Conformational Dynamics and Molecular Recognition: Backbone Dynamics of the Estrogen Receptor DNA-binding Domain

Article No. jmbi.1999.2806 available online at http://www.idealibrary.com on

J. Mol. Biol. (1999) 289, 963±979

Conformational Dynamics and Molecular Recognition: Backbone Dynamics of the Estrogen Receptor DNA-binding Domain Anja WikstroÈm1,3, Helena Berglund1, Charlotta Hambraeus3, Susanne van den Berg2 and Torleif HaÈrd1* 1

Department of Biotechnology Royal Institute of Technology (KTH) 2 Department of Biosciences Karolinska Institutet 3 University of Southern Stockholm Center for Structural Biochemistry, Novum, S-141 57 Huddinge, Sweden

We examined the internal mobility of the estrogen receptor DNA-binding domain (ER DBD) using NMR 15N relaxation measurements and compared it to that of the glucocorticoid receptor DNA-binding domain (GR DBD). The studied protein fragments consist of residues Arg183-His267 of the human ER and residues Lys438-Gln520 of the rat GR. The 15N longitudinal (R1) and transverse (R2) relaxation rates and steady state {1H}-15N nuclear Overhauser enhancements (NOEs) were measured at 30  C at 1H NMR frequencies of 500 and 600 MHz. The NOE versus sequence pro®le and calculated order parameters for ER DBD backbone motions indicate enhanced internal dynamics on pico- to nanosecond time-scales in two regions of the core DBD. These are the extended strand which links the DNA recognition helix to the second zinc domain and the larger loop region of the second zinc domain. The mobility of the corresponding regions of the GR DBD, in particular that of the second zinc domain, is more limited. In addition, we ®nd large differences between the ER and GR DBDs in the extent of conformational exchange mobility on micro- to millisecond time-scales. Based on measurements of R2 as a function of the 15N refocusing (CPMG) delay and quantitative (Lipari-Szabo-type) analysis, we conclude that conformational exchange occurs in the loop of the ®rst zinc domain and throughout most of the second zinc domain of the ER DBD. The conformational exchange dynamics in GR DBD is less extensive and localized to two sites in the second zinc domain. The different dynamical features seen in the two proteins is consistent with previous studies of the free state structures in which the second zinc domain in the ER DBD was concluded to be disordered whereas the corresponding region of the GR DBD adopts a stable fold. Moreover, the regions of the ER DBD that undergo conformational dynamics on the micro- to millisecond timescales in the free state are involved in intermolecular protein-DNA and protein-protein interactions in the dimeric bound state. Based on the present data and the previously published dynamical and DNA binding properties of a GR DBD triple mutant which recognize an ER binding site on DNA, we argue that the free state dynamical properties of the nuclear receptor DBDs is an important element in molecular recognition upon DNA binding. # 1999 Academic Press

*Corresponding author

Keywords: steroid hormone receptors; NMR spectroscopy; protein dynamics; DNA-binding proteins; molecular recognition

Abbreviations used: ER, estrogen receptor; GR, glucocorticoid receptor; DBD, DNA-binding domain; ERE, estrogen response element; NOE, nuclear Overhauser enhancements; TOCSY, total correlated spectroscopy; HSQC, heteronuclear single quantum spectroscopy; NOESY, NOE spectroscopy; DD, dipole-dipole; CSA, chemical shift anisotropy. E-mail address of the corresponding author: [email protected] 0022-2836/99/240963±17 $30.00/0

# 1999 Academic Press

964

Introduction The estrogen and glucocorticoid receptors (ER and GR, respectively) are ligand-activated transcription factors belonging to the superfamily of nuclear receptors (for reviews, see Mangelsdorf et al., 1995; Truss & Beato, 1993; Tsai & O'Malley, 1994). The DNA binding of these proteins is mediated through a small DNA-binding domain (DBD) which is conserved in all members of the superfamily. The conserved core DBD contains about 70 amino acid residues for which proper folding and DNA binding requires the coordination of two zinc ions by conserved cysteine sidechains as illustrated in Figure 1 (Freedman et al., 1988). Full-length ER and GR bind as dimers to estrogen and glucocorticoid response elements (ERE and GRE, respectively) for which optimal sequences consist of two six base-pair half-sites organized as inverted repeats with a three basepair spacing (Lucas & Granner, 1992, and work cited therein). The consensus half-site sequences are 50 AGGTCA30 for the ER and 50 AGAACA30 for the GR. Isolated DBDs bind DNA with reduced af®nity compared to full-length receptors, but with

Backbone Dynamics in the ER DBD

retained speci®city. The differences in DNA-binding af®nity between receptors and isolated DBDs can, to a large extent, be ascribed to the stronger dimerization ability of the full-length receptors due to interactions involving C-terminal (ligand-binding) domains (Brzozowski et al., 1997; Cairns et al., 1991; Kumar & Chambon, 1988). Isolated DBDs are monomeric in solution at high concentrations (see, e.g. Berglund et al., 1997; Schwabe et al., 1993a). Still, part of the overall dimerization activity resides also in the DBDs, where it is localized to the second zinc domain. As a consequence, dimeric binding of the ER and GR DBDs to their response elements on DNA is cooperative (HaÈrd et al., 1990a; Dahlman-Wright et al., 1991; Luisi et al., 1991; Schwabe et al., 1993a). Structures of the free ER and GR DBDs in solution have been determined using NMR (HaÈrd et al., 1990b; Schwabe et al., 1990, 1993a; Baumann et al., 1993; van Tilborg et al., 1995) and several structures of corresponding DBDDNA complexes have been determined using X-ray crystallography (Luisi et al., 1991; Schwabe et al., 1993a, 1995; Xu et al., 1993; Gewirth & Sigler, 1995). The DBD fold (Figure 2) consists of two units with the overall composition zinc

Figure 1. Zinc ®nger representation of the ER and GR core DNAbinding domains indicating locations for backbone dynamics on different time-scales. Residue numbering corresponds to those of the human ER and rat GR. (The amino acid sequence of human GR is identical with that of the rat.) Fragments in continuous-line boxes form helical secondary structure elements in solution (Baumann et al., 1993; Schwabe et al., 1993b). The ER DBD fragment in the broken-line box assumes the distorted helical conformation only upon binding to DNA. Residues for which backbone 15N undergo picoto nanosecond dynamics (50 MHz 1 H-15N NOE < 0.62) are indicated by outline letters in grey boxes. Residues with backbone micro- to millisecond exchange dynamics (calculated Rex > 3 sÿ1) are indicated by arrows. Parentheses indicate that a full set of relaxation data could not be obtained.

965

Backbone Dynamics in the ER DBD

domain - helix - extended region. The two amphipatic helices are packed perpendicular to each other forming the core. The N-terminal (®rst) zinc domain forms a loop which is folded by hydrophobic interactions between conserved side-chains and the core. Residues in the following helix I make direct or water-mediated interactions with base-pairs in the DNA major groove and are primarily responsible for sequence-speci®c binding. Additional residues in this helix and the loop of the ®rst zinc domain interact with DNA phosphate groups. The Cterminal (second) zinc domain begins with the so-called D-box, which forms a loop consisting of ®ve residues (Pro222 to Gln226 in the ER). The D-box together with Thr228 and Ser236 form the protein-protein interface in the DNAbound dimer. The D-box is followed by an extended peptide fragment. In the DNA-bound state, residues Asn232 to Ser236 form a short distorted helix II which is not present in the uncomplexed ER DBD (Schwabe et al., 1993b). Residues Arg234 and Lys235 in this helix make non-speci®c contacts with DNA phosphate groups. The second zinc domain ends with the amphipatic helix III and an extended region which is held to the core by hydrophobic interactions involving conserved side-chains. The present study addresses the internal dynamics in the free ER DBD and how the dynamics compare to that of the GR DBD. The studies were initiated to resolve the origin of apparent differences in the free state structural properties of the second zinc domain in the two proteins. These differences can be summarized as follows. The ER and GR DBDs assume nearly identical conformations in the DNA-bound states (see Discussion). However, the structures of the free and DNAbound states of the ER DBD are different. The differences are primarily located to the second zinc domain which appears to be disordered in the free state, but folded in the bound state. The inference

that a coupled folding reaction occurs upon DNA binding (Schwabe et al., 1993b) is also consistent with the presence of two dimeric DNA complexes in crystals which differ in the conformation of the second zinc domain (Schwabe et al., 1993a). The conformations of the free and DNA-bound states of the GR DBD are also different. Still, with the GR DBD the situation is quite different compared to that of the ER DBD: the solution structure is well de®ned (Baumann et al., 1993) with only limited backbone mobility on short time-scales (Berglund et al., 1992), indicating that the GR DBD fragment which was used in these studies adopts a stable conformation in the free state. In this case, however, DNA-binding and dimerization induce a conformational adaptation of the second zinc domain, including a re-orientation of the D-box (Baumann et al., 1993). The absence of a coupled folding reaction in the GR DBD upon DNA binding is consistent with the thermodynamics for the formation of a complex (LundbaÈck & HaÈrd, 1996b). Thus, it appears that the two very closely related ER and GR DBDs, which assume almost identical structures in the DNA-complexed states, display locally different structures, folding stabilities and/ or conformational dynamics in the free states. Further investigations into the matter are motivated for several reasons. First, a characterization of backbone dynamics by means of 15N relaxation (e.g. see Kay et al., 1989; Palmer et al., 1996; Tjandra et al., 1995) in the free ER DBD is called for to complement the structural studies. These experiments are needed to quantify to what extent some regions, e.g. the second zinc domain, exhibit more dynamics in the uncomplexed ER DBD compared to the GR DBD. Second, and more important, internal dynamics and ¯exibility is a potential factor in the mechanism for how the ER and GR discriminate between their respective binding sites (response elements) on DNA. In this context it is critical to investigate if any internal dynamics in the ER DBD is indeed located at or around sites involved in intermolecular interactions and how these sites compare to the regions of the GR DBD that undergo conformational changes upon dimeric DNA binding.

Results Assignments

Figure 2. Backbone fold representation of the ER DBD in which locations for rapid backbone dynamics in the free state are indicated by darker colours: grey, 50 MHz 1 H-15N NOE < 0.62; black, NOE < 0.50. The structure shown is that of the DNA-bound ER DBD (Schwabe et al., 1993a).

Backbone amide resonances in the ER DBD were assigned by identifying spin systems and sequential nuclear Overhauser enhancements (NOE) connectivities in 3D total correlated spectroscopyheteronuclear single quantum coherence (TOCSYHSQC) and NOE spectroscopy (NOESY)-HSQC spectra, respectively. Complementary information on Ha chemical shifts was obtained from the 3D HNHA spectrum. All backbone amide resonances in the ER DBD fragment Arg183-Arg256 and those of Asp258 and Gly261 could be assigned in the 15N HSQC spectrum (Figure 3). The amide proton

966

Backbone Dynamics in the ER DBD

Figure 3. Assigned 1H-15N HSQC spectrum of the ER DBD. The expanded region illustrates the different lineshapes of some 1H-15N correlation cross-peaks.

chemical shift assignments agree well with those reported by Schwabe et al. (1993b). The backbone amide assignments of the present GR DBD fragment have been reported previously (Baumann et al., 1993; Berglund et al., 1992). Relaxation measurements and qualitative evaluation of dynamics Measured relaxation data are reported in Figures 4 and 5. Spectral overlap prohibited accurate measurements on Asp218, Arg234, Cys245 and Met250 in the ER DBD. The Ile229 resonance is very weak due to broadening and/or proton exchange with water, and relaxation data on this resonance contain larger experimental uncertainties. Measurements on the GR DBD include all important residues in the core DBD except for Asn473, Ile482, Pro493, and Ala494 (the corresponding ER residues are 218, 227, 238 and 239). The heteronuclear 1H-15N NOE is the relaxation parameter which is most sensitive to internal motions on the pico- to nanosecond time-scale. The NOE versus sequence pro®le for the ER DBD shown in Figure 4(a) allows for a qualitative identi-

®cation of regions with more extensive internal dynamics. Restricted dynamics can be observed in the N-terminal region, including the ®rst zinc domain and the ®rst helix where NOE values are uniformly high. The average NOE for the helical region Cys202-Ile213 is 0.72(0.02). Similarly, large NOEs are measured within the fragment Ala223Thr228 in the second zinc domain and for most residues in the second helix. Three regions display NOEs that are signi®cantly lower compared to those of the N-terminal, suggesting more extensive internal dynamics. These are Gln214-Cys221 in the extended peptide strand that follows the ®rst helix, Asp230-Cys237 in the second zinc domain, and the C-terminal region beginning with Lys252. These regions have been mapped onto the structure of the ER DBD in Figure 2. The heteronuclear NOEs measured for the GR DBD (Figure 4(a)) agree well with those reported by us earlier (Berglund et al., 1992). As with the ER DBD, there are large NOEs in most of the N-terminal zinc domain and in the two longer helices I and III, and lower NOE values in the extended strand that follows the ®rst helix and in the Cterminal. However, the NOEs of the GR and ER

Figure 4. Backbone 15N relaxation parameters measured in the ER ( & ) and GR (*) DBDs at 50 MHz 15N resonance frequency. The residue numbering corresponds to that of the human GR DBD sequence (top) and ER DBD sequence (bottom), respectively. (a) Steady-state 1H-15N NOEs. The data represent measurements of I/I0, where I and I0 are measured in the presence and absence of 1H saturation, respectively. (b) 15N R1 relaxation rates. (c) 15N R2 relaxation rates measured with a CPMG refocusing delay of 900 ms.

Backbone Dynamics in the ER DBD

967

968

Backbone Dynamics in the ER DBD

Figure 5. R2 relaxation rates of backbone 15N nuclei in the ER DBD measured at 60 MHz with CPMG refocusing delays of 1848 (*), 900 (&), and 400 ( & ) ms. Experimental errors are similar to those shown in Figure 4(c).

DBDs differ signi®cantly in several regions of the molecules. First, the NOEs of Ser448 and Gly449 in the ®rst zinc domain are signi®cantly lower in the GR DBD than those of the corresponding ER DBD residues (Ser193 and Gly194). This suggests a somewhat larger mobility for GR. Second, measured NOEs do not re¯ect the same extent of internal dynamics in the extended strand (corresponding to ER region Gln214-Tyr219), although the NOEs are clearly lower here than in the ordered regions. The third difference concerns the Asp485-Cys492 fragment of the second zinc domain (ER residues Asp230-Cys237). Here, the NOEs measured for the GR DBD are not as low as those measured in the ER DBD, indicating more extensive rapid internal dynamics in the latter protein. The difference is most pronounced with residues Lys235-Cys237 for which the measured NOEs are 0.69, 0.65 and 0.69, respectively, in the GR DBD compared to 0.49, 0.34 and 0.54 in the ER DBD. The 15N R1 relaxation rates measured at 50 MHz are plotted in Figure 4(b). R1 rates are less sensitive to internal dynamics than the heteronuclear NOEs. The pro®le of R1 versus sequence in the ER DBD is also similar to the NOE pro®le, although the effects of internal dynamics (slower R1 rates) are less pronounced. Still, the three regions of the ER DBD with low NOE values also exhibit slower R1 rates (R1 < 2.0 sÿ1). In the GR DBD the distribution of R1 rates is narrower. Trimmed average R1 rates of 2.17(0.09) sÿ1 and 2.28(0.08) sÿ1 were calculated for the ER and GR DBDs, respectively, by excluding values that differ by more than two standard deviations from the overall R1 averages. Figure 4(c) shows backbone 15N R2 relaxation rates at 50 MHz. These rates are widely dispersed in the ER DBD where they range from 1.6 to

16.8 sÿ1 (His216 and Ala223, respectively). Slower R2 rates for residues Gly215-Tyr219 and the C terminus are consistent with the corresponding pro®les of the NOE and R1 data discussed above. Several residues show R2 rates that are signi®cantly more rapid than those observed e.g. in ordered helical regions. These include most residues within the fragments Tyr195-Trp200 and Cys221-Thr228 (the D-box) in the ®rst and second zinc domains, respectively. Anomolously large R2 rates might be due to either large contributions (Rex) from conformational exchange dynamics on micro- to millisecond time-scales or to anisotropic rotational diffusion of the protein. To directly test the ®rst of these possibilities we studied the dependence of measured R2 rates on the length of the CPMG delay. These measurements were carried out at 60 MHz using three different settings for the delay between refocusing 180  15N pulses: 400, 900 and 1896 ms. The data in Figure 5 reveal a clear monotonic CPMG delay dependence in all high R2 rates, i.e. those of Tyr195, Gly198, Val199, Trp200, Cys221, Ala223, and Cys227. Thus, conformational exchange on micro- to millisecond time-scales contributes to the measured R2 rates in these residues. The R2 values of several other residues also appear to have a CPMG delay dependence. The Rex contribution to these rates is further evaluated below following an assessment of the extent of anisotropic rotational diffusion. We note that the longest of the three CPMG delays used (1896 ms) is too long to completely neglect the evolution of in-phase 15Nx magnetization into antiphase ÿ215N1yHz coherence. These measured R2 rates therefore contain a contribution from the more rapid relaxation of the anti-phase component (Palmer et al., 1992). The average error due to this effect can be estimated (using the theory

969

Backbone Dynamics in the ER DBD

described by Palmer et al., 1991, 1992) to about 3 %, which is too small to affect the conclusions. The R2 rates measured in the GR DBD show a much more limited dispersion than those in the ER DBD (Figure 4(c)). Still, three residues show large R2 rates, possibly re¯ecting micro- to millisecond exchange dynamics. These are Ile483, Ile484 and Cys492 which correspond to ER residues 228, 229 and 237, respectively. The large rates measured for the two Ile residues were not apparent in our previous measurements of R1r, whereas Cys492 did show a high R1r value (Berglund et al., 1992). Our previous R1r data were, on the other hand, not of suf®ciently high precision to allow conclusions regarding the presence or absence of micro- to millisecond dynamics in the GR DBD. Overall rotational diffusion constants and anisotropy Time constants for rotational diffusion were evaluated based on R2/R1 ratios. The extent of anisotropic rotational diffusion was also estimated based on the known solution structures of the two proteins following the procedures described by Tjandra et al. (1995) and Lee et al. (1997). For this analysis we ®rst excluded residues with large amplitude rapid backbone dynamics, i.e. those with 50 MHz 1H-15N NOEs lower than 0.62 (21 and 12 residues in the ER and GR DBDs, respectively). The average R2/R1 ratio of the remaining 48 residues in the ER DBD is 4.35(1.08) and that of the remaining 55 residues in the GR DBD is 3.69  0.80. We then excluded those ratios for which R2 values are most likely to contain exchange contributions. Following Tjandra et al. (1995), we calculated normalized R2ÿR1 differences as (hR2i ÿ R2,n)/hR2i ÿ (hR1i ÿ R1,n)/hR1i. These differences average to ÿ0.001(0.245) and 0.002(0.202) for the ER and GR DBDs, respectively. Residues with R2/R1 values for which the normalized R2ÿR1 difference exceeds the average by more than one standard deviation were then excluded (eight residues in the ER DBD and four residues in the GR DBD).

Optimization of the rotational diffusion tensor in the ER DBD is thus based on 40 known NH vectors in the structure (Schwabe et al., 1993b) and the corresponding R2/R1 ratios. Incidentally, only one of these vectors (Gln226) is located in a region of the protein for which the local backbone conformation could not be well determined (local order parameters for backbone dihedrals <0.6). The exclusion of this residue also has a negligible effect on the outcome of the analysis. The results of ®tting isotropic, axially symmetric and fully anisotropic rotational diffusion models according to Lee et al. (1997) are shown in Table 1. A corresponding test of axially symmetric versus isotropic rotational diffusion according to the alternative procedure, i.e. using equation (5) (see Materials and Methods) without prior calculations of local diffusion coef®cients, leads to very similar results (data not shown). The overall rotation of the present ER DBD fragment appears to be isotropic and the optimal value of the rotational correlation time based on the R2/R1 analysis is tM ˆ 6.42(0.03) ns. The optimization of an axially symmetric diffusion tensor improves the ®t somewhat, but the improvement is not considered statistically signi®cant (F ˆ 2.06, p ˆ 0.12; see Table 1). The optimization of the rotational diffusion tensor in the GR DBD is based on R2/R1 values for 51 residues. In this case the overall rotation appears to be moderately anisotropic because the axial diffusion model represents a statistically signi®cant improvement compared to the isotropic model with F ˆ 6.75 and p ˆ 0.0007 (Table 1). The best-®t parameters are tM,iso ˆ 5.98(0.03) ns, Dpar/ Dper ˆ 1.19(0.03), y ˆ 93(4) and f ˆ 5(5) . A value of Dpar/Dper > 1 indicates that the symmetry of rotation is that of a prolate ellipsoid. The y and f angles are the Euler angles for rotating the PDB entry 1GDC so that the z-axis coincides with the principal axis of the axially symmetric diffusion tensor. The orientation with respect to the structure is such that the principal rotation axis is perpendicular to the ®rst helix and co-linear with a thought extension of the C terminus. As judged by the structure of the ordered regions one would

Table 1. Experimental rotational diffusion parameters for the ER and GR DBDs Protein

Model

tM,iso (ns)a

ER DBD

Isotropic Ax. symm. Anisotropic Isotropic Ax. symm. Anisotropic

6.42  0.03 6.44  0.04 6.46  0.04 5.97  0.02 5.98  0.03 5.98  0.03

GR DBD

yb (deg.)

fb (deg.)

0.86  0.04 1.17  0.05e 1.12  0.05

35  7 105  15

88  13 23  66

25  22

1.19  0.03 1.20  0.03e 1.02  0.03

93  4 94  6

55 4  150

36

Dpar/Dper

Dx/Dy

jb (deg.)

w2red 2.50 2.31 2.32 3.68 2.74 2.84

Fc

pd

2.06 0.91

0.12 0.41

6.75 0.13

0.0007 0.87

Analysis based on R2/R1 values for 40 and 51 NH vectors in the average NMR structures ER and GR DBDs, respectively (PDB entries 1HCP and 1GDC). a Effective correlation time (1/6Diso). b Euler angles describing the orientation of the diffusion tensor in the PDB coordinate frame. (Values obtained for the ER and GR DBDs are not directly comparable because the PDB orientations of the two proteins are different.) c F-ratios (Devore, 1995) for assessing the validity of a reduction in w2red when additional parameters are added to the model. d Probability that the reduction in w2red is achieved by chance. Probabilities were calculated in Mathematica 3.0 (Wolfram, 1996) as p ˆ 1 ÿ CDF[F-ratioDistribution[x, N-m-x], Fx], to compare ®tting of models with m and m ‡ x parameters to N data points. e Dpar/Dper ˆ 2Dz/(Dx ‡ Dy).

970

Backbone Dynamics in the ER DBD

Figure 6. Order parameters (S2) for rapid backbone motions in the ER and GR DBDs.

expect that any anisotropic rotation of the core GR DBD would resemble that of an oblate ellipsoid. However, the GR DBD fragment under study contains an additional ten C-terminal residues which are not included in the PDB entry. This fragment could not be fully assigned and structurally characterized (Baumann et al., 1993) due to lack of sequential NOE connectivities and it is very likely to be disordered. Still, the conformation must be restrained to occupy a volume which is excluded by the ordered core domain. The additional mass of ten (disordered) residues added to the C terminus of the GR DBD core can, therefore, be expected to signi®cantly change the moments of inertia. Because it is disordered it might also change frictional properties for rotation of the connected folded domain. Modeling based on molecular dynamics simulations of a disordered C terminus (not shown) indicated that the diffusion axis is aligned with the core plus an additional C-terminal mass and that the determined diffusion tensor therefore is physically reasonable. Given this result one would perhaps also expect that the ER DBD showed a similar rotational diffusion anisotropy. However, the more extensive internal dynamics (reported here) and disorder (Schwabe et al., 1993b) found in the core ER DBD might affect also the overall rotational tumbling to make it more isotropic than that of the GR DBD. Another possibility which we cannot exclude is that the C-terminal tail of the ER DBD interacts weakly with the core DBD and makes tumbling become more isotropic. Order parameters for rapid internal dynamics The extent of rapid internal dynamics of the ER and GR DBDs was quanti®ed by optimizing parameters of the Lipari-Szabo spectral density function (equation (4)) as described in Materials and

Methods. Initially assuming an overall rotational correlation time of 6.4 ns for the ER DBD, we examined for which residues the inclusion of an Rex contribution to R2 signi®cantly improved the optimization. This was the case for 24 out of 69 residues. Following model selection all dynamics parameters were optimized resulting in an optimal value of tM ˆ 6.16(0.01) ns. Order parameters (S2) obtained from this optimization are shown in Figure 6. Optimized Rex terms are shown in Figure 7 and the location of residues with Rex in the structure is illustrated in Figure 8. Correlation times for internal motions (te) average to 60(32) ps with an average error of 9(7) ps for all residues except 203 (te ˆ 377) and 254 (te > 500). According to the criteria given in Materials and Methods, the optimization resulted in 14 very good, 18 good, 18 acceptable and 19 poor ®ts of local dynamics. Relaxation data for the 19 residues that could not be ®t using equation (4) (see Materials and Methods) were also ®tted to the extended Lipari-Szabo spectral density function (Clore et al., 1990) including internal dynamics on two time-scales, but this did in no case improve the quality of the ®t. The 19 residues which appear to conform less well to Lipari-Szabo-type dynamics, i.e. for which ÿi > 9 w20.05, are residues 185, 188, 190, 211, 214, 216, 233, 239, 243, 246, 248, 249 and 252-261. The poor precision in measured relaxation parameters for Ile229 prohibits a quanti®cation of the backbone dynamics. However, the R2 rate is still clearly more rapid than the average rate. For this residue we therefore calculated Rex based on the difference between the average R2 ˆ 8.4(0.55) sÿ1 (for 29 residues without Rex or low NOE value) and the measured rate, R2 ˆ 14.4(0.9) sÿ1 (Figure 7). Quanti®cation of local backbone dynamics in the GR DBD were carried out with consideration of the

Backbone Dynamics in the ER DBD

971

Figure 7. Conformational exchange (Rex) contributions to R2 relaxation rates in the ER (wide grey bars) and GR (narrow black bars) DBDs.

overall rotational anisotropy. An axial diffusion tensor ratio of Dpar/Dper ˆ 1.19 and an average tM,iso ˆ 5.98 ns closely corresponds to possible local correlation times tM,i in the range 5.81 to 6.36 ns, depending on the orientation of the NH vector with respect to the diffusion symmetry axis. Local correlation times were ®xed to values calculated based on equation (7) and the average NMR

structure (Baumann et al., 1993) rotated into the diffusion frame of reference. Parameters for local dynamics were then optimized and the resulting order parameters are shown in Figure 6. Tests based on F-ratios indicated that inclusion of Rex contributions to R2 signi®cantly improve the optimizations for 25 of 67 residues, although the optimized Rex values (Figure 7) in most cases are small

Figure 8. Structure of a dimeric ER DBD-ERE complex (Schwabe et al., 1993a). Dark backbone segments indicate locations for micro- to millisecond conformational exchange dynamics (calculated Rex > 3 sÿ1) in the free state of the ER DBD.

972 (<1 Hz). The quality of the ®t shows an overall distribution which is similar to that of the ER DBD. Local correlation times obtained average to te ˆ 80(51) ps. The residues for which Rex are included ®t with ÿi ˆ 0 (equal number of parameters and data points). For the remaining 42 residues there are 13, 13, six and ten optimizations which can be classi®ed as very good, good, acceptable, and poor, respectively. Residues with poor ®ts correspond to 449, 452, 459, 467, 471, 475, 503, and 507-510. For comparison we also carried out an optimization of GR DBD data without considering the anisotropic motion, i.e. following the procedures used with the ER DBD. Exchange terms were in this case included for eight residues, and the optimized (isotropic) correlation time is tM ˆ 5.76(0.01) ns. The calculated order parameters are in this case very similar to those obtained when rotational diffusion anisotropy is considered with an average difference
S2 ˆ hS2anisotropic ÿ S2isotropici ˆ 0.004(0.016) and a root-mean-square difference (rmsd) ˆ h[(S2anisotro2 2 1/2 i ˆ 0.012(0.011). The largest and pic ÿ Sisotropic) ] smallest differences are obtained for residues 483 and 452 for which
S2 ˆ 0.038 and
S2 ˆ ÿ 0.063, respectively. The optimized Rex values are also similar in the two models with average and root mean square differences of ÿ0.44(0.32) sÿ1 and 0.43(0.30) sÿ1, respectively. Thus, a quantitative analysis of S2 and Rex parameters is in this case relatively insensitive to effects caused by rotational diffusion anisotropy.

Discussion NMR 15N relaxation measurements allow for investigation of internal dynamics occurring in two time-scale windows: internal dynamics which is more rapid than the overall rotational motion, i.e. the pico- to nanosecond time-scale, and conformational exchange dynamics on the micro- to millisecond time-scale that result in increased R2 rates and excess broadening of 15N NMR resonances. By probing these two time-scales in a direct comparison of the two DBDs, we can reconcile the different observations regarding the free state structural properties. Second, and more important, our studies point to internal dynamics and ¯exibility as a potential factor in the mechanism by which the ER and GR bind to their respective binding sites on DNA. Rapid pico- to nanosecond backbone dynamics in the ER DBD The NOE data and optimized order parameters indicate that three regions of the ER DBD undergo rapid dynamics on the pico- to nanosecond timescale (Figures 1, 2, 4(a) and 6). The regions with low NOEs are: (i) Gln214-Cys221 in the extended peptide strand that follows the ®rst helix; (ii) Asp230-Cys237 in the second zinc domain; and (iii) the C-terminal beginning with Lys252.

Backbone Dynamics in the ER DBD

The Gln214-Cys221 fragment is disordered in the NMR structure of the ER DBD and this is consistent with the rapid dynamics. In the DNA-bound state (Schwabe et al., 1993(a)) it appears to be anchored by hydrogen bonds between the side-chains of Asn217 and Tyr219 and the side-chain hydroxyl and backbone carbonyl groups of Ser212, respectively. Apparently these hydrogen bonds do not form in the free state. Some indication of increased dynamics can be noticed also in the corresponding region of the GR DBD (Figures 4(a) and 6), although it appears to be less extensive. The Tyr474 (ER Tyr219) side-chain in free GR DBD is anchored to the core by hydrophobic interactions involving Phe463, Ala467 and Pro493. This interaction can explain the somewhat more ordered appearance of the extended strand backbone in the GR DBD. Residues Asp230-Cys237 in the second zinc domain show more dynamics in the ER DBD than the corresponding residues in the GR DBD, as judged by lower NOE values (all measured residues; Figure 4(a) and order parameters (primarily Arg233-Cys237; Figure 6). This is the peptide which forms the distorted helix II in the bound, but not the free, state of the ER DBD, and for which two different conformations are found in crystals of dimeric ER DBD-ERE complexes (Schwabe et al., 1993a, 1995). In the GR DBD, on the other hand, it is clearly possible for this peptide to adopt the folded conformation both in the free (Baumann et al., 1993; van Tilborg, 1998) and bound (Luisi et al., 1991) states. The differences in dynamics properties that are observed here are consistent with a different folding stability of helix II in the two proteins. Such a difference is not at all surprising, as helix II is very short and therefore might be less well stabilized. On the other hand, in the dimeric DNA-bound state there is substantial additional stabilization provided by dimerization interactions (Ser236) and electrostatic interactions with the DNA backbone (Arg234 and Lys235 sidechains). Thus, our data support a scenario involving a coupled folding reaction in this region of the ER DBD upon DNA binding. The distorted helix might perhaps be less stable than other secondary structure elements also in the GR DBD. This is because the expected mediumrange NOE connectivities cannot be found in NOESY spectra of a larger GR DBD fragment (the DBD93; van Tilborg et al., 1995), whereas these NOEs are easily identi®ed in the corresponding spectra of shorter GR DBD fragments, i.e. the fragment used here (Baumann et al., 1993) and a 72residue GR DBD fragment (van Tilborg, 1998). We do not, on the other hand, understand why the length of the GR fragment should affect the stability of the distorted helix. Conformational exchange dynamics on the micro- to millisecond time-scales Several regions of the ER DBD backbone display rapid R2 rates (Figures 4(c) and 5). The large R2

Backbone Dynamics in the ER DBD

values cannot be accounted for by anisotropic overall motions and most of the largest values show a monotonic dependency on the CPMG delay. They must therefore contain exchange (Rex) contributions re¯ecting exchange dynamics as illustrated in Figures 7 and 8. The exchange must also re¯ect intramolecular (conformational) and not intermolecular events, because line-widths and chemical shifts in the 15N HSQC spectrum are independent of protein concentration. The exact time constants for the conformational exchange depend on the number of interconverting states, their corresponding chemical shift differences and their relative populations at equilibrium. It is, in principle, possible to estimate many of the involved parameters based on either additional experiments such as the ®eld dependence of offresonance R1r relaxation (Akke & Palmer, 1996) or using only the present limited set of CPMG R2 experiments (Orekhov et al., 1994). Such details might be addressed in further studies. Here, we limit the interpretation to conclude which regions are involved in conformational exchange. We also note that the fact that the exchange term contributes to R2 under the present experimental conditions must imply time constants for exchange in the micro- to millisecond range. The most pronounced exchange dynamics is found within the fragment Tyr195-Trp200 in the ®rst zinc domain, and Cys221-Thr224 (in the D-box), Cys227-Ile229, Lys231, and Ser236 of the second zinc domain, for which the quantitative analysis indicate that 3 sÿ1 < Rex < 8.2 sÿ1 (Figure 7). It is dif®cult to de®nitely conclude if the conformational exchange re¯ects local unfolding events, i.e. order-disorder transitions, or slow interconversions between folded conformations. In the case of the second zinc domain, both of these scenarios might contribute to the fact that very few medium and long-range NOEs are observed in NOESY spectra. For the folding-unfolding alternative, however, one would expect that a signi®cant population of the unfolded (disordered) state also would result in lower values for the heteronuclear NOEs. This is not observed in the ®rst zinc domain, and the alternative scenario of interconverting folded states is therefore more likely in this region. The situation in the second zinc domain might be different, as several of the residues that show exchange dynamics also display low NOE values and belong to those regions that have been concluded to undergo more rapid internal dynamics (see above). Such residues are Cys221, Lys231 and Ser236. It is therefore possible that the exchange terms found in the second zinc domain re¯ect the fact that this domain is not stably folded in the free state. This interpretation is also consistent with the absence of interproton NOEs in NMR spectra and the previous conclusion that the average structure of the second zinc domain in the free ER DBD is largely disordered (Schwabe et al., 1993b). It is conspicuous that the regions which exhibit conformational exchange coincide with those

973 involved in both protein-DNA and protein-protein interactions in dimeric DBD-DNA complexes (Figure 8). Thus, all residues in the loop of the ®rst zinc domain that interact with the DNA phosphate backbone, as well as all of the side-chains or backbone atoms involved in the dimerization interface (Schwabe et al., 1993a), are located close to backbone nitrogen atoms which undergo exchange dynamics in the free state. The conformations of these regions are well resolved in ER DBD-ERE structures. The implication must be that interactions in the bound state select the most appropriate conformation and stabilizes it, as discussed further below. The R2 data on the GR DBD also reveal sites that appear to undergo exchange dynamics, although it is in this case less extensive than in the ER DBD. In particular, three residues in the second zinc domain show rapid R2 rates and optimized Rex terms > 3 sÿ1 (Figure 7). These are Ile483, Ile484 and Cys492 (corresponding to ER residues 228, 229 and 237). The precise nature of this dynamics is not clear. Both side-chain and backbone amides of these residues display a large number of long and medium-range NOE connectivities in the present GR DBD fragment (Baumann et al., 1993), indicating that they on average adopt a stable conformation. It is therefore possible that the observed dynamics in this case instead re¯ects more localized backbone rearrangements. Relation between structure and dynamics of the free DBDs and molecular recognition in DBD-DNA complexes The fact that regions that undergo micro- to millisecond conformational exchange in the free ER DBD are involved in intermolecular interactions in the DNA-bound state implicate that DNA-binding selects for, or imposes, a particular conformation. In the GR DBD there is less dynamics in the free state, but differences between the free and boundstate structures suggest that DNA binding still induces a conformational adaptation (Baumann et al., 1993). This adaptation involves precisely the same regions of the GR DBD that undergo exchange dynamics in the ER DBD: the loop of the ®rst zinc domain, the D-box and the dimerization surface of the second zinc domain. The average rms difference between the backbone structures of the 66-residue core regions of the DNA-bound ER (Schwabe et al., 1993a; Dimer, A) and GR DBDs (Luisi et al., 1991 sequence-speci®cally bound Ê , whereas the experimenmonomer) is only 0.86 A tal rms difference of the two monomers in the Ê . Hence, dimeric ER DBD-ERE complex is 0.66 A the results of the stabilization of a conformation (in the ER DBD) or conformational modi®cation (in the GR DBD) are bound-state structures that can be considered to be identical within the limits imposed by the crystallographic resolution. The identical bound-state structures, but different free state structure and dynamics in combination with

974 different cognate DNA binding sites, hints at a possibility of a relation between free state internal conformational dynamics and DNA-binding sequence speci®city. The issue is intimately related to the concept of allosteric regulation of the GR DBD conformation by DNA as formulated by Sigler and co-workers (e.g. see Luisi et al., 1991). In this context it is important to recall the structural and DNA-binding properties of a triple GR DBD mutant, the GR DBDEGA protein (Berglund et al., 1997). Three mutations in the GR DBD, Gly458Glu, Ser459Gly and Val462Ala (ER residues 203, 204 and 207, respectively), completely switches the speci®city from a GRE to and ERE binding sequence on DNA (Danielsen et al., 1989; Umesono & Evans, 1989; Zilliacus et al., 1991). The function of the Gly ! Glu and Val ! Ala mutations can in this case be understood in terms of sequence-speci®c interactions on the proteinDNA surface. The role of the Ser ! Gly mutation, on the other hand, is more obscure as these sidechains do not contact the DNA in any of the complexes and the presence of the serine actually appears to have a negative effect for binding to both GRE and ERE sites (Zilliacus et al., 1992). We recently examined the free state structural properties of GR DBDEGA (Berglund et al., 1997). We found that the triple mutation results in alterations the DNA-binding surface and in the second zinc domain. For instance, the second zinc domain in GR DBDEGA does not display most of the mediumrange NOE connectivities found with the wild-type protein, indicating that it is less well folded. In addition, several backbone amide protons exhibit increased exchange rates, which also suggests a lower structural stability. Both these properties are reminiscent of the situation in the ER DBD. The key to the connection between the recognition site in helix I and the structure of the second zinc domain appears to be the action of the Ser459 sidechain hydroxyl group (Berglund et al., 1997; van Tilborg, 1998), which can form a hydrogen bond with Arg496 in the second zinc domain. Another property of the GR DBDEGA triple mutant which distinguishes it from the wild-type protein is the DNA-binding cooperativity: the GR DBD displays a weak cooperativity (DahlmanWright et al., 1991; HaÈrd et al., 1990a), whereas that of the GR DBDEGA is approximately ten times stronger (LundbaÈck et al., 1994). The properties of GR DBDEGA is in this regard also more reminiscent of the ER DBD: electrophoretic mobility shift assays of these two proteins with palindromic ERE probes do not show a signi®cant population of a momomeric complex (LundbaÈck et al., 1994; Schwabe et al., 1993b), which is in contrast to a clearly visible monomer band in corresponding experiments with wild-type GR DBD and a GRE probe (Dahlman-Wright et al., 1991; LundbaÈck et al., 1994). Thus, dimeric DNA binding of ER DBD and GR DBDEGA appears to be more cooperative than that of wild-type GR DBD. This effect can also be understood in terms of the free state stability and

Backbone Dynamics in the ER DBD

dynamics of the second zinc domain. In the wildtype GR DBD, the DNA binding forces a relatively stable free state conformation to change in the bound state. This structural change might be costly in terms of free energy and the free energy cost has to be paid also by the second monomer upon dimerization. Hence, the net gain by dimerization (the cooperativity) is attenuated. In case of the ER DBD and the GR DBDEGA there is more dynamics and/or a lower inherent stability in the free state and, consequently, also a lower energy penalty for forming the bound state dimeric structure, resulting in a larger net cooperativity effect. Hence, there appears to be a connection between the structural and dynamics properties of the ER and GR DBDs in the free states and their respective sequence speci®city, and the GR DBDEGA mutant seems to provide an intermediate link. The various observations described above indicate that the sequence-speci®c DNA binding proceeds in two steps: (i) a direct surface-surface recognition of helix I which positions it correctly in the DNA major groove; and (ii) a DNA-induced conformational change (GR DBD) or folding/stabilization (ER DBD) imposed by DNA. Strong dimeric binding fails if initial binding (step (i)) does not orient the DBD precisely so that step (ii) can proceed. These two sequential and conceptually different recognition mechanisms leave considerable room for ®ne-tuning of gene regulation. For instance, the second step provides for DNA to act as the allosteric regulator (Luisi et al., 1991) of the second event, i.e. different response elements might allow or not allow the formation of the active bound state structure, as suggested by Lefstin & Yamamoto (1998) for the allosteric modulation of GR on transcription activation and/or repression activity. If the free state dynamics is linked to DNA recognition, as suggested here, then the micro- to millisecond conformational exchange motions that we observe in the ER DBD should be a general property of receptor superfamily members that recognize the same half-site sequence on DNA. It has very recently also been reported that the DNAbinding domains of RAR and RXR display conformational exchange motions in the second zinc domain (van Tilborg et al., 1999). The sequence versus dynamics pro®les of R2 relaxation rates of these DBDs are not identical and also differ from that of the ER DBD, indicating that the precise nature of the dynamics is different in the three DBDs. However, DNA-induced local folding of dimer interface residues has been observed when the RXR DBD binds as a monomer to an AGGTCA half-site (Holmbeck et al., 1998). Hence, conformational exchange motions and DNA-induced coupled folding appear to be present in these two non-steroid receptor DBDs which, like the ER DBD and the GR DBDEGA mutant, bind to an AGGTCA half-site, whereas they are much less prominent in the GR DBD, which recognizes AGAACA. The RAR and RXR data therefore provide additional support for

Backbone Dynamics in the ER DBD

the emerging picture of a link between free state dynamical properties of steroid receptor DBDs and the DNA response elements that they are designed to recognize.

Materials and Methods Protein production and purification The DNA encoding residues Arg183-His267 of the human ER was cloned into the NdeI and BamHI sites of the pET-11c expression vector (Novagen) and transformed into the Escherichia coli strain BL21(DE3)pLysS. Cell culture conditions for isotopic labeling and puri®cation procedures for the ER and GR DBD (Lys438Gln520 of the rat GR) were identical with those described (Berglund et al., 1997). Sequence-speci®c DNA binding activity of puri®ed proteins was veri®ed using electrophoretic mobility shift assays on agarose gels with ethidium bromide staining (LundbaÈck & HaÈrd, 1996a). The binding af®nity of puri®ed ER DBD to ERE-containing DNA oligomers was found to be very similar to the corresponding GR DBD-GRE af®nity. NMR samples typically contained 1.5-2.0 mM (ER DBD) or 4-5 mM (GR DBD) protein in 20 mM NaH2PO4, 150 mM NaCl (pH 6.0), with 2 mM DTT added to prevent oxidation of cystein residues. The NMR samples were stable for periods of several months or longer. NMR spectroscopy NMR was measured at 30  C on Varian Unity or Inova 500 and 600 MHz spectrometers using 5 mm triple-resonance (1H/13C/15N) pulsed ®eld gradient probes. (When describing 15N relaxation data we have referred to these two magnetic ®elds as the approximate 15 N resonance frequencies of 50 and 60 MHz, respectively). Varian software (VNMR) was used for data processing. All 15N HSQC-based 2D and 3D spectra were recorded using the gradient-based method for coherence selection and sensitivity enhancement developed by Kay and co-workers (reviewed by Kay, 1995) with GARP-1 (Cavanagh et al., 1996) composite pulse sequence for decoupling of 15N during 1H acquisition. The 15N-edited 3D NOESY-HSQC and 3D TOCSYHSQC experiments (Kay, 1995), with cross-relaxation and DIPSI-2 (Cavanagh et al., 1996) isotropic mixing times of 150 and 64 ms, respectively, were used to assign backbone amide resonances in the ER DBD. The 3D experiments were recorded at 500 MHz with spectral widths of 6000, 1800, and 8000 Hz in 1H, 15N and (directly detected) 1H dimensions, respectively, and the corresponding data sets consisted of 128  32  512 complex points. No presaturation of the water resonance was employed. A 3D HNHA experiment (Vuister & Bax, 1993) was also recorded as a complement in the identi®cation of Ha resonances. Sequence speci®c assignments were carried out using conventional procedures. The 1H chemical shifts are referenced to H2O at 4.74 ppm and 15N shifts are referenced to external benzamide at 105.4 ppm. The 15N R1 and R2 relaxation rates and steady-state 1 H-15N NOEs were measured using the inversion recovery (R1), CPMG (R2) and steady-state NOE experiments described by Farrow et al. (1994). Spectral widths were typically 8000 Hz for 1H and 2000 or 2500 Hz for 15N at

975 500 and 600 MHz, respectively, with corresponding data set sizes of 1024  128 complex data points. The water resonance was minimized by the application of gradients in combination with water ¯ip-back pulses. An additional effect of water ¯ip-back is that the water magnetization is aligned along the ‡z-axis before acquisition, which minimizes the effect of (slow) water relaxation on any amide protons in rapid exchange with water. The effect of cross-correlation between dipolar and chemical shift anisotropy relaxation mechanisms is suppressed in these experiments (Farrow et al., 1994). The steady-state NOE experiments were recorded with recycle delays of two seconds followed by a three second 1H saturation pulse train (or an additional three second recycle delay in the control experiments). The R1 experiments were performed with sampling of the relaxation decay at eight times, typically 32, 54, 76, 108, 184, 324, 540 and 864 ms. A duplicate data set was collected at one of these times to estimate experimental errors due to random noise. The CPMG delay in the R2 experiments, i.e. the time between 15N refocusing 180  pulses, was 400, 900 or 1896 ms. Typical relaxation sampling times (with a 900 ms CPMG delay) were 15.7, 31.5, 47.2, 63.0, 78.7, 94.5, 110.2, and 157.4 ms, with a duplicate data set at 47.2 ms. The recycle delay in the R1 and R2 experiments was two seconds. The following set of 15N relaxation experiments was carried out with the ER DBD. The NOE experiments (including controls) were performed twice at 60 MHz and three times at 50 MHz (using two different probes) and the different NOE measurements were averaged (e.g. in Figure 4(a)). The R2 experiments were performed three times with different CPMG delays at 60 MHz and twice at 50 MHz. The second R2 experiment at 50 MHz was carried out using a different 5 mm probe. The R1 experiment was carried out once at 50 MHz. The relaxation experiments on the GR DBD were carried out at 50 MHz, and the NOE, but not the R1 or R2, experiment was repeated. Relaxation data were processed using different apodization functions, cosine bell or shifted Gaussians, to achieve sensitivity or resolution enhancement, respectively. Resonance intensities were measured as peak heights using VNMR routines. Experimental errors in the R1 and R2 experiments were calculated from the duplicate data sets: differences in peak heights (
hi) were measured and the experimental standard deviations (due to random differences between the experiments) were estimated as ((N
h2i )/21/2(N-1))1/2 for N measured peak heights in a spectrum. Relaxation rates Ri were calculated by non-linear curve ®tting of decays to Aexp(ÿRit) using routines available in the ModelFree program package (Dr A.G. Palmer, Columbia University). Reported uncertainties are based on Monte Carlo simulations (Palmer et al., 1991). Errors in the NOE measurements were estimated based on repeated measurements. The average precision of the ER DBD relaxation data sets were as follows: R1(50 MHz), 1.0 %; R2(50), 1.1 %; R2(60 MHz), 2.2 %; NOE(50), 2.3 %; and NOE(60), 2.6 %. The corresponding errors for the GR DBD data set are R1(50 MHz), 0.6 %; R2(50), 1.1 %; and NOE(50), 0.6 %. This high precision primarily re¯ects the signal-to-noise ratio in NMR spectra, i.e. rather small random errors due to thermal noise, but it does not take into account other systematic sources of error such as heating and/or B1 inhomogeneity of probes. The R2 rates of the ER DBD at 50 MHz were also measured using a different probe with a newer design. These rates are on average 0.5(0.5) sÿ1 (6(6)%) slower than those

976

Backbone Dynamics in the ER DBD

measured with older probe indicating a systematic error on the order of 3 %. This latter value is likely to be a more realistic estimate of the precision of reported R2 values than error estimates based only on random noise (note, however, that error bars in Figure 4(c) represent the latter). Relaxation data analysis The relaxation of an amide 15N spin at high magnetic ®eld is dominated by dipole-dipole (DD) interactions with the amide 1H and by chemical shift anisotropy (CSA). The rates for relaxation of longitudinal and transverse magnetization (R1 and R2, respectively) and the 1 H-15N cross-relaxation (NOE) effect are given by (Abragam, 1961): R1 ˆ …d2 =4†‰ J…o†H ÿ oN † ‡ 3J…oN † ‡ 6J…oH ‡ oN †Š ‡ c2 J…oN †

…1†

R2 ˆ …d2 =8†‰4J…0† ‡ J…oH ÿ oN †; ‡3J…oN † ‡ 6J…oH † ‡ 6J…oH ‡ oN †Š ‡ …c2 =6†‰4J…0† ‡ 3J…oN †Š ‡ Rex

…2†

NOE ˆ 1 ‡ …d2 =4R1 †…gH =gN †‰6J…oH ‡ oN † ÿ J…oH ÿ oN †Š

…3†

where J(o) is the spectral density p function, d ˆ (m0hgNgH/ 8p2)h1/r3NHi and c ˆ oN
s/ 3, gi and oi are the gyromagnetic ratio and Larmor frequency of spin i, h is Planck's constant, rNH is the internuclear 15N-1H distance Ê ), and
s is the difference between parallel and (1.02 A perpendicular components of the 15N chemical shift tensor (ÿ160 ppm) (Hiyama et al., 1988). The Rex term in equation (2) is often included to account for additional broadening caused by chemical or conformational exchange. For the quantitative analysis of relaxation data we employed the Lipari-Szabo spectral density function:   2 S2 tM …1 ÿ S2 †t J…o† ˆ ‡ …4† 5 1 ‡ …otM †2 1 ‡ …ot†2 which assumes isotropic overall tumbling characterized by the rotational diffusion constant tM, restricted internal mobility characterized by the order parameter S2 and an ÿ1 internal correlation time te (te < tM), with tÿ1 ˆ tÿ1 M ‡ te (Lipari & Szabo, 1982a,b). Given the spectral density described by equation (4), it can be shown that the ratio R2/R1 is insensitive to internal motions if these are rapid (short te) and limited (high S2), and if R2 does not contain exchange contributions. This provides a way to obtain an estimate of the rotational global correlation time tM, e.g. from a trimmed R2/R1 average (Kay et al., 1989), or local apparent values tM,i from the R2/R1 ratio of residue i (Barbato et al., 1992) if internal dynamics is limited and Rex ˆ 0. The maximum error in tM associated with such a procedure is negligible if te < 10 ps and <3 % if te < 100 ps, for S2 > 0.80 for tM < 7 ns. The dominant dependence of R2/R1 on tM also provides a convenient means to determine whether or not rotational diffusion is anisotropic, given that R2/R1 ratios can be measured with suf®cient precision (Tjandra et al., 1995). The spectral density function for anisotropic rotational motion has been described (Woessner, 1962). In the case of an axially symmetric diffusion tensor this

simpli®es to: J…o† ˆ

2 X Ak tk 5 kˆ1;2;3 1 ‡ …otk †2

…5†

in which A1 ˆ (3 cos2 a ÿ 1)2/4, A2 ˆ 3 sin2 a3 cos2 a, and A3 ˆ (3 sin2 a)/4, where a is the angle between the N-H bond vector and the symmetry axis in the diffusion frame. The time constants in equations (5) are given by t ˆ (6Dper)ÿ1, t2 ˆ (Dpar ‡ 5Dper)ÿ1, t3 ˆ (4Dpar ‡ 2Dper)ÿ1, where Dpar and Dper are diffusion coef®cients for rotations about axis parallel and perpendicular to the symmetry axis, respectively. The effective diffusion constant Diso is de®ned as the average (Dpar ‡ 2Dper)/3. A spectral density function for internal motions in combination with axially symmetric rotational diffusion anisotropy has also been derived (Schurr et al., 1994). Again, it can be shown that the R2/R1 ratio is insensitive to limited internal dynamics as long as the rotational diffusion anisotropy is not large (data not shown). This justi®es the use of equation (5) with an appropriate selection of R2/R1 ratios in combination with known orientations of the corresponding N-H vectors in the protein to explore the possibility that overall rotational diffusion is anisotropic (Tjandra et al., 1995). In our quantitative analysis of relaxation data we ®rst examine the extent of anisotropy of the overall rotational diffusion. The R2/R1 ratios used in this analysis were selected to exclude residues for which the backbone conformation is not accurately determined, display large internal dynamics or for which there are large exchange contributions to R2 (see Results). The average NMR structures of the ER and GR DBDs were used to ®t R2/R1 data to determine optimal diffusion coef®cients and diffusion tensors using computer programs (``R2R1 diffusion 1.1`` and ``quadric diffusion 1.0``) kindly made available by Dr A.G. Palmer, Columbia University. The program R2R1 diffusion 1.1 uses equation (5) and procedures analogous to those described by Tjandra et al. (1995) to discriminate between isotropic and axially symmetric dynamic models. The program quadric diffusion 1.0 instead employs the quadratic approximation approach (BruÈschweiler et al., 1995) and also allows testing of fully asymmetric rotational dynamics in the limit of small anisotropies (Lee et al., 1997). This second procedure requires prior calculations of local diffusion coef®cients from R2/R1 ratios. The average GR DBD structure has the PDB identity code 1GDC and an average ER DBD structure was calculated from the ensemble structures in the PDB entry 1HCP. Following the examination of rotational diffusion anisotropy, we optimized the parameters of equation (4) using Modelfree 3.1 (A.G. Palmer, Columbia University). For the ER DBD this anlysis included the ®ve measured relaxation parameters: heteronuclear NOEs measured at 50 and 60 MHz, R2 values measured at 50 and 60 MHz (both with a 900 ms CPMG delay) and R1 values measured at 50 MHz. For each backbone amide we ®rst tested if a model which also includes exchange contributions to R2 (Rex terms) represent a signi®cant improvement over a model containing the order parameter (S2) and corrrelation time (te) for internal dynamics. The signi®cance of the improved ®t was judged on basis of Fratios (Devore, 1995) and a 90 % con®dence level was required for the inclusion of an Rex > 0 in the model. The (global) rotational correlation time was in this case

977

Backbone Dynamics in the ER DBD assumed to take the value obtained from the average R2/R1 (excluding the same selected R2/R1 values as in the anisotropy analysis). Following model selection all dynamical parameters were optimized, including a global tM. Uncertainties in the optimized parameters were obtained from Monte Carlo simulations using 250 sets of relaxation parameters. These sets were obtained by adding random errors (given experimental uncertainties) to relaxation data back-calculated from the optimized model parameters (Palmer et al., 1991). The quality of the optimization was assessed based on: ÿi ˆ

Mj X jˆ1

exp …Rij

ÿ

2 2 Rcalc ij † =sij

…6†

where Rexp represent experimental values for Mj (usually ij ®ve) different relaxation parameters determined with an uncertainty corresponding to sij, and Rcalc is the backij calculated value given equations (1) to (4) and the optimized dynamics parameters for residue i. The optimized value of ÿi was compared to critical values of the w2a,n distribution for n degrees of freedom. (Critical w20.05,n values determined from Monte Carlo simulations were found to be close to theoretical values.) This procedure ensures a statistically reliable discrimination between good and poor ®ts given that the experimental errors are realistic. To account for less well-determined systematic errors in high-precision data (discussed above) we let sij represent the experimentally determined (random) errors but instead considered a ®t of equation (4) to be very good if ÿi < w20.05,n, good if ÿi < 4w20.05,n, acceptable if ÿi < 9w20.05,n, and poor if ÿi > 9w20.05,n, where the factors four and nine correspond to underestimates of experimental errors by factors of two and three, respectively. This procedure has no effect on model selection based on F-ratios. The quanti®cation of GR DBD data is based on a smaller set of relaxation parameters: R1, R2 and heteronuclear NOE values determined at 50 MHz. As a result of the outcome of the overall rotation analysis, which suggested rotational diffusion anisotropy, we performed this analysis using two different procedures. The ®rst procedure was identical with that used for the ER DBD described above. However, the F-test is not possible because the number of optimized parameters is equal to the number of relaxation parameters if Rex > 0. We therefore required that the Rex term should be added to the model when: (i) ÿi for the reduced model (no Rex) were more than 20 times larger than the w20.05,1 critical value; (ii) ÿi ˆ 0 for the model including Rex, and (iii) the uncertainty in the optimized Rex was smaller than the Rex value. The effect of axially symmetric rotation was taken into account in a second procedure. Here, local values tM,i ˆ 1/6Di were obtained from: Di ˆ Diso ÿ …3 cos2 a ÿ 1†…Dpar ÿ Dper †=6

…7†

where the diffusion coef®cients were those obtained from the analysis of diffusion anisotropy described above. Equation (7) is not exact, but follows from the quadratic diffusion approximation (Lee et al., 1997). However, the error in Di resulting from this approximation is on the order of 3 % or less for Diso ˆ 2.8  107 sÿ1 and Dpar/Dper ˆ 1.2 (data not shown) and is therefore negligible in the present case. The Rex terms were treated and included based on the same three criteria as applied for the isotropic model.

Acknowledgments This work was in part supported by the Swedish Natural Sciences Research Council (NFR). We acknowledge Dr Peter Allard for assistance with NMR spectroscopy and the Swedish NMR Centre (Gothenburg) for providing access to their instruments.

References Abragam, A. (1961). Principles of Nuclear Magnetism, Clarendon Press, Oxford. Akke, M. & Palmer, A. G. (1996). Monitoring macromolecular motions on microsecond to millisecond time scales by R1r-R1 constant relaxation time NMR spectroscopy. J. Am. Chem. Soc. 118, 911-912. Barbato, G., Ikura, M., Kay, L. E., Pastor, R. W. & Bax, A. (1992). Backbone dynamics of calmodulin studied by 15N relaxation using inverse detected twodimensional NMR spectroscopy: the central helix is ¯exible. Biochemistry, 31, 5269-5278. Baumann, H., Paulsen, K., LovaÂcs, H., Berglund, H., Ê . & HaÈrd, T. Wright, A. P. H., Gustafsson, J.-A (1993). Re®ned solution structure of the glucocorticoid receptor DNA-binding domain. Biochemistry, 32, 13463-13471. Berglund, H., LovaÂcs, H., Dahlman-Wright, K., Ê . & HaÈrd, T. (1992). Backbone Gustafsson, J.-A dynamics of the glucocorticoid receptor DNA-binding domain. Biochemistry, 31, 12001-12011. Berglund, H., Wolf-Watz, M., LundbaÈck, T., van den Berg, S. & HaÈrd, T. (1997). Structure and dynamics of the glucocorticoid receptor DNA-binding domain: comparison of wild type and a mutant with altered speci®city. Biochemistry, 36, 1118811197. BruÈschweiler, R., Liao, X. & Wright, P. E. (1995). Longrange motional restrictions in a multidomain zinc®nger protein from aniostropic tumbling. Science, 268, 886-889. Brzozowski, A. M., Pike, A. C. W., Dauter, Z., Hubbard, È hman, L., Greene, R. E., Bonn, T., EngstroÈm, O., O Ê . & Carlquist, M. (1997). G. L., Gustafsson, J.-A Molecular basis of agonism and antagonism in the oestrogen receptor. Nature, 389, 753-758. Cairns, W., Cairns, C., Pongratz, I., Poellinger, L. & Okret, S. (1991). Assembly of glucocorticoid receptor complex prior to DNA binding enhances its speci®c interaction with a glucocorticoid response element. J. Biol. Chem. 266, 11221-11226. Cavanagh, J., Fairbrother, W. J., Palmer, A. G. & Skelton, N. J. (1996). Protein NMR Spectroscopy: Principles and Practice, Academic Press, London. Clore, G. M., Szabo, A., Bax, A., Kay, L. E., Dristoll, P. C. & Gronenborn, A. M. (1990). Deviations from the simple two-parameter model-free approach to the interpretation of nitrogen-15 nuclear magnetic relaxation of proteins. J. Am. Chem. Soc. 112, 3989-4991. Ê. & Dahlman-Wright, K., Wright, A., Gustafsson, J.-A Carlstedt-Duke, J. (1991). Interaction of the glucocorticoid receptor DNA-binding domain with DNA as a dimer is mediated by a short segment of ®ve amino acids. J. Biol. Chem. 266, 3107-3112. Danielsen, M., Hinck, L. & Ringold, G. M. (1989). Two amino acids within the knuckle of the ®rst ®nger specify DNA response element activation by the glucocorticoid receptor. Cell, 57, 1131-1138.

978 Devore, J. L. (1995). Probability and Statistics for Engineering and the Sciences, 4th edit., Wadsworth, Inc./ International Thomson Publishing Inc., New York. Farrow, N. A., Muhandiram, R., Singer, A. U., Pascal, S. M., Kay, C. M., Gish, G., Shoelson, S. E., Pawson, T., Forman-Kay, J. D. & Kay, L. E. (1994). Backbone dynamics of a free and phosphopeptide-complexed src homology domain studied by 15N NMR relaxation. Biochemistry, 33, 5984-6003. Freedman, L. P., Luisi, B. F., Korszun, Z. R., Basavappa, R., Sigler, P. B. & Yamamoto, K. R. (1988). The function and structure of the metal coordination sites within the glucocorticoid receptor DNA binding domain. Nature, 334, 543-546. Gewirth, D. T. & Sigler, P. B. (1995). The basis for halfsite speci®city explored through a non-cognate steroid receptor-DNA complex. Nature Struct. Biol. 2, 386-394. HaÈrd, T., Dahlman, K., Carlstedy-Duke, J., Gustafsson, Ê. J.-A & Rigler, R. (1990a). Cooperativity and speci®city in the interactions between DNA and the glucocorticoid receptor DNA-binding domain. Biochemistry, 29, 5358-5364. HaÈrd, T., Kellenbach, E., Boelens, R., Maler, B. A., Dahlman, K., Freedman, L. P., Carlstedt-Duke, J., Ê . & Kaptein, R. Yamamoto, K. R., Gustafsson, J.-A (1990b). Solution structure of the glucocorticoid receptor DNA-binding domain. Science, 249, 157160. Hiyama, Y., Niu, C.-H., Silverton, J. V., Bavoso, A. & Torchia, D. A. (1988). Determination of 15N chemical shift tensor via 15N-2H dipolar coupling in Bocglycyl-glycyl[15N]glycine benzyl ester. J. Am. Chem. Soc. 110, 2378-2383. Holmbeck, S. M. A., Dyson, H. J. & Wright, P. E. (1998). DNA-induced conformational changes are the basis for cooperative dimerization by the DNA binding domain of the retinoid X receptor. J. Mol. Biol. 284, 533-539. Kay, L. E. (1995). Pulsed ®eld gradient multi-dimensional NMR methods for the study of protein structure and dynamics in solution. Prog. Biophys. Mol. Biol. 63, 277-299. Kay, L. E., Torchia, D. A. & Bax, A. (1989). Backbone dynamics of proteins as studied by nitrogen-15 invers detected NMR spectroscopy: application to staphylococcal nuclease. Biochemistry, 28, 8972-8979. Kumar, V. & Chambon, P. (1988). The estrogen receptor binds tightly to its responsive element as a ligandinducer dimer. Cell, 55, 145-156. Lee, L. K., Rance, M., Chazin, W. J. & Palmer, A. G. (1997). Rotational diffusion anisotropy of proteins from simultaneous analysis of 15N and 13Ca nuclear spin relaxation. J. Biol. NMR, 9, 287-298. Lefstin, J. A. & Yamamoto, K. R. (1998). Allosteric effects of DNA on transcriptional regulators. Nature, 392, 885-888. Lipari, G. & Szabo, A. (1982a). Model-free approach to the interpretation of nuclear magnetic relaxation in macromolecules. 1. Theory and range of validity. J. Am. Chem. Soc. 104, 4546-4559. Lipari, G. & Szabo, A. (1982b). Model-free approach to the interpretation of nuclear magnetic relaxation in macromolecules. 2. Analysis of experimental results. J. Am. Chem. Soc. 104, 4559-1570. Lucas, P. C. & Granner, D. K. (1992). Hormone response domains in gene transcription. Annu. Rev. Biochem. 61, 1131-1173.

Backbone Dynamics in the ER DBD Luisi, B. F., Xu, W. X., Otwinowski, Z., Freedman, L. P., Yamamoto, K. R. & Sigler, P. B. (1991). Crystallographic analysis of the interaction of the glucocorticoid receptor with DNA. Nature, 352, 497-505. LundbaÈck, T. & HaÈrd, T. (1996a). Salt dependence of the free energy, enthalpy and entropy of nonsequence speci®c DNA binding. J. Phys. Chem. 100, 1769017695. LundbaÈck, T. & HaÈrd, T. (1996b). Sequence speci®c DNA-binding dominated by dehydration. Proc. Natl Acad. Sci. USA, 93, 4754-4759. Ê . , CarlstedtLundbaÈck, T., Zilliacus, J., Gustafsson, J.-A Duke, J. & HaÊrd, T. (1994). Thermodynamics of sequence-sopeci®c glucocorticoid receptor-DNA interactions. Biochemistry, 33, 5955-5965. Mangelsdorf, D. J., Thummel, C., Beato, M., Herrlich, P., Schutz, G., Umesuno, K., Blomberg, B., Kastner, P., Mark, M., Chambon, P. & Evans, R. (1995). The nuclear receptor family: the second decade. Cell, 83, 835-839. Orekhov, V. Y., Pervushin, K. V. & Arseniev, A. S. (1994). Backbone dynamics of (1-71) bacteriorhodopsin studied by two-dimensional 1H-15N NMR spectroscopy. Eur. J. Biochem. 219, 887-896. Palmer, A. G., Rance, M. & Wright, P. E. (1991). Intramolecular motions of a zinc ®nger DNA-binding domain from x®n characterized by proton-detected natural abundance 13C heteronuclear NMR spectroscopy. J. Am. Chem. Soc. 113, 4371-4380. Palmer, A. G., Skelton, N. J., Chazin, W. J., Wright, P. E. & Rance, M. (1992). Suppression of cross-correlation between dipolar and anisotropic chemical shift relaxation mechanisms in the measurement of spinspin relaxation rates. Mol. Phys. 75, 699-711. Palmer, A. G., Williams, J. & McDermott, A. (1996). Nuclear magnetic resonance studies of biopolymer dynamics. J. Phys. Chem. 100, 13293-13310. Schurr, J. M., Babcock, H. P. & Fujimoto, B. S. (1994). A test of the model-free formulas. Effects of anisotropic rotational diffusion and dimerization. J. Magn. Reson. ser. B, 105, 211-224. Schwabe, J. W. R., Neuhaus, D. & Rhodes, D. (1990). Solution structure of the oestrogen receptor. Nature, 348, 458-461. Schwabe, J. W. R., Chapman, L., Finch, J. T. & Rhodes, D. (1993a). The crystal structure of the oestrogen receptor DNA-binding domain bound to DNA: how receptors discriminate between their response elements. Cell, 75, 567-578. Schwabe, J. W. R., Chapman, L., Finch, J. T., Rhodes, D. & Neuhaus, D. (1993b). DNA recognition by the oestrogen receptor: from solution to the crystal. Structure, 1, 187-204. Schwabe, J. W. R., Chapman, L. & Rhodes, D. (1995). The oestrogen receptor recognizes an imperfectly palindromic response element through an alternative side-chain conformation. Structure, 3, 201-213. Tjandra, N., Feller, S. E., Pastor, R. W. & Bax, A. (1995). Rotational diffusion anisotropy of human ubiquitin from 15N NMR relaxation. J. Am. Chem. Soc. 117, 12562-12566. Truss, M. & Beato, M. (1993). Steroid hormone receptors: interaction with deoxyribonucleic acid and transcription factors. Endocrine Rev. 14, 459-479. Tsai, M. J. & O'Malley, B. W. (1994). Molecular mechanisms of action of steroid/thyroid receptor superfamily members. Annu. Rev. Biochem. 63, 451-486.

Backbone Dynamics in the ER DBD Umensono, K. & Evans, R. M. (1989). Determinants of target gene speci®city for steroid thyroid hormone receptors. Cell, 57, 1139-1146. van Tilborg, M. (1998). The glucocorticoid receptor DNA binding domain: structure and allosteric modulation. PhD thesis, University of Utrecht. van Tilborg, M. A. A., Bonvin, A. M. J. J., HaÊrd, K., Davis, A. L., Maler, B., Boelens, R., Yamamoto, K. R. & Kaptein, R. (1995). Structure re®nement of the glucocorticoid receptor DNA-binding domain from NMR data by relaxation matrix calculations. J. Mol. Biol. 247, 689-700. van Tilborg, P. J. A., Mulder, F. A. A., de Backer, M. M. E., Nair, M., van Heerde, E. C., Folkers, G., van der Saaga, P. T., Karimi-Nejad, Y., Boelens, R. & Kaptein, R. (1999). Millisecond to microsecond time scale dynamics of the retinoid X and retinoic acid receptor DNA-binding domains and dimeric complex formation. Biochemistry, 38, 1951-1956. Woessner, D. E. (1962). Nuclear spin relaxation in ellipsoids undergoind rotational Brownian motion. J. Chem. Phys. 37, 647-654.

979 Wolfram, S. (1996). The Mathematica Book, 3rd edit., Wolfram Media/Cambridge University Press, Cambridge. Vuister, G. W. & Bax, A. (1993). Quantitative J correlation: a new approach for measuring homonuclear threebond J(HNHa) coupling constants in 15Nenriched proteins. J. Am. Chem. Soc. 115, 7772-7777. Xu, W., Alroy, I., Freedman, L. P. & Sigler, P. B. (1993). Stereochemistry of speci®c steroid receptor-DNA interactions. Cold Spring Harbor Symp. Quant. Biol. 58, 133-139. Zilliacus, J., Dahlman-Wright, K., Wright, A., Ê . & Carlstedt-Duke, J. (1991). DNAGustafsson, J.-A binding speci®city of mutant glucocorticoid receptor DNA-binding domains. J. Biol. Chem. 266, 31013106. Zilliacus, J., Wright, A. P. H., Norinder, U., Gustafsson, Ê . & Carlstedt-Duke, J. (1992). Determinants for J.-A DNA-binding site recognition by the glucocorticoid receptor. J. Biol. Chem. 267, 24941-24947.

Edited by P. E. Wright (Received 10 March 1999; received in revised form 16 April 1999; accepted 20 April 1999)