Structural dissection of an interaction between transcription initiation and termination factors implicated in promoter-terminator cross-talk

Functional cross-talk between the promoter and terminator of a gene has long been noted. Promoters and terminators are juxtaposed to form gene loops in several organisms, and gene looping is thought to be involved in transcriptional regulation. The general transcription factor IIB (TFIIB) and the C-terminal domain phosphatase Ssu72, essential factors of the transcription preinitiation complex and the mRNA processing and polyadenylation complex, respectively, are important for gene loop formation. TFIIB and Ssu72 interact both genetically and physically, but the molecular basis of this interaction is not known. Here we present a crystal structure of the core domain of TFIIB in two new conformations that differ in the relative distance and orientation of the two cyclin-like domains. The observed extraordinary conformational plasticity may underlie the binding of TFIIB to multiple transcription factors and promoter DNAs that occurs in distinct stages of transcription, including initiation, reinitiation, and gene looping. We mapped the binding interface of the TFIIB-Ssu72 complex using a series of systematic, structure-guided in vitro binding and site-specific photocross-linking assays. Our results indicate that Ssu72 competes with acidic activators for TFIIB binding and that Ssu72 disrupts an intramolecular TFIIB complex known to impede transcription initiation. We also show that the TFIIB-binding site on Ssu72 overlaps with the binding site of symplekin, a component of the mRNA processing and polyadenylation complex. We propose a hand-off model in which Ssu72 mediates a conformational transition in TFIIB, accounting for the role of Ssu72 in transcription reinitiation, gene looping, and promoter-terminator cross-talk.

TFIIB is an essential general transcription factor and a necessary component of the RNA polymerase II (pol II) 3 preinitiation complex (PIC) along with TFIIA, TFIID, TFIIE, TFIIF, and TFIIH. TFIIB is a multidomain protein that is composed of an N-terminal zinc ribbon domain for pol II binding (1)(2)(3)(4), a finger/reader domain for transcription start site selection and de novo transcription initiation (1, 4 -6), and two C-terminal cyclin-like domains that form a ternary complex with the TATA box DNA and TATA box-binding protein (TBP) (7,8) (Fig. 1A). TFIIB makes sequence-specific interactions with the TFIIB recognition elements (BREs) found both upstream (BREu) and downstream (BREd) of the TATA box (8 -10), a structural feature that helps impose promoter directionality (11). After a ϳ6 -12-nucleotide nascent transcript is synthesized, TFIIB is ejected from PIC, and pol II escapes from the promoter (12)(13)(14).
TFIIB and TBP are the targets of the acidic activation domain of herpes simplex virus VP16 (VP16AD), which stimulates transcription by interacting with multiple transcription co-activator complexes, histone-modifying complexes, and chromatin remodeling complexes (15). One mechanism of transcriptional activation suggests that VP16AD displaces the intramolecular interaction between the N-terminal region of TFIIB (TFIIB n ) and the cyclin-like domain-containing core region (TFIIB c ), converting TFIIB to an active conformation for PIC assembly (16 -21).
Transcription-coupled functional cross-talk between promoters and terminators has long been noted. The general transcription factor TFIID, consisting of TBP and TBP-associated factors, was shown to recruit cleavage and polyadenylation specificity factor CPSF, to which both Ssu72 and symplekin are also bound on terminators (30,31). In addition to Ssu72, TFIIB also interacts with the CstF64 (homolog of Rna15 in yeast) component of the cleavage stimulation factor (CstF). In particular, TFIIB phosphorylation on serine 65 is crucial for this interaction and for CstF occupancy on promoters as well (32,33). Furthermore, both TFIIB and the general transcription factor TFIIF were found to be associated with and modulate the enzymatic activity of another pol II CTD phosphatase, Fcp1, which is involved in dephosphorylation of serine 2 on CTD (34,35). Deletion of the transcription termination factor Pcf11 depletes TFIIB and TBP from promoters (36). Notably, a point mutation of the polyadenylation site causes defects in transcription initiation (36).
These functional connections between promoters and terminators may be understood in the framework of chromatin looping. Chromatin looping is a phenomenon in which distal genomic elements are brought within close spatial proximity of each other to influence transcription. Looping occurs between enhancer elements and promoters, intergenic regions and promoters, and promoters and terminators. Promoter-terminator gene looping was first discovered in Saccharomyces cerevisiae and was later shown to be present in other organisms, such as human and plants, as well as in mitochondria and HIV (37)(38)(39)(40)(41)(42)(43)(44). Promoter-terminator gene looping is believed to facilitate pol II recycling during transcription reinitiation (45,46). It is also involved in sustaining transcriptional memory and enhancing transcription directionality (47)(48)(49).
TFIIB and Ssu72 occupy both promoters and terminators and are required for gene loop formation (49,50). Mutation of TFIIB or Ssu72 abolishes gene looping in yeast, as demonstrated by chromatin conformation capture (3C) assays (49,50). Loss of Ssu72 leads to defects in TFIIB recruitment to promoters and elimination of TFIIB from terminators (50) (data not shown). To begin to reveal the mechanism of transcription-dependent promoter-terminator cross-talk that may be achieved at least in part through the interaction between TFIIB and Ssu72, we first conducted structural analysis of the core domain of human TFIIB captured in two new conformations. We next carried out systematic, structure-guided mapping of the binding interface of the TFIIB-Ssu72 complex using a series of in vitro binding assays and photocross-linking analysis based on site-specifically incorporated unnatural amino acids. We also generated a knowledge-based structural model for the TFIIB-Ssu72 complex. We propose a handoff mechanism to explain the role of the TFIIB-Ssu72 interaction in facilitating pol II reinitiation.

The core domain of TFIIB displays a large degree of conformational plasticity
We solved the crystal structure of human apo-TFIIB c to 3.4 Å resolution by molecular replacement, with four molecules per asymmetric unit (Table 1 and Fig. S1, labeled as Molecules A-D). TFIIB c forms two distinct, previously uncharacterized conformations (Fig. 1B). Molecules A and B are found in one conformation, and C and D are in the other conformation; for simplicity, we will only discuss molecules A and D further (Fig.  1B). Notably, each conformation is characterized by prominent translational and rotational movement of the two cyclin-like domains along the flexible C-linker between them (Fig. 1, A and  B). Whereas the secondary structure of each individual cyclinlike domain remains the same, their orientation and distance with respect to each other differ (Fig. S2, A and B).
Our structure reveals an unprecedented degree of plasticity of TFIIB c , although we also note that the conformations of TFI-IB c from our structure, like all crystal structures, are partially influenced by crystal packing. Although we cannot say for certain the extent to which these conformations occur in vivo, in the following, we interpret how this plasticity would influence the biological roles of TFIIB. Two separate conserved regions that would contact the TATA box (e.g. residues Arg 189 and Lys 193 on helix H5), the BREu (e.g. residues Ala 281 , Val 283 , Arg 286 , and Arg 290 on helix H5Ј), and BREd (e.g. residues 152-154 between helices H2 and H3) are solvent-accessible in both molecules A and D ( Fig. 2A). However, they are not spatially aligned and hence in an inactive conformation compared with

TFIIB-Ssu72 interaction and promoter-terminator cross-talk
their counterparts in TFIIB c bound to TBP and the TATA box, where the two cyclin-like domains are orientated to allow simultaneous binding of helix H5 to the TATA box, the C-linker to TBP, helix H5Ј to the BREu, and the loop between H2 and H3 to the BREd ( Fig. 2B and Fig. S3A) (8). Interestingly, apo-TFIIB c from Trypanosoma brucei follows essentially the same conformation as human TFIIB c in the active ternary complex, with the exception that the C-linker is replaced by a rigid bent helix (51), implying that T. brucei TFIIB may always assume an active conformation regardless of whether TBP and the TATA box are bound ( Fig. 2C and Fig. S3B). Molecules A and D are also distinct from the inactive conformation of a human apo-TFIIB c determined previously by NMR ( Fig. 2D and Fig. S3C) (20,52). In the latter case, the two cyclinlike domains are stably packed against each other such that about 1900 Å 2 of solvent-accessible surface area is buried. In stark contrast, the same two domains in the current structures barely contact one another, displaying an extraordinary rotational and translational plasticity that is probably subjected to cellular regulation. In line with this proposal, residues Arg 286 and Arg 290 from helix H5Ј of the second cyclin-like domain change conformation when interacting with the archetypal acidic activator VP16 (20,53). Thus, it appears possible that TFIIB c can assume multiple flexible, inactive conformations.
The C-linker of TFIIB is involved in ternary complex formation (7,8). Interestingly, this region varies in length between species ( Fig. 2A). To test whether the relative position of the two cyclin-like domains in TFIIB c is controlled by the C-linker, we appended a GSGS linker directly C-terminal of residue Leu 208 located in the middle of the C-linker. This TFIIB c variant was found to bind TBP and the TATA box DNA to the same extent as wild-type protein (Fig. 2E, WT and GSGS-C), which agrees with the evolutionary divergence of TFIIB, in that S. cerevisiae TFIIB has an extended C-linker beyond residue Leu 208 compared with human and fly TFIIB but still retains the ability to bind TBP and the TATA box ( Fig. 2A). In contrast, the precise arrangement of the first part of the C-linker is critical for formation of the TFIIB c -TBP-TATA ternary complex (8), and replacing residues 200 -208 with a GSG linker or adding a GSGS linker directly N-terminal of residue Lys 200 significantly decreased formation of the TFIIB c -TBP-TATA box ternary complex (Fig. 2E, GSGS-C and GSGS-N). Overall, our results indicate that the semiconservative C-linker is not the only determinant of the active TFIIB c conformation.

Ssu72 binds to the first cyclin-like domain of TFIIB c
Previous studies have implicated a direct physical interaction between TFIIB and Ssu72 (23,24). To understand the molecu-

TFIIB-Ssu72 interaction and promoter-terminator cross-talk
lar basis of this interaction, we first measured the binding affinity between TFIIB c and Ssu72 by isothermal titration calorimetry (ITC), which was estimated to be around 20.2 M (Fig. 3A). Next, we conducted pulldown assays of C-terminal FLAGtagged human TFIIB with His 6 -tagged Ssu72. Human TFIIB constructs used for these assays included full-length TFIIB (TFIIB FL ), TFIIB c (residues 107-316), and the isolated cyclinlike domain 1 (TFIIB c1 ; residues 107-206) and 2 (TFIIB c2 ; residues 207-316). Whereas TFIIB FL , TFIIB c , and TFIIB c1 all bound to Ssu72, TFIIB c2 could not do so (Fig. 3B), indicating that the first cyclin-like domain of TFIIB c serves as the minimal region of TFIIB for Ssu72 binding. TFIIB c2 is required for the full Ssu72-binding activity of TFIIB; in its absence, the binding of TFIIB c1 to Ssu72 became too weak to be detected by ITC (Fig.  S4). Interaction of TFIIB c with Ssu72 did not appear stoichiometric in the pulldown analysis, probably due to the relatively weak interaction, as indicated by the ITC results. However, the interaction is specific, as is detailed in the studies that follow.
Charge reversal mutation of any pair of the positively charged residues Arg 185 , Lys 189 , Arg 193 , and Lys 200 from helix H5 or the linker between helices H4 and H5 in TFIIB c1 decreased binding to Ssu72 compared with wild-type TFIIB c , and the K189E/R193E mutation essentially abolished the interaction (Fig. 3C). All TFIIB c variants were monomeric by gel filtration chromatography analysis, indicating that the mutations did not significantly perturb protein structure (Fig. S5). Thus, a positively charged surface of TFIIB c1 proves to be necessary for Ssu72 binding. Intriguingly, an overlapping set of residues from this region of TFIIB c1 is also implicated in promoter DNA and acidic activator binding (8,20).
In addition, we tested whether the binding interface between TFIIB c and Ssu72 was conserved across species using purified proteins from S. cerevisiae (yTFIIB c and ySsu72) and Drosophila melanogaster (dTFIIB c and dSsu72). In both cases, TFIIB and Ssu72 were found to interact directly (Fig. 3D). Notably, whereas critical interacting residues Lys 189 and Arg 193 of human TFIIB are fairly conserved in yeast (as Lys 201 and Lys 205 , respectively; Fig. 2A), residues Arg 185 and Lys 200 are not conserved (His 197 and Asn 212 in yeast; Fig. 2A). In the pulldown assays, the conserved K201E/K205E mutation of yTFIIB abolished the interaction with ySsu72 in a manner similar to the human mutation. All yTFIIBc mutants were monomeric by gel filtration analysis, indicating that mutation did not disrupt normal protein structure (Fig. S5). Surprisingly, the H197E/N212E mutation of yTFIIB also significantly reduced binding to ySsu72 (Fig. 3D), suggesting that the mode of interaction between TFIIB and Ssu72 is highly conserved from yeast to humans despite differences in protein sequence.

TFIIB c interacts with an Ssu72 surface formed by three discrete sequence patches
To probe Ssu72 residues responsible for TFIIB c binding, we used a site-specific photocross-linking approach based on incorporation of the unnatural amino acid p-benzoyl-L-phenylalanine (pBpa) via suppression of an amber stop codon (54,55). Under UV illumination, pBpa can cross-link to residues on a binding protein within 2.4 -4 Å, allowing for a clear readout of protein-protein interactions (56). Based on analysis of the human Ssu72-symplekin structure (28), several residues of Ssu72 with surface-exposed side chains were chosen for substitution. A cluster of negatively charged residues (positions 171-181) were chosen under the assumption that they may interact with positively charged residues of TFIIB c1 important to interaction (Fig. 3C). Residues around the interface with symplekin (28) were also chosen under the hypothesis that they may form a similar binding interface with TFIIB.
Ssu72 surface residues that were substituted for pBpa included Glu 127 , Gln 128 , Thr 130 , Cys 131 , Pro 133 , Asp 169 , Glu 173 , Leu 188 , Phe 193 , and Tyr 194 . Strong UV-dependent cross-linking to TFIIB was observed for all of the pBpa substitutions except E127pBpa, L188pBpa, and Y194pBpa (Fig. 4A). In agreement with the results from the pulldown assays, no cross-linking was observed with the K189E/R193E mutant of TFIIB c that loses the Ssu72-binding capacity (Fig. 4B). The photocross-linkable substitutions are clustered on an Ssu72 surface on the back of its substrate-binding site, outlining an elongated TFIIB binding surface (Fig. 4C, green dashed outline).
Residues 128 -133 of Ssu72 form a loop that is partially buried in the Ssu72-symplekin complex (28) (Fig. 4C and Fig. S5). Residue Asp 169 sticks out of a helix opposite to this loop and approaches residue Pro 133 . Residue Glu 173 is further down in the same helix and both Asp 169 and Glu 173 are a part of a negatively charged, surface-exposed patch that also includes residues Glu 171 , Glu 174 , and Glu 181 (Fig. 4C, green sticks). When any of these residues are mutated to pBpa, cross-linking to TFIIB c was observed (Fig. 4D). Residue Phe 193 was known to be critical for symplekin binding, and it now appears to contribute to the TFIIB-binding surface as well ( Fig. 4C and Fig. S6) (28). Residues Glu 127 and Leu 188 are located close to but outside of the TFIIB-binding surface, and residue Tyr 194 is partially buried during Ssu72 folding (Fig. 4C, yellow sticks). These facts not only account for the negative photocross-linking results upon UV illumination for the corresponding pBpa substitutions but also support the remarkable specificity of photocross-linking offered by pBpa. Thus, three discrete and spatially proximate patches (i.e. regions centered on residues 128 -133, residues  Fig. 1A) with human TFIIBc from the TFIIB c -TBP c -DNA structure (PDB code 1C9B, TFIIBc in gray schematic, DNA in multicolor schematic, and TBP c omitted). Cyclin-like domain 1 of molecule D was aligned with that of TFIIBc from 1C9B. In the subsequent panels, alignment follows the same scheme, except that the second TFIIB molecule was first aligned with TFIIBc from 1C9B to allow for modeling of the DNA. C, alignment of molecule D with TFIIBc from T. brucei (PDB code 3H4C; magenta). D, alignment of molecule D with human TFIIBc determined by NMR (PDB code 1TFB; cyan). The same TATA box DNA molecule from B is shown in C and D as well to guide the view. The purpose of the structural alignment in B, C, and D is to illustrate the conformational plasticity but not the actual new conformations of TFIIB c . E, electromobility shift assay with E4 promoter DNA, TBP c , and TFIIB variants.
To support the conservation of Ssu72-TFIIB binding, we further showed that the pBpa substitutions of human Ssu72 within the three TFIIB binding patches cross-linked to yeast TFIIB c . The cross-linking was abolished when the K201E/ K205E mutant of yeast TFIIB c was attempted (Fig. 4E), indicating that the binding mode between TFIIB and Ssu72 is highly conserved across species, although one of the three patches exhibits poor sequence conservation (Fig. 4F).

TFIIB and symplekin bind to overlapped Ssu72 surfaces
Ssu72 is known to form a stable complex with the N-terminal region (residues 30 -360) of symplekin (referred to below as Symp-N) (28). In our pulldown assays, TFIIB was unable to interact with Ssu72 in the presence of Symp-N, indicating that the binding sites for TFIIB and Symp-N overlap on the Ssu72 surface (Fig. 5A). To further delineate the competitive binding of TFIIB and Symp-N to Ssu72, we utilized a competition photocross-linking assay. Ssu72 with the pBpa substitutions were prebound with an excess of either TFIIB c or Symp-N, incubated with the competitor protein, and UV-cross-linked. Symp-N strongly cross-linked to Ssu72 C131pBpa (Fig. 5B, lane 2). When cross-linked in competition with TFIIB c , only the crosslinked Ssu72-Symp-N but not Ssu72-TFIIB complex was visible (Fig. 5B, lane 6), which indicates that Symp-N outcompetes TFIIB in Ssu72 binding, probably due to a higher binding affinity (Fig. 5A). Notably, Ssu72 F193pBpa cross-linked to Symp-N in a much weaker manner compared with Ssu72 C131pBpa (Fig. 5B, compare lanes 2 and 8), possibly because the former disturbs Symp-N binding (28). The weaker interaction between Ssu72 F193pBpa and Symp-N allowed TFIIB c to cross-link to Ssu72 to a similar degree in both the presence and absence of an excess of Symp-N (Fig. 5B, compare lanes 10 and 12). In a similar experiment, the K185A mutant of Symp-N that weakens A, cross-linking assays with TFIIBc-FLAG and Ssu72-pBpA substituted mutants, visualized by Western blotting with FLAG and His 6 antibodies. B, cross-linking of TFIIBc-FLAG mutant (Mut) with Ssu72-pBpA variants, visualized by His 6 antibody Western blotting. C, modeling of TFIIB-interacting residues determined by pBpA cross-linking onto the structure of human Ssu72 (from the Ssu72-symplekin structure, PDB code 3O2Q). The TFIIB-interacting surface is labeled by green outline, and residues essential for symplekin binding are labeled by blue outline. D, cross-linking assays of TFIIBc-FLAG and Ssu72-pBpA substituted mutants from the negatively charged helix region as visualized by Western blotting with FLAG and His 6 antibodies. E, cross-species cross-linking of S. cerevisiae TFIIBc (y) with human Ssu72 (h) visualized by Western blotting. The topmost bands in the anti-FLAG blot are bands that are present regardless of UV illumination (Nonspecific). F, sequence alignment of Ssu72 from human (Hs), fly (Dm), and yeast (Sc). Green plus signs indicate pBpA-substituted residues in humans that cross-link to TFIIBc, whereas yellow minus signs are residues that do not cross-link to TFIIB. Orange triangles, residues critical for symplekin interaction.

TFIIB-Ssu72 interaction and promoter-terminator cross-talk
Ssu72 binding also allowed photocross-linking between Ssu72 and TFIIB c (Fig. 5C, lane 6) (28). Overall, our results indicate that Ssu72 contains an overlapping yet not identical binding surface for TFIIB c and symplekin.
Although binding to an overlapped Ssu72 surface, unlike symplekin, TFIIB did not enhance the intrinsic phosphatase activity of Ssu72. In addition, TFIIB did not dramatically influence the stimulation of Ssu72 catalysis by symplekin, even in large excess (Fig. S7), confirming that TFIIB cannot outcompete symplekin for Ssu72 binding, at least at the concentration used in our assays.

Ssu72-binding site on TFIIB c overlaps with the VP16AD-and TFIIB n -binding site
Two additional interactions mediated by TFIIB c were noted to influence transcription initiation, including VP16AD acidic activator binding and TFIIB n binding, in the context of a TFIIB intramolecular complex. Because Ssu72 binds to a similar region of TFIIB c as VP16AD and TFIIB n , we further investigated such functional cross-talk with the photocross-linking competition assays.
We first established that TFIIB c could be photocross-linked to an N-terminal SUMO fusion of TFIIB n (residues 1-106) with a pBpa substitution at residue Trp 52 located in the finger/reader region (Fig. 6A, compare lanes 1 and 2). In the competition assays, we next observed that unlabeled Ssu72 decreased TFI-IB c -TFIIB n cross-linking compared with the GST control, probably due to competitive binding of Ssu72 and TFIIB n to TFIIB c (Fig. 6A, compare lanes 3 and 4 with lanes 6 and 7). In a reciprocal competition cross-linking experiment, GST-tagged TFIIB n could barely disrupt TFIIB c -Ssu72 cross-linking compared with the GST control (Fig. 6B, compare lanes 4 and 6 with lane 12), indicating that TFIIB n displays a comparable or a lower binding affinity toward TFIIB c when compared with Ssu72. Conversely, GST-tagged VP16AD efficiently prevented cross-linking between Ssu72 and TFIIB c (Fig. 6B, compare  lanes 8 and 10 with lanes 4 and 6). Accordingly, we conclude that Ssu72 may open up the inhibitory TFIIB c -TFIIB n intramolecular complex to facilitate pol II transcription with a similar mechanism used by VP16AD (16 -21).

A computational model predicts the interactions of TFIIB c and Ssu72
To further elucidate how the residues identified by our experiments interact on the TFIIB c -Ssu72 interface, we devised a computational approach combining molecular dynamics (MD) simulations and knowledge-based protein-protein docking. Our approach successfully generated an interface model that agrees with all experimental observations. This model suggests a mechanism whereby Ssu72 activates TFIIB for reinitiation. Moreover, the model includes additional, hypothetical TFIIB c and Ssu72 interactions that may represent secondary binding interfaces.

TFIIB-Ssu72 interaction and promoter-terminator cross-talk
We first performed MD simulations of TFIIB c and Ssu72 separately. We observed that whereas Ssu72 is a rigid protein, TFIIB c is highly flexible (Fig. S8), in line with our structural observations analyzed above (Fig. 1B). Therefore, we extracted TFIIB c structures from the MD simulations that may potentially bind to Ssu72 (example in Fig. S9) and then performed protein-protein docking to generate the binding models. Models with low HADDOCK score (less negative) and experimental violation were disregarded, resulting in a final high-quality model that best fits the experimental results.
In this model, the cross-linked residues on Ssu72 are very close to TFIIB c1 . A 60-ns MD simulation of this model revealed that the average D2 distance of all cross-linked residues is well below the threshold value (Fig. 7A), indicating that this binding interface is rather stable. Consistent with the mutagenesis experiment on Lys 200 , Arg 189 , and Arg 193 of TFIIB c , four salt bridges are found on the interface: Lys 200 -Glu 121 , Lys 200 -Asp 117 , Lys 189 -Glu 167 , Arg 193 -Glu 169 , with the TFIIBc residues underlined (Fig. 7B, bottom). These salt bridges are mostly mediated by TFIIB c1 , consistent with its dominant role in Ssu72 binding as shown by our pulldown assay (Fig. 3B). Additional interactions include the salt bridges Lys 152 -Glu 127 , Lys 188 -Asp 173 , Arg 290 -Glu 181 , Lys 272 -Glu 174 , and Arg 286 -Glu 178 . The last three interactions involve residues of TFIIB c2 . Probably, these interactions are weaker compared with those of TFIIB c1 because TFIIB c2 is not sufficient to pull down Ssu72 (Fig. 3B), and mutation of TFIIB c1 abolishes interaction with Ssu72 (Figs. 3C and 4B).

Discussion
Whereas the role of TFIIB in transcription initiation is well established, its role in other processes such as gene looping is not clear. The current study sheds light on possible new roles of TFIIB in this process by revealing a structure of human TFIIB c that adopted two new conformations. We also examined interactions between TFIIB c and Ssu72 and other factors known to interact with these two proteins, such as VP16AD and symplekin, respectively.
The structure of human TFIIB c displays the dynamic nature of the protein and indicates that conformational plasticity of TFIIB may underlie a mechanism for functional control. Apo-TFIIB is in an autoinhibited, "closed" conformation where TFIIB n blocks the TBP and the TATA DNA-binding region of TFIIB c (16 -21) (Fig. 8, "Closed" Autoinhibited state). VP16AD displaces TFIIB n by interacting with an overlapping surface on TFIIB c , promoting the structural conversion of TFIIB to an "open" conformation (20) that is then stabilized by TBP, TATA DNA, and other components of the transcription machinery during initiation. Both conformations of TFIIB c revealed by our study are essentially transcriptionally inactive because the two cyclin-like domains and C-linker are not aligned to interact with TBP and the TATA DNA due to intrinsic conformational flexibility, although structural elements for these functional interactions are all surface-exposed. The structural transition from an inactive conformation to an active conformation may be subjected to additional cellular regulation.
We showed that Ssu72 predominantly interacts with the TATA-binding region of cyclin-like domain 1 of TFIIB c (TFIIB c1 ) and that residues Arg 189 and Arg 193 are critical for binding (Fig.  3C). This region is also a binding site of VP16AD, suggesting that Ssu72 could function to transition TFIIB from a closed to an open conformation in a similar manner (Fig. 8). Indeed, similar to VP16AD, Ssu72 can easily outcompete the binding of TFIIB c to TFIIB n , indicating that it can at least relieve TFIIB from autoinhibition (Fig. 8). In addition, our computational study provides a possibility that Ssu72 may further form a stable complex with the active form of TFIIB c (Fig. 8, "Open" Active state). The open active state of TFIIB c maintained by Ssu72 may facilitate formation of the TFIIB-TBP-TATA ternary complex through a "hand-off" model. This process may occur during transcription reinitiation in the context of a gene loop, through

TFIIB-Ssu72 interaction and promoter-terminator cross-talk
a reinitiation scaffold complex on a TATA promoter that contains TBP but not TFIIB (14). The hand-off model in transcriptional regulation has also been proposed previously for transcriptional activators (57).
Interaction between TFIIB and Ssu72 is well conserved from yeast to flies and humans (Fig. 3D), suggesting a likely functional conservation. TFIIB binds to Ssu72 over an extended surface that includes a negatively charged helix, a loop, and a region near the C terminus of the protein. The latter two TFIIBbinding regions overlap with the binding surface of symplekin, which precludes TFIIB from binding to Ssu72. The Ssu72symplekin interaction is strong, but it can be disrupted by a point mutation of Phe 193 of Ssu72 that does not impede TFIIB binding. Interestingly, residue Phe 193 is close to several conserved serine, threonine, and tyrosine residues. Although we have not found a known phosphorylation motif in this area, it is possible that phosphorylation or other posttranslational modification of the C terminus of Ssu72 could be used to weaken the interaction with symplekin such that TFIIB binding is allowed.
The fact that TFIIB and Ssu72 physically interact yet are involved in disparate steps of transcription, initiation and termination, respectively, is intriguing. In the context of promoter-terminator gene looping, one can imagine a mechanism in which Ssu72 dephosphorylates the CTD of pol II during transcription termination, recruits TFIIB, and keeps it in an open conformation to allow for reassembly of PIC on the promoter for subsequent rounds of transcription. This hypothetical mechanism would explain why both TFIIB and Ssu72 are required for gene looping, and it can be further tested with separation-of-function mutants in future studies.

Protein expression and purification
All proteins were produced recombinantly in Escherichia coli. DNA sequences for each protein were cloned into a pET28a vector (containing an N-terminal His 6 tag), a SUMO-pET28a vector (containing a His 6 SUMO tag), or pGEX4T1 (containing an N-terminal GST tag). All mutations were introduced by site-directed mutagenesis of the parental plasmid. Expression plasmids were transformed into BL21*(DE3)-Rosetta II cells and cultured in LB medium supplemented with 50 g/ml chloramphenicol and either 50 g/ml kanamycin or 100 g/ml ampicillin and cultured at 37°C until reaching an A 600 of 0.6, at which point a final concentration of 0.5 mM isopropyl 1-thio-␤-D-galactopyranoside was added, and cells were incubated overnight at 16°C.
All TFIIB proteins, with the exception of GST-TFIIBn, were purified by nickel chromatography and contained a His 6 -SUMO tag and C-terminal FLAG tag. Cell pellets were resuspended in lysis buffer (20 mM HEPES, pH 8.0, 500 mM NaCl, 10 mM imidazole, and 2 mM ␤-mercaptoethanol) and lysed by sonication. Following centrifugation, the cleared lysate was applied to HisPure nickel resin (Thermo Scientific), the resin was washed with 20 column volumes of lysis buffer, and proteins were eluted with elution buffer (20 mM HEPES, pH 8.0, 500 mM NaCl, 300 mM imidazole, and 2 mM ␤-mercaptoethanol). Proteins that contained the N terminus of TFIIB (i.e. TFIIB FL and TFIIB n ) additionally contained 10 M zinc acetate in all buffers. The His 6 -SUMO tag was removed by Ulp1 cleavage overnight during dialysis into low imidazole buffer (20 mM HEPES, pH 8.0, 300 mM NaCl, 10 mM imidazole, and 2 mM ␤-mercaptoethanol). Proteins were then reapplied to a new nickel column to remove Ulp1 and contaminants, and the flow-through was collected. Proteins were dialyzed into a low-salt buffer (HEPES, pH 7.0 or 8.0, 100 mM NaCl, 10 mM imidazole, and 2 mM ␤-mercaptoethanol) and then applied to either a HiTrap heparin column (human TFIIB FL wild type and mutants) or a MonoS 5/50 column (human TFIIBc, TFIIB cyclin-like 1, and TBP c ) and eluted with a gradient of NaCl (both columns were from GE Healthcare). The crystallized human TFIIBc that lacked a C-terminal FLAG tag was concentrated following the MonoS 5/50 column and then loaded onto a Superdex 200 10/300 GL column (GE Healthcare) in 20 mM HEPES, pH 8.0, 150 mM NaCl, and 1 mM DTT. Because TFIIB cyclin-like 2 and TFIIBc mutants (human K189E/K200E, R185E/R193E, R185E/K200E, K189E/R193E; yeast H197E/N212E and K201E/K205E) failed to bind to heparin, they were passed through a MonoQ 5/50 column (GE Healthcare) to remove contaminants, and the flow-through was collected.
All Ssu72 and N-terminal symplekin (residues 30 -360) proteins contained an N-terminal His 6 tag and were purified by nickel chromatography followed by dialysis into a low-salt

TFIIB-Ssu72 interaction and promoter-terminator cross-talk
buffer (HEPES, pH 8.0, 100 mM NaCl, and 2 mM ␤-mercaptoethanol) and a final MonoQ purification step. To assemble the Ssu72-symplekin complex, Ssu72 was mixed with symplekin in a 2:1 molar ratio, incubated on ice for 1 h, and then separated on a Superdex 200 10/300 GL column in 20 mM Tris, pH 8.5, 200 mM NaCl, and 10 mM DTT. Fractions containing the complex were pooled, concentrated, and stored at Ϫ80°C until use.
Cells containing GST fusion proteins (TFIIBn and VP16AD) were lysed by sonication in GST lysis buffer (50 mM Tris, pH 8.0, 150 mM NaCl, and 5 mM DTT), and clarified lysate was bound to glutathione-agarose (Thermo Scientific). The resin was washed with 20 column volumes of GST lysis buffer and eluted with lysis buffer containing 20 mM reduced glutathione. The proteins were then applied to a MonoQ 5/50 column and eluted over a gradient of NaCl. Following purification, all proteins were concentrated, flash-frozen in liquid nitrogen, and stored at Ϫ80°C until needed.
For Ssu72 proteins containing pBpA mutations, the Ssu72 plasmid containing the amber mutation of interest was transformed into BL21* (DE3) cells containing the pEVOL-pBpA plasmid that encodes for an evolved aminoacyl tRNA synthetase and tRNA that incorporate pBpA into the amber suppression site (55). Cells were cultured at 37°C in 100 ml of 2XYT medium supplemented with 50 g/ml chloramphenicol and 50 g/ml kanamycin. At an A 600 of 0.3, a final concentration of 0.2% arabinose was added to the culture. At an A 600 of 0.6, a final concentration of 1 mM isopropyl 1-thio-␤-D-galactopyranoside and 1 mM pBpA were added to the culture, and the culture was then allowed to incubate overnight at 30°C. Proteins were purified in a manner similar to that described for Ssu72 by nickel and Q Sepharose Fast Flow (GE Healthcare) chromatography. SUMO-TFIIBn-W52pBpA, containing an N-terminal His 6 tag, was expressed in the same manner as Ssu72-pBpA proteins. It was purified by nickel chromatography in a manner similar to other His-tagged proteins, with the addition of 10 M zinc acetate in all purification buffers.

Protein crystallization, data collection, and structure solution and refinement
Human TFIIBc was crystallized at a protein concentration of 7 mg/ml in the hanging drop format, over a well solution containing 100 mM sodium acetate trihydrate, pH 4.5, and 2 M ammonium sulfate. Sucrose at a concentration of 3% (w/v) was used as an additive to improve crystal quality. Crystals were cryoprotected in mother solution supplemented with a final concentration of 30% glycerol and plunged into liquid nitrogen. Diffraction data were collected at beamline 191D of the Advanced Photon Source at Argonne National Laboratory. Data were collected over a total of 180°with oscillation of 1°and an exposure time of 2 s/degree. Diffraction data were processed with HKL2000 (58), and TFIIBc was found to be in space group P2 1 2 1 2 1 (R meas for the lowest-resolution shell (50.00 -9.21) is 6.0%). The structure of TFIIBc was solved by using the separate cyclin-like 1 and cyclin-like 2 domains from the TFIIB c -TBP c -DNA structure (8) using MOLREP (59). Manual building of connecting loops was completed with Coot (60), followed by refinement with PHENIX (61). Figures were created with PyMOL (PyMOL Molecular Graphics System, Version 1.8.6.2., Schroedinger, LLC, New York).

Isothermal titration calorimetry (ITC)
ITC was performed on a VP-ITC instrument (GE Healthcare) at 25°C in 25 mM HEPES, pH 8.0, 150 mM NaCl, 0.5 mM tris(2-carboxyethyl)phosphine. Ssu72 (1 mM) in the syringe was titrated into TFIIBc (50 M) in the cell over 30 injections, and data were background-corrected using a control titration where buffer was injected into Ssu72. Data were analyzed with NITPIC and Sedphat software (62). We noticed slight precipitation after the titration, which might cause an inaccuracy of the protein concentrations and the estimate of K d .

Pulldown assays
For FLAG pulldown assays, an amount of 75 g of TFIIBc variant with a C-terminal FLAG tag was mixed with 150 g of N-terminal tagged Ssu72 in binding buffer (50 mM Tris, pH 8.0, 150 mM NaCl, and 0.1% Tween) and bound to 50 l of anti-FLAG-agarose (Sigma-Aldrich) at room temperature for 1 h with rotation. For the assay conducted in Fig. 5A, TFIIBc-FLAG was prebound to the resin, followed by extensive washing before the addition of other proteins. The resin was then centrifuged for 3 min at 500 rpm at 4°C, the supernatant was removed, and the resin was washed three times with 1 ml of binding buffer. Proteins were then eluted with 100 mM glycine, pH 3.5, neutralized with 1 l of 1 M Tris, pH 8.0, and run on SDS-PAGE. Protein interactions were analyzed by staining with Coomassie Brilliant Blue and/or Western blotting with an anti-His 6 antibody (Abcam).

Site-specific UV-cross-linking analysis
Amounts of 20 g of TFIIBc-FLAG and 40 g of His 6 -Ssu72-pBpA mutant were mixed in assay buffer (20 mM Tris, pH 8.0, 150 mM NaCl, 1 mM DTT) in a final volume of 50 l and incubated on ice for 20 min. An amount of 25 l was removed from each sample as a no UV control. Samples were then UV-crosslinked on ice for four cycles at energy level 9999 in a Spectrolinker cross-linker (Krackeler Scientific), with 5 min of incubation on ice between cycles. 6.25 l of 6ϫ SDS dye was added, samples were boiled, and 12.5 l was run on SDS-PAGE. Crosslinking was analyzed by separate Western blots with an anti-FLAG antibody and an anti-His 6 antibody (Abcam). For competition cross-linking with TFIIBn and VP16AD (Fig. 4), TFIIBc was preincubated with the molar excess of competitor, as indicated in the figure, for 15 min on ice. Samples were then processed as described above. For competition cross-linking between TFIIBc, Ssu72, and symplekin (Fig. 5), 40 g of Ssu72 was preincubated with 30 g of symplekin for 20 min on ice. An amount of 60 g of TFIIBc was then added, and samples were then processed as described above.

EMSA
The template strand DNA oligonucleotide containing the E4 promoter sequence (TTTTTAGTCCTATATATACTCGC TCT) was annealed to the reverse complement strand containing a 3Ј-end Cy5 label (Sigma-Aldrich). TFIIB FL , TBP c , and DNA were mixed at equal molar concentrations (300 nM) in

Molecular dynamics simulation of TFIIB c
The two available crystal structures of TFIIB c were used as initial structures for MD simulation. The first one is TFIIB c in complex with TBP and DNA promoter (PDB code 1C9B) (8). The other one is the crystal structure of apo-TFIIB c (this report).
For TFIIB c from PDB entry 1C9B, only the TFIIB c structure was used. For apo-TFIIB c crystal structure, only molecule D was used. Both structures were solvated in a dodecahedron box with 17,779 TIP3P (63) water molecules. The system also contains 51 Na ϩ and 61 Cl Ϫ ions to create a neutral environment with 0.15 mol/liter salt concentration. The GROMACS version 4.5 (64) simulation package and AMBER 99SB-ILDN (65) force field were used to carry out the succeeding steps. Long-range electrostatic interactions were treated by the particle mesh-Ewald method (66). Electrostatic and van der Waals shortrange interactions were calculated using a cutoff of 10 Å.
Energy minimization was performed using the steepest-descent method, followed by NVT equilibration for 100 ps with position restraints on the heavy atom of TFIIB c at 300 K using V-rescale thermostat (67), which is used in all MD simulations in succeeding steps. At this step, the positions of heavy atoms were restrained with a force constant of 1000 kJ mol Ϫ1 nm Ϫ1 , and all bonds in protein molecules and water were constrained using the LINCS algorithm (68). NPT equilibration was then carried out using the Parrinello-Rahman barostat (69), to a reference pressure of 1 bar until the average pressure reached around 1 bar.
Eight production MD simulations starting from TFIIB c structure from PDB entry 1C9B were performed in the NVT ensemble with the length of each simulation ranging from 50 to 300 ns, with an accumulated time of 1600 ns for TFIIB c structure from PDB entry 1C9B. Five production MD simulations starting from apo-TFIIB c were performed in the NVT ensemble with the length of each simulation ranging from 100 to 400 ns, giving an accumulated time of 1300 ns. In the succeeding MD simulations, we used the same simulation package, force field, non-bonded interaction calculation method, barostat, thermostat, and bond constraint.
TFIIB c is an inherently flexible protein, as shown by the large root mean square fluctuation values for most of the backbone atoms (Fig. S8A). Because we obtained many TFIIB c structures other than the crystal structures, we selected the starting models based on the following criteria. In TFIIB c1 , residues 189, 193, and 200 are likely to interact with Ssu72 (Fig. 3C). Given that residue Ala 281 in TFIIB c2 was shown previously to bind VP16AD and that Ssu72 and VP16AD bind to an overlapping surface of TFIIB c (Fig. 6B), we assumed that residues 281-284 of TFIIB c are located on the binding interface of the TFIIB c -Ssu72 complex. Due to the flexibility of TFIIB c , these two clus-ters of interacting residues on TFIIB c1 or TFIIB c2 can be on the same plane (coplanar) (Fig. S9A) or not (Fig. S9B). To choose TFIIB c structures, clustering for simulations from each crystal structure was performed by K-center clustering (70) to 50 states. Eighteen structures were chosen based on 1) the coplanarity of the interacting residues that mimics the active conformation of TFIIB c when it binds to TBP and the TATA DNA and 2) the distance between the center of mass of TFIIB c1 and TFIIB c2 being less than 40 Å. The crystal structure of TFIIB c from PDB entry 1C9B was also used for protein-protein docking, which resulted in a total of 19 starting structural models of TFIIB c .

Molecular dynamics simulation of Ssu72
The starting structure for MD simulation was taken from PDB entry 3O2Q (28). It was solvated in a dodecahedron box with 11,321 TIP3P water molecules. The system also contains 44 Na ϩ ions and 34 Cl Ϫ ions to neutralize the system and creates a 0.15 mol/liter salt environment. Energy minimization was performed using the steepest-descent method, followed by NVT equilibration for 100 ps with position restraints (force constant ϭ 1000 kJ mol Ϫ1 nm Ϫ1 ) on the heavy atom of Ssu72 at T ϭ 300 K with coupling constant ϭ 0.1 ps. NPT equilibration was then carried out to a reference pressure of 1 bar with coupling constant ϭ 0.2 ps until the average pressure reached around 1 bar. Six production NVT MD simulations with different random initial velocities were performed for 50 ns each, giving an accumulated time of 300 ns. Because Ssu72 is a rigid protein (as shown by the small root mean square fluctuation value of all of the backbone atoms; Fig. S8B), we used the crystal structure to do protein-protein docking.

Protein-protein docking using HADDOCK
The 19 selected structures from MD simulation of TFIIB c were used as input for protein-protein docking, along with the crystal structure of Ssu72 (PDB code 3O2Q). The cross-linked residues on Ssu72 were used as active residues input into the HADDOCK web server (71) (i.e. Pro 133 , Phe 193 , Gln 128 , Thr 130 , Cys 131 , Glu 169 , Asp 173 , Glu 174 , Glu 181 . The interacting residues of TFIIB c , which were either determined by mutagenesis or prior knowledge of interaction with VP16AD, were also used as active residues (i.e. Lys 189 , Arg 193 , Lys 200 , Ala 281 , Asp 282 , Val 283 , Thr 284 ). The intermolecular and intramolecular interactions were determined by CNS 1.3 (72,73) along with PARALL-HDG5.4 (74,75) as force field for protein. For each docking simulation, the two input molecules were separated for 25 Å and rotated randomly around their centers of mass, followed by rigid body minimization and docking. Two hundred best structures from rigid body docking were chosen for semiflexible refinement, based on the weighted sum of several of energy terms (i.e. electrostatic, van der Waals, desolvation, ambiguous interaction restraints, and total buried surface area). Of 1000 structures generated by rigid body docking, 200 structures were used for final solvated refinement and clustering. Clustering was done based on fraction of native contacts using 0.75 similarity as the threshold. For each cluster of every docking simulation, we picked the four models with the lowest HADDOCK scores for the following analysis.

TFIIB-Ssu72 interaction and promoter-terminator cross-talk Selection of docking model
HADDOCK score and experimental results were considered to select the best model. The HADDOCK score used in the succeeding steps excluded the E AIR part of the score. For consideration of the cross-linking experiment, we defined a distance that measured the likelihood of cross-linking reaction (D2 distance, shown in Fig. S10). The D2 distance is defined as the distance between the C-␣ atom of cross-linked residues on Ssu72 and the C-␣ atom of the nearest residue on TFIIB c that can potentially form a cross-linking bond. The cross-linking reaction only occurs between pBpa and a non-quarternary carbon atom (i.e. a carbon atom that binds to at least one hydrogen atom). The threshold for the D2 distance was determined to be 15 Å. Any cross-linked residue that has a D2 distance violating the threshold is considered as 1 violation count. The contacts between residues 189, 193, 200, 281, 282, 283, and 284 on TFIIB c and Ssu72 residues were determined by using the g_contacts module in GROMACS (76), with a threshold of 3 Å. If any of these residues was not in contact, the violation count was 2 for residues 189, 193, and 200, giving priority to the mutagenesis experiment, and 1 for residues 281, 282, 283, and 284. The total experimental violation count was the sum of all of the violations. The plot of HADDOCK score versus experimental violation count is shown in Fig. S11A. All of the models with violation less than 1 were analyzed further. According to the pulldown data, the interaction between TFIIB c1 and Ssu72 is stronger than that between TFIIB c2 and Ssu72. Considering this result, we calculated the ratio of contacts between TFIIB c1 and Ssu72 and contacts between TFIIB c2 and Ssu72. The plot in Fig. S11B clearly shows one model having the largest ratio and a low HADDOCK score (represented as a star), hence fitting best to the pulldown assay experiment (Table S1).

MD simulation of TFIIB-Ssu72 complex
We solvated the TFIIB c -Ssu72 model in a dodecahedron box with a 1-nm distance between the closest point of the protein and the walls of the box. The protein was then solvated in TIP3P water molecules and an equal number of sodium and chloride ions to create a 0.15 mol/liter salt concentration.
The prepared system was then energy-minimized by the steepest-descent method, followed by NVT equilibration with T ϭ 300 K and coupling constant ϭ 0.1 ps as well as position restraints on heavy atoms of the protein (force constant ϭ 1000 kJ mol Ϫ1 nm Ϫ1 ) for 100 ps. Next, NPT equilibration was performed to reference pressure ϭ 1 bar and coupling constant ϭ 0.2 ps, until the average pressure reached 1 bar. The frame from MD simulation that has a volume equal to the average volume was chosen for a 200-ps NVT MD simulation. The last frame of this short MD simulation was the starting ensemble for the production MD simulations. Finally, MD simulation in an NVT environment was performed for 60 ns, with the first 10 ns excluded in analysis.