Transcription Factor-induced, Phased Bending of the E-selectin Promoter*

E-selectin is an endothelial adhesion molecule that is critically involved in neutrophil adhesion and recruit-ment. All DNA elements required for interleukin-1 in- ducibility have been located in the proximal promoter: an NF-ELAMl/ATF site, two NF-ICB sites (I and 111, the NF-ELAM!2 element and a TATA box. We show here that interleukin-1 induced promoter activity is exquisitely sensitive to the spatial arrangements of these elements. Phasing of the ATF and NF-ICB I1 elements indicates that their relative helix orientation is more important than distance per se. This sensitivity is partly due to a re- quirement for correctly oriented, transcription factor-induced DNA-bending. (i) Band shift analyses with per- muted ATF- and NF-& elements show that their associated factors all bend DNA. (ii) One can function- ally replace the NF-ELAMl/ATF element by a subset of a panel of DNA fragments that contain defined bends in various planes. We conclude that the main role of the factors binding at the NF-ELAMl/ATF element is to alter the conformation of the E-selectin promoter, pre- sumably looping distant enhancer elements into each other's proximity.

endothelial and non-endothelial cells (6), whereas tissue-specific expression is associated with promoter hypo-methylation (8). It has been shown that two NF-KB/HMG I(Y) elements within this proximal E-selectin promoter sequence are essential for IL-1 induction (6,9). In addition, we have identified two other sequence elements and their binding factors that co-operate with NF-KB for maximal, induced E-selectin transcription (summarized in Fig. lA). One of these sites, NF-ELA"l/ ATF, is bound by members of the bZIP family of transcription factors: ATF-a, ATF-2, ATF-3, c-Jun (lo), and CREB (11). Another essential DNA element, NF-ELAMB, is bound by nonidentified nuclear proteins (12). These elements and their spacing have been closely conserved among human, mouse and rabbit E-selectin promoters (13,14). Using a panel of promoter mutants we show here that the spacing of the enhancers is highly critical for promoter activity. This spacing dependence can be explained in two ways. (i) Transcription factors need to bind at the same face of the DNA-helix for optimal proteinprotein interaction and co-operation (ii) The bending angle of the promoter induced by transcription factors is changed by 180' when the elements are moved up-or downstream by halfhelix turns, resulting in altered promoter activity. The E-selectin promoter is indeed extensively bent by its binding factors; TF-IID (15,16), various ATF-members (17), HMG I(Y) (181, and NF-KB (19) all have been shown to bend DNA. Furthermore, we show here that the E-selectin promoter is also bent by factors binding the NF-ELA"l/ATF site and a by previously identified p50/p75 form of NF-KB. In order to assess whether this induced E-selectin promoter bending is associated with transcription activity, we replaced the NF-ELAMVATF site by a series of intrinsically bent DNA fragments that are not known to bind transcription factors.

EXPERIMENTAL PROCEDURES
Cloning Procedures-The E-selectin promoter mutants in Fig. lB are derived from 383-CAT (6), in which the CAT gene is driven by 383 bp of the E-selectin promoter. The entire vector was PCR-amplified using the KB-or EL1-oligonucleotides listed in Table I (first 16 on list). Amplification was in 100 pl, total volume, as in (201, using 25 cycles of 1 min at 94 "C, 1.5 min at 50 "C or 60 "C, and 5 min at 72 "C, with 50 ng of template and 2.5 units of Tuq polymerase (Perkin-Elmer). The products were phenol-treated, precipitated, purified from agarose gels, digested with either MZuI (for the KB series) or NotI, ligated and transformed into bacteria. The entire promoter region of the constructs was sequenced. In accordance with the reported low error rate under these conditions (1-2 x see Ref. 21), we did not encounter any unexpected mutations.
The pBEND2 constructs were made by cloning the annealed oligonucleotides EL1-BENDa and -b, and KB-BENDa and -b into the pBEND2 XbaI-Sal1 sites, using standard procedures (22).
The vectors with intrinsically bent DNA are derived from p383-tet (23,24). This vector has 383-bp E-selectin promoter driving expression of the tetracycline-repressorNP1G fusion protein (25). This vector was PCR-amplified using EL1-mutPCR and KB-Ctrla (see Table I). In the product, the NF-ELAMVATF site has been mutated (A to C, G to T, and vice versa), and a unique MZuI site was created at its 3' flank. The curve

T Q T A A T T T T A A Q C A T Q Q Q A A A Q A C
, '

:
NF-ELAM 1 * ns.   ATA ACG CGT GGA TAT TCC CGG GAA AGT TTT TGG  ATA ACG CGT CTT AAA ATT ACA ATG ATG TCA  GAA ACT CTG TCT C  ATA ACG CGT GGA TAT TCC CAG TTT TTG GAT GCC ATT GGG GAT TTC CTC  ATA ACG CGT CCA TGC TTA AAA TTA CAA TGA TGT CAG  AAA CTC TGT CTC TTG AA  ATA ACG CGT GGA TAT TCC CTT GGA TGC CAT TGG GGA TTT CCT CTT TAC TGG A  ATA ACG CGT CTT TCC CAT GCT TAA AAT TAC AAT GAT GTC AGA AAC TCT GTC TCT T  ATA ACG CGT GGA TAT TCC CAG CAT  . The total length of the probes is 137 bp for NF-ELAM1 and 139 bp for NF-KB. Standard probe and nuclear extract preparation, and band shift assays were as described previously (12,20,26).
Plotting of Intrinsically Curved DNA-Coordinates for the molecules in the plane where bending is maximal were calculated using the CUR-VATURE software (27,281. The E-selectin promoter region corresponds to region -169 to -81, the ds curve oligonucleotides were as in Table I. These coordinates were plotted using ORIGIN 3.0 (MicroCal) software.

Moving the E-selectin NF-ELAM-I IATF or NF-KB II Elements Halfhelix lbrns Upor Downstream Periodically Affects
Promoter Activity-A panel of E-selectin promoter mutants was constructed such that the NF-ELAM-1 or NF-KB I 1 elements were moved units of 5 bp (half of a DNA helix turn) up-or downstream. The resulting promoters (containing 383 bp of the upstream E-selectin sequence) were tested as CAT reporter constructs. Fig. L4 gives an overview of the proximal E-selectin promoter, and the transcription factors known to bind at the various DNA elements. Fig. 1B shows in detail how the promoter sequence was modified for each of the mutants. The construction procedure (see "Experimental Procedures") required the insertion of unique restriction sites at the 5' end of the enhancer elements. This modification did not affect promoter activity (data not shown). Fig. 1C summarizes the results of testing the mutants in both primary endothelial (HU-VEC) and established epithelial (HeLa) cells. In general, moving either DNA element 5 bp (half-helix turns) up-or downstream resulted in a strong reduction in promoter activity, whereas shifts of 10 bp in either direction partly or completely restored IL-1 inducibility. An exception to this rule was the NF-ELAMVATF element; moving it further upstream mainly resulted in a decrease of promoter activity.

All Factors Known to Bind the E-selectin Promoter
Bend DNA-Bending of DNA targets has previously been established for TF-IID, various ATF members, HMG I(Y), and NF-KB. All of these have also been shown to bind the E-selectin promoter (summarized in Fig. L4). We have previously shown that the E-selectin NF-KB I element is in fact a target for two het-    Fig. 2 The values pM and pE represent mobilities of complexes in which the DNA element is situated at the middle or at the end of the probe, respectively. The latter values were obtained by extrapolation. The angles a were calculated from standard curves (32). erodimeric complexes: one is the well known NF-KBl/rel-a heterodimer (nomenclature as in Ref. 291, the second form consists of N F -K B~ plus a 75-kDa polypeptide that is serologically related to rel-a. This latter complex is called NF-KB' (30). We decided to investigate whether NF-KB+ and the multiple ATFlike factors binding at the NF-ELAM-lIATF site bend DNA as well. For this purpose, the NF-ELAM1 and NF-KB elements were separately cloned into vector pBend2 (31), between two direct polylinker repeats (see Fig. 2A). From this vector, probes for band shift experiments can be excised, in which the Eselectin promoter element is located at different positions within the DNA fragment ( Fig. 2A, lower half ). The results of these band shifts are shown in Fig. 2B for the NF-ELAM-VATF element, and in Fig. 2C for the NF-KB I element. The relative mobilities (R,) for the various complexes have been plotted in Fig. 2 0 . Several complexes are obtained using the NF-ELA"1 probe. At least for two of them, called NF-ELAMla and -b (121, phasing results in a clear change in R,. The band marked n.s. is not competed away by the short, unlabeled NF-ELA"1 oligonucleotide (C lanes), and therefore does not represent factors binding the E-selectin promoter. Fig. 2C shows that the "short" E-selectin NF-KB probe (&-a and -b in Table I) binds only NF-KB (first three lanes), whereas the same element embedded in a longer probe binds NF-KB' as well. As described previously, NF-KB+ requires extra 3' DNA for binding in uitro, although the particular nucleotide sequence is not relevant (30). As Fig. 2, B, C, and D, clearly shows, NF-KB, NF-KB', NF-ELAM1-a, and -b all bend DNA. The bending angles were calculated by fitting pdpE in a standard curve for 4.5% polyacrylamide gels (32) and are listed in Table 11.

Transcription factor P d P E
Functional Replacement of the NF-ELAM1 IATF Element by Intrinsically Bent DNA-The NF-ELAMUATF element is very sensitive to phasing (Fig. 2) and is also the target of transcription factors that bend DNA. We wished to test the contribution of factor-assisted DNA bending on promoter activity and asked whether an intrinsically bent DNA fragment could replace the NF-ELAMUATF element (the strategy followed is outlined in Fig. 3A). Since we cannot deduce from the data in Fig. 2 in which plane the DNA-elements are bent by its transcription factors, we introduced a set of well-defined DNA-fragments described previously (33) listed as curve A-E in Table I). These molecules all carry an intrinsic bent of about 100". They successively differ, however, by 72" in the plane in which the bend is situated. The spatial coordinates of these fragments, and of the wild-type E-selectin promoter region were calculated using recently estimated wedge angles for all possible dinucleotides (27,28) and are plotted in Fig. 3B. The plane in which the "curve" fragments were plotted is indicated (the one for curve A is arbitrarily set at 0"). It has been shown that mutating the NF-ELAMUATF element nearly completely inhibits E-selectin promoter activity (12). As we anticipated weak activities for the mutant constructs with bent DNA, we used an indirect CAT reporter readout system with increased sensitivity (24,34) (see "Experimental Procedures" for details). The resulting activities of these promoter constructs are shown in Fig. 3C. As expected, the "empty" construct with the mutated NF-ELAMUATF site showed very low activity and was not IL-1 inducible (Ctrl in Fig. 3C). By contrast, insertion of some of the bent DNA-fragments resulted in a large increase in promoter activity, both basal (curues B and E) and IL-1 inducible (curues B, C, and E ) and reached nearly wild-type activity levels (24). The activities imparted by the bent DNA fragments are highly dependent on their orientation (cf: A-andA+) and plane of bending (cf D and E ) . Since the bent DNA-fragments are repetitive, and highly similar in sequence, it is very unlikely that the transcriptional enhancement seen for some of them results from binding of transcription factors. Moreover, band shift assays using the bent fragments as probes failed to yield specific DNA-protein complexes (data not shown).

DISCUSSION
Regulated promoter activity is the result of subtle interplay among a limited number of promoter elements. Nature seems to rely heavily on the use of combinatorial diversity of DNA elements, while using only a relatively limited set of different enhancers and transcription factors. One of the current challenges is to understand by what mechanisms these interactions take place (the problem was recently reviewed in Ref. 35). One important characteristic by which transcription factors synergistically cooperate is by direct protein-protein interaction. Thus, the triggering inducer of the E-selectin promoter, NF-KB, has been shown to interact with ATF-a, c J u n (10, 36, 371, ATF-2 (381, TF-IID (391, and HMG I(Y) (35). Furthermore, direct protein-protein interactions have been demonstrated between HMG I(Y) and ATF-2 (38). Although these interactions are probably weaker than DNA-protein affinities, they may nevertheless assist in the formation of a functional pre-initiation complex.
A second mechanism by which DNA elements may combinatorially affect transcriptional activity is by structural alteration (bending or kinking) of the promoter (recently reviewed in Ref. 40). Protein-induced bending as a transcriptional switch has been well established in prokaryotes (e.g. see Refs. 41 and 42). An example in eukaryotes is E2F, whose association with the retinoblastoma gene product results both in reduction of the DNA bend it causes and its transcriptional activity (43).
We show here that the spacing of the IL-1 inducible E-selectin promoter elements is highly sensitive to upor downstream shifts in a manner that suggests that the relative helical orientation is important. Although this type of result is usually only taken as evidence that protein-protein interactions are important for promoter activity, we decided to take a closer look at induced bending of the E-selectin promoter. The rationale is that phasing of DNA bends results in a change of the plane in which the angle is located. As we have shown here, factors binding the NF-ELA"1 and NF-KB sites all bend their DNA target. The bending angle determined by phasing analysis is highly dependent on polyacrylamide concentration, temperature, buffer system and gel run length (19). Nevertheless, the value we found for NF-KB ( a = 83") fits well with those reported in Schreck et al. (19). It is also clear that NF-KB' (p50/p75) induces a sharper bend than NF-KB (p50/p65). The altered DNA-bending by the two NF-KB complexes is consistent with plotted in the planes where their bending is maximal; the curve B-E fragments are plotted in planes whose angle with curve A is indicated. "Plus" and "minus" signs refer to orientation of the curve inserts ("plus" is as in Table I). Panel C, transcriptional activity of the E-selectin promoters carrying curved inserts (as shown in A ) .

E-selectin Promoter Bending
p65 being responsible for bending in NF-KB (19). Since p65 is also responsible for NF-KB'S trans-activating potential (44), one may speculate that p50/p75 plays an important role in E-selectin promoter activity. We have previously shown that p50/p75 requires extra DNA contacts 3' of the NF-KB binding site (30). Since the center of bending is the same for NF-KB and NF-KB' (Fig. 2 0 1 , this might mean that NF-KB' partially "wraps" the DNA around the 75-kDa subunit, resulting in increased apparent bending with conservation of its bending center. Like the NF-&-elements, the NF-ELA"1 site has imperfect symmetry (TGACATCA), and this site is a well conserved deviation from the symmetrical CREB site (TGATATCA). It is therefore likely that this site is mainly the target for heterodimeric bZIP factors. In this context it is interesting that ATF-2 homodimers hardly bend DNA, whereas ATF2Ic-Jun heterodimers do (17).
We have shown here that the NF-ELAMUATF element can be functionally replaced by intrinsically bent DNA. The NF-ELAMUATF element has no enhancer activity on its own, even when multimerized, but is able to stimulate neighboring enhancers (12). These findings are consistent with the view that, like YY1(45), the only role of the transcription factors binding the NF-ELAM1 element is to change the spatial configuration of the promoter and the other transcription factors binding to it. The result of this alteration may be "looping" to bring various proteins in closer proximity of each other or of the transcription initiation site (46,471. Looping was recently proposed for the p-interferon promoter ( 3 9 , which is very similar t o the E-selectin promoter in that is also contains cooperating ATF-2 and NF-KB binding sites. Our work supports this view, and emphasizes the importance of transcription factor-induced promoter bending as an important control mechanism for transcriptional activity.