Inter- and Intrasubunit Interactions during the Formation of RNA Polymerase Assembly Intermediate*

We used yeast two-hybrid and in vitroco-immobilization assays to study the interaction between theEscherichia coli RNA polymerase (RNAP) α and β subunits during the formation of α2β, a physiological RNAP assembly intermediate. We show that a 430-amino acid-long fragment containing β conserved segments F, G, H, and a short part of segment I forms a minimal domain capable of specific interaction with α. The α-interacting domain is held together by protein-protein interactions between β segments F and I. Residues in catalytically important β segments H and I directly participate in α binding; substitutions of strictly conserved segment H Asp1084 and segment I Gly1215 abolish α2β formation in vitro and are lethal in vivo. The importance of these β amino acids in α binding is fully supported by the structural model of the Thermus aquaticus RNAP core enzyme. We also demonstrate that determinants of RNAP assembly are conserved, and that a homologue of β Asp1084 in A135, the β-like subunit of yeast RNAP I, is responsible for interaction with AC40, the largest α-like subunit. However, the A135-AC40 interaction is weak compared with the E. coli α-β interaction, and A135mutation that abolishes the interaction is phenotypically silent. The results suggest that in eukaryotes additional RNAP subunits orchestrate the enzyme assembly by stabilizing weak, but specific interactions of core subunits.

Cellular RNA polymerases (RNAPs) 1 are large, multisubunit enzymes. A typical prokaryotic RNAP core contains 5 polypeptides with a total molecular mass of ϳ400 kDa. Core RNAP from eukaryotes and archaea contain 10 -14 subunits with a total molecular mass in excess of 500 kDa. Sequence alignments of RNAP subunits reveal extensive similarities; each of the two largest RNAP subunits, which are the most evolutionarily conserved, contains 8 -9 colinear segments with many invariant amino acids (1,2). RNAPs from different sources are also homologous structurally; low resolution (16 -35 Å) threedimensional models of Escherichia coli, and RNAP II and RNAP I from yeast obtained by means of electron crystallography reveal significant similarities (3)(4)(5).
Evolutionarily conserved subunit segments probably form distinct functional domains common to all RNAPs. Genetic data support this notion; mutational changes of conserved residues selectively destroy distinct partial functions of the enzyme (e.g. the transition from abortive initiation to productive elongation (Ref. 6), or phosphodiester bond synthesis (Ref. 7)), but leave other functions unperturbed. However, isolated subunits themselves do not possess any of the partial functions of the whole enzyme (8). Therefore, RNAP functional sites are either formed allosterically upon the enzyme assembly, or are located at subunit interfaces. Thus, understanding inter-and intrasubunit interactions should provide insights in RNAP mechanism and regulation.
The ␤ and ␤Ј homologues are naturally fragmented in some archaea and cyanobacteria (14,15). The assembly pathway should be more complex in organisms with split ␤, ␤Ј homologues, but this has not been investigated. Our own work with E. coli defined four separable domains in ␤ and three in ␤Ј (16,17), that correspond to natural fragmentation sites in some of the archaea and cyanobacteria. The ability to generate functional E. coli RNAP using subunit domains greatly aided the study of RNAP assembly and biochemical functions of assembly intermediates. A combination of fragmented RNAP reconstitution, limited proteolysis, and protein-protein co-immobilization assays was used to demonstrate that determinants of specific ␣ binding reside in the C-terminal assembly-competent structural module of ␤, containing amino acids 907-1342 (conserved segments H and I; Ref. 18). However, similar analysis performed by the Ishihama group (19) suggested that the primary determinants of ␣ binding include conserved segments F and G, corresponding to E. coli ␤ amino acids 800 -900.
The discrepancy between the two sets of data illustrates problems associated with in vitro coimmobilization assays, which can be artifact-prone. Thus, alternative approaches are needed to characterize the ␣-␤ interaction in molecular detail. Here, we used the method of yeast two-hybrid analysis to analyze the interaction of the ␣ and ␤ subunits during ␣ 2 ␤ formation. Our results support the conclusion of Wang et al. (18) and directly implicate two strictly conserved amino acids in ␤ segments H and I in interactions with ␣.

MATERIALS AND METHODS
Proteins and Plasmids-Gal4-based two-hybrid plasmids pPC97 (bait plasmid) and pPC86 (prey plasmid) were described in Ref. 20.
␣ Derivatives-The two-hybrid plasmids expressing ␣ NTD (rpoA codons 1-235) were constructed by PCR cloning using E. coli expression plasmids containing either wild-type rpoA or rpoA harboring the R45A mutation as templates (18). E. coli plasmid expressing His 6 -␣NTD, and conditions for protein expression and purification are described in Ref. 21.
␤ Derivatives-The two-hybrid plasmids expressing various fragments of ␤ used in this work were constructed by PCR cloning using E. coli pMKSe2 expression plasmid (22) as a template. Site-specific PCR mutagenesis and standard recloning steps were used to introduce point mutations in yeast two-hybrid of E. coli expression plasmids. Fulllength ␤ and ␤Ј used in in vitro RNAP assembly studies were obtained from expression plasmids pMKSe2 and pT7␤Ј (23), respectively. Overexpression was induced with 1 mM isopropyl-␤-D-thiogalactoside at 30°C for 3 h. The proteins in inclusion bodies were then prepared according to Ref. 21.
Yeast RNAP Subunit Derivatives-The two-hybrid plasmids expressing various fragments of yeast RNAP subunits were constructed using appropriate primers by high fidelity PCR with yeast genomic DNA as a template. Plasmid pNOY302, which harbors the entire A135 gene, was provided by M. Nomura and is described in Ref. 24.
Yeast Strains and Techniques-Yeast were transformed using standard lithium acetate procedure, and transformants were selected on appropriate drop-out media. For two-hybrid assay, we used the LNY 327 tester strain provided by L. Neigeborn. This strain is identical to the PCY2 strain of Chevrai and Nathans (20). The interaction between two-hybrid constructs was scored using filter assays, or quantitative colorimetric assay using ortho-nitrophenyl-␤-galactoside as a substrate (25). The following formula was used to calculate ␤-galactosidase production: ␤-galactosidase activity (Miller units) ϭ 10,000 ϫ (OD 420 Ϫ 1.75 ϫ OD 550 )/(OD 600 ϫ t), where OD 420 , OD 550 , and OD 660 are optical density of the reaction at 420, 550, and 600 nm, respectively at time t (min) of the reaction.
Yeast tester strain NOY302-1a was provided by M. Nomura and has been described previously (26).
Ni 2ϩ -NTA-Agarose Co-immobilization Binding Assays-Protein complexes (about 10 g) were mixed with pre-equilibrated Ni 2ϩ -NTA-agarose beads (Qiagen) in 50 l of reconstitution buffer and incubated for 10 min at room temperature with gentle mixing. The beads were pelleted by centrifugation and washed twice with 500 l of buffer with 10 mM imidazole. The protein samples were then eluted from the beads with buffer containing 100 mM imidazole and analyzed by SDS-PAGE.
In Vitro Transcription-10 l of RNAP reconstitution reaction was incubated for 15 min at 37°C in the presence of 1 g of recombinant 70 subunit (23), and 50 ng of the T7A1 promoter-containing DNA fragment (22), 0.5 mM CpA, with or without 25 g/ml rifampicin. Reactions were initiated by the addition of NTP (final concentration 25 M ATP, CTP, and GTP, 1 M [␣-32 P]UTP (1000 Ci/mmol). Reactions proceeded for 15 min at 37°C, and were terminated by the addition of urea containing loading buffer. Reaction products were resolved by denaturing 20% PAGE and revealed by autoradiography.

RESULTS
The Interaction between E. coli RNAP ␣ and ␤ Can Be Detected in Yeast Two-hybrid System-For our research, it was critical to establish that the interaction between ␤ and the dimeric ␣ subunit could be studied by two-hybrid approach, and that mutations that decrease this strong interaction can be detected in yeast. As an initial test, we used the N-terminal domain of ␣ (NTD, amino acids 1-235) and the ␤ subunit fragment containing amino acids 711-1246, which specifically interact with each other in vitro (18). ␣-NTD was fused to the GAL4 DNA binding domain (Gal4-DB) of the bait plasmid pPC97, and ␤ 711-1246 was fused to the GAL4 activation domain (Gal4-AD) of the prey plasmid pPC86 (20). High levels of ␤-galactosidase were observed when both plasmids were present (intense blue color developed in less than 30 min in a standard filter assay; 440 Miller units of ␤-galactosidase activity in a quantitative liquid assay, see "Materials and Methods"); no activity was observed when either fusion plasmid was used separately, as expected (cells remained white after overnight incubation on a filter, less than 1 Miller unit of ␤-galactosidase activity). Importantly, when pPC97␣ 1-235 , carrying a mutation that changes ␣ Arg 45 for Ala and weakens the interaction with ␤ (12), was used, no activity was detected even in the presence of pPC86␤ 711-1246 (white color, less than 1 Miller unit of ␤-galactosidase activity, data not shown). Thus, we can detect the ␣-␤ interaction and amino acid changes that affect this interaction using the two-hybrid system. This result sets the stage for more detailed analyses of RNAP assembly described below.
Defining the Minimal Fragment of ␤ Capable of Interaction with ␣-Our starting ␤ two-hybrid construct, pPC86␤ 711-1246 , contained 535 rpoB codons and corresponded exactly to the smallest ␤ fragment that specifically interacted with ␣ in vitro (18). The ␤ 711-1246 fragment spans four universally conserved ␤ segments, F, G, H, and part of I. To determine the smallest ␤ fragment capable of interacting with ␣, we engineered pPC86␤ 711-1246 derivatives with deletions at either the beginning or the end of the ␤ moiety, and tested these new plasmids for their ability to elicit ␤-galactosidase production in yeast cells harboring pPC97␣  . The results are schematically presented in Fig. 1.
As can be seen, ␤ 800 -1246 hybrid interacted with the ␣ hybrid. However, further deletion in segment F, ␤ 812-1246 , abolished the interaction. At the C-terminal side, removal of ␤ amino acids 1231-1246 had no effect on interaction with ␣. However, the removal of additional 6 ␤ amino acids (1226 -1231) abolished the interaction. The results suggest that ␤ amino acids 711-800, which contain 80 non-conserved amino acids, and 10 amino acids from conserved segment F, and ␤ amino acids 1231-1246, which contain most of ␤ conserved segment I, including residue His 1237 , which forms the 5Ј-face of the catalytic center (27), are not necessary for specific interaction with ␣. We note, however, that these regions of ␤ may still contribute to the overall strength of the binding, since the two-hybrid interaction may be saturated and insensitive to minor changes in the strength of binding.
The two-hybrid construct containing ␤ amino acids 800 -1231 was the smallest construct that interacted with ␣ strongly (520 Miller units) and specifically (0.5 Miller units when used with ␣ R45A two-hybrid construct) (Fig. 1). Shorter constructs, constructed by site-directed mutagenesis, ␤ 812-1231 , ␤ 832-1231 , and ␤ 8oo-1225 , failed to interact with ␣ hybrid (Ͻ1 Miller unit). We also attempted to further narrow the interacting domain by performing nested Bal31 deletion mutagenesis at either the 5Ј or the 3Ј ends of the rpoB portion of pPC86␤ 800 -1231 , transforming mutated plasmids in yeast cells harboring pPC97␣, and selecting interacting (Lac ϩ ) colonies. The DNA sequence of several interacting ␤ plasmids was determined, and in all cases Bal31 deletions extended into the vector part of pPC86␤ 800 -1231, while rpoB codons 800 and 1231 were retained (data not shown). We conclude that ␤ 800 -1231 is the minimal ␤ fragment capable of ␣ binding.
The minimal interacting fragment of ␤, ␤ 800 -1231 , contains dispensable region II (␤ amino acids 907-1050) that is missing in some bacterial ␤ subunits and in homologues from eukaryotes and archaea (28). This region tolerates an artificial split at around position 950 (16), and is highly susceptible to trypsin attack at position 907 and 911 (28). Thus, dispensable region II likely demarcates two structurally independent ␤ modules. Earlier studies indicated that assembly-competent ␤ module, ␤ 950 -1342 , which contains conserved segment H and I, interacted with His 6 -␣-NTD (18). However, ␤ 950 -1342 binding was weak compared with the binding of the larger, ␤ 711-1342, fragment or full-sized ␤. Moreover, ␤ 950 -1342 interacted with His 6 -␣ WT and His 6 -␣ R45A in vitro with the same (low) efficiency (18). On the other hand, Nomura et al. (19) mapped the primary ␣-binding site to ␤ amino acids 737-904. We constructed pPC86-based two-hybrid plasmids with ␤ 800 -911 (conserved segments F and G), and ␤ 907-1231 , (conserved segments H and a small portion of I) and tested them for their ability to interact with ␣. Neither pPC86␤ 711-911 plasmid, nor pPC86␤ 907-1231 elicited ␤-galactosidase activity with pPC97␣ 1-235 suggesting that ␤ segments F and G, or segment H and I, alone do not interact with ␣ in our assay (data not shown).
Our results suggest that ␤ amino acids 801-811 and 1226 -1231 in conserved regions F and I, respectively, contribute, either directly or indirectly, to the interaction with the ␣ dimer. In the former scenario, both segment F and segment I residues interact with ␣ to result in strong binding; neither segment binds ␣ strongly enough for the interaction to be detected by two-hybrid assay. The latter scenario may involve an intrasubunit interaction between the two regions of ␤, one containing segment F and G, and another containing segments H and I. One can envision, then, that the interaction of the two ␤ fragments results in the formation of the structure which is fully capable of ␣ binding. Two-hybrid constructs containing ␤ 800 -911 and ␤ 907-1231 strongly interacted with each other (Fig. 2). In contrast, construct containing ␤ 812-911 did not interact with ␤ 907-1231 . Likewise, no interaction was detected between twohybrid constructs containing ␤ 800 -911 and ␤ 907-1225 . Thus, the residues required for ␣ binding are also required for intrasubunit interactions between two ␤ modules, consistent with the idea that segment F and I amino acids contribute to ␣ binding indirectly.
Point Mutations in ␤ Segment H and I Affect ␣ Binding in Vivo and in Vitro-In ␣, substitution of the strictly conserved Arg 45 interferes with the ␤ binding (12). We hypothesized that Arg 45 interacts with a negatively charged, conserved amino acid in ␤, and that this interaction is responsible for the ␣ 2 ␤ formation. Accordingly, we inspected the E. coli ␤ fragment between amino acids 800 and 1231 to see if such amino acids could be found. Segment G contains no conserved negatively charged amino acids; segment F contains one such residue, Glu 813 . This position was previously mutated to an Arg (29); the resulting RNAP had a catalytic defect but had no assembly defect, and therefore we did not study this position further. As can be seen from the alignment presented in Fig. 3A, segment H contains 4 negatively charged amino acids, Asp 1064 , Asp 1084 , Asp 1095 , and Glu 1114 , which are strictly conserved and thus fulfill our criteria for potential candidates involved in ␣ binding. Residue Asp 1064 was mutated previously in the course of a mutational study of the RNAP initiating site (30). The results of that study revealed that ␤ harboring the D1064A mutation was assembly-proficient, but the resulting RNAP was catalytically defective. Based on these results, position 1064 was excluded from our analysis. Segment I contains no conserved negatively charged amino acids. However, there are three Gly residues, at positions 1215, 1218, and 1228, which are strictly conserved. Previously, we had demonstrated that substitutions of evolutionarily conserved glycines for aspartic residues in the ␤Ј subunit result in assembly defects (31). Therefore, in addition to the three site-specific mutations changing two aspartates and one glutamate from segment H for alanines, we also substituted each of the three segment I glycines for aspartates. The mutations were engineered in pPC86␤ 800 -1231 background, and their effect on ␣ binding in two hybrid system was studied. The results are shown above the alignment on Fig. 3A. Of the six mutations tested, only two, D1084A and G1215D, completely abolished the interaction as judged by both the filter and the liquid ␤-galactosidase assays. G1218D showed weak, but clearly positive, interaction. The remaining mutations had no effect on ␣ binding.
To confirm the ␣ binding defects directly, we recloned all mutations into the pMKSe2(S531F) ␤-overproducing plasmid (22). The ␤ subunit expressed from pMKSe2(S531F) contains a rifampicin (Rif) resistance mutation changing Ser 531 to Phe. Therefore, Rif-sensitive host cells transformed with pMKSe2(S531F) grow on medium containing Rif. In contrast, cells transformed with pMKSe2(S531F) harboring the engineered segment H and I mutations did not form colonies in the presence of Rif, but were unaffected for growth in its absence. Thus, the rpoB mutations behaved as recessive lethals, a phenotype expected of assembly mutations.
The binding defect of ␤ mutants was also showed directly, by studying their ability to co-immobilize on Ni 2ϩ -NTA-agarose through protein-protein interactions with hexahistidine-tagged ␣NTD (Fig. 3B). In this experiment, the wild-type or mutant ␤ subunits were prepared from inclusion bodies and combined with recombinant ␣NTD in the presence of high concentration of denaturing agent guanidine HCl. The denaturant was dialyzed away at conditions favoring RNAP assembly, and reactions were allowed to bind to Ni 2ϩ -NTA-agarose beads. The beads were washed, and the bound protein was eluted with a buffer containing high concentration of imidazole. As can be seen, the wild-type ␤ was efficiently immobilized on the sorbent in the presence of tagged ␣NTD (Fig. 3B, lane 4), as expected. In contrast, most of ␤ harboring the D1084A and G1215D substitutions was found in the flow-through (lanes 6 and 10) and did not associate with ␣NTD (lanes 8 and 12). G1218D also caused a defect in ␣NTD binding, but, in agreement with the two-hybrid results, the extent of this defect was weaker than those of either D1084A or G1215D (compare lane 20 with lanes  8 and 12). Similar analysis revealed that in agreement with the two-hybrid result substitutions at positions 1095, 1114, and 1228 had no effect in the co-immobilization assay (Fig. 3B, lane 24; and data not shown). We conclude that ␤ positions 1084, 1215, and, to a lesser degree, 1218 are involved in ␣ binding, and that the lethal in vivo phenotypes exhibited by the rpoB genes harboring substitutions in these positions are likely explained by the defect in the ␣ 2 ␤ formation. In contrast, the lethal phenotype caused by substitutions at rpoB positions 1095, 1114, and 1228 must be due to alterations of some other essential RNAP function(s).
In principle, two classes of mutations can abolish proteinprotein interaction. The mutations of the first class distort the native structure, perhaps through an allosteric mechanism, and thus prevent the interaction. The second, most interesting class, is true interaction mutations, which affect favorable interactions that bring the two proteins together. We investigated the ability of the ␤ subunit harboring the D1084A and G1215D mutations to assemble into active RNAP in vitro in the presence of the wild-type ␤Ј and hexahistidine-tagged ␣NTD. The assembly reactions were examined by Ni 2ϩ -NTA coimmobilization (Fig. 3C). The addition of ␤Ј corrected the ␣NTDbinding defect caused by the D1084A mutation, and RNAP core enzyme was formed in good yield (Fig. 3C, lane 8). A steadystate in vitro transcription experiment revealed transcription activity in the RNAP assembly reactions containing ␤ with D1084A substitution (Fig. 3D, lane 3). This transcription was rifampicin-resistant (lane 4), proving that the activity was due to the mutant RNAP, which also carries the S531F Rif resistance mutation, and not due to contaminating wild-type RNAP from the host, which is rifampicin-sensitive (see lanes 5 and 6). This result implies that the D1084A mutation does not grossly distort ␤ structure and may be directly involved in interactions with ␣. In contrast, no assembled RNAP was detected in the in vitro RNAP assembly reaction containing His-tagged ␣, ␤Ј, and ␤ carrying the G1215D mutation (Fig. 3C, lane 12), and there was no transcription activity in G1215D assembly reactions (Fig. 3D, lanes 7 and 8). Thus, Gly 1215 appears to play a structural role, and its involvement in ␣ binding may be indirect.
Evolutionary Conservation of the ␤and ␣-like RNAP Subunit Interactions-In yeast, two ␣-like subunits, AC40 and AC19, form a heterodimer that is functionally equivalent to the ␣ 2 homodimer in prokaryotes (32). Interestingly, AC40 and AC19 are shared by RNAP I and RNAP III. Thus, interaction of the RNAP I ␣-like subunit, A135, and its RNAP III homolog, C128, with the AC40/AC19 heterodimer could determine the relative amount of RNAP I and RNAP III in the cell. We wanted to know (i) whether C-terminal A135 and C128 fragments contained determinants for interaction with the ␣-like subunits, as would be expected from our E. coli results, and (ii) whether C128 and A135 interact with the same ␣-like subunit in the heterodimer. Two-hybrid plasmids expressing the N-and C-terminal halves of A135 and C128 as defined by the archaeal split were constructed. The two-hybrid plasmids were then tested for their ability to elicit ␣-galactosidase production with complementary plasmids expressing either AC19 or AC40 hybrids. As control, we used the AC40 and AC19 pair, which had

FIG. 2. A strong intrasubunit interaction between ␤ segments F and I.
The structure at the top represents assembly-competent E. coli ␤ subunit fragment corresponding to archaeal ␤Ј subunit (17). The Gal4-AD hybrids containing the fragments of ␤ shown at the left were tested against the ␣-Gal4-DB ␤ fragments shown at the right, and the results are shown at far right.
previously been shown to interact in two-hybrid assay. The results can be summarized as follows. First, as expected, AC40 fused to Gal4-DB showed strong positive interaction with AC19-Gal4-AD (120 Miller units). Second, the N-terminal fragments of the ␤-like subunits, containing conserved segments A, B, C, D, and E (residues 1-728 and 1-717 in C128 and A135, respectively) did not interact with the ␣-like subunits. Third, C-terminal fragment of C128 (residues 729 -1149 conserved segments F, G, H, and I) also did not interact with either AC19 or AC40 hybrids, and thus no conclusions about the formation of RNAP III ␣ 2 ␤-like structure could be made. Finally, weak (13 Miller units) interaction between the C-terminal fragment of A135 (residues 717-1203) with the AC19 hybrid, but not with the AC40 hybrid was detected.
Since the ␣-like subunits AC19 and AC40 heterdimerize with high efficiency (32, see also above), the observed interaction between the A135 FGHI and AC19 two-hybrid constructs is likely due to A135 FGHI -Gal4DB interaction with AC40/AC19-Gal-AD heterodimer. We constructed an AC19 two-hybrid plasmid, which contained a double-alanine substitution at AC19 positions 78, and 79, corresponding to E. coli ␣ positions 44 and 45, and should have abolished A135 binding to AC19. The resulting construct was unaffected in A135 FGHI interaction (15 Miller units), suggesting that conserved residues in the ␣-motif Mutant ␤ subunits were overproduced, purified, and combined with hexahistidine-tagged ␣NTD at conditions favoring RNAP assembly. Reactions were loaded onto Ni 2ϩ -NTA-agarose beads (L) and the unbound protein removed (F). The beads were then washed with buffer containing 25 mM imidazole (W), and then the bound proteins were eluted with buffer containing 100 mM imidazole (E). The protein fractions were then analyzed by SDS-PAGE on a 10% gel. C, experiment was performed as in B in the presence of the recombinant ␤Ј subunit. Reactions were analyzed by SDS-PAGE on a 10% gel (top) or a 6% gel (bottom), to separate the ␤ and ␤Ј subunits. D, RNAP was assembled as in C; reactions were provided with purified 70 subunit, and the T7 A1 promoter-containing DNA fragment, in the presence, or in the absence of 25 g/ml rifampicin. Transcription was initiated by the addition of CpA primer, ATP, CTP, GTP, and [␣-32 P]UTP. Reaction products were resolved by denaturing 20% PAGE and visualized by autoradiography.
of AC19 are not involved in A135 binding, or A135 interacts with AC40. To establish the specificity of two-hybrid interaction between AC40/AC19-Gal-AD and A135 FGHI -Gal4DB, we engineered a two-hybrid construct that expressed the C-terminal fragment of A135 with D935A substitution, which corresponds to E. coli rpoB D1084A (Fig. 3A). No ␤-galactosidase was produced when the AC19 and A135 D935A hybrids were combined in the same yeast cell (1 Miller unit). We conclude that (i) the C-terminal portion of A135 interacts with the largest ␣-like subunit, AC40, and not with AC19, and (ii) A135 Asp 935 is required for this interaction.
In E. coli, rpoB mutations that abolish the strong ␣-␤ interaction behave as haploinviable (see above). To test the functional consequence of D935A substitution in A135, which abolishes the A135-AC19/AC40 interaction in the two hybrid assay, yeast plasmid expressing the mutant A135 gene was constructed. The phenotype of this gene was tested using haploid NOY408 -1a yeast tester strain (26). In this strain the chromosomal copy of A135 has been deleted, and the cells are kept viable because of the presence of a multicopy plasmid harboring an rDNA repeat under the control of GAL7 (RNAP II) promoter. Therefore, NOY408 -1a is only viable when galactose is present and does not grow in the presence of glucose, when the GAL7 promoter is repressed. However, when a plasmid expressing a functional copy of A135 is introduced, NOY408 -1a can grow in the presence of glucose, since rDNA is now transcribed by RNAP I. As can be seen, plasmid-borne A135 D935A , allowed robust growth on glucose (Fig. 4). Additional experiments established that the A135 D935A plasmid allowed growth on glucose at low temperature of 16°C, and at high temperature of 37°C (data not shown). We conclude that the A135 D935A gene is functional. DISCUSSION The impetus for this study came from the apparently contradictory results of Nomura et al. (19) and Wang et al. (18), who localized the ␣-binding determinants into non-overlapping fragments of E. coli ␤. Nomura et al. claimed that ␤ conserved segments F and G, harbor the primary ␣-binding site; Wang et al. localized the binding site to ␤ segments H and I; however, four conserved regions, F, G, H, and I, were required for strong and specific ␣ binding. The results of our work are consistent with Wang et al., and demonstrate that ␤ amino acids 1084 (conserved segment H) and 1215 (conserved segment I) are involved in ␣ binding. We also demonstrate that there exists a strong intrasubunit interaction between ␤ segments F and I, and that this interaction appears to be necessary for ␣ binding. Finally, we present evidence that, in yeast, the second largest, ␤-like RNAP I subunit A135 interacts with the larger ␣-like subunit, AC40, and, unexpectedly, that the A135 mutation that abolishes this interaction is viable. While this research was in progress, a 3.3-Å structure of RNAP core from Thermus aquaticus (Taq) appeared (33). In addition, a medium resolution structure of yeast RNAP II also became available (34). Below, we interpret our findings in the context of the structural model of the Taq enzyme.
Intrasubunit Interaction of between Segment F and I Residues Is Important for ␣ Binding-Our results demonstrate that the smallest fragment of E. coli ␤ capable of specific ␣ binding spans amino acids 800 -1231. The structural context of the corresponding fragment of Taq ␤ is shown in Fig. 5 (A-C). As can be seen, the fragment has an elongated shape and consists of two domains. One domain is formed entirely by conserved segment G and contains the so-called "flexible flap" element, which is thought to interact with RNA at hairpin-induced pause sites (35), and harbors the site of trypsin attack (28). The second domain contains the entire conserved segment H as well as segments F C and I N , which are far away from each other in the primary sequence. The second domain is involved in protein-protein interactions with one ␣ monomer, ␣1. The second ␣ monomer, ␣2, does not interact with ␤ and instead contacts ␤Ј.
The ␣-binding domain is held together by a 6-strand antiparallel ␤-sheet. Conserved segments F, H, and I contribute to this ␤ sheet, and stretches of amino acids corresponding to E. coli 800 -812 and 1226 -1231 each form interacting strands (Fig. 5D). We registered this intrasubunit interaction as a strong ␤-galactosidase production when hybrids containing E. coli ␤ 800 -907 and ␤ 911-1231 were combined. Removing either segment F or segment I amino acids disrupts the structure of the entire ␣-binding domain, and as a consequence abolishes ␣ binding.
Segment H and I Amino Acids Participate in ␣ Binding-The prototypical E. coli mutation that decreases the avidity of ␣ binding to ␤ changes ␣ Arg 45 which is evolutionary conserved (12). We hypothesized that, in ␤, the likely counterpart of ␣ Arg 45 is negatively charged and evolutionarily conserved. Out of three possible candidates, all of which were located in segment H, only one, Asp 1084 , when mutated affected ␣ binding both in two-hybrid assay, and in in vitro co-immobilization experiments. In addition, substitution of segment I Gly 1215 also abolished ␣ binding.
The structure reveals that Taq Asp 857 , which is homologous to E. coli ␤ Asp 1084 , is indeed in direct contact with Arg 42 (E. coli Arg 45 ) in ␣1 (Fig. 5E). Taq Gly 977 , which is homologous to E. coli ␤ Gly 1215 , is located at the tip of a loop formed by segment I and buttresses Asp 857 (E. coli Asp 1084 ) to position it such that it can correctly interact with Arg 42 of ␣1. The G1215D substitution that we engineered probably causes unfavorable electrostatic interaction with Asp 1084 and may significantly alter the conformation of the ␣-binding domain. In agreement with this idea, G1215D abolished the ␤Ј entry in the complex. In contrast, RNAP harboring ␤ D1084A mutation assembled efficiently and was active. Thus, ␤Ј can stabilize the ␣ 2 ␤ subassembly, presumably through independent proteinprotein interactions with the second ␣ monomer, ␣2, and ␤. Of the remaining four mutations that we engineered, one (G1218D) substantially decreased ␣ binding, while others had no effect. Taq Gly 980 (E. coli Gly 1218 ) is in the same loop as Taq Gly 977 (E. coli ␤ Gly 1215 ); it makes no contacts with ␣ and is at least 10 Å away from Arg 42 (E. coli Arg 45 ). Thus, G1218D substitution inhibits ␣-binding indirectly, probably by affecting the relative position of the stabilizing Gly 1215 . The importance of precise positioning of ␣ Arg 45 and ␤ Asp 1084 for ␣ 2 ␤ formation is underscored by the fact that no interaction was detected when a pair of mutants containing an Asp in place of Arg 45 and an Arg instead of Asp 1084 was tested (data not shown). FIG. 4. Yeast A135 harboring mutation that abolishes A135 interaction with AC40/AC19 in two-hybrid system is functional. Yeast tester strain NOY308 -1a, which lacks A135 and is kept viable by the presence of Gal7-rDNA plasmid (26), was transformed with vector plasmid or plasmids expressing wild-type A135, or mutant A135 with D935A substitution. Transformants were spotted on rich medium containing galactose (top row) or glucose (bottom row) as carbon source. The results of growth after a 62-h incubation at 30°C are presented.
The ␣-binding domain of ␤ also contains two strictly conserved residues that are implicated in catalysis, and that are exposed into the DNA-binding channel of the enzyme, close to the catalytic Mg ion. In segment H, Taq Lys 838 (Lys 1065 in E. coli) contacts the ␣-phosphate of the initiating purine nucleotide (6); in segment F Taq Glu 685 (Gly 813 in E. coli) is probably required for proper interaction with incoming NTP (29). Binding of ␣ thus stabilizes the active center conformation. In agreement with this idea, we had found that in the absence of ␤Ј, ␣ is necessary to induce specific binding of initiating purine nucleotide by the ␤ subunit. 2 Yeast RNAP Assembly-Our studies of yeast RNAP subunit interactions reveal several important points. First, as expected, the determinants of ␣ binding reside in the C-terminal portion of A135, the ␤-like subunit of RNAP I, and Asp 935 , which is homologous to E. coli ␤ Asp 1084 is required for this interaction. This result is in agreement with recent RNAP interaction mapping by Flores et al. (36). These authors reported that A135 segment 678 -1055, containing conserved segments E, F, G, H, and I C (corresponding to E. coli ␤ positions 642-1253), bound both AC19 and AC40 in two-hybrid assay. Our interacting fragment of A135 contained conserved segments F, G, H, and entire I (amino acids 716 -1201), and corresponded to E. coli ␤ residues 681-1342. Based on mutational analysis, A135 appears to interact with the larger ␣-like subunit, AC40, and not with the smaller AC19. This is consistent with structural analysis of yeast RNAP II by the Kornberg group (34), who observed that the largest ␣-like subunit, Rpb3, contacts the ␤-like Rpb2.
Our data indicate that, in yeast, the interaction between the ␤-like and the ␣-like subunits of RNAP I is much weaker than the corresponding interaction in E. coli, and A135 mutations that abolish this interaction are viable. The corresponding interaction may be even weaker in case of RNAP III. We and others (36) were unable to detect any interaction between RNAP III ␤-like subunit C128 and AC40/AC19 using twohybrid assay. Additional RNAP subunits may strengthen the weak ␣ 2 ␤-like complex in eukaryotes. Indeed, the AC40/AC19 heterodimer was shown to bind ABC10␤ (Rpb10) shared RNAP subunit (32). More recently, crystallographic analysis revealed that ABC10␤ and ABC10␣ (Rpb12) bind to the larger ␣-like subunit Rpb3 in yeast RNAP II and anchor it on the ␤-like Rpb2 (34). An alternative possibility would be that the eukaryotic ␣ 2 ␤-like complex is stabilized by the entry of the largest, ␤Ј-like subunit in the complex, similar to the situation observed with E. coli ␤ D1084A mutation. This latter scenario envisions that the relative amount of RNAP I and III in the cell is chiefly determined by the joint availability of catalytic ␤ and ␤Ј-like FIG. 5. Structural context of the minimal ␤ domain capable of ␣ binding and ␤ assembly mutations. The RIBBONS diagram of the three-dimensional structure of Taq RNAP core enzyme is shown (33). The ␤ subunit is indicated in blue, ␤Ј in magenta, ␣1 in green, ␣2 in yellow, and in white. The active center Mg 2ϩ is shown in red and spacefill. The domain of ␤ corresponding to E. coli resides 800 -1231 is shown in cyan. A, view roughly parallel with the main axis of the RNAP channel. The active site Mg 2ϩ is seen through the secondary channel. B, view A rotated 90°clockwise about the vertical axis. C, view A rotated 180°about the vertical axis, giving a view down the opposite end of the main channel. D, the RIBBONS diagram of Taq ␤ fragment corresponding to E. coli ␣-binding domain 800 -1231(cyan) complexed with ␣1 (green). The catalytic Mg 2ϩ is shown in spacefill and red. Parts of ␤ segments F and I important for intrasubunit interaction are shown in magenta and orange. The ␣1 residue corresponding to E. coli Arg 45 is shown in yellow and spacefill. E, the view in D was rotated approximately 90 o counterclockwise about the horizontal axis. Taq ␤ residues corresponding to E. coli residues corresponding to E. coli positions that when mutated alter ␣ binding, as well the catalytic residues involved in substrate interactions are indicated, and shown in color and spacefill. Residues indicated in white correspond to E. coli ␤ residues 1095, 1114, and 1228, and are not involved in ␣ binding. subunits, which are specific, and not by the shared ␣-like subunits.