Regulation of mammalian transcription by Gdown1 through a novel steric crosstalk revealed by cryo-EM

Single-particle analysis of native bovine RNAPII and RNAPII(G) complexes in negative stain

RNAPII and RNAPII(G) display different behaviour in solution. As shown in an EM image (Figure 2A), RNAPII(G) predominantly formed monomers. By contrast, RNAPII mainly formed dimers (Figure 2D). Images of RNAPII dimeric particles or RNAPII(G) monomeric particles were aligned using the SPIDER (Frank et al, 1996) and clustered with XMIPP (Sorzano et al, 2004). 7689 RNAPII(G) particle images conferred a set of class averages resembling the 2D projections of yeast RNAPII X-ray structure (Figure 2B; Supplementary Figure 1A). By the common-line method (Penczek et al, 1996), those class averages were used to generate an initial model, which was used to guide the angular reconstruction (Penczek et al, 1994) of RNAPII(G) to obtain a volume with ∼30 Å resolution (Supplementary Figure 1B). As the EM structure of RNAPII(G) was superimposed with the 12-subunit yeast RNAPII (Armache et al, 2005; PDB: 1WCM) (Figure 2C), good agreement was found, while Gdown1 density was virtually undetected. As to RNAPII dimers, alignment and clustering of selected dimeric particle images resulted in very few numbers of different classes, indicating the dimmers had preferred orientations. A class average of the dimer images (Figure 2E) carried the feature of yeast RNAPII dimers previously identified in 2D crystal (Darst et al, 1991; Asturias et al, 1998). To investigate the interface for dimerization of RNAPII, a dimer model was built from the 3D reconstruction of a negative-stained RNAPII(G) to suit the 2D dimer class average, by which the Rpb3 or Rpb4 subunit is suggested to participate in the dimer contact (Figure 2F). That Gdown1 can break the dimer formation suggests its binding sites on RNAPII include those interfacial areas. Interestingly, antibody against the Rpb3 subunit could induce a large fraction of RNAPII (∼70%) to form monomers (Supplementary Figure 1C). As suggested by our low-resolution negative-stain EM study together with the folding analysis, Gdown1 may dwell on RNAPII in an extended form, leading us to pursue a higher quality and more detailed structure by cryo-EM. Furthermore, though native RNAPII and RNAPII(G) appear relatively stoichiometric, we were unable to achieve a complete separation of native RNAPII(G) so that some amounts, though very minimal, of 12-subunit RNAPII certainly contaminate our RNAPII(G) preparation. Therefore, for the cryo-EM study, we generated RNAPII(G) particles by reconstituting RNAPII with a three- to four-fold excess of recombinant Gdown1 (Hu et al, 2006).

Open in a separate window

Figure 2

Image analyses of native bovine RNAPII complexes. (A) Native bovine RNAPII(G), enriched in monomers (white circles), were preserved in negative stain (2% uranyl acetate, 50 nm bar scale). (B) Upper row: class averages of bovine RNAPII(G) particles obtained after reference-free alignment and clustering reveal renowned RNAPII features such as groove and stalk (Rpb4/7). Lower row: re-projections from the 3D reconstruction of native RNAPII(G) that best match the class averages above. (C) EM structure of the native bovine RNAPII(G) superimposed with the yeast RNAPII X-ray structure (PDB: 1WCM). (D) Native bovine RNAPII, enriched in dimers (white circles), preserved in negative stain (50 nm scale bar scale). (E) A representative class average of RNAPII dimer. (F) 3D model of the RNAPII dimer built from reconstruction RNAPII monomers. An orientation best matches the class average in (E) shows subunit Rpb3 and Rpb4 likely participate the dimer interface.

Cryo-EM of bovine RNAPII elongation complex

For a rigorous assessment of Gdown1 on RNAPII, it was desired to directly compare a 13-subunit RNAPII(G) with a 12-subunit RNAPII, both derived from bovine. Therefore, a cryo-EM structure of bovine RNAPII (12-subunit) reconstructed from its monomer projections was pursued. However, the dimeric form of RNAPII as opposed to the monomeric form of RNAPII(G) was not suited for a direct comparison. In an attempt to generate monomers from the bovine RNAPII dimer, the screening of salt conditions was exhausted but none of them succeeded to dissociate the 12-subunit RNAPII dimers into monomers. Interestingly, as RNAPII was supplemented with a nucleic acid scaffold (Kettenberger et al, 2004; Chen et al, 2009) (Supplementary Figure 2A), RNAPII predominantly formed monomeric particles in cryo images (Supplementary Figure 2B). The 12-subunit bovine RNAPII reconstituted with dsDNA/RNA is herein designated as the RNAPII elongation complex. We first generated a cryo-stained structure of bovine RNAPII elongation complex (Supplementary Figure 2C) by the angular reconstruction method using the negative-stained EM volume of RNAPII(G) as an initial model (Figure 2C). The cryo-stained structure of bovine RNAPII was consistent with that of the human RNAPII EM structure (EMD-1284) (Kostek et al, 2006) obtained by a similar approach (Supplementary Figure 2D). The cryo-stain reconstruction was employed to guide the angular parameters of ∼20 000 unstained cryo-EM images of bovine RNAPII elongation complex to a resolution ∼19 Å (Figure 3). As the resultant cryo-EM map of bovine RNAPII elongation complex was properly contoured according to the molecular mass of RNAPII, it gave no sporadic densities on the surface that did not belong to RNAPII but agreed nicely with that of the X-ray structure of the yeast RNAPII elongation complex filtered to the same resolution (Kettenberger et al, 2004; PDB: 1Y1W) (Figure 3). Since we did not inject any X-ray model of RNAPII for reconstructing the RNAPII cryo-EM images, the remarkable match between the cryo-EM maps of RNAPII with the X-ray structure strongly indicated that our reconstruction algorithm was reliable and could be applied to other RNAPII complexes in this study.

Open in a separate window

Figure 3

Cryo-EM reconstruction of the reconstituted bovine 12-subunit RNAPII elongation complex. Three views of the cryo-EM reconstruction of bovine RNAPII elongation complex at ∼19 Å resolution (FSC 0.15). Superimposed with the EM envelop is the X-ray structure of yeast 12-subunit RNAPII (coloured ribbons) (PDB: 1Y1W). The threshold for rendering the EM reconstruction is chosen based on a molecular weight of ∼520 kDa.

Cryo-EM of the 13 subunit bovine RNAPII–Gdown1 elongation complex

To assure a complete stoichiometric presence of Gdown1 in the reconstituted RNAPII(G) samples, we added four-fold recombinant human Gdown1 (rGdown1) to the 12-subunit bovine RNAPII to form the RNAPII–rGdown1 complex. RNAPII–rGdown1 was previously determined to be functionally equivalent to native bovine RNAPII(G) (Hu et al, 2006). The ratio of four for reconstitution was determined by titration of rGdown1 to RNAPII followed by negative-stain EM observation to assess the minimal amount of rGdown1 required to turn the majority of RNAPII dimmers into monomeric particles (Supplementary Figure 3B). Such ratio of Gdown1 to RNAPII was found to completely inhibit promoter-specific transcription (Jishage et al, 2012). The RNAPII–rGdown1 complex is herein termed RNAPII–Gdown1. In addition, the RNAPII–Gdown1 complex was reconstituted with dsDNA/RNA to form the RNAPII–Gdown1 elongation complex, whose reconstruction would be used to compare with that of RNAPII elongation complex. The 3D cryo-EM reconstruction of RNAPII–Gdown1 (Supplementary Figure 3C) and RNAPII–Gdown1 elongation (Figure 4A) were both obtained to a resolution of 19 Å (Supplementary Figure 3E) from ∼25 000 unstained particle images respectively via the same route as for the RNAPII elongation complex. By subtracting the cryo-EM volume of RNAPII elongation complex from that of RNAPII–Gdown1 elongation complex, a positive difference map was obtained and scored based on thresholding in units of standard deviation (σ). Those densities above 5σ spread on RNAPII extensively with the majority found in the vicinity of the DNA cleft (Figure 4B); they were estimated to account for a mass of ∼40 kDa, fairly close to that of Gdown1. We divide those on the top of RNAPII into ‘a’ through ‘e’ according to where they dwell (Figure 4B). Remarkably, in region a, a bulky domain with mass of ∼6 kDa above 10σ appeared on the Rpb5 shelf and connected to the Rpb1 jaw. We named this domain Gdown1-a. To verify the localization of Gdown1 on RNAPII, antibody labelling experiments were performed. By negative-stain EM, a polyclonal antibody against Gdown1 was located near the Rpb1 jaw where the Gdown1-a is (left panel in Figure 4D). To further locate the terminal regions of Gdown1, a glutathione S-transferase (GST) was fused to the N-terminus of Gdown1 and a monoclonal antibody against GST was found also near the Rpb1 jaw (right panel in Figure 4D). Since our structural results indicated Gdown1 direct contacted the Rpb5 subunit of RNAPII, we tested if Gdown1 and Rpb5 would bind to each other with an in vitro pull-down assay. Indeed, recombinant Gdown1 and Rpb5 co-purified through two distinct affinity steps, supporting the notion that Gdown1 and Rpb5 would physically associate (Supplementary Figure 3F).

Open in a separate window

Figure 4

Cryo-EM reconstruction of reconstituted bovine RNAPII–Gdown1 and antibody labelling experiments. (A) Front and top views of the cryo-EM reconstruction bovine RNAPII–Gdown1 elongation at ∼19 Å resolution (FRC 0.15) are depicted as solid deep green surface models. The threshold for rendering RNAPII–Gdown1 elongation is chosen based on a molecular weight of ∼560 kDa. (B) A positive difference map was calculated between the bovine RNAPII–Gdown1 elongation and bovine RNAPII elongation complex (grey mesh) by using volumes that were both filtered to 15 Å, and shown in yellow hue above 5 σ. The most conspicuous density above 10 σ is in the gap between the Rpb5 shelf and the Rpb1 jaw is shown in deep green. (C) A negative difference map was calculated between the bovine RNAPII–Gdown1 elongation and bovine RNAPII elongation complex and shown in pink hue. The additional densities attributed to the RNAPII elongation complex are likely to be the result of conformational changes (outer densities). (D) The right panel presents RNAPII–Gdown1 EM analysis with a polyclonal antibody against Gdown1 (gift from Dr David Price, University of Iowa, USA). From top to bottom are raw negative-stained images, images filtered to 50 Å to reveal RNAPII gross features, matching projections of RNAPII at 50 Å, and the corresponding 3D models. The left panel shows a similar analysis employing a monoclonal antibody recognizing GST, fused to the N-terminus of Gdown1. Antibody densities are encircled in the top rows, and their location denoted with arrows in the lower 3D models.

Cryo-EM of bovine RNAPII–TFIIF elongation complex

The structure of RNAPII–Gdown1 established herein has immediate functional implications. If common sites of association with RNAPII exist for other transcription factors, Gdown1 may poise a challenge for them to access RNAPII because Gdown1 binds to RNAPII so tightly as it does not dissociate either in high salt or in urea (Hu et al, 2006), in contrast to other transcription factors that readily dissociate from RNAPII in the presence of high salt (Cheng and Price, 2009).

Based on the reasons to be described, TFIIF was thought to be susceptible. In higher eukaryotes, TFIIF is composed of a large and a small subunit, RAP74 and RAP30, and their yeast homologues are dubbed Tfg1 and Tfg2 respectively. Yeast also contains a third TFIIF subunit Tfg3 not present in other eukaryotes (Henry et al, 1994; Kimura and Ishihama, 2004). In the previous cryo-EM study using the yeast proteins (Chung et al, 2003), the Tfg2 subunit was assigned in the DNA-binding cleft, mainly inferred from the crystal structures of bacteria RNAP holoenzyme (Murakami et al, 2002; Vassylyev et al, 2002), while the Tfg1 subunit was interpreted to reside largely alongside or on the RNAPII stalk, formed by subunits Rpb4 and Rpb7.

However, an alternative approach utilizing cleaving reagents mapped the residues of Tfg1 in the Tfg1/Tfg2 dimerization domain on the lobe/protrusion region of Rpb2 (Chen et al, 2007; Eichner et al, 2010). Recently, high-resolution cross-linking followed by mass spectroscopy allowed for positioning of almost all TFIIF residues on RNAPII (Chen et al, 2010), in which an extended track of Tfg1 was found on Rpb2, from the Rpb1 jaw-Rpb5 shelf to the Rpb2 lobe/protrusion. Providing that the interactions between RNAPII and TFIIF are largely conserved between yeast and mammals, RAP74—the mammalian homologue of Tfg1 would associate with mammalian RNAPII in a similar region.

To resolve the controversies as to the location of TFIIF on RNAPII, we re-investigated the cryo-EM structure of RNAPII–TFIIF but in the context of mammalian proteins by reconstituting bovine RNAPII with recombinant human TFIIF. The updated RNAPII–TFIIF cryo-EM map has direct relevance to mammals and would eliminate any imprecise inference from the yeast.

As the addition of recombinant TFIIF to bovine RNAPII did not turn the dimeric RNAPII into the monomeric form as effectively as Gdown1, we supplemented the RNAPII–TFIIF with nucleic acid scaffold, which is herein termed RNAPII–TFIIF elongation complex. The cryo-EM structure of the RNAPII–TFIIF elongation complex was reconstructed to a resolution of 19 Å (Supplementary Figure 4D) from ∼25 000 monomeric particle images (Supplementary Figure 4B) using the similar approach for RNAPII(G). The 3D cryo-EM envelop of RNAPII–TFIIF elongation (Figure 5A) also matches well with that of the bovine RNAPII elongation cryo-EM structure. However, by subtracting RNAPII from RNAPII–TFIIF, densities attributed to TFIIF were revealed. Those scored above 5σ mainly appeared around the DNA cleft but not in the cleft and were distributed in regions labelled ‘a’ through ‘e’, used for describing Gdown1 (Figure 5B). Regions ‘a’ to ‘c’ accord with the localization of TFIIF on RNAPII by cross-linking experiments (Chen et al, 2010); for example, ‘a’ on the Rpb1 jaw domain corresponds to the charged domain of Tfg1 (RAP74), ‘b’ around the Rpb2 lobe area encircles the dimerization domain of Tfg1–Tfg2 (RAP74–RAP30) together with the N-terminus of Tfg1 (RAP74), and ‘c’ next to the Rpb2 protrusion corresponds to the linker of Tfg2 (RAP30). To further confirm the localization of RAP74 independently by EM, the same GST strategy was employed. The GST protein fused to the N terminus of RAP74 was detected on the Rpb2 side of RNAPII, near the jaw or lobe (Figure 5D), agreeing nicely with the finding by cross-linking (Chen et al, 2010). Importantly, as soon as the positive difference ascribed to TFIIF is overlaid with that of Gdown1 (Figure 5E), it is evident that most of Gdown1’s sites clash with TFIIF. The conflict occurs on the jaw/shelf (region a in Figure 5E), where Gdown1-a meets with the charged region of RAP74 (Chen et al, 2010) and also likely with part of RAP30 (Wei et al, 2001; Le et al, 2005); near the lobe/protrusion (regions b and c), where Gdown1 clashes with the N-terminus of RAP74 and the RAP74–RAP30 dimerization domain (Chen et al, 2010); and on the clamp (region e), where Gdown1 can run into RAP30, according to the X-ray study of yeast RNAPII-Tfg2 (Kornberg, 2007). Such findings strongly suggest that Gdown1 and TFIIF would mutually exclude each other from accessing RNAPII.

Open in a separate window

Figure 5

Cryo-EM reconstruction of RNAPII–TFIIF. (A) Top and front views of the cryo-EM reconstruction of bovine the RNAPII–TFIIF elongation complex at a resolution of ∼19 Å (FRC 0.15). The threshold for rendering the complex is chosen based on the total mass of the complex (∼620 kDa). (B) A positive difference map in blue hue is overlaid on the RNAPII elongation complex (grey mesh). The difference was obtained by subtracting RNAPII with nucleic acids from RNAPII–TFIIF with nucleic acid, both filtered to 15 Å. (C) A negative difference map in red hue is overlaid on the RNAPII elongation complex (grey mesh). (D) RNAPII–TFIIF images with a monoclonal antibody recognizing the GST fused to the N-terminus of TFIIF. From top to the bottom are raw negative-stained images, images filtered to 50 Å to reveal RNAPII gross features, matching projections of RNAPII at 50 Å, and the corresponding 3D models. (E) Common, overlapping regions attributed to both Gdown1 in Figure 4B (green) and regions attributed to TFIIF in Figure 5B (blue) are shown in yellow.

RNAPII(G) nonspecific promoter assay

The transcription elongation assays were performed as previously described (Gnatt et al, 1997). Briefly, nonspecific initiation on a tailed template was employed. The tailed template was formed by annealing a nontemplate strand: 5′-AACACCAGCGAGCAAGGCGTTTCGGGGAAGAAAAA, and a template strand: TTTTTCTTCCCCGAAACGCCTTGCTCGCTGGTGTTCCCCCCCCCCCC. 0.4 or 0.8 μg of polymerase, RNAPII or RNAPII(G), were employed, and α-amanitin, an RNAPII-specific inhibitor, was used as negative control. Quantification of bands was performed using the NIH free ImageJ software.

Expression of recombinant human Gdown1 and TFIIF

Recombinant Gdown1 expression plasmid, originally constructed in the pET151 vector (Hu et al, 2006), was subcloned into a pET21a vector using NdeI and NotI restriction sites. The Gdown1 protein was overexpressed in BL21 cells (Invitrogen) by IPTG induction (0.4 mM) and grown overnight at 20°C. The protein was purified by using a nickel-NTA column with standard procedure as described by the manufacturer (Sigma). Further HPLC purification was depicted in Supplementary Figure 3A. The purified proteins were stored at −80°C. A bacterial expression plasmid encoding human TFIIF was constructed in the pACYCduet1 vector. To express TFIIF, Escherichia coli BL21 (DE3)-RIL cells (Merck) transformed with the TFIIF containing duet plasmid was grown at 37°C and the temperature was lowered to 30°C when OD reached 0.4 for induction with 0.84 mM IPTG. The cells were further grown for 3 h at 30°C before harvest. TFIIF was purified using a nickel-NTA column in the same manner as Gdown1 was, with slight modifications in the washing step by raising the concentration of NaCl to 300 mM and that of imidazole to 40 mM (see also Supplementary Figure 4A).

Limited proteolysis analysis of Gdown1

Purified recombinant Gdown1 was digested with trypsin (7500:1) at 30°C. After 40 min, the reaction was quenched by heating at 95°C for 10 min. The digestion products were subsequently analysed by LC mass spectrometry.

Size exclusion chromatography assays

To form RNAPII–Gdown1 or RNAPII–TFIIF complex, 10 μg of bovine RNAPII was mixed with individual recombinant proteins, rGdown1 or rTFIIF, and incubated for 60 min at 20°C, respectively. For rGdown1, 4 μg was used, representing a rGdown1 to RNAPII ratio of 4:1. As to the formation of RNAPII–TFIIF, about eight-fold recombinant TFIIF (16 μg) was used. The size exclusion chromatography was carried out using the AKTA purifier system (GE Health/Amersham Biosciences). A Superose 6 column PC 3.2/30 (GE Healthcare) was equilibrated with equilibrating buffer (50 mM Tris–HCl (pH 7.5 at 4°C), 100 mM NaCl, 1 mM TCEP, and 1% protease inhibitor cocktail). The protein sample (∼30 μl) was loaded onto the column for fractionation with a total elution volume of 3.2 ml at a flow rate of 0.06 ml/min. The fractions were collected with 60 μl per fraction. The protein complex in each fraction was concentrated with nickel beads (Sepharose High Performance, GE Healthcare). In brief, 10 μl pre-equilibrated Ni-beads were added to individual fraction from (15–35) respectively, and incubated with mixing at 4°C overnight. The supernatant was discarded; the same volume of 2 × SDS–PAGE sample buffer was added to the beads, boiled on a heating block for 10 min. The denatured protein samples were separated on a NuPAGE 4–12% Bis-Tris gel using MES running buffer and visualized with silver stain. In all, 5 μl of 10-fold diluted protein ladder was loaded onto an SDS–PAGE gel. To challenge RNAPII–Gdown1 with rTFIIF, RNAPII was first incubated with four-fold rGdown1 at 20°C and mixed at 550 r.p.m. for 60 min. After the first incubation period, eight-fold rTFIIF was added, and another 60 min of incubation was performed, followed by loading the protein samples onto a Superpose 6 PC 3.2/30 size exclusion column. To challenge RNAPII–TFIIF with rGdown1, RNAPII was first incubated with four-fold rGdown1, and eight-fold rTFIIF was added after the first incubation period for an additional incubation period before size exclusion chromatography was carried out.

EM sample preparation

In brief, a frozen aliquot of bovine RNAPII or RNAPII(G) (1 mg/ml) was freshly thawed and diluted with de-ionized water to a final concentration of 50 μg/ml for direct EM usage or for reconstitution. To reconstitute bovine RNAPII complexes, incubation of bovine RNAPII with additional proteins were performed at 20°C for 60 min before the EM sample was prepared. To form a RNAPII–Gdown1 complex, the molar ratio of recombinant Gdown1 to RNAPII was adjusted to 4:1. For RNAPII–elongation complexes, three-fold of nucleic acid scaffold containing the bubble DNA and a 35-nt RNA was used (Supplementary Figure 2A). For RNAPII–TFIIF, six-fold recombinant TFIIF together with three-fold nucleic acid scaffold was added. In this study, three different techniques for preserving RNAPII complexes were employed, including negative-stain, cryo-staining and unstained cryo. To make an EM specimen, 3 μl of protein solution was applied to Cu grid (300 mesh) or quantifoil grid coated with thin carbon film to enhance the Thon ring observation. The carbon film was made by evaporating a carbon rod onto a cleaved mica sheet with an evaporator (Edwards Auto 306), deposited on the grid to dry, and freshly discharged with amylamine in a glow-discharger (EM Science, EM-100) before applying protein solution. For the negative-stain technique, 2% uranyl acetate was used; the sample was either sandwiched by another layer of thin carbon (Tischendorf et al, 1974) or deep-stained (Stoops et al, 1991). For cryo-staining, the protein-adsorbed grids were first stained with 1% uranyl acetate and not allowed to dry; the sample was immediately blotted with a filter paper (Wattmann No. 1, smooth side) and plunged into liquid ethane with a customer built plunger in a cold room with 80% humidity (Golas et al, 2003; Ohi et al, 2004). The frozen EM samples were then stored under liquid nitrogen before EM. For unstained cryo preservation, the procedure was the same as cryo-staining except no staining agent was used.

Electron microscopy

Negative-stain preserved native bovine RNAPII or RNAPII(G) were collected on Kodak SO-163 films by a 200-kV HRTEM (JEOL 2011, LaB₆ filament, Cs=1.0 mm) at magnification of × 40 000 with ∼10 e⁻/Ų, defocus of ∼2–3 μm. The micrographs were developed for 12 min at 25°C in full strength with the Kodak D-19 solution, dried and digitized on a Zeiss/Integraph flat-bed densitometer using a step size of 14-micron, corresponding to 3.5 Å on the object scale. For cryo-staining, Cu grids (300 mesh) were loaded with a layer of carbon film and glow discharged with amylamine; the specimens were observed using a Gatan 914 quick-load cryo-holder on a 200-kV field-emission microscope (JEOL 2010F, Cs=3.0 mm) at magnification of × 52 000 and images of the reconstituted RNAPII complexes were collected on a 4K × 4K CCD (Gatan 895) with the doses controlled at ∼10–20 e⁻/Ų and defocus near ∼2.5 μm. At such magnification, each CCD pixel corresponded to 2.9 Å. For cryo-EM imaging of unstained specimens, Quantifoil R2/2 grids (Quantifoil Micro Tools GmbH) were loaded with a very thin layer of carbon film and glow discharged with amylamine; images of the reconstituted RNAPII complexes were collected on a 4K × 4K CCD (Gatan 895) by the same 200 kV field-emission microscope with the magnification adjusted to × 78 000 and the doses ∼10–20 e⁻/Ų with defocus controlled in the range ∼2.0–4.0 μm. At such magnification, each CCD pixel corresponded to 1.9 Å.

Immuno-electron microscopy

For antibody decoration against the GST-tagged Gdown1 in the RNAPII–Gdown1 complex or against the RAP74 subunit within the RNAPII–TFIIF complex, RNAPII complexes were incubated with two-fold antibody and then subjected to gel filtration for fractionating the RNAPII–Gdown1 antibody or the RNAPII–TFIIF antibody complex. Application of the polyclonal antibody against Gdown1 also followed the same procedure. To disclose the location of the antibody in complex, a 120-kV TEM (JEOL 1400, LaB₆ filament, Cs=3.4 mm) and 4K × 4K CCD camera (Gatan 895) were used; the magnification used for visualizing antibody was × 104 000 and the defocus was∼1.5 μm.

Single-particle reconstruction of RNAPII complexes

To analyse RNAPII(G) preserved in negative-stain, a total of 7689 particle images were interactively selected by using EMAN BOXER (Ludtke et al, 1999). The images were pre-aligned on SPIDER (Frank et al, 1996), transferred to XMIPP for classification by the CL2D method (Sorzano et al, 2004) into 128 classes (Supplementary Figure 1A). Subsequently, representative class averages were selected and an initial model was derived by the common-line method (Penczek et al, 1996). To reconstruct a 3D structure of RNAPII(G) by back-projection, the initial model was injected for assigning the initial orientation parameters to the individual raw images by the projection matching method (Penczek et al, 1994), by which 10–20 cycles of angle refinement were used until convergence occurred.

Before we pursued cryo-EM of RNAPII complexes in the unstained state, reconstruction of each complex preserved by cryo-staining was generated. More precisely, the negative-stain reconstruction of RNAPII(G), lowpass filtered to 40 Å, was used as initial model for calculating the cryo-stained reconstruction of RNAPII–Gdown1 from a total of 13 000 cryo-stained images. The same procedure was repeated for obtaining the cryo-stained reconstruction of RNAPII in an elongation form also using 13 000 images, and for obtaining cryo-stained reconstruction of RNAPII–TFIIF, 13 000 images were employed as well. The resolution of reconstruction was estimated by the degree of similarity between the two reconstructions derived from each half-set using Fourier shell correlation (FSC) method and the reported resolution is based on FSC=∼0.15 (Kostek et al, 2006).

A total of 25 000 cryo-EM images of RNAPII–Gdown1 in the unstained state were used to calculate the cryo-EM reconstruction of RNAPII–Gdown1 by the back-projection method using the cryo-stained reconstruction of RNAPII–Gdown1 as the initial model. The same approach was used for obtaining the cryo-EM reconstruction of RNAPII in elongation form (∼20 000 particles) and for that of RNAPII–TFIIF (25 000 particles) by using the respective cryo-stained reconstruction as the initial model. The resolution of the cryo-EM reconstruction was estimated by the FSC method as described. The difference mapping was performed using the RobEM program (R Ashmore and T Baker, unpublished data) and two thresholds (5σ and 10σ) were used to present the positive and negative difference maps.

Docking of the X-ray structure and volume rendering

UCSF Chimera (Goddard et al, 2007) was used to analyse and display the EM reconstructions. The PDB ribbons of yeast RNAPII (1WCM or 1Y1W) were first roughly aligned into the reconstruction volumes by visualization. Further optimization of the fitting was achieved by using ‘Fit Model in Map’ of Chimera. Each reconstruction was rendered with a threshold yielding a volume consistent with the molecular weight by assuming an average protein density of 1.21 ų per dalton (Harpaz et al, 1994).

We acknowledge the Instrument Center of Academia Sinica (AS), Institute of Chemistry of AS, and YK. Hwu at Institute of Physics of AS for cryo-electron microscopes. W-HC thanks Yao-Yin Chuang and Yi-Yum Chen for assistance on TEM, Chen-Yu Li for helping trypsin digestion and Ting-Yi Wu for boxing cryo-EM images, and the Mass Spectroscopy Center at Institute of Chemistry of AS. W-HC is grateful for Dr Burton for TFIIF plasmids. W-HC and AG are indebted to Dr Price for anti-Gdown1 antibody and to Dr Price and Dr Roeder for communicating unpublished results of Gdown1. W-HC was supported by AS Thematic Grant (AS-99-TP-A03), AS Nanoscience programme, and National Science Council (NSC) of Taiwan (NSC99-2113-M-001-022-MY3 and NSC98-2113-M-001-020) for this work; Dr Y-M Wu and Y-C Lin have been supported by NSC postdoctoral fellowships (NSC98-2811-M-001-141, NSC96-2811-M-001-083 respectively). XH was supported by Natural Science Foundation of China (Nos. 30800169). AG was supported by the National Institutes of Health Grant (GM64474) and by the DOD Breast Cancer IDEA Award BC083495.

Author contributions: YM Wu did cryo-EM and image reconstruction; JW Chang did gel-filtration competition assays; JW Chang, SH Huang, and PL Wu did antibody EM; YC Lin compiled reconstruction algorithm; CH Wang and CC Chang did GST fusion and subcloning of recombinant TFIIF and Gdown1; X Hu did pull-down assay; X Hu and A Gnatt purified RNAPII and RNAPII(G). YM Wu and WH Chang designed the experiments and analysed the data; YM Wu, JW Chang, X Hu, A Gnatt, and WH Chang wrote the manuscript. Results and opinions herein represent those of the authors alone and not of the funding institutions unless otherwise specified.