Sox2–Pax6 on DC5 and Sox2–Oct4 on FGF4 use the same HMG interface. (A) EMSA shows that the same mutation in HMG (m1) that interferes with POU/Sox2 on FGF4 also impairs complex formation with Pax6 on DC5, indicating that Sox2 uses the same interface for ternary complex assembly on these elements. (B) Schematic representation of the Sox2 and Pax6 DNA-binding site in the DC5 enhancer.

DNA-sequence-dependent protein interaction surfaces of Sox2 and Oct factors. Interface areas involved in protein—protein interaction in a DNA-sequence-specific manner are marked on the sequence of Sox2, Oct1, and Oct4 (nonvariant residues of Oct4 and Oct1 are shown with a dot). Several of these areas interact with several other protein partners, resulting in various combinations of homo- and heterodimers. The loop region between the first and second helices in the POU domain, for example, is involved in interactions with Sox2 on the FGF4 and UTF1 enhancers as well as in the homodimerization with a second Oct factor on the PORE element. Also, the C-terminal region the POU domain interacts not only with a second Oct factor on the MORE but, in the case of Oct1, also interacts with coactivator (Obf1) on the PORE as well as on the octamer motif(Chasman et al. 1999; Reményi et al. 2001b).

Oct4 and Sox2 differentially interact on FGF4 and UTF1. (A) EMSA assay of Oct4 and Sox2 with radiolabeled DNA oligonucleotides of FGF4 and UTF1. (Lane 1) No protein, DNA alone. (Lane 2) Oct4 alone + DNA (* indicates that no Sox2 protein was added to this lane). (Lanes 3–7) Increasing amounts of Sox2-HMG protein mixed with equal amounts of Oct4 + DNA. Although Oct4 binds alone to UTF1 significantly more weakly than to FGF4 (cf. the two lane 2s), heterodimerization on UTF1 is more pronounced with even lower amounts of Sox2 (e.g., cf. the two lane 4s or 5s). (Remark: Sox2 binds FGF4 and UTF1 with similar affinity; see Supplemental Material.) The significantly lower degree of Oct4/DNA interaction in the absence of Sox2 on UTF1 compared with that on FGF4 is very likely caused by the nonoptimal octamer motif sequence for POU binding within this DNA element (Nishimoto et al. 1999). (B) Differential spacing of binding sites for Oct4 and Sox2 in FGF4 and UTF1.

Comparison of POU/HMG complexes formed on FGF4 and UTF1. (A, top) Model of POU/HMG/FGF4 (POU domain of Oct1 or Oct4; left), compared with the model of Oct4-POU/HMG/UTF1 (right). The figure illustrates that different spacing between the binding sites for the POU and HMG domains within FGF4 and UTF1 enhancers causes formation of different heterodimeric interfaces. (Bottom) Close-up views of the HMG/POU_S interfaces on FGF4 (left) and UTF1 (right). The POU/HMG/UTF1 model suggests involvement of helix 3 instead of the C terminus of the HMG domain to form the HMG-POU_S interface, whereas the same surface patch of the POU_S domain appears to be involved in both interfaces. The DNA molecules are depicted with a transparent surface and are brown. Coloring of the FGF4 and UTF1 DNA sequences is according to their POU- and Sox2-HMG-binding sites. Notice the difference in spacing of these two sites in the two enhancers. (B) To validate the homology models in A, mutations in Sox2-HMG were designed to selectively interfere with ternary complex formation on the FGF4 (m1) or on the UTF1 enhancers (m2 and m3); (m1) R75E; (m2) K57E,R60E; (m3) R60E,M64E; (mut) mutant version of Oct4 POU (I21Y,D29E). In agreement with both models, m1 specifically impaired heterodimerization on FGF4, whereas m2 and m3 specifically abrogated heterodimerization on UTF1, but Oct4 mut affected both.

Crystal structure of the Oct1/Sox2/FGF4 ternary complex. (A) The centrally positioned POU-specific domain (POU_S) interacts with the HMG domain of Sox2. The HMG domain of Sox2 is blue and the POU domain of Oct is green (POU_s: light green; POU_H; dark geeen). Some part of the POU linker region connecting POU_s and POU_H is invisible in the crystal structure (shown as a dotted line). (B) Protein—DNA interactions between the HMG, POU, and the FGF4 enhancer. The figure shows the sequence of the oligonucleotide used for crystallization. The HMG and the POU domain binding sites are blue and green, respectively. For simplicitty, only DNA-sequence-specific hydrogen bonds are shown. The hydrophobic side chains of M7, F10, and M11 are inserted between base pairs T5 · A45 and T6 · A44, causing an ∼45° bend of the DNA axis. These and other amino acid residues that play a role in bending the DNA at three different base stack levels are either underlined or written in black. (C) EMSA results of Octl/Sox2/DNA ternary complex formation on the FGF4 enhancer using mutant versions of the HMG domain of Sox2 (m1) and POU domain of Oct1 (mut). The mutants were designed to interfere with the POU-HMG interface formation based on the Oct1/Sox2/FGF4 crystal structure. The R75E mutation in Sox2 (m1) and the I21Y, D29R mutation in Oct1 (mut) were introduced to reverse the charges or increase the bulkiness of important amino acids at the POU/HMG interface. (wt) Wild-type protein.

Comparison of differentHMG domains. (A) Structures of different HMG domains from HMG-D (Murphy et al. 1999), Lef-1 (Love et al. 1995), Sry (Werner et al. 1995), and Sox2 (taken from the Oct1/Sox2/FGF4 crystal structure) in complex with DNA. Sox2 generates a similar bend angle (90°) in DNA as reported for other HMG/DNA complex structures (HMG-D, 111°; Lef-1, 117°; Sry, 75°). Helix 1 and helix 2 form extensive contacts with the DNA in the widened minor groove and are involved in bending the DNA in all structures. However, the proteins show significant differences in how their C-terminal regions interact with the DNA molecule. In HMG-D, which binds DNA without sequence specificity, only helices 1 and 2 interact with the DNA. The C terminus of the Lef-1 HMG domain lies in the compressed major groove and stabilizes the bent DNA conformation. In the Sry/DNA structure, the C-terminal part is mainly disordered and is not positioned in the minor groove. The Sox2-HMG C-terminal region fits tightly into the compressed minor groove when in the presence of the POU_S domain. The DNA sugar-phosphate backbone is brown, and the bases belonging to different chains of the DNA molecule are depicted by differentcolors (yellow and orange). Helix 2 (α2) is behind the DNA molecule and, therefore, cannot be seen from this orientation. (B) Multiple sequence alignment of HMG domains of several proteins from differentorganisms. (y) Yeast; (c) Caenorhabditis elegans; (d) Drosophila melanogaster; (m) mouse; (h) human. The order of the proteins on the list reflects their sequence similarity. HMG domains from the first six proteins (dHmgd–hHmg2b) are known to bind DNA in a non-sequence-specific manner, whereas the others (hLef1–hSox13) bind DNA according to a specific sequence. Secondary structure elements of Sox2 from the Oct1/Sox2/DNA crystal structure are shown beneath the alignment. Protein residues that are highly conserved are boxed in gray. It is notable that the protein residues that play an important role in the ordering of the C-terminal region of the HMG from Sox2 (V3, R5, P6, H63, H67, P68, Y70, Y72, R75, and R76) are conserved in almostall members of the SOX family. However, these residues are divergent in HMG domains that bind DNA in a non-sequence-specific manner (Murphy and Churchill 2000). Residues P68, Y72, and P74, which play the most prominent role in positioning the C terminus in the compressed minor groove (cf. Fig. 4A), are red.