Multiple DNA-binding modes for the ETS family transcription factor PU.1

The eponymous DNA-binding domain of ETS (E26 transformation-specific) transcription factors binds a single sequence-specific site as a monomer over a single helical turn. Following our previous observation by titration calorimetry that the ETS member PU.1 dimerizes sequentially at a single sequence-specific DNA-binding site to form a 2:1 complex, we have carried out an extensive spectroscopic and biochemical characterization of site-specific PU.1 ETS complexes. Whereas 10 bp of DNA was sufficient to support PU.1 binding as a monomer, additional flanking bases were required to invoke sequential dimerization of the bound protein. NMR spectroscopy revealed a marked loss of signal intensity in the 2:1 complex, and mutational analysis implicated the distal surface away from the bound DNA as the dimerization interface. Hydroxyl radical DNA footprinting indicated that the site-specifically bound PU.1 dimers occupied an extended DNA interface downstream from the 5′-GGAA-3′ core consensus relative to its 1:1 counterpart, thus explaining the apparent site size requirement for sequential dimerization. The site-specifically bound PU.1 dimer resisted competition from nonspecific DNA and showed affinities similar to other functionally significant PU.1 interactions. As sequential dimerization did not occur with the ETS domain of Ets-1, a close structural homolog of PU.1, 2:1 complex formation may represent an alternative autoinhibitory mechanism in the ETS family at the protein-DNA level.

The differentiation of distinct lineages of blood cells from a single progenitor species occurs in a multistep process, termed hematopoiesis, that is intricately controlled at the transcriptional level. The ETS family transcription factor PU.1 ranks among the most essential hematopoietic regulators in ensuring the continued self-renewal of this progenitor, the hematopoi-etic stem cell (1). PU.1 is also essential for directing correct differentiation of the hematopoietic stem cell to various cell lineages. Crucially, PU.1 governs cell fate specification and functions in a transient, dosage-and cell stage-dependent manner (2). In mature cells, graded PU.1 activity is also required for key cellular processes and the specialization of mature cells into function-specific subtypes. Aberrant PU.1 activity is linked to a spectrum of diseases, including rheumatism (3), hematologic cancers (4 -6), and Alzheimer's disease (7). Clearly, knowledge of the regulatory mechanisms of PU.1 is essential to an understanding of its role in normal biology and in disease.
Biological modulation of PU.1 activity is generally attributed to up-or down-regulation at the level of expression. With a metabolic half-life of ϳ50 h, a period that spans the entire cell cycle (8), the cellular persistence of PU.1 means that downregulation of its own expression cannot alone provide a complete description of PU.1 regulation, as additional dampening mechanisms are required to prevent open-ended escalation of PU.1 activity during its lifetime in the cell.
Outside of down-regulated expression, few inhibitory mechanisms have been described for PU.1. The best understood is the mutual antagonism between PU.1 and the zinc finger transcription factor GATA-1, wherein each protein inhibits DNA binding by the other during myelopoiesis (9). In addition, PU.1 is one of only a few ETS family members that lack so-called autoinhibition, a regulatory mechanism in which helices adjacent to the ETS DNA-binding domain allosterically reduce DNA-binding affinity (Fig. 1A) (10). In the case of Ets-1, the archetypal autoinhibited ETS paralog, interactions with partner proteins, such as Pax5 (11) and AML1/RUNX1/CBF␣2 (12), relieve autoinhibition and restore high-affinity binding. Thus, in the absence of lineage-specific inhibitory partners such as GATA-1 or some intrinsic regulatory alternative to autoinhibition, PU.1 would be continuously poised in a functionally uncontrolled, transcriptionally permissive state.
In previous work, we have observed in calorimetric titrations the potential for the eponymous DNA-binding domain of PU.1 to dimerize at a single cognate DNA-binding site (13). Whereas dimerization of ETS domains of other ETS family homologs bound to two sites (i.e. 2:2 complexes) has been reported (14 -18), self-association at a single site is poorly understood. We have carried out an extensive series of spectroscopic and biochemical experiments to directly characterize the variable binding modes of PU.1 as a function of DNA sequence and site size. The results show a site-specific 2:1 complex in exchange between free PU.1 on the one hand and the 1:1 site-specifically bound state on the other, while contacting the DNA over an extended interface beyond the single helical turn observed in the 1:1 co-crystal structure. Sequential dimerization imposes the dual requirements of specific DNA as well as a site size longer than 10 bp. Nonspecific DNA forms oligomeric but not 1:1 complexes at equilibrium. Sequential dimerization of sitespecifically bound PU.1, which sequesters excess circulating PU.1 from accessible DNA target sites, suggests itself as a potential mechanism of negative feedback in the absence of inhibitory binding partners.

Results
A 1:1 protein/DNA site stoichiometry is universally observed in co-crystal structures of ETS family transcription factors, including PU.1 (19,20) (Fig. 1A). Nevertheless, calorimetric measurements of DNA binding by the PU.1 ETS domain (encoded by the C-terminal fragment, PU.1⌬N167) showed that PU.1 bound site-specific targets with non-1:1 stoichiome-try (13). When site-specific DNA was titrated into PU.1⌬N167, the protein bound the DNA in a negatively cooperative manner (Fig. 1, B and C). Dimerization was strictly noncovalent, as PU.1⌬N167 harbored no cysteine residue. To broaden our observations and determine whether 2:1 binding was a class property of ETS domains, we measured high-affinity site-specific binding by the ETS domain of Ets-1 (encoded by the C-terminal fragment Ets-1⌬N331), which contained two free cysteines. Although PU.1⌬N167 and Ets-1⌬N331 represent sequence-divergent ETS members, their backbones are superimposable in their DNA-bound states (21). At comparable concentrations as PU.1⌬N167 (ϳ40 M) and under reducing but otherwise identical conditions, Ets-1⌬N331 bound site-specific DNA at strictly 1:1 stoichiometry (Fig. 1, D and E). Moreover, the inclusion of flanking elements known to autoinhibit Ets-1 (Ets-1⌬N280) did not affect the binding stoichiometry (Fig. 1, F and G). Thus, dimerization at a single site was not shared by Ets-1 but was particular to PU.1 and possibly other sequencesimilar ETS homologs.
Although the ITC 7 titrations could be fitted empirically with model-dependent profiles, the high concentrations (Ͼ10 Ϫ5 M) required for the experiments precluded an accurate quantitative determination of binding affinities due to the strong dissociation constant of the 1:1 complex (10 Ϫ9 M). We therefore titrated a 20-bp Cy3-labeled duplex oligonucleotide harboring the same high-affinity site as used in the ITC experiments and measured binding from changes in fluorescence polarization of the DNA probe ( Fig. 2A). The binding data yielded a biphasic profile to which a sequential binding model (22) was fitted with dissociation constants of 7.0 Ϯ 1.3 nM and (1.2 Ϯ 0.8) ϫ 10 3 nM, or a (negative) cooperativity parameter of ϳ170. Constraining the model to 1:1 binding significantly compromised the fit to the data (green curve in Fig. 2A). To determine whether a single helical turn of contact interface, as observed in the co-crystal PU.1/DNA structure (23), was sufficient to support sequential binding of PU.1, we repeated the titration with a DNA construct in which only the core 10 bp of the cognate site (5Ј-AGCGGAAGTG-3Ј) was duplex. Binding to the 10-bp duplex exhibited monophasic binding that was well-described by a 1:1 model with a ϳ2-fold reduction in dissociation constant (12 Ϯ 2 nM). To further determine whether the excess binding to the 23-bp construct represented nonspecific interactions, we measured PU.1⌬N167 binding to an isomer of the 20-bp DNA in which the core 5Ј-GGAA-3Ј consensus was mutated to 5Ј-GAGA-3Ј. In contrast with the specific site, binding to the nonspecific site was Ͼ100-fold weaker (2.1 Ϯ 0.2 M) than either site-specific DNA and yielded a Hill coefficient of well above unity, indicative of concerted binding of two or more equivalents of PU.1⌬N167 (Fig. 2B). The titrations therefore showed that sequential dimerization of the PU.1 ETS domain, wherein excess protein self-titrated the canonical 1:1 PU.1/DNA complex to form a 2:1 complex, was exclusive to site-specific DNA longer than 10 bp. Although nonspecific DNA bound PU.1 in Figure 1. Dimerization at a single cognate binding site is intrinsic to the ETS domain of PU.1, but not its structural homolog Ets-1. A, the co-crystal structures of PU.1 (gold; PDB code 1PUE) and Ets-1 (green), the latter with (1MDM) or without (1K79) part of its autoinhibitory domain (blue). All three structures show the canonical 1:1 binding stoichiometry with oligomeric DNA harboring a core 5Ј-GGAX-3Ј consensus (red), as labeled. Note that the cognate DNA sequences in the co-crystal structures are not sequence-identical to the experimental sequences in this study. B, D, and F, representative ITC thermograms at 25°C of DNA-into-protein titrations for the ETS domains of PU.1 (B) and Ets-1 (minimal ϭ ⌬N331 (D); autoinhibited ϭ ⌬N280 (F)). The ordinate is baseline-subtracted and normalized to the amount of DNA delivered per injection to aid comparison; exothermic response is upward. C, E, and G, the titration data for PU.1⌬N167 was empirically fitted as a negatively cooperative interaction. The two phases in the PU.1 titration (marked R␣ and R␤ in C) represent the successive formation of a protein/DNA ϭ 2:1 complex (protein in excess) followed by the 1:1 complex. For Ets-1⌬N331 (E) and Ets-1⌬N280 (G), a 1:1 model was empirically fitted to the data. The stronger and more complex apparent heats associated with the PU.1⌬N167 titrations included the dimerization and binding of PU.1⌬N167 as a 2:1 complex, which dissociates to two 1:1 complexes as DNA reached molar equivalence, in addition to more enthalpically driven 1:1 binding than Ets-1. The details of the thermodynamic deconvolution are provided under "Materials and methods." Self-titration by PU.1 multiple equivalents, it did not form a 1:1 complex at equilibrium (Fig. 2B).
To evaluate site-specific PU.1 dimerization in the presence of excess nonspecific DNA, as would be encountered under genomic conditions, we titrated a 209-bp fragment harboring a single high-affinity PU.1-binding site with PU.1⌬N167. Resolution of the DNA by native gel electrophoresis (Fig. 2C) showed a PU.1-bound band that transitioned to a less mobile band at ϳ100-fold excess protein with respect to the specific site (1 nM), in agreement with the sequential site-specific dimer observed in the fluorescence polarization titrations. The low mobility and broadness of the dimer peak suggested that this complex was interconverting between free and bound states at rates comparable with electrophoretic separation of the two complexes (24). The dimeric peak occurred in advance of a final nonspecific complex that was not detected in the gel. The latter state was verified with an isomeric nonspecific DNA fragment that failed to yield PU.1⌬N167 at any defined stoichiometry (Fig. 2D). Thus, the electrophoretic data showed that the sequential dimerization of PU.1 at a single embedded cognate site effectively resisted competition from excess nonspecific DNA.

Hydrodynamic characterization of PU.1/DNA complexes by NMR spectroscopy
To characterize the solution behavior of the PU.1/DNA complex directly, we interrogated PU.1⌬N167 with site-specific and nonspecific DNA oligonucleotides hydrodynamically by diffusion-ordered NMR spectroscopy (see supplemental Methods). At 250 M protein, we measured a self-diffusion coefficient of (9.1 Ϯ 0.1) ϫ 10 Ϫ11 m 2 /s for unbound PU.1⌬N167 (13.0 kDa) in D 2 O at 25°C. Comparison with a computed value (25) under equivalent conditions for a PU.1⌬N167 monomer derived from the co-crystal structure of the 1:1 PU.1 ETS/DNA complex (see "Materials and methods") found good agreement (8.8 ϫ 10 Ϫ11 m 2 /s). An unbound PU.1 dimer modeled as a pair of rigid spheres would exhibit a diffusion coefficient at 75% of the monomer (26), or ϳ7 ϫ 10 Ϫ11 m 2 /s. Thus, although unbound PU.1 was known to dimerize at very high concentrations (13,27) in vitro, PU.1⌬N167 was monomeric under the conditions of the DOSY experiments.
We tracked the self-diffusion coefficients of PU.1⌬N167 at graded stoichiometric ratios of DNA (supplemental Figs. S2 and S3; parametric values in supplemental Table S1). Titration with 16-bp high-or low-affinity site-specific DNA lowered the apparent diffusion coefficient to a minimum of (5.9 Ϯ 0.1) ϫ 10 Ϫ11 m 2 /s at, within experimental uncertainty, a DNA/PU.1 ratio of 0.5 (i.e. PU.1/DNA ϭ 2:1). The subsequent addition of site-specific DNA past this point increased the diffusion coefficient to a stable value of (7.5 Ϯ 0.2) ϫ 10 Ϫ11 m 2 /s at 1:1 equivalence and beyond (supplemental Figs. S3A). This biphasic behavior was consistent with the other titration data by ITC, fluorescence polarization, and gel mobility shift. This change in DOSY intensity was not due to simple contributions from added DNA because we had carefully avoided peaks that overlapped with DNA (supplemental Fig. S2). The measured diffusion coefficient upon reaching molar equivalence also agreed with a computed value (25) of 7.3 ϫ 10 Ϫ11 m 2 /s based on the 1:1 PU.1/DNA co-crystal structure. Finally, the sequential transitions in diffusion coefficients at half and unit molar equivalence were incompatible with a 2:2 complex. Thus, the DOSY titrations indicated that PU.1 formed exclusively a 2:1 complex with site-specific DNA at PU.1/DNA up to 2:1, followed by a 1:1 complex at molar equivalence and above.
In contrast with site-specific DNA, nonspecific binding by PU.1⌬N167 showed qualitatively different behavior (supplemental Fig. S3C). Specifically, titration of PU.1⌬N167 with 16-bp nonspecific DNA yielded only a single inflection point at a DNA/PU.1 ratio of 0.5 and a stable diffusion coefficient of (6.6 Ϯ 0.2) ϫ 10 Ϫ11 m 2 /s, between the site-specific 1:1 and 2:1 complexes. Thus, the DOSY titration data pointed to the exclusive formation of a defined dimeric nonspecific 16-bp complex. An alternative scenario in which PU.1⌬N167 formed a mixture of 2:1 and 1:1 complexes was not likely, as the apparent diffusion coefficient would be composition-dependent and change upon continued titration of DNA.
As the fluorescence polarization titration showed that the 10 bp of site-specific DNA bound PU.1⌬N167 in a 1:1 complex but was insufficient to invoke sequential dimerization, we repeated the DOSY titrations with a 10-bp duplexes. In stark contrast  (22) harboring the high-affinity site (A) 5Ј-AGCGGAAGTG-3Ј or its isomeric nonspecific variant (B) 5Ј-AGCGAGAGTG-3Ј with PU.1⌬N167. The 10-bp duplex in A was constructed by annealing the labeled strand with a 10-bp complement encoding only the core ETS-binding site. Symbols represent data from independent replicate experiments. Curves represent a least-square fit of the data from triplicate experiments to a sequential 2:1 binding model (red) or constrained to a 1:1 model (green). The latter afforded a significantly poorer fit of the data (p Ͻ 1 ϫ 10 Ϫ4 , Fisher's F-test on sums of squares). The nonspecific data were fitted with the Hill equation (black). Error bars, S.E. C, electrophoretic mobility shift titration of a 209-bp DNA fragment (1 nM, marked 0) harboring a single copy of the high-affinity site with PU.1⌬N167. Following formation of the 1:1 complex (marked 1), a discrete, low-mobility species was present at 0.1 M protein (labeled 2). At 1 M protein (10 3 -fold excess), a nonspecific complex finally formed, which did not enter the gel, as confirmed with a fragment harboring the nonspecific sequence (D). The shadows lining the wells in C represent an irregular imaging artifact of the stained gel, not protein-bound DNA, as it was observed even in the negative-control lane containing no PU.1 (marked 0).

Self-titration by PU.1
with their 16-bp parents, all of the 10-bp complexes regardless of sequence gave indistinguishable hydrodynamic profiles showing single inflections (supplemental Fig. S3, D-F). If the site-specific 10-bp 1:1 complexes maintained the structure observed in the co-crystal structure, their computed (25) diffusion coefficient would be 7.9 ϫ 10 Ϫ11 m 2 /s under the present experimental conditions. Thus, the measured diffusion coefficients of the 10-bp complexes (ϳ6.5 ϫ 10 Ϫ11 m 2 /s), which were significantly lower even than their 16-bp 1:1 counterpart, were unexpected for a 10-bp 1:1 complex. We confirmed the 1:1 stoichiometry of the 10-bp PU.1/DNA complexes by examining the 1 H spectra of the 10-bp high-affinity DNA in the imino region at graded PU.1⌬N167 concentration. Resonances corresponding to free DNA were exhausted by unit molar protein/ DNA ratio (supplemental Fig. S4). Thus, PU.1 bound 10-bp DNA exclusively as monomers even at excess concentrations, and sequential dimerization of PU.1 was limited to site-specific DNA longer than 10 bp. Moreover, the data implied that the protein underwent significant conformational changes to hydrodynamically larger structures than the same protein bound to 16-bp site-specific DNA.

Structural properties of the site-specific PU.1 ETS dimer
We recorded 1 H-15 N HSQC spectra of uniformly 15 N-labeled PU.1⌬N167, which showed a structured protein with well-dispersed cross-peaks in the absence of DNA (Fig. 3A). Upon the addition of 16-bp high-affinity DNA (Fig. 3, B-D) to a DNA/PU.1 ratio of 1:2, we observed the immediate disappearance of ϳ80% of the NH resonances and a marked loss of chemical shift dispersion, but no sign of precipitation even after prolonged incubation (Ͼ24 h). The addition of a second halfequivalent of DNA to DNA/PU.1 ϭ 1:1 promptly restored the NH resonances, with significant chemical shift perturbations compared with the free protein. The further addition of excess DNA produced no further changes to the HSQC spectrum.
When we repeated the HSQC titration using the same highaffinity 10-bp construct as in the DOSY experiments, we observed a progressive disappearance of resonances past the half-equivalence point and no further change past the 1:1 equivalence point (Fig. 3, E-H). The monotonic transition for the 10-bp DNA tracked the changes in diffusion coefficient (Fig. 3, E-H), in contrast with the sharply biphasic behavior of the 16-bp site.
To better understand the effect of binding site size on the bound protein's conformation, we probed DNA-bound PU.1⌬N167 with 8-anilino-1-naphthalenesulfonate (ANS), an indicator dye of solvent-exposed hydrophobic moieties. Unbound PU.1 at 50 M induced strong blue-shifted ANS fluorescence associated with a significant number of untitrated basic residues, which paired with the anionic dye (28), in the absence of DNA (Fig. 4A). DNA alone induced a negligible effect on ANS fluorescence. After subtraction of a dye-only control, ANS fluorescence of PU.1 was reduced about 3-fold upon binding a half-equivalent of the 16-bp site-specific DNA (2:1 complex), and another 8-fold at unit equivalence (1:1 complex) (Fig. 4B). The higher sensitivity to ANS, together with the NMR DOSY data, suggest that the DNA-bound PU.1 dimer may be less structured than in the 1:1 complex.

Topology of the site-specifically bound PU.1 dimer
The attenuation in NMR signal from the 2:1 complex suggests intermediate exchange between these two states. As a result, although the disappearance of 80% of cross-peaks in the DNA-bound PU.1 dimer (cf. Fig. 3B) precluded a direct identification of the residues involved in 2:1 complex formation, the remaining resonances still provided valuable clues to the location of the dimerization interface. We overlaid the HSQC spectra for free and bound PU.1 to 16-bp DNA and identified resonances that showed strong overlap throughout the titration (Fig. 5, A-E). Using a reported 1 H-15 N HSQC assignment of the PU.1 residues 167-260 (27), which applied well to PU.1⌬N167 (supplemental Fig. S5), we identified well-resolved, well-overlapped resonances for Arg 173 , Ala 210 , Lys 224 , Gly 238 , Gly 241 , Lys 247 , and Ser 255 . Resolvable resonances that overlapped only in the free and 1:1-bound states (i.e. no detectable signal in the 2:1 state), including Ser 184 , Trp 192 , Trp 193 , Asp 197 , and Thr 200 , mapped to solvent-exposed surfaces away from the DNA (Fig.  5F). Thus, the HSQC data implicated the distal surface of PU.1⌬N167 opposite the DNA-binding site as a major part of the dimerization interface of the site-specific 2:1 complex.
To further understand how the distal surface of the PU.1 ETS domain was involved in dimerization, we examined the amino

Self-titration by PU.1
acids that mapped to that surface and noticed a sequence of four alternately charged residues, 195 DKDK 198 , that comprise part of a ␤-pleated sheet. These residues include (Asp 197 ) or are proximal to residues (Trp 192 , Trp 193 , and Thr 200 ) whose resonances became reversibly undetectable in the 2:1 complex (Fig.  5, A and D). The 195 DKDK 198 sequence gave rise to a charge distribution that suggested an electrostatically complementary interface for dimerization (Fig. 5G). This hypothesis was further motivated by the low level of sequence conservation in Ets-1 ( 357 TGDG 360 ) and within the ETS family in general (29). We therefore cloned a PU.1⌬N167 mutant harboring 195 NINI 198 , which abrogated the charges but maintained similar side-chain structures and secondary structure propensities (30). In fluorescence polarization and gel mobility shift experiments, the mutant gave titration profiles that showed a single binding mode at up to 10 M, a concentration at which DNAbound wild-type PU.1 had undergone two binding transitions ( Fig. 5H; cf. Fig. 2). The anisotropy and electrophoretic mobility of the bound mutant corresponded to the 1:1 complex formed by wild-type protein. Thus, the 195 NINI 198 mutant confirmed that the distal surface was involved in PU.1 dimerization. In addition, the mutant bound DNA ϳ10-fold more weakly (88 Ϯ 11 nM) than wild-type PU.1⌬N167, suggesting coupling between DNA binding and dimerization of the bound state.

The 2:1 site-specific PU.1/DNA complex occupies an expanded DNA-binding site
To define the contact interface of the 2:1 PU.1/DNA complex, we carried out hydroxyl radical ( ⅐ OH) footprinting titration of a 130-bp radiolabeled DNA fragment harboring the same high-affinity binding site used in the other experiments. Previous footprinting studies of the 1:1 site-specific PU.1/DNA complex by our group (13,31) and others (32,33) have established that two spaced clusters of minor-groove contacts flanking the 5Ј-GGAA-3Ј core consensus generate a highly characteristic ⅐ OH footprint on the 5Ј-TTCC-3Ј strand. We therefore used this biochemical signature to track changes in the DNAbinding site at graded concentration of wild-type PU.1⌬N167 (Fig. 6A). Upon reaching PU.1⌬N167 concentrations of ϳ10 Ϫ7 M that saturated 1:1 complex (marked P1 and P2 in Fig. 6B),

Self-titration by PU.1
additional PU.1⌬N167 gave rise to protected positions near P2 (marked P*). In total, the footprint of the 2:1 complex spanned ϳ20 bp of DNA. Quantitation of the protection from ⅐ OH at the protected bases as a function of PU.1 concentration clearly recapitulated the sequential formation of the 1:1 followed by the 2:1 complex observed in the other experiments (Fig. 6C). In addition, the hypersensitive positions between P1 and P2, which is also diagnostic of site-specific 1:1 binding, was preserved in the 2:1 complex and tracked the titration profiles produced by fluorescence anisotropy (cf. Fig. 2A).
Thus, the ⅐ OH footprints showed, at single-nucleotide resolution, that the site-specific PU.1⌬N167 dimer made extended contacts with the DNA minor groove although the dimerization interface was distal from the DNA. The extended footprint exerted by the 2:1 complex also explained the 10-bp specific site's apparent incompatibility with PU.1 dimerization, which required several more flanking bases downstream of the 5Ј-GGAA-3Ј core consensus, although its affinity for the 1:1 complex was only modestly compromised relative to longer DNA (cf. Fig. 2A).

Discussion
When bound to sufficiently long site-specific DNA, the ETS domain of PU.1 self-associates sequentially to a defined dimer, a behavior that is heretofore unknown for ETS transcription factors (19). The reversibility of dimerization is demonstrated by its independence on directionality. In gel mobility shift, fluorescence anisotropy, or DNA footprinting titrations in which excess protein was titrated into site-specific DNA, the 2:1 complex was produced subsequently to the canonical 1:1 complex. When the titration was reversed (DNA into protein), as was the case in the NMR and ITC studies, limiting concentrations of DNA directly yielded the 2:1 complex. Impressively, the NMR titrations showed that even after prolonged co-incubation, the A singly end-radiolabeled DNA fragment was titrated at equilibrium with PU.1⌬N167 and digested with ⅐ OH under single-hit conditions. A C ϩ T reaction was included to index the digested DNA following denaturing electrophoresis. A, image of the sequencing gel. N and U denote DNA digested without protein and undigested DNA, respectively. A second footprint was observed at a cryptic binding site (5Ј-ATGGGAATTC-3Ј) encoded by pUC19 vector further downstream from the cloned high-affinity site. The lower affinity of this site (48) meant that it did not generate the sequential 2:1 complex beyond the 1:1 footprint at the maximum PU.1 concentration used. B, traces of the indicated lanes. Brackets and red dots denote protected and hypersensitive positions at the indicated and higher protein concentrations, respectively, relative to a distal control peak marked with a hollow dot (E). C, titration of the summed integrated intensities of the protected bases marked P1 and P2 (white squares) associated with the 1:1 complex and P* (black squares) produced by the 2:1 complex in B, normalized to the control peak intensity and scaled to (0, 1). Curves represent empirical fits to the Hill equation. D, titration of the summed integrated intensities of the hypersensitive peaks (red circles), scaled to (0, 1) but normalized to the intensity at the highest PU.1⌬N167 concentration tested. The curve represents a fit by a sequential 2:1 binding model.

Self-titration by PU.1
further addition of site-specific DNA converted the 2:1 complex rapidly and quantitatively into its 1:1 counterpart. Structurally, the identification of the solvent-exposed surface distal from the DNA as the dimerization interface on the one hand, and the expansion of the DNA footprint of the 2:1 complex on the other, suggested an allosteric coupling between the PU.1 dimers and their bound DNA.
Although many DNA-binding domains are known to selfassociate when they bind to site-specific DNA (34), this behavior is associated with systems in which the protein protomers bind multiple DNA subsites independently, such as the Trp repressor (35), or with positive cooperativity, such as the p53 core domain (36). Ets-1 and several other non-PU.1 ETS members can also bind as homodimers, but only to two tandem DNA sites (14 -18). To our knowledge, negatively cooperative binding to a single DNA site, as the PU.1 ETS domain is able to execute, has not been reported previously.

Flanking sequence length as a specificity determinant of PU.1/DNA binding
We tested a range of DNA lengths to define the site size requirements for PU.1 dimerization in the bound state and to probe the relevance of dimerization in the presence of excess nonspecific DNA. We observed that the binding modes accessible to the ETS domain of PU.1 depended on a threshold length of bases flanking the core 10-bp binding site. Within the range of DNA lengths tested in the various experiments, 16-bp and longer DNA invoked sequential dimerization of bound PU.1. In stark contrast, in the absence of flanking bases, the 10-bp DNA bound PU.1⌬N167 exclusively with 1:1 stoichiometry (Figs. 2B and 3). Thus, 10 bp of site-specific DNA was insufficient to elicit the full site-specific behavior of the PU.1 ETS domain. Available evidence indicates that flanking sequence identity is not a determinant because we had observed two other sitespecific DNA sequences yielding the same ITC profiles for PU.1⌬N167 (13).
The 10-bp complex represented a distinct binding mode as the bound PU.1 monomer was structurally different from its 16-bp counterpart as judged by their HSQC spectra (Fig. 3). This observation was unexpected, given the single turn of contacted double helical DNA in the co-crystal structure of the high-affinity PU.1/DNA complex (23) and ETS/DNA structures more generally. Of relevance is the report that DNA with staggered ends was absolutely required for diffraction-quality crystals of the PU.1/DNA complex (37). The overhangs, which paired end-to-end between asymmetric units, would result in essentially continuous DNA in the crystal. Additional interactions with flanking bases that are not part of the core sequence therefore stabilize the bound protein, and without this stabilization, dimerization becomes prohibitive. In summary, flanking sequence length represents an essential additional determinant to fully specify cognate binding by PU.1 in solution.

Functional relevance of self-titration as a potential negative feedback mechanism for PU.1 transactivation
PU.1 is a highly inducible protein, ranging from Ͻ10 to Ͼ200 copies of mRNA per cell in murine bone marrow progenitors, depending on the stage of hematopoietic development (38).
Under physiologic induction, PU.1 mRNA levels matching and even exceeding that of glyceraldehyde 3-phosphate dehydrogenase, an abundant glycolytic housekeeping enzyme, have been measured in cultured (39) and primary (40) human cells. This inducible expression profile suggests that interactions spanning a large range in affinity are likely to be biologically relevant. For instance, NMR characterization of the functionally essential PU.1/GATA-1 interaction estimated its dissociation constant to be no stronger than 10 Ϫ4 M in vitro (41) and did not appear to involve (as judged by chemical shift changes) the dimerization interface of PU.1.
Many ETS family transcription factors, such as Ets-1, ERG, and members of the ETV subfamily, are regulated at the protein/DNA level by inhibitory helices that pack against their DNA-binding domain in the unbound state (Fig. 1A). Perturbing these helices imposes an energetic penalty on DNA-binding that maintains, by default, an autoinhibited state. Binding partners that disrupt the autoinhibitory interactions thus induce a transcriptionally permissive state (19). ETS paralogs, such as PU.1, that lack this mechanism would therefore be locked in a permissive state in the absence of some mechanism for negative regulation. Whereas functionally antagonistic binding partners, such as GATA-1, would serve such an inhibitory role, their expression profiles only partially overlap with that of PU.1 (the common myeloid progenitor in the case of GATA-1 (42)). An intrinsic negative feedback mechanism is hitherto unknown in PU.1. Our observation that PU.1 forms a reversible, negatively cooperative 2:1 complex with site-specific DNA suggests "self-titration" as a potential mechanism of negative feedback: even if the 2:1 complex retains the functional activity of the 1:1 complex, removal of circulating PU.1 alone would attenuate transactivation of target genes. Consistent with this notion, we observed self-titration only with site-specific DNA and not nonspecific DNA. Moreover, we did not observe dimer formation with the structural homolog Ets-1, with or without its autoinhibition helices, when its cysteines were maintained in a reduced state (Fig. 1, D-G). Interestingly, a 2:1 Ets-1/DNA complex was reported under non-reducing conditions (43), reflecting the strong propensity for its two cysteine residues (which are not present in the PU.1 ETS domain) to form nonnative disulfide linkages.
The dissociation constant for binding to oligomeric nonspecific sites (ϳ10 Ϫ6 M) (31), such as that used in our NMR experiments, is only ϳ10-fold higher than the sequential affinity of the second equivalent of PU.1⌬N167. It might therefore appear that the abundance of nonspecific DNA relative to specific sites would overwhelm self-titration of specific complexes. Our gel mobility data on binding to polymeric DNA (cf. Fig. 2C) provide a useful insight into this question. Compared with titration of oligomeric site-specific DNA, formation of the 2:1 complex at an embedded binding site flanked by substantial nonspecific DNA (ϳ100 bp on each side) occurred at ϳ10-fold lower concentration (ϳ10 Ϫ7 M) and clearly preceded any nonspecific binding. The footprinting data showed the same behavior at a shorter (ϳ130-bp) DNA fragment. This difference reflected the favorable contribution to reaching an embedded site from linear diffusion that was absent for an isolated counterpart. Thus, a complete description of the effect of excess nonspecific flank-

Self-titration by PU.1
ing DNA (as would be expected under genomic conditions) includes a competitive effect that is more than offset by favorable contributions from linear diffusion.

Nonspecifically bound PU.1 is oligomeric
The 16-bp nonspecific site, involving only the isomeric reversal of two adjacent positions in the core consensus (5Ј-GGAA-3Ј to 5Ј-GAGA-3Ј), forced the exclusive formation of a dimeric complex. No 1:1 complex was detectable at equilibrium. In the context of self-titration as a potential negative regulatory mechanism, this behavior suggests that the role of sitespecific DNA (i.e. sequences harboring the core consensus) is not only to provide a much higher-affinity binding site for PU.1 but, perhaps more importantly, to "unlock" the transcriptionally active 1:1 conformation. It may therefore be more appropriate to consider dimeric PU.1, as the default autoinhibited state, which becomes activated, by a coupled dissociation/ order transition, upon encountering a specific DNA site at permissively low protein concentrations.

Conclusion
We report, for the first time, a 2:1 complex formed by PU.1 with a single cognate binding site. This complex forms negatively cooperatively with respect to the canonical, transcriptionally active 1:1 complex and resists competition from nonspecific DNA. It is kinetically stable (on the order of many hours) and interconverts efficiently with the 1:1 complex (within minutes) upon the addition of DNA. These biophysical properties of self-titration of PU.1 at site-specific DNA are biologically compatible and, indeed, physiologically appropriate given the significant accumulation of PU.1 under induction (Ͼ10 Ϫ6 M), when negative feedback would be most required to dampen its transcriptional response. Self-titration therefore represents a potential buffering mechanism for self-regulation in ETS paralogs, such as PU.1, that lack autoinhibitory elements in their structures.

Fluorescence polarization titrations
ETS protein binding to fluorescently labeled DNA sites was measured using a Molecular Devices Paradigm plate reader as described (29). In brief, DNA probe (0.5 nM) was incubated to equilibrium with graded concentrations of purified PU.1⌬N167 in a total volume of 30 l of 10 mM Tris-HCl (pH 7.4) containing 150 mM total Na ϩ and 0.1 mg/ml bovine serum albumin. Steady-state fluorescence parallel and perpendicular to the incident polarized light was acquired at 595/35 nm upon excitation at 535/25 nm. Each data point represents the mean Ϯ S.E. of five consecutive measurements as an indication of instrumental noise. Anisotropy data were fitted with a 1:1 or sequential binding model (22) to directly estimate the dissociation constants of the PU.1/DNA 1:1 and 2:1 complexes.

ANS fluorescence
ANS (ammonium salt, Alfa Aesar) was prepared at 2 mM in 10 mM NaH 2 PO4/Na 2 HPO 4 (pH 7.4) buffer containing 150 mM NaCl and stored in the dark at 4°C. Triplicate samples of PU.1 with or without 16-bp high-affinity DNA plus various controls were prepared in the same buffer before the addition of ANS to 200 M. Final concentrations of PU.1 and DNA were as indicated under "Results." After incubation for 30 min, the Self-titration by PU.1 fluorescence intensity of each sample was read at 370/530 nm or scanned from 400 to 750 nm with a Paradigm plate reader.

Hydroxyl radical DNA footprinting
A 130-bp DNA fragment harboring a copy of the high-affinity PU.1-binding site 5Ј-AGCGGAAGTG-3Ј was generated by PCR using two primers, of which the one encoding the 5Ј-CACTTCCGCT-3Ј strand had been 5Ј-end-labeled with [ 32 P]ATP. After purification by agarose gel electrophoresis, the radiolabeled fragment (Ͻ1 nM) was incubated to equilibrium with graded concentrations of PU.1⌬N167. Each sample was digested with hydroxyl radical, purified, resolved by denaturing polyacrylamide gel electrophoresis, and digitized by phosphorimagery as described (31). Lane traces were constructed, and bands were indexed using a C ϩ T chemical sequencing reaction. Peaks were fitted as a superposition of Gaussian distributions, numerically integrated, and normalized to a band outside of the binding site to quantify fractional protection relative to the unbound sample.

Structure-based calculations
Self-diffusion constants for unbound and 1:1 DNA-bound PU.1⌬N167 were computed using the software HydroPRO (25). DNA-bound and unbound PU.1 structures were templated from the co-crystal structure with DNA (PDB code 1PUE) (23), appended with additional residues present in PU.1⌬N167, and relaxed by all-atom molecular dynamics simulation for 200 ns following our established protocol (29). Computations were carried out using volumetric values for D 2 O at 25°C, namely a density of 1.107 g/ml, viscosity of 1.25 centipoises, and partial specific volume of 0.70 ml/g. Continuum electrostatics of PU.1 in the co-crystal structure were computed using APBS (45). Calculations were performed for an aqueous solution containing 0.15 M NaCl at 25°C and rendered on the solvent-accessible surface from Ϫ1 to ϩ1 kT/e.

ITC
Purified PU.1 or Ets-1 was dialyzed extensively together with 23-bp DNA harboring the protein's respective optimal target (5Ј-AGCGGAAGTG-3Ј for PU.1; 5Ј-GCCGGAAGTG-3Ј for Ets-1) in separate compartments, against 50 mM NaH 2 PO 4 / Na 2 HPO 4 , pH 7.4, 150 mM total Na ϩ , 0.1 mM EDTA, and 1 mM dithiothreitol. Titrations were performed by injecting DNA (initial concentration ϳ500 M) into protein (ϳ50 M) in a Nano ITC instrument (TA Instruments). Data fitting to empirical 1:1 and cooperative models was performed as described (46) only to demonstrate the models qualitatively, not for quantitative estimation of the binding affinities, due to the very strong dissociation constant of the 1:1 complex (10 Ϫ9 M).
To compare the calorimetric enthalpies for DNA binding by PU.1 with those for Ets-1, which exhibited strictly 1:1 behavior, the calorimetric enthalpies for PU.1/DNA binding were decomposed as follows to extract the enthalpy changes for the 1:1 complex. Whereas PU.1⌬N167 dimerizes in both unligated and DNA-bound states, the former occurs at considerably higher concentrations (near 10 Ϫ3 M) (13,27) than those used in the "reverse" DNA-into-protein titrations shown in Fig. 1 (A and  B). Under these conditions, the biphasic profile arises from the 1:1 complex being strongly favored and yielding a 2:1 complex only in excess protein at the initial phase (R␣) of the titration as shown in Scheme 1, where P and D represent PU.1⌬N167 and site-specific DNA in their various free and bound states. The two phases R␣ and R␤ are marked in the reverse titration shown in Fig. 1B. Because the transition from the 1:1 to 2:1 PU.1/DNA complex occurs sequentially, both phases are well-defined and extracted according to the technique of "total association at partial saturation" (47). To compare the enthalpy changes meaningfully with the manifestly 1:1 binding for Ets-1 in Fig. 1 (C-F), the complex heats in the reverse titrations are dissected to account for the thermodynamics of coupled dimerization and dissociation of PU.1 as shown in Scheme 2. The calorimetric enthalpy marked F␤ has been measured previously for PU.1⌬N167 under the same solutions (13). Based on ⌬H R␤ ϭ Ϫ44.2 Ϯ 1.4 kJ/mol (cf. Fig. 1B) and ⌬H F␤ ϭ 17.1 Ϯ 0.7 kJ/mol (13) at 25°C, the enthalpy change for the formation of the canonical 1:1 complex was Ϫ27.1 Ϯ 1.6 kJ/mol. Thus, the enthalpy change for formation of the 1:1 complex from unbound constituents was larger in magnitude than that for minimal (⌬H 1:1 ϭ Ϫ12.0 Ϯ 0.4 kJ/mol) and autoinhibited Ets-1 (⌬H 1:1 ϭ Ϫ8.1 Ϯ 0.4 kJ/mol) at 25°C.