Conserved strategies of RNA polymerase I hibernation and activation

RNA polymerase (Pol) I transcribes the ribosomal RNA precursor in all eukaryotes. The mechanisms ‘activation by cleft contraction’ and ‘hibernation by dimerization’ are unique to the regulation of this enzyme, but structure-function analysis is limited to baker’s yeast. To understand whether regulation by such strategies is specific to this model organism or conserved among species, we solve three cryo-EM structures of Pol I from Schizosaccharomyces pombe in different functional states. Comparative analysis of structural models derived from high-resolution reconstructions shows that activation is accomplished by a conserved contraction of the active center cleft. In contrast to current beliefs, we find that dimerization of the S. pombe polymerase is also possible. This dimerization is achieved independent of the ‘connector’ domain but relies on two previously undescribed interfaces. Our analyses highlight the divergent nature of Pol I transcription systems from their counterparts and suggest conservation of regulatory mechanisms among organisms.

T ranscription of the ribosomal RNA (rRNA) precursor by RNA polymerase (Pol) I is a prerequisite for the biosynthesis of ribosomes in all known eukaryotes 1 . To allow transcription initiation in baker's yeast Saccharomyces cerevisiae (Sc), monomeric Pol I is bound by the initiation factor Rrn3, enabling recruitment to the rDNA promoter via the heterotrimeric core factor (CF), TBP and upstream activating factor (UAF) [2][3][4] . Structures solved from Sc Pol I crystals revealed inactive polymerase dimers with widely expanded active center clefts in three similar conformations [5][6][7] , matching biochemical observations in extracts 8 , initial and recent Electron Microscopy studies 9,10 . Such dimerization relies on a 'connector' domain at the C-terminus of Pol I subunit A43. In detail, the 'stalk' subunit complex of one monomer is inserted into the 'cleft' of the other monomer from the upstream side. This allows formation of a connector α-helix in subunit A43 which attaches to the clamp core helices of subunit A190 in the second monomer. Furthermore, a C-terminal connector β-hairpin traps the lid of subunit A190, completely inactivating both polymerases 5,6 . Activation then requires Pol I monomerization, cleft contraction and stabilization of monomers by Rrn3. Further cleft contraction takes place upon promoter melting and Pol I interaction with the DNA/RNA hybrid [11][12][13] .
Structures of monomeric states and of actively elongating Pol I from Sc were solved by single particle Electron Cryo Microscopy (cryo-EM) 14,15 . Furthermore, differential fluorescent allele tagging demonstrated that Pol I dimerization is associated with inactivation (hibernation) under certain starvation conditions in an elegant in vivo approach 16 . In such experiments, deletion of the connector resulted in dimer disruption. To date, it is unclear which mechanisms and structural features of Pol I regulation are specific to Sc and which are conserved among organisms 17 .
Here, we use single particle cryo-EM to solve the structure of Pol I from Schizosaccharomyces pombe (Sp) in a monomeric (apo) and an actively elongating form at high resolution, showing that Pol I cleft contraction upon transcription activation is common to both organisms. Strikingly, we also uncover that Pol I dimerization can take place independent of the A43 connector domain in vitro and solve the cryo-EM reconstruction of such a dimer. Our results allow discussing the evolutionary conservation of structural features, hibernation and activation mechanisms.

Results
Preparation and cryo-EM of Schizosaccharomyces pombe Pol I. To study the structure of its Pol I, we generated an Sp strain carrying a C-terminal flag-his tag on subunit AC40. Purification using established protocols 5 yielded pure Pol I that shows protein bands for all 14 subunits ( Supplementary Fig. 1). Sp Pol I is active in elongation and cleavage of an RNA primer from synthetic constructs in vitro similar to Sc Pol I ( Supplementary Fig. 2). To determine the structure of a monomer, we stabilized Pol I by crosslinking with BS3, followed by quenching and size exclusion chromatography (Methods). In an independent experiment, we established an Sp Pol I elongation complex (EC) in vitro similar to its Sc counterpart 15 . Using three locked nucleic acid (LNA) bases at the 3′ end of the RNA primer reduced sample heterogeneity by preventing transcript cleavage via subunit A12.2 (Methods). Two cryo-EM datasets were collected on a Titan Krios Electron Microscope equipped with Falcon III direct electron detector: one from non-crosslinked, LNA-containing EC particles and one from BS3-crosslinked Pol I, both following size exclusion chromatography (Supplementary Table 1).
Pol I architecture is conserved but sub-complex occupancy varies among organisms. The cryo-EM densities reveal secondary structure elements for the entire polymerase and side chain orientations in many regions ( Fig. 1 Table 2). Both, EC and monomeric Pol I, lack cryo-EM density for the A49/A34.5 heterodimer, indicating that the subcomplex is flexible or was lost during grid preparation (Fig. 1). The presence of the A49/A34.5 subunit-complex in Pol I preparations was confirmed by Coomassie−stained SDS-PAGE ( Supplementary Fig. 1a) and mass-spectrometry analysis.
Differences between Sc and Sp Pol I include the lack of density for the Pol-I-specific helix α0 in subunit Rpb6 and insertions in the 'Dock' and 'Foot' domains of Sp subunit A190. Helix α0 forms only in Sc Pol I and was presumed to strengthen stalk attachment compared to Pol II 5,6 . A divergent region (residues 549-559) and an insertion (residues 591-600) in the 'external 2' domain of Sp subunit A135 are adjacent to the lobe-region responsible for tight association of the A49/34.5 sub-complex 18 and may be responsible for its reduced affinity (compare Supplementary Figs. 3b and 4). Poor stalk density is observed, indicating a high degree of flexibility, similar to human Pol II 19 and potentially linked to a lack of Rpb6 α0 formation. Nevertheless, fitting of a stalk homology model created from the crystal structure of the Sc A14/A43 complex 20 was possible. In addition, local resolution is reduced in the jaw regions and the tip of the clamp domain of subunit A190 in Sp Pol I EC and monomer reconstructions, also indicating a high degree of flexibility (Supplementary Figs. 1 and 2). The C-terminal domain of subunit A12.2 is mostly flexible in ECs and monomers, although low-resolution density in the Sp EC ( Supplementary Fig. 3a) may indicate a position outside subunit A135 as observed in some Sc ECs 21 .
Sp Pol I contains an expander element that is flexible in ECs. In ECs, we observe density for DNA and RNA, allowing us to build and refine a model for 25 template bases, 11 non-template bases and 7 RNA bases (Fig. 1d). Interactions with Pol I, positioning in the cleft and relative orientation of the scaffold are similar to Sc structures 14,15,21 . We do, however, observe a density that we attribute to the 'expander' ('DNA-mimicking') element in Pol I monomers (Fig. 1b). This density is located between the 'protrusion' domain of subunit A135 and the 'clamp core' region in subunit A190 on the upstream region of the active center cleft and appears to prevent unspecific DNA binding. Due to flexible connections to the 'jaw' domain of subunit A190, we refrained from modeling the exact residues of the section in the Sp monomer. Location of presumed expander density in the cleft is comparable to that observed in Sc Pol I dimers ( Supplementary  Fig. 3d), while the element is flexible in Sc Pol I monomers.
Sp Pol I dimerizes in vitro independent of the 'connector' domain. Surprisingly, we observed many dimer-particles in the monomer dataset ( Supplementary Fig. 1). As described above, Pol I subunit A43 does not contain a connector element in Sp and therefore should be unable to form dimers, as inferred from connector deletions in Sc 16 . Previous studies described that Sc Pol I dimers and monomers are in equilibrium in solution. Buffer conditions 8 and concentration of highly purified Pol I 5 can shift this equilibrium in vitro. To test whether dimerization is possible in solution or may result from chemical crosslinking, we carried out analytical size exclusion chromatography (SEC) in different buffers and in-solution dynamic light scattering (DLS) at different concentrations of Sp Pol I. A shift to earlier SEC elution volumes indicates an increased particle size under high-salt conditions (Fig. 2a). Negative stain EM analysis of SEC peak fractions revealed the presence of dimeric particles independent of the buffer ( Supplementary Fig. 5). In a high-salt buffer, about 32% of identified particles (49% of Pol I molecules) appear to be in a dimeric state as indicated by unsupervised 2D-classification. In contrast, dimer-prevalence is below 2% of particles in low-salt conditions. In line with this, DLS using the Prometheus Panta technology (Nanotemper) indicates an increase of hydrodynamic particle radius with increasing Sp Pol I concentration (Fig. 2b). Taken together, we concluded that reversible, connectorindependent dimerization of Sp Pol I is possible. To understand the underlying molecular principles, we reconstructed the dimer architecture from its cryo-EM density.
Inactive Sp Pol I dimers arrange in a divergent architecture. A three-dimensional reconstruction of Sp Pol I dimers at 4.5 Å resolution was obtained in C2 symmetry ( Fig. 3 Fig. 1 Structure of monomeric and elongating S. pombe Pol I. a Sub-domain architecture of the Pol I subunits A190 and A135 in Sp. Insertion/Deletion of more than five residues compared to Sc are highlighted. b Cryo-EM reconstruction of monomeric Sp Pol I. Transparent, unsharpened cryo-EM density overlaid with ribbon model shows 12 subunits but lacks the A49/A34.5 subcomplex. The cleft is expanded and density (purple, space-filling) indicates the location of the 'expander' element. c Cryo-EM reconstruction of an Sp Pol I Elongation complex. The cleft is contracted and the expander is displaced by the hybrid. d Schematic representation of the artificial bubble construct used to establish an EC. Nucleotide bases included in the EC-density are highlighted. A model of the template-DNA/product-RNA hybrid overlaid with sharpened EC density is shown, the 3′ nucleotide of the RNA is present but shows poor density and was thus not modeled. and the Pol I specific helix α12a) with the wall subdomain of subunit A135, subunit AC40 and the common subunit Rpb12 of the second monomer. These two interfaces allow a tight association via the establishment of contacts between both Pol I upstream-faces independent of the connector element but utilizing organism-specific regions. Low-resolution density on the lobe of subunit A135 in both molecules in the dimers may indicate the presence of A49/A34.5 ( Supplementary Fig. 6). This is in line with the suggestion that dissociation of this subcomplex is independent of Pol I inactivation in Sp 22 .
Cleft contraction is common to active Pol I. Comparative structural modeling reveals that, similar to Sc Pol I reconstructions, the cleft is extended in monomers and contracted in ECs ( Fig. 4; Supplementary Movie 1). Consequently, the 'bridge' helix of subunit A190 is mostly disordered in monomers (poor density despite higher overall resolution) and forms a well-structured helix in ECs ( Supplementary Fig. 3e). The

Discussion
In this work, we described three single particle cryo-EM reconstructions of Sp Pol I, representing the only structures of this enzyme from an organism other than Sc to date. Availability of a second in vitro transcription system allows cross-validation of structural and functional investigations in future studies. Whereas we find that the general architecture of Pol I is conserved, structural details vary among organisms.
In contrast to Sc, density for the expander element was observed in Sp Pol I monomers, hinting at an additional possibility to prevent unspecific DNA-binding to Pol I monomers. Absence of cryo-EM density for the A49/A34.5 heterodimer may result from a loss of the subcomplex due to stress on the airbuffer-interface during freezing 28 , a flexible association in Sp, or may indicate functional relevance. The A49/A34.5 subunitcomplex plays an important role in Pol I initiation by stabilizing the non-template strand during promoterDNA melting and may support promoter escape in Sc 12,29 . The subcomplex shows similarities to initiation factors TFIIF and TFIIE in the Pol II transcription system, as suggested from homology to crystal structures and native mass spectrometry analysis 30 . However, the A49/A34.5 subcomplex has also been described as important for Pol I elongation 20,31,32 and is constantly attached to the Pol I core throughout elongation in vivo 33 . Dissociation of the subcomplex from Sc Pol I was observed under specific experimental conditions in vitro 20 or in ECs established using a nucleotide analogon 21 . Notably, the EC scaffold used in this study contains modified RNA which may lead to the loss of A49/A34.5 21 , even though this was not the case in an Sc Pol I EC structure determined with the identical scaffold 15 . Taken together, it can be speculated that the A49/A34.5 subcomplex not only carries out multiple functions in Pol I initiation and elongation (as recently reviewed 34 ), but also that these roles may vary in importance a b Fig. 2 Sp Pol I can dimerize in vitro dependent on protein concentration and buffer condition. a Analytical Size Exclusion Chromatography (SEC) of Sp Pol I in buffers containing 1500 mM (S1500; dotted), 600 mM (S600; dashed) or 300 mM (S300; continuous) Potassium Acetate. The main peak (A 280nm ) elutes at higher apparent molecular weight in the high salt buffer. The average elution peaks of reference proteins are indicated: black square -Tyroglobulin; Black diamondß-Amylase; Black triangle -Albumin; Black dot -Carboanhydrase (exact retention volumes at the bottom). Compare Sup. Fig. 5 for negative-stain EM characterization of S300 and S1500 peak fractions and methods for details. b Dynamic Light Scattering (DLS) analysis of Sp Pol I particles (S600 buffer) measured using the Panta technology (Methods). Each mark represents an individual measurement (n = 80 for each Pol I concentration). At higher Pol I concentration, the hydrodynamic radius of particles in solution increases, indicating a shift of equilibrium towards dimerization. The boxes show the median (horizontal line inside box) and span from the 25 th to the 75 th percentiles, the whiskers reach up to the 90 th and down to the 10 th percentile, measurements beyond these are shown as individual points (* one outlier at 32.72 nm is not displayed; ** two outliers at 19.98 nm and 21.04 nm are not displayed).
Comparison of Pol I cleft expansion states showed that transcription activation by active center cleft contraction is conserved in Sc and Sp. This contraction distinguishes the Pol I transcription system from its Pol II/III counterparts 2,36 . Strikingly, well-defined dimers are present in Sp Pol I preparations that form independent of the A43 connector domain, which is required for dimerization in Sc 16 . Dimer prevalence in vitro is influenced by buffer conditions and Pol I concentration in a dynamic equilibrium in both organisms and not an artefact of chemical crosslinking or crystallization. We find that different organism-specific regions are involved in formation of inactive dimers in both, Sp and Sc Pol I, occluding the binding site for Rrn3 in both organisms. Hence, conserved Rrn3-association 23,37,38 can likely stabilize monomers as required for activation 39 . Regulated binding of Rrn3 40-43 then allows promoter recruitment and DNA-melting to take place 29,44 .
Apparently, dimerization of Pol I molecules is possible but involves divergent primary and secondary structure elements. Dimers adopt a variable architecture, while their formation still results in a hibernating state. In vivo relevance of these findings and the reason for the dynamic association of the A49/A34.5related subunits are still under debate awaiting further investigation. Nevertheless, we conclude that the principles underlying Pol I regulation 'activation by contraction' and 'hibernation by dimerization' are conserved among organisms.

Methods
Construction of AC40-tagged S. pombe strain. A construct for genomic insertion of a 10xHis/Flag tag was ordered as plasmid (Gene Art). The construct was amplified and genomically inserted into the haploid S. pombe strain 972h-: A 100 ml YPD culture was started at optical density (OD 600 ) of 0.25 from an over night culture at 30°C. After 5-6 h, OD 600 was at 1.0 and cells were harvested in 250 ml conical tubes (1361 g, 5 min). Cells were resuspended in 25 ml sterile water by vortexing and again centrifuged. Cells were resuspended in 1 ml of sterile 100 mM     Li 2 Ac solution. The suspension was then transferred into a 1.5 ml reaction tube and centrifuged for 15 s (tabletop centrifuge, full speed). Supernatant was removed and the pellet resuspended in 400 µl of fresh 100 mM Li 2 Ac. In parallel, 500 µl of salmon sperm DNA (2 mg/ml) were boiled at 95°C for 5 min and quickly chilled on ice. The cells were then split into 100 µl aliquots, pelleted and the supernatant removed. To a pellet, the following transformation mix was added in the following order: (1) 240 µl sterile PEG3350 (50% w/v), (2) 36 µl 1 M Li 2 Ac, (3) 50 µl salmon sperm DNA (2 mg/ml), and (4) 34 µl PCR product of the insertion construct. Tubes were vigorously vortexed for more than 1 min and incubated at 30°C for 30 min under shaking. Subsequently, reactions were transferred to 42°C and incubated under shaking for 25 min. Cells were then pelleted (tabletop centrifuge at 6000 g for 15 s), the supernatant removed and cells were resuspended in 1 ml YPD medium. Cells were transferred into 15 ml conical tubes and shaken at 30°C for 3 h. After centrifugation at 1361 g for 5 min, the pellet was resuspended in 500 µl sterile water and plated on YPD plates with Kanamycin/G418. The plates were incubated at 30°C for 3-4 days, single colonies picked and re-plated on fresh plates. For verification of correct genomic insertion, the respective regions were amplified by PCR and sequenced. RNA elongation and cleavage assays. Purified Sc or Sp Pol I (1, 0.5 or 0.25 pmol) was pre-incubated with 0.25 pmol of pre-annealed minimal nucleic acid scaffold (template DNA: 5′-CGAGGTCGAGCGTGTCCTGGTCTAG-3′, non-template DNA: 5′-CGCTCGACCTCG-3′; RNA: 5′-FAM-GACCAGGAC-3′) in transcription buffer (20 mM HEPES pH 7.8, 60 mM (NH 4 ) 2 SO 4 , 8 mM MgSO 4 , 10 µM ZnCl 2 , 10% (v/v) glycerol, 10 mM DTT) for 20 min at 20°C. For RNA elongation, NTPs (1.4 mM end concentration each) were added and the reaction was incubated for 30 min at 28°C. To examine cleavage activity, the pre-incubated reaction with a twofold molar excess of Pol I compared to scaffold was incubated for 30 min at 28°C without the addition of NTPs. To stop the reaction an equal amount of 2x RNA loading dye (8 M Urea, 2× TBE, 0.02% bromophenol blue, 0.02% xylene cyanol) was added and the sample was heated to 95°C for 5 min. As control 0.25 pmol of scaffold was treated identically, without the addition of polymerase and NTPs. 0.125 pmol of FAM-labeled RNA product (as well as a marker containing 9 nt, 15 nt and 21 nt long FAM-labeled RNAs: 5′-FAM-GACCAGGAC-3′, 5′-FAM-AACGGAGACCAGGAC-3′, 5′-FAM-UGUUCUUCUGGAAGUCCA GTT-3′) was separated by gel electrophoresis (20% polyacrylamide gel containing 7 M Urea) and visualized with a Typhoon FLA9500 (GE Healthcare).

Active center cleft expansion is conserved between Sc and Sp Pol I but most pronounced in Sc
Preparation of Pol I elongation complex. Synthetic DNA (IDT) and RNA (Qiagen) oligonucleotides were designed and assembled as described 15 , with the scaffold sequence for the template DNA (5′-AAGCTCAAGTACTTAAGCCTGGT CATTACTAGTACTGCC-3′), non-template DNA (5′-GGCAGTACTAGTAAAC TAGTATTGAAAGTACTTGAGCTT-3′), and RNA (5′-UAUCUGCAUGUAGAC CAGGC-3′; for the underlined nucleotides a methylene bridge between the 2′-O and the 4′-C of the ribose ring has been formed, thus creating a locked nucleic acid, LNA). Annealing was achieved by equimolar mixing (40 µM), then heating to 95°C, and gradually reducing the temperature to 20°C over 90 min. Pol I (1 mg/ml) was incubated with a 1.35-fold molar excess of pre-annealed EC-scaffold for 30 min at room temperature.
Single-particle cryo-EM. Images were collected on a Titan Krios Electron Microscope (Thermo Fisher Scientific) at 300 keV. Movies of 40 frames were acquired on a Falcon III direct electron detector at 75,000x magnification (pixel size 1.0635 Å). The movies were recorded in linear mode with a dose rate of~19 e − /px/s and a total dose of around 86 e − /Å 2 . The defocus span from −1.4 µm to −2.4 µm alternating in 0.2 µm intervals with a total of four exposures per hole.
Data processing. The EC dataset was processed using the RELION 3.0 suite 46 ( Supplementary Fig. 2). Movie frames were aligned and dose weighted using Relion's own implementation of MotionCor and Contrast Transfer Function (CTF) parameters were estimated using GCTF. A total of 3,598 movies were chosen based on accumulated motion, visual inspection and CTF values, astigmatism, defocus and maximal resolution. A set of 100 randomly picked micrographs throughout the dataset was chosen for reference-free auto-picking using the Laplacian-of-Gaussian (LoG) routine and yielding 2,829 particles. Two-dimensional classification resulted in templates for reference-based auto-picking yielding 299,038 particles. Two-fold binned particles (128 pixel boxes) were subjected to reference-free 2D classification (250 Å mask). Following removal of contaminants, a total of 156,493 unbinned particles were selected and aligned in 3D using an initial model generated in RELION as reference. These particles then underwent CTF refinement, bayesian polishing, followed by another round of CTF refinement. Masked Auto-refinement resulted in a reconstruction at 3.89 Å overall resolution (0.143 FSC). Removal of particles showing increased flexibility of the Jaw and Clamp subdomains were removed by 3D Classification, resulting in 61,954 particles that allow reconstruction of an Sp Pol I EC at 4.00 Å resolution.
The 'monomer' dataset was processed using the RELION 3.0 suite 46 unless stated otherwise (Supplementary Fig. 1). After importing pre-averaged movie frames (sums) the Contrast Transfer Function (CTF) parameters were estimated with the embedded Gctf program. Pre-processing was performed as described for the EC dataset. LoG picking resulted in a total of 11,594 particles from which 1,142 were chosen for template-based auto-picking following 2D classification. Removal of~90% of initially picked particles can be attributed to contamination and damage resulting from stress on the air-buffer interface. Initial auto-picking identified 874,753 particles from 4,333 micrographs, of which~50% were discarded as contamination based on 2D-Classification. A total of 477,791 particles were subjected to 3D classification using PDB 5M3M as reference. This allowed the removal of damaged particles and Pol I particles with highly flexible subdomains. The remaining 79,313 particles were subjected to CTF-refinement and bayesian polishing as for EC particles. The final 3D-reconstruction of monomeric Sp Pol I at a nominal resolution of 3.84 Å shows an even orientational distribution of particles and some flexibility in the peripheral jaw, clamp and stalk regions.
In initial 2D classifications, minor dimer-classes were noticed. Thus, autopicked particles were re-extracted in larger boxes of 360 pixels and analyzed in a second, independent processing tree. From 510,315 Pol -like particles selected by 2D classification, a class of 17,552 particles could be attributed to well-defined dimers. Particles were centered on the interface between both Pols, re-extracted and another 450 poor particles removed based on 2D classification without sampling. Final 3D auto refinement imposing C2 symmetry yielded a reconstruction of Sp Pol I dimers at an overall resolution of 4.5 Å.
Model building. At nominal resolutions of 3.8-4.0 Å, we derive near-atomic models for most regions of Sp Pol I monomers and the EC. To commence model interpretation, we constructed homology models of subunits A190, A135, AC40, AC19, A43, A14 (ker1 in Sp) and A12.2 based on sequence comparison with their Sc homologs, alignment of actual and predicted secondary structures, and domain searches using HHPRED 47 . To construct homology models, the MODELLER 48 implementation of UCSF Chimera was used 49 . Structures of the general subunits Rpb5, Rpb6, Rpb8, Rpb10 and Rpb12 were imported from the crystal structure of Sp Pol II 50 . Subdomain boundaries were defined based on Sc homology ( Fig. 1 and Supplementary Fig. 4). Subdomains were then rigid body fitted into EC densities (which were first obtained) using COOT 51 . While many regions allowed accurate fitting of sidechain orientations in the sharpened cryo-EM map, others suffered from poor main-chain tracing. Hence, density-guided modeling was performed in the clamp head, dock-insertion, foot-insertion and part of the jaw regions in subunit A190, as well as the subunit A12.2 and the toe domain of subunit AC40. Modeling of stalk-subunits A43 and A14 was limited to rigid body fitting of trimmed homology model domains in the unsharpened cryo-EM density. As a final step, real-space refinement was carried out using phenix.refine 52 . The Sp Pol I monomer was built by placement of the EC model and adjustment of subdomains in COOT, followed by manual inspection and real-space refinement using phenix. refine. The dimer model was constructed by placement of two monomers, rigid body fitting of subdomains and refinement in phenix-refine using NCS restraints.
Concentration-dependent dimerization using dynamic light scattering. Frozen Sp Pol I was thawed and diluted (1.5 µM, 1.0 µM, 0.75 µM, 0.5 µM) in S600 buffer (10 mM HEPES pH 7.8, 1 mM MgCl 2 , 0.01 mM ZnCl 2 , 5 mM DTT, 1.5 % (v/v) glycerol, and 0.6 M KAc) to a final volume of 20 µl. Technical duplicates of 10 µl of each Pol I concentration were loaded into glass capillaries (Prometheus NT.48 Series nanoDSF Grade Standard Capillaries) and each capillary mounted into a Prometheus Panta (NanoTemper Technologies GmbH). Fourty consecutive DLS measurements of each capillary were taken at 75 % LED power, 100% Laser power and 15°C (total measurements per condition: n = 80). Calculation and visualization were carried out using GraphPad Prism version 8.0.1 for Windows, GraphPad Software, La Jolla California USA, www.graphpad.com. The boxes extend from the 25th to 75th percentiles 53 . The whiskers in Fig. 2b are drawn down to the 10th percentile and up to the 90th. Points below and above the whiskers are drawn as individual points.
Negative staining, EM data collection and image processing. Analytical SEC peak fractions were diluted to 20% (v/v) and 10% (v/v) in their respective buffers and were centrifuged (4°C; 21,130 g; Eppendorf tabletop centrifuge) for 5 min. Five µl of the samples were then applied to 400-mesh copper grids (G2400C; Plano) with a selfmade carbon film of~7 nm thickness (self-made). After 30 s, grids were washed in 200 µl ddH 2 O for 30 s, and stained three times in 20 µl saturated uranyl formate solution (30 s). After each step, excess liquid was removed with a filter paper. Images were collected on a JEOL 2100-F Transmission Electron Microscope operated at 200 keV and equipped with TVIPS-F416 (4kx4k) CMOS-detector at 40,000x magnification (pixel size 2.7 Å) with alternating defocus (−2.5 to −4.5 µm).
The images were processed using RELION 3.1 (see above). For the high-salt SEC peak fraction (S1500), a total of 90 out of 98 collected micrographs were analyzed. A set of 10 randomly chosen images was used to train and optimize the reference-free auto-picking using Laplacian-of-Gaussian (LoG) routine. These settings were then applied on all 90 images yielding 40,544 particles that were applied to reference-free 2D classification (380 Å mask). After removal of junk, a total of 20,054 particles were classified into 16 classes.
For the shown low-salt SEC peak fraction (S300), 129 micrographs were analyzed. A set of 10 randomly chosen images was used to train and optimize the reference-free auto-picking using Laplacian-of-Gaussian (LoG) routine. These particles underwent selection based on 2D classification and 3D centering yielding 3,532 particles that were subsequently used as a template for a reference-based auto-picking from all 129 images resulting in 36,733 particles. After 3D centering using the filtered density of PDB 5M3M as reference and removal of junk particles by 2D classification (380 Å mask), 24,462 particles remained. The outcome of a 2D classification into 16 classes was then compared to the high-salt particles ( Supplementary Fig. 5).
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The cryo-EM density of Sp Pol I monomer, dimer and EC have been deposited in the Electron Microscopy Data Bank under accession codes EMD-11840, EMD-11841 and EMD-11842, respectively. Coordinates of the Sp Pol I monomer, dimer and EC were deposited with the Protein Data Bank under accession codes 7AOC, 7AOD and 7AOE, respectively. The data underlying Fig. 2 and unprocessed gel scans are provided in a separate Source Data File. Further material can be obtained from the corresponding author upon reasonable request. Source data are provided with this paper.