QDE-1Mth), show that all three enzymes can synthesize RNA, but the precise modes of their action differ considerably. Unlike their QDE-1Ncr counterpart favoring processive RNA synthesis, QDE-1Tte and QDE-1Mth produce predominantly short RNA copies via primer-independent initiation. Surprisingly, a 3.19 Å resolution crystal structure of QDE-1Tte reveals a quasisymmetric dimer similar to QDE-1Ncr. Further electron microscopy analyses confirm that QDE-1Tte occurs as a dimer in solution and retains this status upon interaction with a template. We conclude that divergence of orthologous RdRPs can result in functional innovation while retaining overall protein fold and quaternary structure.

Many RdRPs contribute to maintenance and amplification of gene silencing signals initiated by primary sRNAs originating from viral genomes, endogenous sequences, or dsRNAs experimentally delivered into a cell (16 -23). In such cases, long single-stranded RNA (ssRNA) targets of an initial round of silencing become templates for RdRP-catalyzed RNA synthesis, ultimately giving rise to secondary sRNAs. Moreover, some RdRPs may trigger a silencing response with no apparent need for preexisting sRNAs (22)(23)(24)(25)(26)(27). This may involve RdRP recruitment to transcripts with aberrant or unusual molecular features generated by other RNA polymerases (23,24,28,29).
In the filamentous fungus Neurospora crassa, silencing, or "quelling," of transgenic arrays relies on RdRP QDE-1, PIWI/ PAZ protein QDE-2, DNA helicase QDE-3, and several other components, including the ssDNA-binding protein RPA (2). Interestingly, QDE-1 may trigger quelling by producing long aberrant ssRNA copies (aRNAs) of ssDNA intermediates that frequently arise in tandem-duplicated genomic sequences (30 -32). This reaction depends on QDE-3 and RPA and is stimulated by DNA damage. QDE-1 can subsequently use its aRNA product as a template to produce dsRNA intermediates. These are subsequently converted into sRNAs that associate with QDE-2 and target complementary sequences. Thus, at least some RdRPs may initiate gene silencing de novo.
In line with their diverse biological functions, individual RdRPs and their protein complexes isolated from various sources show different enzymatic properties in vitro. These include polymerase-specific preferences between ssRNA versus ssDNA templates and primer-dependent versus primer-independent initiation of RNA synthesis (33)(34)(35)(36)(37). Some RdRPs can also function as template-independent terminal transferases (34,35). Notably, two distinct template-dependent polymerization modes have been described for previously studied RdRPs: (a) processive synthesis of long double-stranded products, which is typically initiated at or close to the 3Ј-end of a singlestranded template using either primer-independent or socalled "back-priming" mechanisms, and (b) non-processive synthesis of sRNA copies initiated in a primer-independent manner at internal positions of a single-stranded template (33).
Different RdRPs appear to utilize the two modes with markedly different efficiencies. For example, Rdr1 from Schizosaccharomyces pombe or RDR6 from Arabidopsis efficiently synthesizes long products (34,38), whereas RRF-1 from Caenorhabditis elegans specializes in production of sRNAs (39). These biochemical differences are consistent with in vivo evidence; long dsRNAs generated by Rdr1, RDR2, and RDR6 must be processed by Dicer/RNAse III-like endoribonucleases to generate functional small interfering RNAs (siRNAs), whereas sRNA products of RRF-1 apparently do not require further processing for their secondary siRNA function (16,19,20).
Purified QDE-1 from N. crassa can use both primer-independent and "back-priming" mechanisms in vitro (33,35). The crystal structure of QDE-1 catalytic fragment, thus far the only known structure of a cell-encoded RdRP, suggests that this enzyme is a homodimer with the two subunits adopting either "closed" or "open" conformation (40). It has been proposed that the two structurally distinct conformations may help this remarkably versatile enzyme choose between different activities (40). However, in the absence of structural information for corresponding enzyme-substrate complexes, whether QDE-1 in fact remains a dimer upon template binding is unknown. Moreover, it remains to be seen whether other RdRPs can form homodimers, an important question, given that at least some RdRPs behave as monomers in solution (36).
On a more fundamental level, how new functional properties evolve in the RdRPs (and in other protein families for that matter) is poorly understood. Gene duplication followed by paralog divergence is a major driving force in protein evolution (41,42), and it clearly contributed to RdRP diversification. Indeed, many species encode more than one distinct RdRP, with three paralogous genes present in N. crassa (QDE-1, SAD-1, and RRP-3), four in C. elegans, and six in Arabidopsis (14). The last eukaryotic common ancestor might have contained three functionally distinct RdRPs giving rise to the ␣, ␤, and ␥ branches of the RdRP genealogy, an arrangement that was further modified by lineage-specific gene duplications and losses (43).
Species-specific members of individual paralogous groups, referred to as orthologs, are typically assumed to have similar biological activities (44). However, it has been alternatively proposed that divergence of orthologous sequences might frequently result in acquisition of novel functional properties (45). Until recently, it has been difficult to investigate these possibilities experimentally because genomes of just a few distantly related model organisms have been sequenced completely. Here we took advantage of the increasing number of wholegenome sequences available for fungal species and examined evolutionary trends in QDE-1 orthologs using phylogenetic, biochemical, and structural approaches.

Experimental Procedures
Phylogenetic Analyses-Amino acid sequences of fungal polymerases were downloaded from OrthoDB (46) and aligned using MUSCLE (47). Phylogenetic trees were constructed in MEGA6 (48) by computing evolutionary distances using Pois-son correction and inferring evolutionary history by the neighbor-joining method. Tree topology was tested using bootstrapping. Amino acid sequence conservation profiles were plotted in EMBOSS/plotcon using a 50-amino acid sliding window (49). Protein structures were color-coded according to interspecies conservation using Chimera (50). Chimera was also used to predict ligand positions in N. crassa QDE-1 apoenzyme (Protein Data Bank code 2J7N) (40) based on the known structure of the polymerase II elongation complex (Protein Data Bank code 1R9T) (51).
Protein Expression and Purification-Recombinant proteins were expressed and purified as described elsewhere (52,53).
Protein Thermostability Assay-Protein thermostability was determined by monitoring temperature-induced fluorescence changes, as described elsewhere (55). Purified proteins were incubated at 1 mg/ml in the gel filtration buffer with 1,000-fold diluted SYPRO Orange stock (Life Technologies) in 96-well PCR plates (Bio-Rad) sealed with optical sealing tape (Bio-Rad). Fluorescence was measured using an iCycler iQ5 real-time PCR detection system (Bio-Rad) with excitation and emission wavelengths set at 490 and 575 nm, respectively. The temperature was increased from 20 to 90°C with 1°C increments, and the mixture was incubated for 12 s at each temperature. Protein melting temperatures were calculated using the iQ5 Optical System software, version 2.1 (Bio-Rad).
RNA Polymerase Assays-A synthetic single-stranded DNA (ssDNA) template (5Ј-CTGACTGCTTCCTGTTTCTGTTT-TCTCTCCCCTCTTTTTCCTCATGTCCCACACCCCAAC-GGTCCCTTCATTTGTCTGTCTACCCTGTTGACAATTA-ATCATCGGCA-3Ј) was synthesized by Sigma. An ssRNA template corresponding to the sϩ transcript of bacteriophage 6 was produced from pLM659 (56) linearized with SmaI using the mMessage mMachine T7 transcription kit (Ambion). QDE-1 assays were carried out in 50 mM HEPES-NaOH, pH 7.8, 0.1 mM EDTA, 2% (v/v) Triton X-100, 100 mM NH 4 OAc, 2 mM MgCl 2 , 0.1 units/l recombinant RNasin (Promega), 0.4 mM each of ATP, CTP, and GTP, 0.2 mM UTP, and 25 Ci/ml [␣-32 P]UTP (PerkinElmer Life Sciences). Reactions were initiated by adding a corresponding QDE-1 polymerase to a final concentration of 0.25 g/l followed by 1-h incubations at 30 -65°C. Reaction products were separated as described (33) using native or denaturing formaldehyde-containing agarose gel electrophoresis or urea-containing PAGE and analyzed by phosphorimaging (Typhoon Trio, GE Healthcare). An RNase protection assay was carried out by stopping the polymerase reactions with 250 mM NH 4 OAc and 10 mM EDTA followed by incubation with various concentrations of RNase ONE (Promega) for 1 h at 30°C. The reaction products were analyzed by formaldehyde-containing agarose gel electrophoresis as described above. To quantify radioactivity incorporated into newly synthesized RNA products, reaction mixtures were spotted onto Whatman 3MM paper pretreated with 10% trichloroacetic acid (TCA) solution, washed in ice-cold 10% TCA solution for 10 min, rinsed twice with ice-cold 10% TCA solution and once with 95% ethanol, and allowed to air-dry. To determine the sum of incorporated and non-incorporated radioactivity, samples were spotted on Whatman 3MM paper and airdried with no additional treatment. Radioactive signals were then analyzed by phosphorimaging.
Protein Crystallization, Diffraction Data Collection, and Structure Determination-Crystals of the QDE-1 Tte catalytic fragment were obtained at 30°C by mixing 2 l of the protein solution at 2 mg/ml with 1 l of crystallization solution (100 mM Tris-HCl, pH 8.0, 75 mM NaCl, 10% (w/v) PEG 10,000, 5 mM MgCl 2 , and 6 mM spermine). Crystals were transferred to a drop containing the crystallization solution supplemented with 10% (v/v) glycerol and incubated at 12°C overnight. Before freezing, crystals were dehydrated at room temperature in two steps of 15 min each in the crystallization solution supplemented with 20 and 30% glycerol. X-ray diffraction data were collected to 3.19 Å resolution at 100 K at the PXIII beamline of the Swiss Light Source (Villingen, Switzerland) using a Pilatus 6M detector (Dectris). The crystal belonged to the P2 1 space group with the following unit cell dimensions: a ϭ 84.23 Å, b ϭ 165.84 Å, c ϭ 173.83 Å, and ␤ ϭ 90.10°. The data collection and structure refinement parameters are listed in Table 2. The structure was determined by molecular replacement using the known structure of the QDE-1 Ncr catalytic fragment (Protein Data Bank code 2J7N) (40) as a search probe. The model was built interactively using Coot (57), and the structure was refined using REFMAC from the CCP4 package, with tight non-crystallographic symmetry restraints between the four independent monomers (each monomer was considered as a group) with individual atom isotropic temperature factors and TLS refinement (58). Each monomer of the QDE-1 Tte fragment used for crystallization contains 1,034 residues, including the His tag and the tobacco etch virus protease cleavage site. Of these, 924/ 922 could be built in dimer A/B and 922/922 in dimer C/D in the two non-crystallographic dimers of the asymmetric unit ( Table 2). Missing residues in the model belong to the N and C termini and flexible loops ( Table 2). Because QDE-1 Tte dimer A/B is better ordered than dimer C/D in the electron density map, we use it for subsequent comparisons. Electron Microscopy (EM)-Purified QDE-1 Tte was diluted to 10 g/ml with 100 mM Tris-HCl, pH 7.5, 75 mM NaCl, 5 mM MgCl 2 , and 5% (v/v) glycerol. A volume of 4 l of protein sample was applied to a glow-discharged carbon-coated transmission EM grid and stained with 2% (v/v) uranyl acetate. RdRP-ssDNA substrate complex was formed by incubating 1.3 M QDE-1 Tte with 1.7 M (i.e. ϳ10-fold higher than the apparent K d determined in the filter binding assay) DNA oligonucleotide 5Ј-CCTTAATTGTATAG-3Ј before transferring acutely diluted mixture to a grid and subsequently staining with uranyl acetate as above. Images were recorded at a magnification of ϫ66,350 using a FEI T12 transmission electron microscope equipped with a 4K CCD camera (FEI) under low dose conditions. Single particles were selected and processed with the EMAN2 image-processing package (59). Initially, a total of 3,000 particles each of the QDE-1 Tte apoenzyme and of QDE-1 Tte preincubated with ssDNA were used for two-dimensional classifications and three-dimensional reconstructions. For both data sets, 10 initial models were generated with EMAN2 and compared with the 40 Å resolution filtered crystal structure of the QDE-1 Tte dimer. The initial model structurally most similar to the crystal structure was selected for further refinement for both data sets to obtain a three-dimensional EM map at about 31 Å resolution without imposing any symmetry constraint. To obtain a higher resolution map of QDE-1 Tte preincubated with ssDNA, a data set of 11,992 particles was used and processed as described above, which led to a "reference-free" three-dimensional EM map at a resolution of 20 Å. In parallel, the same data set for QDE-1 Tte preincubated with ssDNA particles was processed as described above but with the crystal structure of apo-QDE-1 Tte as a reference.

Orthologs of N. crassa QDE-1 Belong to a Rapidly Evolving
Protein Group-To gain insights into RdRP evolution, we analyzed corresponding genes from 40 taxonomically diverse fungi with completely sequenced genomes (Table 3 and Fig. 1). Most species encoded RdRP proteins clustering with N. crassa QDE-1, SAD-1, or RRP-3, except for three species from the Eurotiales and Hypocreales orders that had QDE-1 and SAD-1 but not RRP-3. S. pombe and Schizosaccharomyces japonicus had a single RdRP (Rdr1) related to SAD-1, and Saccharomyces cerevisiae had no RdRPs, as expected. Within each of the three groups, RdRPs clustered according to their taxonomic origin (Fig. 1). This topology suggested that the last common ancestor of fungi might have contained at least three RdRP paralogs that evolved as orthologously related lineages or were occasionally lost during speciation.
Notably, the QDE-1 orthologs tended to be connected by longer branches than their SAD-1 and RRP-3 counterparts (Fig. 1). To identify regions accounting for their apparently accelerated evolution, we examined a QDE-1 conservation plot (Fig.  2). The non-catalytic N-terminal part showed extremely low conservation scores, as pointed out previously (33). However, even within the generally less divergent C-terminal part, a prominent peak of sequence conservation was detected only in the vicinity of the catalytic DYDGD motif (14,33). This contrasted with the SAD-1 and RRP-3 plots that contained substantially broader regions of relatively high conservation. Conservation was even more uniform for two polymerase II subunits, RPB1 and RPB2, distantly related to cell-encoded RdRPs (40) (Fig. 2A). Quantitative analyses of amino acid substitution rates showed that QDE-1 was significantly more divergent than SAD-1, RRP-3, RPB1, and RPB2 (Fig. 2B). Interestingly, conserved amino acid residues showed prominent clustering around the active center of the known crystal structure of the QDE-1 Ncr apoenzyme interacting with one of the two catalytic Mg 2ϩ ions (Mg 2ϩ A ) (Protein Data Bank code 2J7N) (40) (Figs. 2C and 3). We modeled positions of other molecules participating in RNA polymerization, including an incoming ATP monomer, the second Mg 2ϩ ion (Mg 2ϩ B ), the template, and the RNA product based on the structure of the polymerase II elongation complex (Protein Data Bank code 1R9T) (51). This placed the ATP and Mg 2ϩ B near the conserved surface of the nucleotide pore and the nascent 3Ј-end of the RNA product along with the corresponding template nucleotides near the conserved DYDGD loop and its Mg 2ϩ A ligand (Figs. 2C and 3). On the other hand, a more distal segment of the template-product duplex egressing from the active center was surrounded by substantially more divergent QDE-1 surfaces (Figs. 2C and 3). Notably, all contacts made by the template and the nascent RNA with the RPB1 and RPB2 subunits of the polymerase II complex appear highly conserved in evolution (Fig.  2C). We concluded that the QDE-1 group is generally more divergent than its RdRP paralogs and DdRP relatives and that it shows unusually strong sequence variability outside of the active center and the NTP-interacting surfaces.
QDE-1 Orthologs Generate Markedly Different Combinations of Long and Short RNA Products-To examine whether divergent QDE-1 proteins had distinct functional properties, we purified catalytic fragments of QDE-1 Ncr and its ortholog from Thielavia terrestris (QDE-1 Tte ) from the Chaetomiaceae family distantly related to N. crassa (Sordariaceae family) using a standardized protocol (see "Experimental Procedures") and analyzed the RNA polymerase activity of these two proteins (Figs. 4 and 5). QDE-1 Ncr is known to accept either ssRNA or ssDNA templates and generate their continuous end-to-end RNA cop-   , an incoming ATP, a template, and an RNA product are also indicated. Note that highly conserved amino acid residues cluster in the vicinity of the catalytic center and the NTP pore. . The reaction products were separated by native agarose gel electrophoresis and visualized using ethidium bromide staining or 32 P phosphorimaging, as indicated. Note that QDE-1 Tte is substantially more efficient than QDE-1 Ncr in generating sRNA products that migrate either at a low molecular weight position or in a template base-paired form. C, RNA polymerase activities of QDE-1 Ncr and QDE-1 Tte were assayed at 45°C in the presence of a ssRNA template and analyzed by denaturing agarose gel electrophoresis. D, RNA products from C were incubated with increasing concentrations of RNase ONE or RNase ONE reaction buffer, as specified under "Experimental Procedures." Positions of the 1ϫ full template-length products of processive end-to-end polymerization initiated in a primer-independent manner and sRNA products of non-processive polymerization are indicated on the right. Also shown is an expected position of "back-primed" 2ϫ template-length products, which QDE-1 Ncr can generate for some but not all ssRNA templates (33).
Similar to its N. crassa ortholog, the newly analyzed QDE-1 Tte was a highly active RNA polymerase (Fig. 4A). However, it showed a striking bias toward generating short RNA products rather than full-length copies (Fig. 4A). This activity produced readily detectable amounts of template-product duplexes migrating slower than the ssRNA template but faster than the corresponding full-length dsRNA on EtBr-stained gels (Fig.  4A). Gel autoradiography additionally revealed a prominent low molecular weight band probably corresponding to short RNA copies released from the template (Fig. 4A). A similarly migrating band of short RNAs was also a major reaction product in the ssDNA-programmed reactions containing QDE-1 Tte but not QDE-1 Ncr (Fig. 4B). On the other hand, both enzymes gave rise to full-length DNA-RNA template-product duplexes with comparable efficiencies (Fig. 4B).
To better understand the nature of RNA products, we analyzed ssRNA-programmed reactions by denaturing agarose gel electrophoresis (Fig. 4C). Under these conditions, short RNA copies quantitatively dissociated from the template and migrated at the expected low molecular weight position. This analysis additionally revealed RNA products of 1ϫ template length, thus suggesting that, under conditions used in our RdRP assays, both QDE-1 Ncr and QDE-1 Tte can initiate end-to-end RNA synthesis in a predominantly primer-independent manner (33,35) (Fig. 4C). Consistent with the native gel analyses, the ratio between short and long RNA products was noticeably higher for QDE-1 Tte than for QDE-1 Ncr (Fig. 4C). Importantly, the bias for short RNA products was consistently detected for several independently purified batches of QDE-1 Tte and observed over a wide temperature range (Fig. 4, A and B). This ruled out the trivial explanation that the enzymatic differences between QDE-1 Tte and QDE-1 Ncr were related to a higher growth temperature limit of T. terrestris compared with N. crassa (60 -62). In fact, the thermal stability of the QDE-1 Tte protein exceeded that of QDE-1 Ncr by only 3.7°C (T m ϭ 54.0 Ϯ 0.0°C versus T m ϭ 50.3 Ϯ 0.6°C; Fig. 5B). The two polymerases also had comparable single-stranded template binding properties (Fig. 5, C and D). Thus, distinct QDE-1 orthologs may have markedly different RNA polymerization properties.
Evolutionary Divergence between QDE-1 Family Members Modulates Their Product Preferences-To ensure that functional differences between QDE-1 Ncr and QDE-1 Tte resulted from evolutionary innovation, we purified a catalytic fragment of QDE-1 from Myceliophthora thermophila, a Chaetomiaceae fungus more closely related to T. terrestris than to N. crassa (Figs. 1 and 5). The T m of QDE-1 Mth was comparable with that of QDE-1 Ncr and QDE-1 Tte (T m ϭ 52.3 Ϯ 0.6°C versus T m ϭ 50.3 Ϯ 0.6°C and T m ϭ 54.0 Ϯ 0.0°C, respectively; Fig. 5). To compare enzymatic properties of QDE-1 Ncr , QDE-1 Tte , and QDE-1 Mth , we incubated the three polymerases with either ssRNA or ssDNA template for 1 h at 45°C and separated the reaction products by urea-containing PAGE, affording simultaneous detection of long RNAs migrating at the top of the lane and sRNA products visualized at single-nucleotide resolution (Fig. 6). In ssRNA-programmed reactions, QDE-1 Mth polymerase was clearly more efficient in producing short RNA products than QDE-1 Ncr , albeit not to the same extent as QDE-1 Tte (Fig.  6A). Moreover, QDE-1 Ncr synthesized detectable amounts of Յ23-nt and ϳ28 -31-nt products but virtually no 24 -27-ntlong sRNAs (Fig. 6A). On the other hand, both QDE-1 Tte and QDE-1 Mth efficiently produced 24 -27-nt sRNA in addition to shorter products (Fig. 6A). Interestingly, each of the three polymerases generated a unique blend of sRNA products in ssDNA-programmed reactions (Fig. 6B). As an additional control, we purified a D607A mutant QDE-1 Mth , where the first Asp residue of the DYDGD motif was mutated to Ala. As expected, the mutant protein lacked detectable nucleotidyltransferase activity (Fig. 7). We concluded that sequence diver-gence between QDE-1 orthologs appears to underlie differences in their activities.
QDE-1 Tte Is Structurally Similar to QDE-1 Ncr -We next wondered whether distinct functional properties of QDE-1 orthologs might be due to major differences in spatial structures of these enzymes. This appeared plausible, given the degree of amino acid sequence divergence outside of the active center (see Fig. 2). To this end, we determined QDE-1 Tte threedimensional structure using x-ray crystallography. A 3.19 Å resolution trace of the QDE-1 Tte polypeptide chain revealed two closely similar QDE-1 Ncr -like homodimers in the asymmetric unit (root mean square deviation of 1.06 Å between dimer A/B and C/D for 917 ␣-carbon atoms). In each dimer, the two monomers are related by a non-crystallographic dyad (Fig. 8A).
Other structural details of QDE-1 Tte also resembled those of QDE-1 Ncr (40). For example, the QDE-1 Tte dimer had a pyramidal shape with a pair of double-␤-barrel domains (DPBB1 and DPBB2; residues 312-413 and 535-639) located at the base of each subunit (Fig. 8, A and B). The "head" (residues 457-507 and 822-1006 of the catalytic fragment) and the "neck" domains (residues 428 -532 and 784 -821) of the two subunits adopted a slightly tilted back-to-back position (Fig.  8A), and the "slab" domain (residues 1-244) of each subunit protruded from the catalytic domain toward the corresponding head domain. The five domains formed an extensive groove, probably accommodating the template and the nascent RNA product, with the catalytic site located at the bottom (Fig. 8, A and B). In general, C␣ atoms in the QDE-1 Ncr and QDE-1 Tte dimers superimposed very well with a root mean square deviation score of 1.34 Å for 917 ␣-carbon atoms (Fig. 8C).
As found previously for QDE-1 Ncr , one of the two QDE-1 Tte subunits adopted a slightly more closed conformation than the other. Superposition of 922 C␣ atoms between the two monomers (dimer A/B) using the SSM server returned a root mean square deviation of 2.15 Å (Fig. 8D). The largest conformational . QDE-1Tte crystal structure. A, QDE-1 Tte is an asymmetric dimer with each subunit containing a DPBB1 (shown in cyan for subunit A), a DPBB2 (green), a "head" (magenta), a "neck" (blue), and a "slab" domain (red). The entire B subunit is colored in gray. B, magnified view of the DPBB1 and DPBB2 domains with the three catalytic aspartate side chains shown as balls and sticks. C, structural alignment between QDE-1 Tte and previously solved QDE-1 Ncr structure showing a considerable overlap between the two proteins. D, structural alignment between subunits A and B QDE-1 Tte demonstrating that subunit A adopts a more "closed" conformation than subunit B. The two subunits are color-coded as in A.
differences between the two monomers occurred closer to the C-terminal end of the polypeptide chain. The homodimer interface is stabilized by a total of 119 and 115 residues of monomer A and B, respectively (63). Most residues at the interface originate from the upper "neck" and "head" domains, where the subunits display the largest structural differences (Fig. 8C). These data suggest that QDE-1 Ncr and QDE-1 Tte have remarkably similar tertiary and quaternary structures despite their primary sequence differences. QDE-1 Tte Can Function as a Dimer in Solution-Extensive contacts between the A and B subunits observed in the QDE-1 Ncr and QDE-1 Tte crystals suggested that these proteins may form catalytically active dimers in solution with the two monomers oscillating between the closed and open conformations. To test this possibility directly, we used EM to compare nega-tively stained images of the QDE-1 Tte apoenzyme and its complex with a ssDNA template (Fig. 9). using reference-free twodimensional class averaging (Fig. 9B). These initial EM reconstructions carried out using 3,000 apoenzyme and 3,000 ssDNA bound particles revealed largely similar pyramidshaped homodimer structures, each containing two quasi-symmetric grooves and showing at this resolution a generally good fit with the QDE-1 Tte crystal structure obtained without bound DNA (Fig. 9C). As in the crystal structure, one of the two template-product grooves in the QDE-1 Tte dimer adopted a more open conformation in the EM reconstructions (Fig. 9C). To obtain a better resolved solution structure of QDE-1 Tte in the presence of ssDNA, we extended our EM data analysis to include a total of 11,992 particles. This allowed us to visualize this complex to a resolution of 20 Å (Fig. 10, A and B). As  A negatively stained electron micrograph shows particles of QDE-1 Tte dimers in complex with an ssDNA template. B, representative reference-free two-dimensional class averages of the particles in A. C, overlay of QDE-1 Tte structures determined by reference-free EM reconstruction in the presence (orange) or absence of a ssDNA template (gray). The crystal structure of the unbound QDE-1 Tte dimer is overlaid, with its subunits colored in cyan and gray, respectively. Note that at this resolution of about 31 Å (1/0.032 Å Ϫ1 shown by the Fourier shell correlation map as an inset, both the apoenzyme and the ssDNA complex appear as structurally similar dimers. expected, the structural fit between the EM and crystal structures of QDE-1 Tte further improved when the EM map was prepared using the crystal structure as a reference (Fig. 10B). Of note, even this improved resolution was not sufficiently high to unambiguously assign density for the ssDNA. Likewise, understanding of the subtle conformational changes in the QDE-1 Tte dimer upon template binding will require a higher resolution structure of the RdRP-ssDNA complex, a challenge that will be addressed in the future. Nonetheless, our present data strongly suggest that QDE-1 orthologs function as dimers in solution.

Discussion
This study suggests that fungal QDE-1 orthologs underwent an unusually rapid sequence divergence outside of the catalytic center and the nucleotide-binding site. This corresponds to a significantly higher amino acid substitution rate in this group compared with its RdRP paralogs and distant DdRP relatives (Fig. 2). Consistent with their stronger divergence from QDE-1 Ncr than from each other (Fig. 1), QDE-1 Tte and QDE-1 Mth differ from QDE-1 Ncr in their enhanced ability to generate sRNA copies and reduced ability to synthesize long RNA products (Fig. 4). Because QDE-1 Ncr , QDE-1 Tte , and QDE-1 Mth originate from a single taxonomic order (Sordariales), these data argue that protein functions can undergo substantial changes over relatively short periods of evolutionary history. Notably, this effect was not due to a recent gene duplication event followed by functional specialization of the newly emerged paralogs because our BLAST searches did not reveal any additional RdRP gene in  the N. crassa, T. terrestris, and M. thermophile genomes besides QDE-1, SAD-1, and RRP-3.
What could be the biological significance of such accelerated functional evolution? One possible answer relates to the role of RNA silencing in cellular defense against viruses and transposable elements (3,5,6). Different species encounter distinct sets of pathogens, which necessitate corresponding changes in cellular defense mechanisms (64). Interestingly, N. crassa is a mesophilic fungus colonizing burnt vegetation and occasionally living plants (60,61), whereas T. terrestris and M. thermophila typically inhabit self-heating composts (62). Exposure to diverse environments and correspondingly different sets of biohazards might have exerted substantial evolutionary pressure on cellular immunity factors, including QDE-1. An interesting direction for future studies would be to test whether this also rewired RNA silencing pathways in a more fundamental manner (e.g. incorporating sRNA products of QDE-1 Tte and QDE-1 Mth into signal amplification loops similar to the secondary siRNA pathway in C. elegans (16, 19 -21, 39)).
The striking structural similarity between QDE-1 Tte and QDE-1 Ncr uncovered in this work (Fig. 8) suggests that evolution in this group probably proceeded through alteration of functionally important surface residues rather than major changes in the protein fold. In other words, QDE-1 orthologs show a combination of structural robustness and functional innovability (65). Although further work will be required to identify specific structural elements underlying functional divergence in the QDE-1 group, these will probably include amino acid residues directly interacting with the templateproduct duplex or affecting protein flexibility. Indeed, structural alignment of QDE-1 Ncr or QDE-1 Tte with the yeast polymerase II elongation complex (Protein Data Bank code 1R9T) (51) using structural similarity between the QDE-1 DPBB2 domain and the only DPBB domain of the polymerase II RPB1 subunit suggests that the QDE-1 enzyme may have to undergo major conformational changes to allow egress of templateproduct duplexes longer than ϳ10 base pairs.
The EM reconstructions presented here provide an unprecedented insight into solution structure of an RdRP enzyme (Figs. 9 and 10). In line with the earlier prediction (40), QDE-1 assembles into a pyramid-shaped homodimer with each of the two subunits containing a groove suitable for template-product binding. One of the two subunits in the EM density maps adopts a more "closed" conformation than the other. This supports the "two-stroke motor" model of QDE-1 activity that was proposed earlier based on the x-ray structure of QDE-1 Ncr apoenzyme (40). Importantly, our data indicate that QDE-1 retains its dimeric form in the presence of a single-stranded DNA template ( Figs. 9 and 10).
We finally note that the efficient synthesis of sRNA products by the newly isolated QDE-1 enzymes might facilitate a range of research and diagnostic applications. QDE-1 Tte appears to generate more than one copy of a given template sequence (Fig. 4, A  and B). Therefore, this enzyme might be especially useful for amplifying either an entire nucleic acid target or its parts accessible to the polymerase. Combined with deep sequencing technology, this may open up new possibilities in high-throughput analyses of the transcriptome composition, RNA conformation and ribonucleoprotein complex structure. Enzymatic properties of QDE-1 Tte could be further improved by knowledgebased mutagenesis of its evolutionarily variable parts or by "shuffling" corresponding sequences with their counterparts from other QDE-1 orthologs.
In conclusion, our study argues that acquisition of novel enzymatic properties through divergence of orthologous sequences could be a more common evolutionary scenario than anticipated previously. This work also improves our understanding of molecular mechanisms underlying RdRP functions and expands the existing molecular biology toolkit. We predict that further comparative analyses of this remarkably diverse class of enzymes will be a rewarding experience for evolutionary biologists and biochemists alike.