Hierarchical mechanisms build the DNA-binding specificity of FUSE binding protein

Results

FBP and Its KH-Domains Do Not Read Sequence Stringently.

Previous studies with FUSE; i.e., the 3-d NMR structure of FBP(KH3+KH4) binding to a FUSE subsequence (15); assumed that the DNA binding by FBP fragments mirror the specificity of the full-length protein. To compare DNA binding by full-length FBP versus FBP(KH3+KH4), a series of 27-mer oligonucleotide probes (Fig. 1A) scanning through FUSE were tested for protein-binding by electrophoretic mobility shift assay (EMSA) (Fig. 1B). FBP(KH3+KH4) bound a subset of the probes bound by the full-length protein (probe 2 binds the latter but not the former) (Fig. 1C). Surprisingly, although probes 1 and 2 each included sequences recognized by KH3 and KH4 (5) in the reported structure, they shifted less well than probes 3–5 that included only the KH3-binding segment and lacked the KH4-binding sequences (Fig. 1B). Moreover, probe 6, lacking the contacts seen in the structure, still bound efficiently with both proteins. With the same panel of subsequences, FBP(KH2+KH3) showed binding to a much more restricted subset of probes than full-length FBP; this subset was distinct but overlapping with the probes bound by FBP(KH3+KH4) (Fig. 1B). Straightforward rules explaining FBP target recognition could not be inferred from these data.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

EMSA scanning of FBP binding with FUSE oligonucleotides. (A) The eight 27-mer single-stranded probes used in this gel-mobility shift assay are aligned with the FUSE region upstream of the c-myc P2 promoter. The KH3 element (5′-ATTTTT-3′) and the KH4 element (5′-TATTCC-3′) reported from the NMR study (5) are shown as boldface italic. (B) A total of 0.5 pmol of each FUSE probe was incubated with 5 pmol of each of affinity-purified GST-FBP(KH2+KH3) or GST-FBP(KH3+KH4), or with 1.25 pmol of full-length FBP. Binding reactions lacking protein are shown at Left. (C) A diagram showing how FBP's KH domains are situated in the full-length protein.

These results raised several issues concerning target site selection by FBP. Sequence specificity might be slight, so that the principle determinant of FBP selectivity in vivo is the accessibility, size, and persistence of single-stranded zones. In contrast, each KH-domain might have a hierarchy of preferred sequences, so minimizing the sum of the binding free energies for each KH-domains sets the net affinity with each probe. In the latter case, each KH-domain might accept a suboptimal sequence to allow another KH-domain to settle upon an energetically compensatory site. In this situation it is important to illuminate the spectrum of acceptable KH-binding sequences and to define the rules governing their spacing and arrangement. In another scenario, binding sites are defined by protein-backbone contacts with alternative DNA structures; FBP-binding would be indirectly sequence-specific insofar as the sequence defines the alternative structure. Thus, we sought to tease apart the roles of sequence, arrangement, and DNA conformation in FBP-DNA interactions.

Selection of FBP-Binding Consensus Sequences.

SELEX (Systematic Evolution of Ligands by Exponential Enrichment) is an unbiased method to elucidate the sequence preferences of the binding specificities of dsDNA- and RNA-binding proteins (17, 18). As diagramed in supporting information (SI) Fig. S1, we adapted SELEX for ssDNA-binding proteins such as FBP. To provide a platform to amplify selected sequences while minimizing interference from PCR primer binding sites, 22 random nucleotides were sandwiched between imperfect hairpins designed to fold under selection conditions but to bind primer during PCR (19). Hairpin formation occludes KH-domain binding. Comparison of the electrophoretic mobility of similar length oligonucleotides with and without these imperfect hairpins confirmed hairpin folding during EMSA (Fig. S2).

Prior experiments in vivo and in vitro have suggested that FBP contacts a 40-nucleotide segment of FUSE. Because it is impossible to synthesize a pool of all possible 40 mers, we elected not use full-length FBP for selection. SELEX using contiguous pairs of KH domains (1 + 2, 2 + 3, and 3 + 4) yielded the results shown in Fig. 2. Each KH-pair enriched a highly nonrandom set of sequences. The tetrads “TTGT” and “TCGT” were highly over-represented and are shown in red and blue, respectively. TTGTs were especially highly represented at the 5′-end of sequences selected by (KH2+KH3). Selection by (KH1+KH2), the weakest binding pair of domains (20), enriched “GTGT” and “GTGC” (both green). Adjacent to the randomized segment of the selection template, rare mutations in the primer/primer binding sequences likely represented oligonucleotide synthesis or PCR errors.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

SELEX-based enrichment of binding sites in ssDNA. (Upper) Diagram of method. (Lower) Sequences selected by paired KH domains. (KH3+KH4) Sequences of the input degenerate oligonucleotide and “winners” after 7 rounds of binding to GST-FBP(KH3+KH4). (KH2+KH3) Winners after 6 rounds of binding to GST-FBP(KH2+KH3). (KH1+KH2) Winners after 4 rounds of binding to GST-FBP(KH1+KH2). Mutations in the nonrandomized flanking sequence are shown in lowercase; deletions are indicated with dashes. Some output sequences were longer than the 22 bases randomized. Selected tetrads TTGT are red, TTGT are blue, GTGT and GTGC are green ([KH1+KH2] only); overlapping tetrads are purple or yellow.

Analysis of the (KH3+KH4) and (KH2+KH3) selected sequences with the program “Consensus” (21) revealed a weak preference for As at a fifth position expanding the tetrad to a quintet. The SELEX data alone did not assign unique optimal binding sites to individual KH domains. However, the enrichment of TGTG or GTGT only with (KH1+KH2), but not with (KH2+KH3) or (KH3+KH4), suggests that KH1 binds (g)TGTG(t). The sequences TTGT and TCGT likely bind well with KH-domains 2, 3, and 4.

All of the selected sequences were highly enriched for T's (Table 1). This prevalence is not explained by composition of the consensus tetrads; excluding the tetrads, the residual sequences are still T-rich. Nor is the excess due to FBP binding stably to strings of T's; FBP does not bind to oligo(dT) (data not shown). Embedding the KH-cognate elements in T-rich DNA may create a molecular environment conducive for recognition (22).

Spacing Between KH Domain-Binding Sequences Tunes FBP Binding.

SELEX did not reveal the preferred spacing between KH domain-binding tetrads. Based on the flexibility seen in the solution structure of FBP(KH3+KH4) and the lower rigidity of ssDNA than dsDNA, there was no a priori prediction for the spacer lengths separating the selected sequences. To determine the effect of spacer-length, a series of oligonucleotides (Fig. 3A) with variable separation between consensus elements was tested by competition against a 29-base FUSE probe for binding FBP fragments (SI Methods). Because A was by far the least enriched base in the SELEX assays (Table 1), KH-binding sequences were embedded in oligo-dA to maintain the overall length while minimizing confounding interactions with the spacer or flanking sequences.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

Determination of the Optimal Spacing Between KH-domains by EMSA. (A) Sequence of spacer oligonucleotides. (B and C) A standard oligonucleotide was end-labeled and gel-shifted by an excess of (B) GST-FBP(KH3+KH4) (compare lanes 1 and 2) or (C) GST-FBP(KH2+KH3). Unlabeled competitor oligonucleotides (17 μM) in excess relative to the protein (≈40 nM) were added to most binding reactions (lanes 3–16), before protein. “Spacer4,” (lane 8) competed most effectively among the spacer oligonucleotides (lanes 6–14). All spacer oligonucleotides competed for protein binding more effectively than control oligonucleotides containing a single (lanes 4, 5, 15, and 16) or no (lane 3) consensus elements. Shifted bands were quantified by Imagequant software, version 1.2, and charted in Fig. S3.

The results of this assay were similar using (KH3+KH4) (Fig. 3B) or (KH2+KH3) (Fig. 3C). Competition occurred over a range of spacer-lengths, but was most efficient at 4- or 5-bases (Fig. 3B, lanes 9 and 10; also Fig. 3C, lanes 8 and 9; Fig. S3). Oligonucleotides with single KH-binding sequences competed inefficiently (Fig. 3B, lanes 4, 5, 15, and 16; Fig. 3C, lanes 4, 5, 14, and 15), arguing that high affinity protein binding required both embedded SELEX sequences. Surprisingly, the no-spacer oligonucleotide (Spacer0) still competed efficiently with the probe for binding (Fig. 3B, lane 6). The structures of DNA-engaged FBP(KH3+KH4) and other KH-domain-nucleic acid complexes do not allow 2 KH-domains to simultaneously engage immediately juxtaposed pentameric sequences; therefore, Spacer0 must either toggle between KH-domains by sliding, transiting through out-of-register intermediates, or some unanticipated conformational change must allow simultaneous cooccupancy by both KH-domains. When embedded in T's, pairs of KH-binding sequences bound FBP more strongly and were relatively insensitive to spacing (data not shown). These results show that spacer composition and length tune FBP binding.

FBP Is Positioned at FUSE Through SELEX-Like Sequences.

Together the SELEX results and EMSA analysis of FUSE subsequences suggest that FBP KH domains 2–4 each interact with a similar set of partially degenerate pentamers, and KH1 recognizes a different set of pentamers. The local nucleotide composition and spacer distances modify the net binding affinity.

FBP may bind in many registers to a series of energetically degenerate, overlapping and nested binding sites. To determine whether FBP has a preferred register within the FUSE region, and to identify which FUSE-bases were important to bind full-length FBP, a missing-base interference assay was developed similar to previous techniques using depurinated/depyrimidinated probes to identify contacted bases (23). This assay (Fig. S4) used a single end-labeled 87-mer FUSE oligonucleotide with random abasic sites occurring on average once per molecule (SI Methods).

Oligodeoxynucleotides competent to bind FBP were recovered from protein-DNA complexes separated by EMSA from unbound/incompetent probe (Fig. 4A Left), cleaved at abasic sites, and compared on sequencing gels (Fig. 4B). The FBP-bound probe (lanes 2 and 3) showed 2 regions of strong interference bracketing 2 regions of lesser interference. The 4 regions of interference matches the number of KH domains in FBP. These regions were approximately evenly spaced, and the total span of all contacts was 30–32 bases. If each KH-domain bound 5-bases (as suggested by SELEX), and each region is separated by 4 or 5 bases (optimal), then the span of a FUSE footprint (four pentamers and 3 spacers) would be 32–35 nt. Note that both regions of strong interference included the triplet “TGT” present in all of the SELEX consensus sequences; moreover, the most upstream contacts occurred within the sequence “TGTGT” highly enriched by KH1 in SELEX reactions. The weaker regions lacking TGT were poorly delimited as if the internal KH-domains (KH2+KH3) were more mobile on the DNA than the bracketing domains (KH1, KH4).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

Missing-base interference footprinting. (A) Preparative gel shift of FBP-FUSE (Left) (3 μmol of FBP; ≈0.1 μmol of probe), and FIR-FUSE or FIR-FBP-FUSE (Right) (0.5 μmol of FIR ± 29 nmol of FBP; ≈0.1 μmol of probe). (Left) A longer exposure than shown at Right. Probe was recovered from all bands and processed through the remainder of the assay (SI Methods). (B) Interference footprint of FBP on FUSE. Compare samples that bound FBP (lane 3) with free (lane 4) and input (lane 2) DNA. Regions with strong interference (thick bars) and weak interference (thin bars) are shown on both the gel and the sequence (5′ to 3′ starting at the bottom). No other interfering regions were seen on a 20% gel (data not shown). (C) Interference footprints of FIR alone (annotated in red, lane 3) and FIR+FBP (green, lane 6) on FUSE 87-mer; FBP unsupershifted by FIR is also shown (blue, lane 5). The EMSAs for lanes 2–3 and lanes 5–7 used the same input probe (lane 4). Reinforcing (“hypersensitive”) sites are marked with bullets. The same sample was run on 7% (Left) and 20% (Right) denaturing PA gels. (D) Comparison of footprints. Strongly footprinted regions from B are shown in gray.

Based on contact points, SELEX and spacing analyses, we surmise the optimal sequence for full-length FBP to be: TTGTa(N)_4/5TYGTa(N)_4/5TYGTa(N)_4/5KTGY (Y = T or C; K = T or G).

FIR Modifies FBP's Interaction with FUSE.

FIR represses transcription after binding with FBP and FUSE (9), although in vitro FIR binds with FUSE weakly. FBP augmented FIR binding (Fig. S5), presumably through their previously described protein–protein interaction (20, 24). Missing-base interference identified nucleic acid contact points distinguishing FBP-FUSE, FIR-FUSE, and FBP-FIR-FUSE complexes (Fig. 4 A Right and C). 4C, lane 3 (both gels) shows the interference pattern of FIR alone; this footprint enveloped FBP's and consisted of 3 regions of interference, 1 entirely 5′ of FBP's footprint (Fig. 4D). Most of the contacted bases were Ts. Interestingly, “reinforcement” was seen at several purines (bullets). The increased the stability of FIR-FUSE complexes missing these purines indicate steric hindrances with these bases or alternatively, FUSE may form a FIR-nonbinding structure requiring these purines to fold.

FBP-FIR-FUSE (ternary) complexes were separated by EMSA from the FBP-FUSE (binary) complexes (Fig. 4A Right).

Surprisingly, the footprint of the ternary complex (Fig. 4B, lanes 6) was contracted relative to the FIR-alone footprint (Fig. 4B, lane 3). Although the upstream, i.e., 3′, FBP+FIR boundary was extended slightly, the strong downstream FIR contacts were entirely lost. Apparently, FBP provoked either a major reorganization or repositioning of FIR on FUSE. Equally surprising, the contacts within the binary complexes formed in the presence of FIR (Fig. 4B, lane 5) were distinguishable from those of the binary complexes formed in its absence (Fig. 4C). These differences were not easily explained by combination or superimposition of the FBP-FIR-FUSE and FBP-FUSE (in the absence of FIR) footprints. We speculate that dissociation of FIR from an FBP-FIR-FUSE ternary complex yielded a metastable FBP-FUSE complex that slowly relaxed to the FIR-independent, FBP-FUSE ground state.

Site Selectivity of Full-Length FBP on Full-Length FUSE.

Tight binding of FBP to FUSE might reflect the summation of binding at multiple degenerate sites, or alternatively, binding at 1 or 2 predominant sites. In the case of multiple, equivalent sites, the FBP-contact point analysis would reveal the overlap of all registers, and (i) maximal overlap would be expected in the central region of FUSE, (ii) shortening the probe would decrease the number of contributing frames at the ends and reduce overall binding, (iii) single point mutations would leave most frames competent and marginally reduce overall binding, and (iv) mutations within the central portion of the FUSE region where more degenerate binding sites overlap would attenuate binding more than mutations at the fringes of the melted zone, where there would be fewer sites because of the dsDNA-ssDNA boundary. If, however, FBP binds mainly at a single site, then (i) maximum protection would occur within the predominant site at critical bases to bind KH-motifs, (ii) shortening the probe would have little effect unless encroaching on that site, and (iii) point mutations of the critical bases within the predominant site should dramatically lessen binding.

A complex formed between FBP and an 87-mer FUSE probe including the entire melting segment was challenged with a 47-mer competitor that included all of the bases contacted by FBP or by cold 87-mer (SI Methods). Because the 87-mer and 47-mer were similarly effective competitors (data not shown), the longer probe must lack additional good FBP binding sites. These data show that although FBP binds with most short overlapping probes from FUSE by EMSA, when presented with the entire region, FBP settles on a single primary site.

In Vivo Test of FUSE Contacts and an Optimal FBP Binding Sequence.

To test the in vivo relevance of the predominant FBP binding site defined in vitro, episomes with wild type or mutant FUSE elements sandwiched between divergently transcribing metallothionein promoters were used to drive EGFP expression. This arrangement insures sufficient transcriptional supercoiling to melt FUSE upon the addition of Zn²⁺s. The wild-type FUSE-episome, pMT2(FUSE) has been demonstrated to bind FBP and FIR in vivo and to program a pulse of reporter expression (9). The mutant episome, pMT2(mutFUSE) (Fig. 5A), contained 2 T-to-A FUSE mutations, 1 in each of the 2 major contacted regions collinear with KH1 and KH4 described above. This double mutation decreased FBP-binding by >15-fold by EMSA (Fig. 5B, compare lane 2 to lane 8). The T-to-A transversions preserve AT-content and the melting profiles of both were calculated to be indistinguishable using the highly reliable Stress-Induced Duplex Destabilization (SIDD) algorithm (Fig. S6) (25).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 5.

Double point mutation in FUSE reduces FBP binding in vitro and FUSE function in vivo. (A) Schematic of pMT2 (+FUSE) with pMT2(mutFUSE) mutation sites indicated (partial FUSE sequence is shown). In pMT2(-FUSE), bacterial sequence replaces FUSE to maintain the length. (B) In vitro competition for FBP binding by (wt)FUSE 87-mer and (mut)FUSE 87-mer oligonucleotides. A small amount of end-labeled wtFUSE87-mer was gel-shifted by an excess of FBP (compare lanes 1 and 11). Included in the binding reactions were increasing amounts: 1× (> [FBP]), 3×, 10×, or 30× unlabeled wtFUSE87-mer (lanes 2–5); or 1×, 3×, 10×, 30×, or 100× unlabeled mut-FUSE87-mer (lanes 6–10). Comparing the quantified (ImageQuant) shifts in lanes 2 and 8, wtFUSE binds FBP ≈15-fold more strongly than mut-FUSE87-mer. (C) Histograms of flow cytometry testing FUSE function. EGFP expression was measured 10 h after induction with 60 μM Zn²⁺ (x axis). The histogram for uninduced pMT2(FUSE) is overlaid in green.

Raji cells were transfected with pMT2(-FUSE), pMT2(+FUSE), or pMT2(mutFUSE) and transcription from the divergent metallothionein (MT) promoters induced with Zn²⁺ torsionally strained the upstream DNA (2). MT-IIA promoter-driven EGFP was monitored by flow cytometry (Fig. 5C). As reported, cells with pMT2(+FUSE) expressed much more EGFP than those with pMT2(-FUSE) (9). EGFP-expression from pMT2(mutFUSE) was indistinguishable from pMT2(-FUSE), showing that mutation of 2 key FBP contacts disabled the FUSE-effect.

To test the function of an optimal FBP-binding sequence, FUSE was swapped for a perfect consensus in the pMT2 episomes (SI Methods) and transfected into Raji cells. The reporter genes were expressed even without induction (Table S1), and were further induced with low levels of Zn²⁺. The increased potency of the consensus was expected because of tighter binding by FBP and because the consensus by SIDDs, melts at lower levels of supercoiling than FUSE.