A 3D puzzle approach to building protein–DNA structures

ABSTRACT

Despite recent advances in structural analysis, it is still challenging to obtain a high-resolution structure for a complex of RNA polymerase, transcriptional factors, and DNA. However, using biochemical constraints, 3D printed models of available structures, and computer modeling, one can build biologically relevant models of such supramolecular complexes.

To understand mechanism, one needs structure. Yet despite an explosion of X-ray, NMR, and EM structures for thousands of proteins, it is still a challenge to obtain the structure of multiple protein subunits associated with DNA, such as an activated transcription complex. Even today, only one crystal structure for RNA polymerase (RNAP) with an activator and DNA has been reported.¹ Advances in high-resolution cryo-EM techniques² promise to reveal many more such structures in the near future. However, artifacts and inactive complexes can be obtained by both X-ray crystallography and cryo-EM. Consequently, it will always be necessary to use supplemental methods to confirm that one has obtained the structure of the functional complex.

Despite the scarcity of structures for complete transcription/regulator/DNA complexes, separate structures exist for bacterial, archaeal, and eukaryotic multi-subunit RNAPs as well as hundreds of transcriptional regulators or regulator domains with and without DNA. Even when no structure is available for a particular regulator of interest, many times the structure of an ortholog is available. Fortunately, for many transcription factors there are also extensive footprinting analyses, proximity labeling techniques, and crosslinking experiments, which have been successfully applied to determine the position of a regulator relative to the DNA and/or RNAP.

The trick then is to figure out a way to intelligently take these “puzzle pieces” and fit them together. During the past few years, we have used a combination of techniques to do just that, resulting in a biochemically grounded structural model of E. coli RNAP together with a bacteriophage activator (T4 MotA protein), a co-activator (T4 AsiA protein), and promoter DNA. Using a combination of various available structures, FeBABE (iron bromoacetamidobenzyl-EDTA) cleavage analyses, 3D printed models, crosslinking, modeling software, and biochemical constraints, my laboratory in collaboration with Tim Cardozo at New York University has visualized this complex for the first time.³ Importantly, the biochemical analyses were performed using the transcriptionally competent species, giving us confidence that the model represents an active complex. In this point-of-view, I describe this process in hopes that others with similar interests may consider this approach.

Transcription initiation by E. coli RNA polymerase and the process of sigma appropriation

Years of genetic, biochemical, and structural work have resulted in a deep understanding of the process of transcription and the structures of various RNAPs (reviewed in Refs.^4-6). The structure of E. coli RNAP, which has been a favorite subject of biochemical studies, was published in 2013.⁷ This RNAP shares features with multi-subunit polymerases present in bacteria, archaea, and eukaryotes: a highly conserved “core” of proteins (β, β’, 2α's, and ω in bacteria) that synthesizes RNA and a specificity factor that recognizes promoter DNA and sets the transcription start site. In bacteria, these specificity subunits are called σ factors. σ's are equivalent to TFIIB together with TBP (or TBP-related proteins) of eukaryotic Pol II.^{8, 9} Hundreds of σ factors have been identified (reviewed in Refs.^10,11). Most belong to the σ⁷⁰ family, sharing homology with E. coli σ⁷⁰, the primary σ needed for vegetative growth.

Host promoter recognition arises from specific interactions between domains of σ⁷⁰ (while it is present within RNAP) and the promoter DNA (reviewed in Ref.¹²). Crucial contacts occur between residues in σ⁷⁰ Regions 3 and 2 with the extended −10 element (⁻¹⁵TGn⁻¹³) and the −10 element (⁻¹²TATAAT⁻⁷), respectively, and between residues in a helix-turn-helix within Region 4 and the −35 element (⁻³⁵TTGACA⁻³⁰)(Figs. 1A, 2A, 2B). This binding is facilitated by contact between the C-terminus of σ⁷⁰ (called H5) and a structure within core called the β-flap¹³ In addition, the C-terminal domains (CTDs) of the α subunits of core RNAP can interact with AT-rich UP elements that are located between −40 and −60 [consensus sequences of ⁻⁵⁸nnA(a/t)(a/t)(a/t)(a/t)(a/t)TTTTT⁻⁴⁶ and ⁻⁴⁵AAAAAA(a/g)n(a/g)⁻³⁷].¹⁴ When activators are present, the α^CTDs can move to positions even farther upstream (reviewed in Ref.¹⁵). This is because a flexible linker connects the α^CTD's back to the α N-terminal domains (NTDs), which are embedded within core.

Figure 1.

Activation through sigma appropriation. (A) At an E. coli host promoter, σ⁷⁰ residues in Regions 2/3 contact the −10 element (and extended −10 element), while residues in Region 4 contact the −35 element. UP elements between −40 and −60 may also interact with α^CTD's of core. (B) At a T4 middle promoter, contacts between σ⁷⁰ Regions 2/3 and the −10 element remain, but the remodeling of σ⁷⁰ Region 4 by AsiA eliminates the ability of Region 4 to contact the −35 element. MotA binds to the MotA box centered at −30 and to the C-terminus of σ⁷⁰ (H5). (C) Structures of AsiA bound to σ⁷⁰ Region 4 (PDB ID: 1TLH),³⁴ MotA^CTD (PDB ID: 1KAF)²³ showing the unusual double wing motif, and MotA^NTD (PDB ID: 1I1S)²⁶ showing the cleft that binds σ⁷⁰ H5. In each germane panel σ⁷⁰ is shown in yellow, α^CTD's are in blue, MotA in green, and AsiA in gray.

Figure 2.

Structural models of E.coli RNAP at a host promoter and E. coli RNAP/MotA/AsiA at a T4 middle promoter. (A) Schematic of σ⁷⁰ showing the positions of Regions 1–4 and H5. (B) Top, E. coli RNAP (PDB ID: 4IGC)⁷ that has been modeled with host promoter DNA as seen in PDB ID: 1L9Z.³⁵ Core polymerase subunits (β,β’, 2α's, ω) are shown in ice blue, σ⁷⁰ is shown in yellow. (α^CTD's are not seen). DNA is in red/orange, and positions of −10 and −35 elements are indicated. The positions of σ⁷⁰ H5, which is held by the β-flap, and Regions 2, 3, and 4 are shown. Bottom, structure has been turned 180° on the y-axis to show σ residue 581 (in red) and β’ residues 80–87 (in Carolina blue). (C) Top, generated structural model of E. coli RNAP with MotA, AsiA, and T4 middle promoter DNA. MotA is in green, AsiA is in gray, and other colors as in (A). The position of the β mutation G1249D is indicated. Bottom, as in (B).

Most characterized bacterial activators fall into one of two categories (reviewed in Ref.¹⁵). Class II activators bind to the DNA just upstream of or overlapping the −35 element and interact with σ⁷⁰ Region 4 and/or a promoter proximal α^CTD. Class I activators bind farther upstream and interact with one or two α^CTD's that also sit on the upstream DNA. Thus, the interaction of the activator with the DNA and with a portion of RNAP acts as additional “glue” to hold polymerase to the promoter as the complex is being formed or isomerizing into the transcriptionally competent species.

Bacteriophage T4 is a double-stranded DNA lytic phage that infects E. coli and then lyses the cell, releasing ∼150 phage after 20 min at 37°C.¹⁶ T4 expresses its 169 kbp genome in three stages, early (∼0–4 min after infection), middle (∼2–8 min), and late (∼8–13 min). Because T4 does not encode an RNAP, it must rely on the host RNAP throughout infection. Due to this hierarchical program of gene expression, transcription of T4-encoded genes has been an excellent model system for examining promoter recognition by RNAP. Promoters for T4 early genes have excellent matches to the σ⁷⁰ recognition elements and thus compete extremely well for bacterial RNAP. Middle promoters have excellent matches to the −10 element, but lack the −35 element of ⁻³⁵TTGACA⁻³⁰. Instead they share a 10 bp motif, a MotA box, centered at −30 [consensus sequence of 5′ (a/t)(a/t)(a/t)TGCTTtA3′; Fig. 1B]. Late promoters lack either element, but two T4 middle genes encode a new σ that replaces σ⁷⁰ and allows recognition of the late promoter motif. In addition, several T4 early genes encode proteins that result in the specific degradation of the host DNA, essentially eliminating its ability to compete for RNAP shortly after infection.

The switch from host and T4 early to T4 middle promoter recognition is called sigma appropriation.¹⁷ This process is fundamentally different from class I or class II activation. It requires the products of two T4 early genes, motA and asiA. AsiA is a co-activator that binds very tightly to σ⁷⁰ Region 4¹⁸ (Fig. 1B). The structure of AsiA/σ⁷⁰ Region 4¹⁹ (Fig. 1C) revealed that AsiA dramatically reconfigures this portion of σ⁷⁰ such that a helix-turn-helix motif within Region 4 that normally interacts with the −35 element is now one continuous helix. This reconfiguration also places H5 in a very different position. MotA is a transcriptional activator that binds to the MotA box element centered at −30 in the middle promoter DNA.^20,21 Early work indicated that the CTD of MotA has DNA-binding activity.²² However, the structure of MotA^CTD revealed an unusual “double wing” that had not been previously observed as a DNA-binding motif²³ (Fig. 1C). Other analyses indicated that the NTD of MotA also binds to σ⁷⁰.²² In this case, the acidic and hydrophobic residues of H5 interact with a basic/hydrophobic cleft of MotA^NTD,^24,25 which is seen in the structure of MotA^NTD26 (Fig. 1C). Thus, in sigma appropriation, remodeling of σ⁷⁰ Region 4 by AsiA relocates Region 4 away from the −35 element. The relocation then places σ⁷⁰ H5 in a position so that it can interact with its new protein target, MotA^NTD, rather than the β-flap.

Building a structural model of sigma appropriation

Despite all of this information, it was still not possible to generate a complete picture of the sigma appropriated complex. Multiple attempts to obtain a structure of MotA or MotA^CTD with the DNA were unsuccessful. Consequently, we did not know how the unusual double wing motif of MotA interacts with the DNA. Furthermore, there were no structures of orthologous proteins that could be used as a starting point. Given these problems, we decided to use the chemical cleaving reagent FeBABE to determine the architecture of MotA relative to the DNA. FeBABE can be conjugated to a protein through a cysteine residue. After incorporation of the conjugated protein into a multi-subunit/DNA complex, addition of H₂O₂ and ascorbate then generates a burst of hydroxyl radicals that cleaves DNA and protein within ∼20 Å.^27-29 Importantly, the resolution is excellent due to the short distance over which the hydroxyl radicals survive, allowing localization within just 3–5 bp. In our case, we conjugated FeBABE to 12 different MotA mutant proteins that we had engineered to contain a single cys residue. A great advantage of this technique is that we assembled complete transcription complexes and determined that they were indeed functional for transcription in vitro. This allowed us to determine the FeBABE cleavage sites in the context of the functional complex.³⁰

To visualize the cleavage results in three dimensions, we built physical 3D models, using 3D printing technology, of the known structures of MotA^CTD (Fig. 1C) and MotA^NTD (Fig. 1C), in which we color-coded the conjugated residues.^30,31 We then correspondingly color-coded the DNA cleavage sites within a 3D model of the B-form of the MotA box DNA. B-form DNA was chosen since previous biochemical analyses suggested that MotA binding did not significantly distort the DNA.

Generating these physical models was an important step in this process. It allowed us to manually manipulate MotA^CTD relative to the DNA and determine that only one orientation “fit” on the sequence in a way that was consistent with the FeBABE cleavages. This position put both wings of MotA within a single major groove of the DNA, which could not have been predicted from the structure of MotA^CTD alone.

Although this technique of using 3D models might seem unconventional or crude, it is a relatively quick and affordable way for one to “see” how a protein is positioned relative to DNA (or to another protein) if one has sufficient biochemical information. When I showed the models to my colleagues, even my hard-core structure friend was amazed at how well this works. The manually determined structure can then be transferred to the computer by using a program that allows one to move structures relative to each other. In our case, we used the Molsoft program to generate a model, and then we confirmed that the predicted FeBABE cleavages matched our observed cleavages.

Unlike MotA^CTD, MotA^NTD does not interact with the DNA. However, it is still close enough to the DNA that cleavages were obtained when MotA^NTD residues were conjugated. Using these cut sites, we obtained a likely location for MotA^NTD relative to DNA and MotA^CTD. Fortunately, the distance between these two domains could easily accommodate the 8 aa linker that must connect the domains. Interestingly, this linker is an arginine/lysine-rich sequence (KRATRKAR). Arginine-rich sequences are known to interact with the minor groove of DNA,³² and previous work had indicated that the upstream portion of the MotA box is recognized through base determinant(s) within the minor groove.³³ Thus, our model predicted that the complete DNA-binding region of MotA consists of the linker, which interacts within the minor groove of the upstream portion of the MotA box, together with MotA^CTD, which interacts within the major groove of the downstream portion of the MotA box.

To complete the picture: (1) we removed σ⁷⁰ Region 4-H5 within the RNAP structure and replaced it with the AsiA/σ⁷⁰ Region 4-H5 structure.¹⁹ This could be done by simply overlapping four residues at the beginning of σ⁷⁰ Region 4 seen in the E. coli RNAP structure⁷ with the same portion of σ⁷⁰ at the beginning of the AsiA/σ⁷⁰ Region 4 structure.³⁴ (2) We modeled the DNA into E. coli RNAP using the same path that host promoter DNA takes within the Thermus aquaticus RNAP/DNA structure.³⁵ This was acceptable given that footprinting analyses had indicated that MotA and AsiA do not affect RNAP/promoter interactions downstream of about −20.³⁶ We then replaced the −35 region of the promoter with the MotA box sequence. (3) We placed MotA on the DNA based on the FeBABE/Molsoft analyses described above.³⁰ Thus, at this point we had put all the pieces of the puzzle together without introducing any distortion of the proteins or the DNA. It was gratifying then to see that in this first structure/model, σ⁷⁰ H5 was pointing toward its correct partner, MotA^NTD.³

However, we know that σ⁷⁰ H5 and MotA^NTD must actually “handshake” so it was necessary to determine the correct way to insert σ⁷⁰ H5 into the hydrophobic cleft of MotA^NTD. This required additional biochemical constraints since many differing modeling solutions could accomplish this result. Consequently, we determined the position of σ⁷⁰ residue 581 present in Region 4.³ FeBABE conjugation at a cys introduced at 581 has previously been used to confirm that the helix-turn-helix in Region 4 lies within the major groove of the −35 element of host promoter DNA.³⁷ We found that the presence of AsiA shifts this position considerably downstream (about 8 bp).

We also introduced the photocrosslinkable amino acid analog BpA (benzoyl-phenylalanine) at σ⁷⁰ position 581. This was relatively easy using a technique developed by Peter Schultz and colleagues.³⁸ In this system, one first changes the codon for the residue of interest to a TAG stop. The protein is then synthesized in cells that have an aminoacyl-tRNA synthetase/suppressor tRNA pair that recognizes the TAG and inserts BpA that has been added to the culture. After purification of the protein and assessment of its activity by functional assays, one can then use the protein for photocrosslinking. In our case, we observed a new species on denaturing SDS protein gels after photocrosslinking of the transcription complex containing σ⁷⁰ 581BpA. Mass spectrometry of the tryptic peptides from this band revealed that in the AsiA-associated RNAP, σ⁷⁰ residue 581 is within 11 Å of β’ residues 80–87. No such crosslinking occurs in the absence of AsiA. With this information about the position of σ⁷⁰ residue 581 relative to the DNA and to β’, we were now able to intelligently model the complex so that sigma H5 was buried within the cleft of MotA^NTD. This generated the final complete structural model³ (Fig. 2C). It is important to note that the final modeling required that the AsiA/σ⁷⁰ Region 4-H5 be shifted within RNAP so that it would be closer to the downstream DNA and the N-terminal region of β’ (Compare bottom panels of Figs. 2 B and C).

Why should we want to develop such a structure-based model? First, it is tremendously satisfying to be able to “visualize” a transcription complex whose biochemistry has been elucidated by decades of work in multiple laboratories. Given that this type of activation is fundamentally different from the mechanism of class I or II, the visualization of sigma appropriation provides the molecular features of a different paradigm for transcription regulation. Second, our modeling suggested that the movement of AsiA/σ⁷⁰ Region 4-H5 to its correct position could impact the environment of σ⁷⁰ residues just N-terminal of Region 4. This was particularly interesting because a mutation within β (G1249D) that is near these residues was known to specifically impair sigma appropriation.³⁹ β G1249D is far from the action of AsiA, MotA, σ⁷⁰ Region 4-H5, and the DNA (Fig. 2C). However, its effect could now be explained. Several negatively charged sigma residues are within 9 Å of β 1249. Thus, it is reasonable to infer that a glycine to aspartic substitution at β 1249 would interfere with the correct positioning of AsiA/σ⁷⁰ Region 4-H5. Third, the positioning of MotA on the DNA is particularly useful because it harbors such an unusual DNA-binding motif. Since the structure of MotA^CTD was first reported, similar structures for other proteins of unknown function have been solved. Thus, the visualization of how MotA^CTD and the MotA linker interact with the DNA may provide insights into structure/function of these other factors. Finally, and perhaps most importantly, the procedure may provide a guide for others who have been stymied in their efforts to generate an experimentally guided structure for their particular large complex of interest.

Disclosure of potential conflicts of interest

No potential conflicts of interest were disclosed.

Acknowledgments

The author thanks Scott Stibitz, Jennifer West, Leslie Knipling, and Kumaran Ramamurthi for helpful discussions.

Funding

This research was supported by the Intramural Research Program of the NIH, The National Diabetes and Digestive and Kidney Diseases (NIDDK).