Solution structure and backbone dynamics of the DNA-binding domain of mouse Sox-5

Abstract

The fold of the murine Sox-5 (mSox-5) HMG box in free solution has been determined by multidimensional NMR using ¹⁵N-labeled protein and has been found to adopt the characteristic twisted L-shape made up of two wings: the major wing comprising helix 1 (F10–F25) and helix 2 (N32–A43), the minor wing comprising helix 3 (P51–Y67) in weak antiparallel association with the N-terminal extended segment. ¹⁵N relaxation measurements show considerable mobility (reduced order parameter, S²) in the minor wing that increases toward the amino and carboxy termini of the chain. The mobility of residues C-terminal to Q62 is significantly greater than the equivalent residues of non-sequence-specific boxes, and these residues show a weaker association with the extended N-terminal segment than in non-sequence boxes. Comparison with previously determined structures of HMG boxes both in free solution and complexed with DNA shows close similarity in the packing of the hydrophobic cores and the relative disposition of the three helices. Only in hSRY/DNA does the arrangement of aromatic sidechains differ significantly from that of mSox-5, and only in rHMG1 box 1 bound to cisplatinated DNA does helix 1 have no kink. Helix 3 in mSox-5 is terminated by P68, a conserved residue in DNA sequence-specific HMG boxes, which results in the chain turning through ∼90°.

In mammalian embryogenesis the male determining factor is the SRY protein encoded on the Y chromosome, which switches development from the default (female) pathway to male development (Sinclair et al. 1990). The characteristic feature of the SRY protein is the presence of a sequence-specific DNA-binding HMG box, and a number of other proteins have been described containing HMG boxes with sequences closely homologous to that in SRY (called Sox proteins, where Sox stands for SRY-related HMG box). Alternative splicing of the mouse Sox-5 (mSox-5) gene product results in a short form of the protein, found in the nuclei of postmeiotic spermatids (Denny et al. 1992; Connor et al. 1994) and a long form (L-mSox-5) expressed in chondrocytes (Lefebvre et al. 1998). Both forms of the protein contain the single HMG box. This HMG box recognizes the 6-bp sequence AACAAT (Denny et al. 1992) with an association constant K_a of 6.2 × 10⁷ M⁻¹ (Privalov et al. 1999). Several other Sox proteins have been studied in some detail: for example, defective human Sox9 has been shown to be responsible for the condition campomelic dysplasia and can be a cause of autosomal XY sex reversal (Wagner et al. 1994), whereas the mouse transcription factor Sox-4, which contains an HMG box closely related to that of mSox-5, plays an important role in early lymphoid and heart development (van de Wetering et al. 1993). Drosophila Sox70D (dichaete, fish-hook) encodes a protein required for embryonic segmentation (Nambu and Nambu 1996; Russell et al. 1996). Sequence-specific DNA-binding HMG boxes not classed as Sox proteins by virtue of somewhat divergent amino acid sequences have been recognized in other mammalian proteins, such as human and mouse LEF-1 (Giese et al. 1991; Waterman et al. 1991), a transcription factor that binds to the T-cell α-chain enhancer. It is clear that the sequence-specific HMG box is a DNA-binding domain found in many transcription factors playing key developmental roles (for review, see Wegner 1999).

A second class of HMG boxes that bind DNA are those that show no obvious DNA sequence-specific recognition but prefer to bind to preformed specific DNA structures such as the 4-way junction (Wright and Dixon 1988; Bianchi et al. 1989, 1992; Ferrari et al. 1992; Webb and Thomas 1999), DNA bulges (Payet et al. 1999), and cisplatin-modified DNA (Pil and Lippard 1992; Ohndorf et al. 1999). These non-sequence-specific HMG boxes frequently occur as multiple boxes and include those from the mammalian chromosomal proteins HMG1 and HMG2 (Johns 1982), Upstream Binding Factor UBF (Jantzen et al. 1990; Bazett-Jones et al. 1994), and mtTF1 (Parisi and Clayton 1991). The term ``architectural transcription factors'' has been applied to these HMG proteins (Groschedl et al. 1994; Wolffe 1994), to indicate their ability to manipulate the structure of the DNA to which they bind.

The structures of several HMG boxes have been determined for both the non-sequence-specific (Read et al. 1993; Weir et al. 1993; Jones et al. 1994; Hardman et al. 1995; Allain et al. 1999) and sequence-specific (van Houte et al. 1995) classes. Additionally, the structure of HMG boxes in complex with DNA is known, both non-sequence-specific (Allain et al. 1999; Murphy et al. 1999; Ohndorf et al. 1999) and sequence-specific (Love et al. 1995; Werner et al. 1995). In all cases the protein fold is observed to be a 3-helix bundle in the form of a somewhat asymmetric L-shape with helices 1 and 2 in one arm (the major wing) and most of helix 3 together with the most N-terminal residues in the other arm (the minor wing). The structures of the sequence-specific HMG box of hSRY bound to 8-bp DNA (Werner et al. 1995) and of mLEF-1 bound to 15-bp DNA (Love et al. 1995) showed that the L-shaped protein takes the minor groove of the DNA in a pincer-like grip and causes the DNA to bend away from the protein, toward the major groove, with considerable unwinding. DNA bending is brought about by a series of contacts on the inside of the L-shaped HMG box fold and in part by partial intercalation of a single hydrophobic sidechain between two adjacent adenines.

The mLEF-1/DNA complex additionally revealed that residues C-terminal to the minimal HMG box crossed over the major groove on the inside of the bend to make contact with the far end of the DNA (Love et al. 1995). This feature may explain the exceptionally large bend angle (130°) generated by mLEF-1 (Giese et al. 1992), shown to be dependent on the presence of a C-terminal extension (Lnenicek-Allen et al. 1996). In the model of the non-sequence-specific HMG box of NHP6A bound to DNA, a basic N-terminal extension of the folded domain wraps around the major groove (Allain et al. 1999).

Absent from our understanding of sequence-specific HMG box structures and their interactions with DNA is knowledge of the structure of the same HMG box both free in solution and bound to DNA, since the free-solution structures of neither hSRY nor mLEF-1 have been determined. This is a matter of particular interest since calorimetric measurements of the mSox-5 HMG box have shown that significant levels of protein refolding occur on association, in addition to the DNA bending (Privalov et al. 1999). The present study of mSox-5 represents the first part of a project to make a detailed comparison of the structure of a free and DNA-bound sequence-specific HMG box. We also take the opportunity to compare the structure of the mSox-5 HMG box with other HMG box structures both free and bound to DNA.

Results

NMR resonance assignments of the mSox-5 HMG box

Sequential assignment of the amino acid spin systems was achieved by a combination of 2d ¹H–¹H NOESY and ¹H–¹H HOHAHA NMR spectra obtained at pH 5.4 and 6.2 in 90% H₂O and 100% ²H₂O, plus ¹H NOESY ¹⁵N–¹H HMQC and ¹H HOHAHA ¹⁵N–¹H HMQC 3d spectra at pH 6.2 in 90% H₂O. All spectra were recorded at 25°C. By comparing spectra, all the backbone connectivities were completed, and 82.5% of all protons were assigned. Owing to chemical shift overlap in the 2.8–3.3-ppm region of the proton spectrum it was not possible to assign the arginine C_δH protons of residues 5, 19, 58, 75, and 78 nor the lysine C_ɛH protons of residues 4, 15, 35, 42, 49, 61, 66, 71, 73, and 77. The assignment of the backbone ¹⁵N resonances was complete except for N30 and S31, for both of which the amide proton is in fast exchange with the solvent. The sidechain ¹⁵N resonances of all the asparagine, glutamine, and tryptophan residues were completely assigned, plus the ¹⁵N_ε resonances of residues R18, R19, and R40. The remaining sidechain nitrogen resonances could not be identified.

The NOE restraints

From the 2d and 3d NOESY spectra, a total of 1424 nonredundant distance restraints were established, of which 277 were intraresidue and 1147 were interresidue—452 sequential (i, i + 1), 145 (i, i + 2), 159 (i, i + 3), 70 (i, i + 4) and 321 long range (i, i > 4). Figure 1 ▶ shows the distribution of NOE contacts in the mSox-5 HMG box. The (i, i + 3) and (i, i + 4) contacts run parallel to the diagonal and indicate the three α-helical regions F10 to F25, N32 to A43, and P51 to Y67. The location of helices can also be seen from Figure 2B ▶, which documents the runs of d_αN (i, i + 3) and d_αN (i, i + 4) cross-peaks that define these regions as being α-helical. This assignment of helices is supported by the Chemical Shift Index of the C_αH protons (Fig. 2C ▶). Together these represent 45 out of 81 residues (55%) in an α-helical conformation, in good agreement with CD data (Connor et al. 1994). The position and length of the three α helices in mSox-5 are similar to those observed in other sequence-specific HMG box structures (Love et al. 1995; Werner et al. 1995).

Fig. 1.

Distribution plot of NOE restraints observed for each residue of the mSox-5 HMG box. Each square represents at least one NOE contact between backbone protons (above diagonal, filled squares) or between backbone–sidechain protons and sidechain–sidechain protons (below diagonal, open squares). The sequence of the expressed mSox-5 HMG box (less the N-terminal Gly–Ser) together with the positions of the α helices determined in this work are shown above the plot. The line at residue 16 represents the location of the kink in helix 1.

Fig. 2.

(A) Histogram of the total number of NOE restraints used in the structure determination plotted against position in the mSox-5 HMG box sequence. (B) The distribution of observed d_αN(i, i + 3) (upper) and d_αN(i, i + 4) (lower) cross-peaks showing the α-helical regions of the mSox-5 HMG box. (C) C_αH chemical shift index according to Wishart et al. (1992), where +1 = β sheet, 0 = irregular, and −1 = α helix. (D) Variation in the relative peak volumes of backbone amide protons, as measured from a ¹⁵N–¹H HSQC spectrum. (E) Normalized NOE cross-peak volumes of amide NH peaks at the H₂O chemical shift position from a ¹H NOESY ¹⁵N–¹H HMQC spectrum. (F) Normalized peak volumes of amide NH peaks at their NH chemical shift position from a ¹H NOESY ¹⁵N–¹H HMQC spectrum. The horizontal arrows indicate regions of slow amide exchange.

For mSox-5, no backbone-to-backbone NOE contacts could be observed between N-terminal residues 1–12 and helix 3. There were also fewer and weaker backbone-to-sidechain and sidechain-to-sidechain contacts between these two segments of mSox-5 than were observed for hamster HMG1 box 2 (Read et al. 1993). This indicates a weaker association between the N-terminal segment and helix 3 in the Sox-5 HMG box than in HMG1 box 2.

A plot of the number of NOE restraints against position in the sequence (Fig. 2A ▶) shows that the most ill-defined regions correspond to P1–H2, R5–M7, N30–S31, T45–L47, and P74–T79. Lack of definition in the R5–M7 region was partly caused by chemical shift overlap, which also prevented observation of sequential amide NOEs between residues Y70–K71 and Y72–K73. The Chemical Shift Index (Fig. 2C ▶) suggests that residues 1–8 and 73–79 are in an irregular conformation (CSI = 0).

The volume of peaks arising from each amide proton in the domain was measured in a ¹⁵N–¹H HSQC spectrum. A plot of the relative peak volumes (Fig. 2D ▶) shows that these are smaller for residues I3–M7, H29–N32, T45–L47, and H63, in comparison with the remainder. This is probably because of fast solvent exchange of these amide protons and could account for the fewer NOEs observed in these regions. Measurement of NH proton cross-peak volumes at the NH and water chemical shift positions in 3D NOESY spectra confirmed this rapid exchange (see Fig. 2E,F ▶). Helix 3 (P51–Y67) was found to contain slowly exchanging amide protons up to residue L59, beyond which more rapid exchange occurs.

³ J_HNα coupling data

Coupling constants were measured at two separate temperatures (12° and 25°C, Fig. 3A,B ▶). A series of values between 2.5 and 5.0 Hz was observed in the helical regions, as expected. However, in helix 1, ³J_HNα values of 6.5 Hz (at 12°C) and 7.5 Hz (at 25°C) were obtained for residue D16, which suggests a break or kink in the helix at this position. In helix 3 it is notable that the ³J_HNα values steadily increase from 3.0 to 7.0 Hz between residues Q56 and Y67, suggesting a gradual loosening or stretching toward the C-terminal end of the helix. Regions with ³J_HNα values generally greater than 6.0 Hz at 25°C are P1–N8, A24–N32, A43–K49, and K66–T79. A lower temperature of 12°C did not significantly affect these values in regions A43–K49 and Y67–T79. However, ³J_HNα coupling constants for the C-terminal end of helix 3 (S60–K66) and for the N-terminal residues K4, R5, and M7 are reduced at 12°C by 0.5 to 2.0 Hz (Fig. 3C ▶). This reduction represents an increase in structural order in this region as the temperature is lowered. Increased structural order at lower temperatures has previously been observed for the mSox-5 HMG box using CD and DSC (Crane-Robinson et al. 1998).

Fig. 3.

Histogram of ³J_HNα values recorded at (A) 25°C and (B) 12°C for the mSox-5 HMG box plotted against position in the sequence. Residues for which no values are shown correspond either to prolines or to residues for which coupling constants could not be extracted. (C) Difference plot of the ³J_HNα values (25°–12°C).

Relaxation measurements

Figure 4A–C ▶ shows the experimentally determined ¹⁵N longitudinal (T₁) and transverse (T₂) relaxation times, with the heteronuclear {¹H}–¹⁵N NOE values, obtained at 25°C, for 64 out of a possible 72 backbone amides. The combination of ¹⁵N T₁ and T₂ relaxation times, with the {¹H}–¹⁵N NOE values were then analyzed within a framework of isotropic rotational diffusion, using the Lipari–Szabo model-free formalism (Lipari and Szabo 1982a,b). In this procedure the internal dynamic properties of the protein are described in terms of residue-specific measures of the generalized order parameter (S²; see Fig. 4D ▶), a time constant (τ_i) for fast (picosecond) internal motion, and where necessary a factor R_ex introduced for a good fit. The R_ex term may be taken to indicate a contribution to the transverse relaxation rate arising from slow (milli- to microsecond) interconversion processes (Clore et al. 1990a) or from specific self-association phenomena (Pfuhl et al. 1999). Where necessary for a good fit an additional order parameter (S_f²) was introduced, which may be taken to indicate motion on an intermediate time-scale (Clore et al. 1990b; Mandel et al. 1995). For the most important parameter to assess fast internal motion, a limiting value of S² = 1 indicates total restriction and S² = 0 would indicate no restriction of the internal motion of the amide N—H bond.

Fig. 4.

Values of (A) ¹⁵N T₁; (B) ¹⁵N T₂; (C) steady-state {¹H}–¹⁵N NOE; and (D) the derived order parameter S², obtained at 25°C and plotted against position in the mSox-5 HMG box sequence. Residues for which no values are shown correspond either to prolines or to residues for which relaxation data could not be extracted. Errors are shown either as bars or not shown, in which case the error is smaller than the plot symbol.

The relaxation data (Fig. 4 ▶) for residues between F10 and Y67 show rather uniform values in both the relaxation parameters and extracted S² values (with a mean S² = 0.83 and mean {¹H}–¹⁵N NOE = 0.63) as expected for a folded domain. Outside of this region, from residue A9 to the N terminus and from residue P68 to the C terminus, the relaxation parameters and S² values indicate a gradually increasing flexibility of the polypeptide chain. In addition, derived S_f² values (data not shown) were observed in the N- and C-terminal regions of mSox-5, as expected for dynamic disorder having both fast and intermediate time-scale components.

Helices 1 (F10–F25) and 2 (N32–A43) have mean S² values of 0.85 and 0.86, with mean {¹H}–¹⁵N NOE values of 0.67 and 0.68, respectively. However helix 3 (residues P51–Y67) exhibits mean S² and {¹H}–¹⁵N NOE values of only 0.78 and 0.52, respectively. These lower values, as compared to the rest of the domain, arise from a series of low S² and {¹H}–¹⁵N NOE values for residues Q62 to Y67, in the C-terminal end of helix 3. Calculation of the mean S² and {¹H}–¹⁵N NOE for just the N-terminal end of helix 3 (residues P51–K61) yielded values of 0.85 and 0.64, values similar to that found for helices 1 and 2. This indicates that helices 1 and 2 and the N-terminal segment of helix 3 do not display fast internal motion, whereas the six residues at the C-terminal end of helix 3 show a degree of dynamic disorder.

In addition, helix 1 exhibits a trend toward lower S² values along the helix, suggesting the N-terminal end of the helix has a slightly greater degree of dynamic order than the C-terminal end. Isolated low values occurred near the ends of helices 1 and 2 at residues F25, I33, and W41. For loop 1 slightly low values for the mean S² and {¹H}–¹⁵N NOE were observed (0.81 and 0.61, respectively), although this was the not case for loop 2 (0.85 and 0.67, respectively). This could suggest that the degree of dynamic order in the loops is as great as in the helices.

The derived R_ex terms (data not shown) were widely dispersed throughout mSox-5 and were all small in value (<3.30 sec⁻¹) with the exception of residues L22 (R_ex = 4.64) and K49 (R_ex = 8.79). Since no concentration dependence on chemical shifts in the ¹⁵N–¹H HSQC spectra was observed, this would argue against self-association and implies that slow conformational interconversion processes occur at residues L22 and K49, in loops 1 and 2, respectively.

Determination of structures

Structures were initially determined using a restraint file for the complete 79-residue mSox-5 HMG box. Families of consistent structures were clearly observed, but these structures showed that beyond residue 70 there was a high degree of conformational variability. Further sets of structures were therefore obtained using a restraint file for residues 1–70. From 50 calculated structures, the 30 of lowest energy (<1300 kcal mole⁻¹) were selected and further refined in the presence of selected hydrogen bond and dihedral angle restraints, which were based on hard-to-exchange amide proton data (Fig. 2E,F ▶) and ³J_HNα coupling-constant data (Fig. 3 ▶), respectively. After the final minimization, all selected structures contained no distance restraint violations greater than 0.5 Å for both backbone and sidechain distance restraints and no dihedral angle violations greater than 10°.

Deviation from a standard α helix was found within helix 1. The ³J_HNα coupling constant for D16 is 7.5 ± 0.5 Hz (Fig. 3A ▶), a value compatible with a phi angle of −90 ± 10°. Furthermore, the d_αN(i, i + 4) NOE between residues V12 and D16 was absent from the NOESY spectra. Taken altogether, this indicates that the hydrogen bond between the carbonyl O of V12 and the NH of D16 is very extended and nonlinear, leading to distortion of helix 1 at this point. This hydrogen bond restraint was therefore not included in the final modeling.

In the C-terminal part of helix 3 (H63–Y67), the ³J_HNα values for K66 and Y67 are 7.2 and 8.2 Hz, respectively, and all the d_αN(i, i + 3) and d_αN(i, i + 4) NOE cross-peaks were very weak. Furthermore, an increase in exchange volumes of the amide protons of residues 60–67 at the water chemical shift position was observed (Fig. 2E ▶). The more mobile C-terminal end of helix 3 is also manifested in the longer T₂ and lower S² values for residues 62–67 (Fig. 4B,D ▶). Hydrogen bond restraints were therefore applied only to residues E54–Q62 of helix 3.

Residues 1–70 were further examined for multiconformational states using the Xplor v3.851 ensemble program for cross-validation of structures (Bovin and Brunger 1995,1996). Twenty structures were collected, and the averages were determined for the existence of 1, 2, or 3 conformers. The number of violations greater than 0.2 Å were 38.1, 2.3, and 1.3 (with standard deviations of 3.1, 2.0, and 1.3, respectively) for nonrefined 1-, 2-, and 3-conformer models, respectively. The considerable reduction in violations with increasing number of possible conformers, in particular from 1 to 2, could result either from the fact that there are genuinely two conformers or because in a fold having significant flexibility in parts of the structure, an increase in the number of allowed conformers inevitably leads to fewer violations. We therefore compared the two averaged conformers in the 2-conformer model for clear signs of differences at particular points. In loop 2 there were no significant conformational differences, but in loop 1 the two conformers were different. To decide if this was caused by flexibility or the presence of two genuine conformers, we compared the 3 structures of loop 1 in the 3-conformer model: The third average conformer was intermediate between the first two, and we conclude that loop 1 is indeed flexible. This conclusion accords with the finding that the amide NHs of N30 and S31 are in very rapid solvent exchange (Fig. 2F ▶).

Description of the mSox-5 HMG box structure

Figure 5A ▶ shows a stereo backbone view of the 30 final structures comprising residues 1–70 of the mSox-5 HMG box. The structures were overlaid by best-fit superposition of the backbone heavy atoms (amide N, C_α, carbonyl C and O) of residues F10–F25 and N32–K42. These two regions, comprising helices 1 and 2, were chosen for overlay since they were the best defined (number of NOEs per residue >21, with an average of 58) and consistent in position relative to each other, that is, low average pairwise RMSDs. The average pairwise RMSD for the heavy backbone atoms in this set was 0.27 ± 0.13 Å. Superposition of residues F10–F25, N32–K42, and P51–Y67 gave a pairwise RMSD of 1.56 ± 0.90 Å, whereas superposition of all heavy backbone atoms for residues 1–67 gave a pairwise RMSD of 2.3 ± 1.7 Å.

Fig. 5.

(A) Stereo-view of the final set of 30 lowest energy structures for the mSox-5 HMG box (residues 1–70) obtained by best-fit superposition of the C_α, C, O, and N backbone atoms of helices 1 and 2 (F10–F25 and N32–K42, respectively). (B) Sidechain positions of selected residues in the major wing of the mSox-5 HMG box. (C) The final set of 30 lowest energy structures (residues 50–70 and 1–9) obtained by superposition of the C_α, C, O, and N backbone atoms of residues 51–67 and 1–9. The co-ordinates for the mSox-5 HMG box have been submitted to the Protein Databank.

The overall L-shaped conformation of the mSox-5 HMG box is similar to that found for other HMG boxes. The major wing (containing the majority of the amino acids) is the short arm of the fold and contains A9, helix 1 (F10–F25), loop 1 (P26–S31), helix 2 (N32–A43), loop 2 (M44–Q50), and the N-terminal part of helix 3 (P51–Y53) that projects sidechains into the hydrophobic core between helices 1 and 2. Helices 1 and 2 make an angle of 56° to each other. In loop 1, P26–H29 makes a type III β turn, although the distance between the NHs of D27 and M28 is ∼1.0 Å larger than expected for a canonical turn. The following two residues, N30 and S31, are in an extended conformation. In loop 2, M44–L47 are in a partially extended conformation, and the following 4 residues, E48–P51, have an approximately α-helical conformation with a similar set of NOE contacts to the equivalent residues in HMG1 box 2 (Read et al. 1993; Weir et al. 1993; Read et al. 1995). The minor wing is the long arm of the fold and consists of the extended N-terminal segment P1–N8 running alongside and antiparallel to most of helix 3 (E54–Y67).

Figure 5B ▶ shows a view of the major wing, highlighting selected sidechains in all of the 30 final structures. A consistent hydrophobic core (F10, W13, L37, W41, M44, and Y52) is seen with the main apolar contacts having a well-defined geometry. In contrast, the sidechains of M11, N30, and S31 point out into the solvent and show no fixed conformation.

The residues involved in forming the apolar core maintain the orientation of the surrounding three α helices (Fig. 5B ▶). The aromatic rings of F10, W13, and W41 stack onto one another and orient helices 1 and 2. Residues L22 and L37 form hydrophobic contacts between helix 1 and the β turn of P26–H29 and between helices 1 and 2, respectively. The other aromatic residue, F25, makes close contact with the methyl groups of M28 and I21: These serve to maintain the angle between helix 1 and helix 2, as well as stabilizing the β turn between helices 1 and 2. The orientation of helix 1 to helix 3 depends on apolar contact between the methyl group of A9 and the sidechain of Y52 in helix 3.

In the minor wing, NOEs were observed between the sidechains of L59, H63, and L64 and the backbone residues of M7/N8, R5, and K4, respectively. There were also weak NOEs between the sidechains of H63 and Y67 and the sidechain of I3. These fix the position of the N-terminal segment of the box alongside helix 3. At the C-terminal end of helix 3, residues Y67–Y70 form a type III β turn similar to that found in loop 1. In order to better assess the degree of local order in the minor wing, superpositions were made of residues 51–67 (helix 3) and residues 1–9 in the N-terminal segment. Figure 5C ▶ shows that helix 3 exhibits considerable regularity, whereas the nine N-terminal residues are significantly less ordered but more ordered than appears to be the case from the helix 1/2 superpositions of Fig. 5A ▶. The disorder in the central part of the N-terminal segment, R5–M7, is owing to a lack of NOEs to residue P6, which in turn is caused by chemical shift overlap. However, the apparent decrease in structural order from residues 9 to 1 (Fig. 5C ▶) is in good accord with the reduction in S² shown in Figure 4D ▶.

Conformation of the C-terminal segment

Since a regular NOE intensity pattern was observed for residues lying beyond helix 3, P68–R75, that is, medium/strong d_NN, d_αN(i, i), very strong d_αN(i, i + 1), and very weak d_αN(i, i + 2), a new set of structures were calculated that included restraints observed for residues 71–79. The conformation of the C-terminal region, including and beyond helix 3, was then examined using the following superpositions: 51–67, 51–77, and 68–77. This procedure is similar to that carried out for rHMG1 box 2 (Weir et al. 1993) and for mSox-4 (van Houte et al. 1995). From 50 calculated structures, 30 low-energy structures were selected and refined. The final structures when overlaid by superposition of all backbone heavy atoms for residues 51–77 (Fig. 6A ▶) gave a pairwise RMSD of 3.78 ± 1.0 Å, whereas superposition utilizing only helix 3 (P51–Y67, Fig. 6B ▶) gave an RMSD of 0.99 ± 0.30 Å, a value typical for an ordered structure with some variability. The large difference between these two RMSD values indicates that residues beyond the C-terminal end of helix 3 do not have a particularly fixed position. In order to see if there is any regular structure at all in the region beyond helix 3, residues 68–77 were superimposed (Fig. 6C ▶). The backbone heavy atom RMSD for this region was 2.9 ± 1.1 Å. Figure 6C ▶ shows that the overall conformational envelope of these C-terminal residues (68–79) is a hooked shape with a poorly defined turn at P68 (the exit to helix 3) and a second turn centred around P74.

Fig. 6.

The final set of 30 lowest energy structures using only restraints for residues 48–79 of the mSox-5 HMG box. Backbone views of the structures overlaid using the heavy atoms of (A) helix 3 and the C-terminal region (P51–K77); (B) helix 3 only (P51–K67); and (C) the C-terminal region only (P68–K77). In view C only residues Y67–T79 are shown to aid clarity.

Discussion

Comparison with other HMG box structures

A comparison of the lowest-energy mSox-5 HMG box structure was made with other HMG box structures whose coordinates are available. These are: (1) box 2 of HMG1, of which there are two examples: our own derived from the hamster sequence (Read et al. 1993, 1995) and that derived from the rat sequence (Weir et al. 1993); (2) box 1 from rat HMG1 (Hardman et al. 1995) and the same box complexed with cisplatinated 16-bp DNA (Ohndorf et al. 1999); (3) Drosophila HMG-D (Jones et al. 1994) and the same box complexed with 10 bp of DNA (Murphy et al. 1999); (4) the HMG box of human SRY complexed with 8 bp of DNA (Werner et al. 1995); (5) the mouse LEF-1 HMG box complexed with 15 bp of DNA (Love et al. 1995); and (6) the yeast HMG protein NHP6A, both alone and complexed with 15 bp of DNA (Allain et al. 1999). In the comparison that follows, the residue numbering of mSox-5 is used for the equivalent residues in all structures.

It is striking that superposition of the various HMG box folds using the heavy backbone atoms of helices 1 and 2 (residues 10–25 and 32–42) yields little variation, with the RMSD values ranging only between 1.00 and 1.72 Å (Table 1, column 2). When helix 3 (residues 52–67 in mSox-5) is included in the superposition, the RMSD values rise to between 1.30 and 3.12 Å (Table 1, column 4), indicating that the position of helix 3 in the minor wing is not well defined relative to the major wing (which includes helices 1 and 2). Removal of the last 5 residues of helix 3 from the superpositions, that is, using only residues 52–62 for helix 3, significantly reduced the RMSD values for some of the boxes (Table 1, column 3) and reflects the increased uncertainties in the fold between residues 62 and 67.

Table 1.

Comparison of mSox–5 with ten other HMG box folds1

	RMSD values for the three helices (in Å)2			Angles in and between helices3			Pairwise RMSD values of aromatic residues (in Å)4
HMG box	For residues 10–25, 32–42	For residues 10–25, 32–42, 52–62	For residues 10–25, 32–42, 52–67	Bend angle in helix 1	Angle of Helix 1 to Helix 2	Angle of helix 2 to Helix 3	Positions in mSox-5 HMG box	RMSD for Cβ & Cγ atoms	RMSD for all sidechain heavy atoms
mSox–5				35°	56°	60°
rHMG1 box 1	1.32	1.25	1.30	30°	55°	58°	10, 13, 41, 52	0.83	1.38
chHMG1 box 2	1.15	2.48	2.69	35°	52°	71°	10, 13, 41, 52	0.76	1.31
rHMG1 box 2	1.02	2.17	2.59	40°	55°	80°	10, 13, 41, 52	0.86	1.60
dHMG–D	1.23	1.61	1.66	55°	65°	70°	10, 13, 41, 52	0.84	1.40
yNHP6A	1.00	1.87	2.22	35°	55°	70°	10, 13, 41, 52, 53	0.90	3.39
hSRY/DNA	1.70	2.59	3.12	33°	60°	87°	10, 13, 41, 52	2.33	3.30
hSRY*/DNA							10, 13, 41, 52	2.43	1.59
mLEF–1/DNA	1.22	1.54	1.49	40°	72°	55°	10, 13, 41, 52, 53	1.36	1.88
yNHP6A/DNA	1.13	2.08	2.38	35°	55°	70°	10, 13, 41, 52	0.86	0.84
dHMG–D/DNA	1.07	1.51	1.62	50°	65°	65°	10, 13, 41, 52	0.48	1.47
rHMG1 box 1/DNA	1.72	1.65	1.63	0°–5°	62°	64°	10, 13, 41, 52	1.78	1.28

¹ Protein Databank accession numbers laab model 22 (rat HMG1 box 1), 1hsm (hamster HMG1 box 2), 1hme (rat HMG1 box 2), 1hma model 1 (D HMG–D), 1hry (human SRY/DNA), 1qrv (Drosophila HMG–D/DNA), 1 ckt (rat HMG1 box 1/cisplatin DNA). Mouse LEF–1/DNA coordinates were provided by Dr. Peter Wright. Yeast NHP6A and yNHP6A/DNA coordinates were provided by Dr. Julie Feigon.

² RMSD values were obtained by superimposing the heavy backbone atoms of the above residues with respect to the lowest-energy mSox–5 HMG box structure (residue numbering as in mSox–5).

³ The bend angles in helix 1 are measured in the plane containing both parts of the helix. The included angle between helices 2 and 3 was measured by projecting helix 2 onto the plane through the long axis of helix 3 that lies parallel to the long axis of helix 2. In several cases the long axes of both helices lie approximately in the same plane. The angle between helices 1 and 2 was measured in the same way as that between helices 2 and 3.

⁴ Sidechain superpositions were carried out as pairwise alignments with the lowest energy structure of the mSox–5 HMG box. hSRY/DNA and mLEF–1/DNA HMG boxes have an aromatic residue at position 53. hSRY*/DNA differs from hSRY/DNA by the removal of residue F53 (to make it more readily comparable with the other HMG boxes).

Most structures show a considerable kink in helix 1 at residue 16 (Table 1, column 5), which does not vary greatly in magnitude, averaging 39.0 ± 8.0°. However, in the case of box 1 of rHMG1 bound to 16 bp of cisplatinated DNA, the bend in helix 1 is absent. This change could be caused by the presence of the cross-linked cisplatinated DNA, changed solvation, or crystal packing, since this is a structure obtained by X-ray crystallography (Ohndorf et al. 1999). Helix 3 of mSox-5 is defined as starting at P51, but the preceding 4 residues are in an approximately helical conformation and have been defined as included in helix 3 in some HMG box structures. We do not include residues L47–Q50 in helix 3 since that turn is inclined at an angle of about 20° to the axis of helix 3 in mSox-5, and is similarly inclined in the other HMG box structures. The angle between helices 1 and 2 is fairly constant (Table 1, column 6), although there is a small difference between the average angle for free and for bound HMG boxes. The angle between helices 2 and 3 (Table 1, column 7) represents the angle between the two wings of the HMG box fold and likewise shows little variation between free and DNA bound boxes. The suggestion that the angle between the two wings of the fold might alter on binding DNA, that is, that there might be a considerable degree of induced fit in the protein (Ferrari et al. 1992) is therefore not supported.

Figure 7A ▶ shows the hydrophobic core for all 6 HMG box folds determined in the absence of DNA, and it is seen that the relative positions of the 4 principal aromatic sidechains (residues 10, 13, 41, and 52) are the same in all 6 folds. Figure 7B ▶ shows the hydrophobic core for the five HMG box/DNA structures (together with the free-solution mSox-5 structure). For these structures, again, the 4 principal aromatic sidechains are in similar orientations, except for hSRY/DNA, for which the positions of the W41 and W13 sidechains are inverted such that it is the sidechain of W13 that contacts residues F52 and F53, rather than the sidechain of W41. A detailed comparison of the conserved aromatic sidechains in the hydrophobic core was made by superimposing them onto those of mSox-5. The pairwise RMSD values obtained for the C_β and C_γ atoms (Table 1, column 9) and for all sidechain heavy atoms (Table 1, column 10) in the 10 HMG box structures do not differ greatly, with the notable exception of hSRY/DNA. This divergence of hSRY is unexpected because the amino acid sequence of mSox-5 is closer to hSRY than to any of the other 9 structures, and there are considerable differences between the sequence of mSox-5 and the non-sequence-specific HMG boxes.

Fig. 7.

(A) Comparison of aromatic sidechain positions in the main hydrophobic core of the HMG box of mSox-5 (this work) with yeast NHP6A, rat HMG1 box 1, Drosophila HMG-D, hamster HMG1 box 2, and rat HMG1 box 2. (B) Comparison of aromatic sidechain positions in the main hydrophobic core of the HMG box of mSox-5 (this work) with the DNA complexes of yeast NHP6A, rat HMG1 box 1 (with cisplatinated DNA), Drosophila HMG-D, mouse LEF-1, and human SRY. In both A and B each structure was superimposed on that of mSox-5 using the heavy backbone atoms of residues equivalent to F10–F25 and N32–K42 of the mSox-5 HMG box.

DNA binding and bending by HMG boxes

HMG boxes all have very similar L-shaped structures, so what distinguishes the sequence-specific and structure-specific HMG boxes and their binding to DNA? Recently, a binding mode has been shown for non-sequence-specific HMG boxes bound to DNA (Allain et al. 1999; Murphy et al. 1999; Ohndorf et al. 1999) in which the residue at position 30 is hydrophobic and partially intercalates into the DNA. For box 1 of rHMG1 and yNHP6A, residue 30 is phenylalanine, and in dHMG-D it is valine. In other non-sequence-specific boxes, for example, mammalian HMG1 box 2, residue 30 is isoleucine. In sequence-specific HMG boxes this residue is polar (asparagine in hSRY and Sox proteins, serine in LEF-1 and TCF1). Partial intercalation of residue 30 is thus not expected for sequence-specific HMG boxes, but could be the norm for non-sequence-specific boxes.

The primacy of the minor wing in establishing DNA sequence specificity was shown by a subdomain swap experiment using hLEF-1 and HMG1 box 2 (Read et al. 1994), and this was borne out by the structures of the two sequence-specific HMG box/DNA complexes (Love et al. 1995; Werner et al. 1995). Several amino acids in the minor wing have been singled out as critical for sequence-specific DNA binding: N8 is involved in hydrogen bonding and electrostatic interactions with three bases and forms the stem of an amino acid wedge that forces intercalation of the sidechain of residue 11 between two adenine rings (Werner et al. 1996), and it is likely that this mechanism operates for mSox-5/DNA binding and involves N8–W13, with M11 as the intercalating residue. In non-sequence-specific HMG boxes, residue 8 is normally serine and never asparagine. At position 11 in sequence-specific boxes the residue is either methionine or isoleucine (occasionally phenylalanine), and in non-sequence-specific boxes it is one of several large hydrophobics, with the notable exception of box 1 of mammalian HMG1 and HMG2, in which it is alanine. Indeed, this residue was found not to be intercalated into the cisplatinated DNA of the complex with rHMG1 (Ohndorf et al. 1999). It seems therefore that for most non-sequence-specific HMG boxes there are two DNA-intercalating residues but only one for sequence-specific boxes.

Residues V3, Y67, and Y70 in hSRY pack together so as to present a precise surface of the N-terminal region to the DNA, which appears critical for DNA sequence-specific recognition (Werner et al. 1995, 1996): In DNA non-sequence-specific boxes, residue 3 is normally proline (Read et al. 1995) and never valine or isoleucine (as in mLEF-1). For this hydrophobic cluster of 3 sidechains to form it appears necessary for the backbone to turn through an approximate right angle at the end of helix 3, a conformation very evident in the structures of hSRY and mLEF-1 bound to DNA (Love et al. 1995; Werner et al. 1995). A proline residue (P68 in mSox-5) seems essential for this change in backbone direction, and proline is conserved at this position in all sequence-specific HMG boxes but absent from all non-sequence-specific boxes (Ner 1992). In free-solution mSox-5, P68 is part of a turn (Y67–Y70) that also results in a change of chain direction by ∼90° (Fig. 6 ▶).

The minor wing of the mSox-5 fold

The structural independence of the minor wing of the HMG box fold from its major wing is indicated by several criteria. In our early studies of the domain structure of HMG1 using limited trypsin digestion (Cary et al. 1983) it was found that box 1 was preferentially cut at the C-terminal sides of residues 6 and 60, positions now seen to correspond to the boundary between the two wings. Importantly, the remaining major wing fragment was shown to be fully folded, demonstrating that the major wing does not require the minor wing in order to fold (Cary et al. 1983). In later experiments we showed that it was possible to construct a folded chimeric HMG box from the major wing of HMG1 box 2 and the minor wing of hLEF-1 that preserved the sequence-specific DNA-binding characteristics of hLEF-1 (Read et al. 1994).

A difference between mSox-5 and non-sequence-specific boxes in free solution is the increased flexibility in helix 3 of mSox-5. The relaxation data in Figure 4 ▶ show a steady decrease in the order parameter S² beyond residue Q62, 6 residues before the end of helix 3. In marked contrast, the equivalent residues in helix 3 of non-sequence-specific boxes (both boxes from HMG1 [Broadhurst et al. 1995] and dHMG-D [Jones et al. 1994]) showed no increase in dynamic disorder. This might be related to a requirement for a higher level of induced fit in the binding of sequence-specific HMG boxes to DNA. The importance of protein refolding in DNA sequence recognition has recently been discussed (Wright and Dyson 1999).

A DSC and CD study demonstrated that the mSox-5 HMG box denatures as two separate subdomains. The lower melting subdomain was assigned to the minor wing on the basis of changes in the intrinsic fluorescence and NMR spectrum (Crane-Robinson et al. 1998). DSC and CD melting studies of HMG1 box 2 also show the presence of two subdomains, but the stability of the lower melting minor wing is substantially greater than for mSox-5 (P.D. Cary, C.M. Read, C. Crane-Robinson and P.L. Privalov, unpubl.). Deconvolution of the calorimetric Cp/T function observed for mSox-5 suggested that the melting of the minor wing is a cooperative transition with a T_m of 34°C that partially overlaps the melting of the major wing (T_m = 46°C). The relaxation measurements obtained at 25°C (Fig. 4 ▶) are revealing in this respect. If a subdomain denatures cooperatively in a two-state process, then at some defined intermediate temperature one would expect the mobility of all residues in the subdomain to be the same and, in a fast exchange situation, to reflect the relative proportions of the native and denatured states at that temperature. But that is not what is seen in Figure 4 ▶ for the minor wing (residues 54 to the C-terminus and 1–8): The degree of dynamic order of residues 54–61 is high, but that for the succeeding C-terminal residues gradually decreases. The degree of dynamic order of the first 8 N-terminal residues also varies with its position in the chain. This is what would be expected if temperature increase resulted in a gradual unfolding of the minor wing (from the N- and C-terminal ends) rather than a cooperative melting of the whole wing. Relaxation measurements at several temperatures would be required to absolutely prove this point, but such considerable differences in mobility along the minor wing measured at a single temperature leave little doubt that its melting is a continuous process.

Materials and methods

Protein expression, purification, and characterization

The HMG box of mSox-5 (residues 182–260, Denny et al. 1992) was expressed using the pGEX-2T system in Escherichia coli BL21 (DE3) plysS cells, essentially as described (Read et al. 1993, 1994). Following thrombin cleavage of the box from the glutathione–agarose beads, the protein was purified by reverse phase HPLC on a C4 column. The domain was refolded by addition of 20 mM potassium phosphate (pH 6.2), and an NMR spectrum was obtained to confirm that the box folded correctly (as compared with a sample that had never been denatured). The ¹⁵N uniformly labeled mSox-5 HMG box was obtained by growth of the cells in an ampicillin-containing M9 minimal medium, with 10 mM ¹⁵N ammonium chloride as the sole nitrogen source. When the cell concentration reached 1.0 OD₆₀₀, IPTG induction was followed by a further 4 hr of growth. Labeled protein was purified by the same means as used for the unlabeled.

The domain studied represents an 81-residue peptide having two additional amino acids (Gly–Ser) at the N terminus of the 79-residue HMG box of mSox-5 (for amino acid sequence, see Fig. 1 ▶). The purified recombinant mSox-5 HMG box showed a single band on both SDS and acetic acid/urea polyacrylamide gels. Electrospray mass spectrometry gave the expected mass of 9804.6 Da. CD spectroscopy of the refolded mSox-5 HMG box indicated an α-helical content of ∼55%. Gel retardation studies showed a band of reduced electrophoretic mobility on binding the protein to a 12-bp DNA duplex containing the recognition sequence 5′-AACAAT-3′, and a circular permutation assay gave a bend angle of 70° (Privalov et al. 1999).

NMR spectroscopy

Two- and three-dimensional NMR spectra were obtained on a home-built General Electrics Omega 600-MHz spectrometer (OCMS, University of Oxford) in 90% H₂O/10% ²H₂O at 25°C, pH 6.2. Further 2d NMR spectra were obtained at 25°C and 32°C at pH values of 6.2 and 5.4 in 90% H₂O/10% ²H₂O and in 100% ²H₂O in order to resolve problems of cross-peak overlap. The spectra obtained at pH 5.4 were recorded on a Bruker AM600 at Oxford University. Protein concentrations of ∼3 mM and 2 mM were used for the 2d and 3d spectra, respectively, in 75 mM potassium phosphate, 0.5 mM DTT. ¹H–¹H NOESY spectra (Jeener et al. 1979; Kumar et al. 1980; Kieffer et al. 1994) were collected in the phase-sensitive manner by the time proportional increments method, with mixing times of 120 msec (pH 6.2) and 130 msec (25°C, pH 5.4)/160 msec (32°C, pH 5.4). ¹H–¹H HOHAHA spectra (Braunschweiler et al. 1983; Davis and Bax 1985^; Bax et al. 1987^; Kieffer et al. 1994) were collected with a mixing time of 32 msec (pH 6.2) and 35 or 40 msec (pH 5.4), as previously described. Care was taken to optimize baseline flatness in the spectra by appropriate choice of the initial t₁ and preacquisition delays (Marion and Bax 1988; Bax et al. 1991). The solvent signal was removed in the F2 dimension by time domain deconvolution (Marion et al. 1989b). Three-dimensional ¹H NOESY ¹⁵N–¹H HMQC and ¹H HOHAHA ¹⁵N–¹H HMQC experiments (Marion et al. 1989a; Messerle et al. 1989; Driscoll et al. 1990) were recorded with 120-msec and 40-msec mixing times, respectively, using 128 × 32 × 512 complex points with spectral widths of 6 kHz (F1), 2 kHz (F2), and 12 kHz (F3). NMR data were processed using the program Felix v2.30 (MSI/Biosym Technologies) and displayed using NMRview (Johnson and Blevins 1994).

Further NMR data were acquired on a Varian UnityPlus 500-MHz spectrometer at University College London. ³J_HNα coupling constants, and their temperature dependence, were determined from ¹⁵N–¹H HMQC-J spectra sequentially recorded at 12°C and 25°C. ³J_HNα coupling constants were evaluated with a nonlinear least-squares curve fit to the observed peaks using an analytical expression for the cross-peak lineshape, including the effects of antiphase dispersive character, cross-correlation, and apodization, according to previously described procedures (Norwood et al. 1992). The ¹⁵N–¹H HMQC-J experiments were also used to define the solvent exchange rate of backbone amide NH protons.

The ¹⁵N T₁, T₂, and {¹H}–¹⁵N heteronuclear NOE data (Kay et al. 1989) were recorded with a 1 mM mSox-5 sample at 25°C, at a ¹⁵N frequency of 50.6 MHz. The data were analyzed using the Lipari–Szabo model-free formalism (Lipari and Szabo 1982a,b) according to the procedures detailed in Pfuhl et al. (1999) and Kristensen et al. (2000) and are essentially identical to those of Palmer and co-workers (Mandel et al. 1995). The ¹⁵N T₁ and T₂ values were estimated by fitting the heights from any given NH peak to a two-parameter exponential decay function. Steady-state {¹H}–¹⁵N NOE values were calculated as NOE = Ī_sat/Ī_unsat, where Ī_sat and Ī_unsat are the average signal heights in the presence and absence of ¹H presaturation, respectively. All spectra were processed with Felix v2.30, and signals were integrated with Xeasy (Bartels et al. 1995). The uncertainties of the signal heights were estimated from the rms noise level in the spectra. Uncertainties in T₁ and T₂ times were estimated from Monte Carlo simulation, and uncertainties of the steady-state {¹H}–¹⁵N NOE were calculated by error propagation.

Structure calculations

Calibration of NOE-derived distances was based on known interatomic spacings within aromatic rings (F10, W13, F25, W41, and Y52) and checked using the well-defined α helices (averaged distances for i to i + 1, i + 2, i + 3, and i + 4). All assigned NOE cross-peaks were then classified as either strong, medium, or weak and for the 3d spectra the distance restraints for the 3 categories were set at 1.8–2.7, 1.8–3.5, and 1.8–5.0 Å, respectively. Since a number of 2d spectra involved mixing times of 120–160 msec, the possibility of spin diffusion effects meant that these spectra had to be recalibrated. It was found, for example, in the case of 2d spectra recorded in ²H₂O, that interactions of up to 7.0 Å could be observed; accordingly, distance categories of 1.8–2.7, 1.8–3.5, 1.8–5.5, and 1.8–7.0 Å were applied to these spectra. Stereospecific assignment (Gronenborn et al. 1991) of the methyl protons of V12, L22, and L39 was achieved from 2d NOESY data in which clear chemical shift differences were observed in conjunction with differential intensities.

Xplor v3.851 (Brunger 1993) was used to calculate structures from an initial pseudorandom coordinate template file. Distance NOE restraint files were written as described by Wuthrich et al. (1983), and redundant NOE restraints were eliminated with Aqua v2.0 (Laskowski et al. 1996). Hydrogen bonds and dihedral angles within expected helices were initially set on the basis of: (1) hard-to-exchange backbone NH protons [NOE volumes measured at the NH (¹⁵N–¹H HSQC) and NH and H₂O (3d NOESY) chemical shift positions]; (2) continuous stretches of d_αN(i, i + 3) and d_αN(i, i + 4) NOE cross-peaks; (3) C_αH proton chemical shifts (Wishart et al. 1992); and (4) ³J_HNα coupling constants. Restraints for phi dihedral angles were determined using the equation ³J_HNα = 6.98 cos² θ − 1.38 cos θ + 1.72. Psi dihedral angles were not restrained. This led to the use of 1424 (1383) distant NOE restraints, 68 (61) dihedral restraints, and 24 (24) pairs of hydrogen bond restraints, on structures modeled for residues 1–79 and 1–70, respectively.

Simulated annealing was carried out using center averaging for pseudo atom positions using the soft potential for all initial calculations, and these were refined until all NOE restraints, including hydrogen bond violations, were less than 0.7 Å for all structures. Low-energy structures selected from these were further refined to produce a consistent family of structures with distance NOE violations less than 0.5 Å. Structures with energies less than 1300 kcal mole⁻¹ were then refined, from which 30 were selected having energies less than 500 kcal mole⁻¹. Multiconformer analysis of the 1––70-residue domain was carried out using Xplor v3.851 with the lowest energy refined structure as initial template and the same restraint files as used previously. Ensembles of 20 structures were examined for each conformer using a violation setting of 0.2 Å, for a maximum of 3 conformer ensembles.

Abbreviations

• NOESY, nuclear Overhauser enhancement and exchange spectroscopy

• HOHAHA, homonuclear Hartmann Hahn

• HMQC, heteronuclear multiple quantum coherence

• HSQC, heteronuclear single quantum coherence

• CD, circular dichroism

• DSC, differential scanning calorimetry.