THE JOURNAI. OF E I O ~ X C A L CHEMISTRY 0 1994 by The American Society for Biochemistry and Molecular Bioiogy, Inc.

Vof. 269, No. 2, Issue of January 14, pp. 1493-1500,1994 Printed in U.S.A.

Analyses of P-Thalassemia Mutant DNA Interactions with Erythroid Kriippel-like Factor (EKLF), an Erythroid Cell-specific Transcription Factor*

(Received for publication, June 17, 1993, and in revised form, September 13, 1993)

Waldo C. Feng$, Cherie M. Southwoodl, and James J. Biekerln /I From the ~ p a F t ~ n t s of $Physiology and Biophys~cs and Wiochemistry and SThe ~ r ~ k ~ l e Center for Mo~ecu~ar ~ ~ o l o g y , Mount Sinai School of Medicine, New YO&, New York 10029

We describe functional tests and molecular modeling of erythroid Kriippel-like factor (EKLF) interactions with its DNA binding site. EKLF, a zinc finger-contain- ing, erythroid-specific transcription factor, binds and transactivates from the CACCC element, an evolution- arily conserved DNA sequence present within a large number of erythroid-specific promoters and enhancers. This DNA binding element is the site of naturally occur- ring point mutations that give rise to 8-thalassemia. We have directly tested whether CAC site point mutations (including two of the @-thalassemia mutants) affect EKLF t r a ~ c t i v a t i ~ n and DNA binding function. In vivo analyses demonstrate that EKLF is unable to transacti- vate a reporter plasmid that contains these mutations. In vitro analyses reveal a 40-100-fold decrease in bind- ing affinity for these sites that accounts for the in vivo observations. The homology between the three EKLF and Zif268 zinc fingers and their conserved sequence- specific contacts to their target site allowed us to for- mulate a molecular model of the EKLFICAC site com- plex, based primarily on energy minimization/ refinement of the Zit268/i)NAco-crystal structure. These models suggest that both specific and nonspecific hydro- gen bonding play a critical role in the ability of EKLF to prefer binding to its cognate site. Analysis of sequence- specific contacts by EKLF to its target site within the @-globin promoter verified the residues predicted to be important by the functional and modeling data. To- gether these results demonstrate that EKLF displays a strong discriminatory ability among potential DNA tar- get sites consistent with the P-thalassemia data. They also suggest that lack of EKLF binding to these sites may play a determining role in its phenotype, and they strengthen the evidence in favor of EKLF’s proposed role in erythroid-specific transcriptional activation through the CACCC elements.

Dissecting the intracellular controls involved in establish- ment and maintenance of the erythroid cell is an intense area of study. A primary focus has been on DNA-binding proteins, which, by virtue of the critical DNA sequences to which they bind, may be important in restricting pluripotent progenitor

* This work was supported by funds from the Brookdale Foundation and United States Public Health Service Grants HD23250 and DK46865 (to J. J. BJ. This is publication 138 from The Brookdale Center for Molecular Biology at the Mount Sinai Medical Center. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked “adver- tisement” in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

I/ To whom correspondence should be addressed: The Brookdale Cen- ter for Molecular Biology, Mount Sinai School of Medicine, Box 1126, One Gustave L. Levy PI., New York, NY 10029. Fax: 212-860-9279.

cells to express only those proteins diagnostic of the terminal red cell phenotype (1). A limited number of cell-specific consen- sus motifs (e.g. GATA (2), NF-E2 (31, and CACCC (4, 5)) are present within a large number of proximal erythroid-specific promoters and the more distantly located cy- and @-globin locus control regions. I t is thought that these bind transcription fac- tors, the coordinated presence of which may be crucial for cor- rect tissue-specific expression of downstream target genes (6-9).

Erythroid Kfiippel-like factor (EKLF)’ is a recently identi- fied erythroid-specific transcription factor that binds to and activates transcription from the CACCC site located a t approxi- mately -90 in the murine and human @-globin promoters ( 10). This evolutionarily conserved target site (e.g. Ref. 11) and its variants (including the “GTG” sites) have been shown to be essential elements for appropriate and optimal promoter func- tion of both globin (11-16) and erythroid-specific non-globin (17-19) genes. Some of these sites also bind proteins in vivo only in erythroid cells (15, 19-23). The possibility that EKLF is involved in regulating the expression of these and other red cell-specific genes through this important site makes it of in- terest to characterize the specificity of binding by EKLF.

The predicted protein sequence of EKLF contains three tran- scription factor IIIA-like zinc fingers at its extreme carboxyl end. First described in transcription factor IIIA of Xenopus laevis (24--26), zinc fingers are now recognized as a common and important structural motif found in a wide variety of tran- scription factors and proteins that regulate gene expression (27-30). Each zinc finger comprises approximately 30 amino acids, which form independent folding regions within the DNA binding domain of the protein. The structural information that determines the characteristic anti-parallel P-sheet at the N terminus and the a-helix at the C terminus of a correctly folded zinc finger resides in the conserved sequential arrangement of the invariant cysteines and histidines as well as aromatic and hydrophobic residues in the zinc finger polypeptide (31-34). Thus it is possible to construct novel zinc fingers based solely on their amino acid sequence information if they adhere to the canonical YIFXCX,,4CX~F/YX5LXzHx,fi motif, where X , rep- resents any amino acid.

The modular nature of zinc fingers suggests they may form recognition units whereby each finger reads a particular se- quence of DNA, independent of the context of its surrounding fingers. Therefore, zinc fingers within the DNA binding domain from the same or different proteins can be exchanged, resulting in specific binding to novel DNA sequences. Such zinc finger “swapping” experiments (35-37) suggest that the independent nature of the zinc finger/DNA interaction can be understood

* The abbreviations used are: EKLF, erythroid mppel-like factor; CAT, chloramphe~col acetyltransferase; CCiHH, CysZ-Hisz.

1493

1494 EKLF J P-Thalassemia Mutant DNA Interactions

and applied to novel zinc finger complexes. However, it is not yet possible to accurately predict the target DNA binding site given the primary amino acid sequence of a zinc finger. At- tempts to alter DNA binding specificity by changing specific amino acids of the zinc finger to accompany changes in the DNA sequence have met with varying success (3840). The difficulty is that an accurate prediction of the binding se- quence requires a complete understanding of the molecular details of specificity in zinc finger-DNA recognition. Neverthe- less, certain formulations of the Kriippel family zinc finger- DNA interactions derived from these experiments and from the high resolution crystallographic structure of Zif268 (41) enabled us a priori to restrict the potential EKLF binding site to a limited consensus sequence in the absence of any site-se- lection strategy (10).

The complete EKLF DNA-binding element (5"CCACAC- CCT-3') in the mammalian 13-globin promoter is the site of three point mutations that give rise to p-thalassemia ( p + ) in humans (42-44). This attests to its importance in main- taining appropriate levels of P-globin transcripts in vivo. Additionally, it provides a stringent test for any protein that potentially binds to that site, as each of these mutations should have a deleterious effect on EKLF transactivation and DNA binding.

We describe a series of experiments using the recently iso- lated EKLF protein and its binding site. First, we have directly tested whether EKLF transactivation and DNA binding func- tion is affected by three point mutations within its target site, two of which correspond to the previously described p-thalas- semia mutants. Second, we have tested whether our theoretical molecular models of zinc finger/DNA interactions are able to correctly predict the effect of these three mutations upon EKLF binding. Finally, we have examined the importance of specific nucleotide contacts between EKLF and its binding site within the murine P-globin promoter.

MATERIALS AND METHODS Functional Tests-Cell transfections into CV-1 cells, chloramphenicol

acetyltransferase (CAT) assays, and normalizations relative to cotrans- fected growth hormone were performed as previously described (10). These were camed out using pSG5mKLF effector plasmid and wild type (pCAC-tkCAT) or mutant reporter plasmids. One copy of each mutant oligonucleotide was inserted into the Hind111 site of pBLCAT2 (ptkCAT). These constructions resulted in the following variants (the point mutation is in boldface type (43, 44)): pMUT6-tkCAT (CCA- CATCCT), pMUT7-tkCAT (CCACACGCT), and pMUT8-tkCAT (CCA- CACCGT). All constructs were verified by sequence analysis.

Gel shift assays were performed as previously described using bac- terially derived, affinity-purified glutathione S-transferase/EKLF and radiolabeled double-stranded, wild type CAC-containing oligonucleo- tide (10). Binding reactions (in 25 mM HEPES, pH 7.5, 16 m v KCI, 50 mM NaCI, 2 ZnC12, 0.6 mM p-mercaptoethanol, 8% glycerol) con- tained cold competitor double-stranded oligonucleotides a t the molar excess indicated in the figure. Competitor DNA included the wild type CAC and Zit268 oligonucleotides described previously (10,41) and other CAC sequences mutated at the single sites described above and desig- nated mut6, mut7, and mut8. Shifts were resolved on 4% polyacryl- amide gels.

Methylation interference assays were performed by using a cloned fragment containing the murine p-globin promoter (-168 to +37; Ref. 45) that was end-labeled on the noncoding strand by a Klenow polymerase- mediated fill-in reaction (46) and partially methylated with dimethyl sulfate (47). The gel shift assay with glutathione S-transferaselEKLF protein was as described above except that 1 ng ofthe promoter fragment was used and the shift was resolved on a 3.5% acrylamide gel. After electroblottingin 0.5 x Tris boratelEDTAto NA45 filterpaper(Sch1eicher & Schuell), regions containing the "free" and "bound" fragments were localized via autoradiography, and filters were sliced at those regions. The DNAfragments were then eluted from these slices (as recommended by the manufacturer), precipitated, cleaved with piperidine (47), and resolved on 8% polyacrylamide/urea sequencing gels. Formic acid- cleaved promoter fragment was electrophoresed alongside as a marker.

4 r .

' 0

- 0

EKLF , - + I , - + I I - + I I - +, ,- + I

"_ CAC mut7 mu18 mu16

t h I- o < t " o n w c1 A - < z 0 z

0 EKLF ,- + I I- + , ,- +I , - + I I - + I - CAC mut6 mu17 mute

FIG. 1. In vivo activation of mutant CACCC promoters by EKLF. Cotransfection experiments were performed with 2 pg of re- porter plasmid ptkCAT (- - -), pCAC-tkCAT (CAC), pMUT6-tkCAT (rnut6), pMUT7-tkCAT (rnut7), or pMUT8-tkCAT (rnut8) and 10 pg of either pSG5 (-EKL.F) or pSG5/EKLF (+EKLF) effector plasmid as in- dicated. The autoradiograph of the thin layer plate for one experiment is shown in A, and the normalized results averaged from three experi- ments are shown in B.

The dried gels were then exposed to autoradiography. All chromatograms and gel shifts were quantified by analysis on a

Phosphoimager (Molecular Dynamics). Molecular Modeling-Initial coordinates of the Zit268 complex were

taken from the crystallographic coordinates a t a resolution of 2.1 A (crifZIF; c = crystal, DNA component in italics, protein component in bold). The starting structures of the EKLF DNA and the EKLF protein were constructed with the Quanta (release 3.2) program (Molecular Simulations, Waltham, MA). Strict molecular placement of identical amino acids or base pairs between Zit268 (41) and EKLF were done wherever possible. The Cartesian coordinates of these atoms in the Zit268 complex were kept when forming the novel EKLF-DNA model. If exact replacements with equivalent atoms were not possible, only the main-chain Q carbon atoms of the Zif268 zinc fingers and the phosphate and sugar moieties of the Zit268 DNA were used as a structural tem- plate on which to construct the EKLF-DNA complex. The initial orien- tation of these EKLF protein side chains were derived from a standard residue topology file available in the AMBER4.0 program (48). Un- equivalent DNA base pairs of Zit268 were replaced with EKLF B-con- formation DNA base pairs.

Structural refinement of the model complexes was performed with an energy minimization algorithm (steepest descent) available in A" BER4.0. Energy minimization seeks to lower the potential energy of the system by relieving unfavorable molecular contacts or enhancing favor- able ones through displacements of atoms. Minimization simulta- neously takes account of all the molecular interactions in the protein- DNA complex; however, it does not allow for the structure to overcome local energy bamers. Energy minimization of each modeled complex was halted once convergence was established. Convergence was defined as a change in the potential energy of less than 1.0 kcaVmol in 100 steps. A distance-dependent dielectric constant was used to approxi- mate the electrostatic screening effect of the solvent.

RESULTS

Mutated CAC Elements Are Not Sites for Dansactivation in Vivo-We tested the effects of three point mutations within the

EKLFI S-Thalassemia Mutant DNA Interactions 1495

A 1 2 3 4 5 6 7 1 2 3 4 5 6 7

Yuuutll

analysis of EKLF binding to mutant FIG. 2. Competitive gel retardation

CACCC oligonucleotides in vitro. Gel shift assays were performed as described under “Materials and Methods” using ra- diolabeled wild type CAC oligonucleotide and non-radioactive competitor oligo- nucleotides a t 0-, 20-, 50-, loo-, 200-, and 400-fold molar excess (lanes 2-7, respec- tively). Lane 1 contained no protein added to the incubation. Competitor oligonucleo- tides were either wild type CAC, Zif268, mut6, mut7, or mut8 as indicated. The autoradiograph of the gel resulting from all the assays is shown in A, and its quan- titation is shown in B, defining the amount of shift seen without any competi- tor as 100%. The point a t which each of these curves cross the “50% signal re- maining“ line was used as the basis for estimating the competitive ability of each oligonucleotide for binding to EKLF rela- tive to wild type CAC.

COMPETITOR: CAC Zif268

1 2 3 4 - 5 6 7 1 2 3 4 5 6 7 1 2 3 4 5 6 7 - c” !? .

COMPETITOR: mut6 mu17

B mut8

i l

- * * *

wild type: CCA CAC CCT

Zif268

rnut6: CCA CAT CCT

rnut7: CCA CAC GCT

mu@: CCA CAC CGT

1 10 100 1000

-fold excess competitor

EKLF DNA binding site. mut6 and mut7 were chosen because they correspond to the locations of known non-deletion P-thal- assemia mutations mapping to the human CACCC site (43,441. Oligonucleotides containing each of the mutated sites were in- serted into the ptkCAT reporter plasmid at the same location and orientation as that used previously to generate pCAC- tkCAT (10). Cotransfection with pSG5/EKLF into CV-1 cells (Fig. 1) indicate that each of these point mutations has no effect on reporter activity, as CAT levels are no higher than that seen in the absence of any CAC-related site. As previously observed (lo), the wild type CAC site requires cotransfected EKLF for activation. These results indicate that each of the point mu- tants within the EKLF DNA-binding site disrupts the ability of EKLF to transactivate a reporter gene in vivo.

EKLF Protein Has a Dramatically Decreased Afinity for Mu- tated CAC Sites in Vitro-The most likely explanation for the ineffective transactivation demonstrated in Fig. 1 is that the mutated CAC sites are not efficient EKLF DNA binding ele- ments. To test this, and at the same time to ascertain the relative competitive affinity of each of the mutated elements compared to the wild type site, competitive binding assays were performed (see ”Materials and Methods”). Under the conditions of the assay, an 11-fold excess of wild type CACCC sites was required to compete the EKLF-CAC shift by 50% (Fig. 2). This value provided the basis against which all other oligonucleo- tides were compared. A Zif268 site-containing oligonucleotide,

one that does not bind EKLF well (10) yet binds to another member of the Kriippel family of zinc finger proteins, competed at an approximately 8-fold lower level than wild type CACCC site. Each of the mutated CAC site oligonucleotides were more drastically impaired than Zif268 in their ability to compete the EKLF-CAC binding (Fig. 2). mut6 was the “strongest” of the three, although it competed at an approximately 40-fold lower level than wild type. mut7 and mut8 were a t least 100-fold less effective than wild type. These experiments indicate that EKLF is not able to transactivate the mutant site-containing reporter plasmids due to its inability to bind efficiently to these altered target DNA binding sites. They also exhibit the drastic effect these single point mutations can have on DNA binding and a partial extent of the specificity required by EKLF to bind to its target site.

Molecular Modeling: Rationale and Basis for Construction of an EKLF-DNA Model-Molecular models of protein-DNA in- teractions were tested for their ability to predict the relative affinity of EKLF for each of the mutated sites. The abundance of the classical Cys2-His2 (CC/HH) zinc fingers and their con- sistency in supersecondary structure among a variety of orga- nisms highlight their importance as a structural motif for spe- cific DNA interactions. Structural comparison of helix-turn- helix DNA-binding proteins from numerous sources (e.g. ACro, A434, Escherichia coli CAP, etc.) revealed common features shared in the protein-DNA complex (49).

1496 EKLFI P-Thalassemia Mutant DNA Interactions

' I

FIG. 3. Molecular modeling of protein-DNA interactions. Fin- gers 1,2, and 3 p m e d h m the N terminus at the bottom lefi of each figure. Zinc atoms are depicted as orange spheres. A, interaction of zin68 protein (blue ribbon) with ita target DNA site (purple backbone) through specific amino acid side chains (yellow) based on the co-crystal structure (41). B, interaction of EKLF protein (green ribbon) with the CACCC element (white backbone) through specific amino acid side chains (yellow) after performing the energy minimization procedure described in the text.

Recurrent properties of zinc finger-DNA contacts have also been observed in the high resolution co-crystal of Zif268. Analy- ses of the complex suggested that each finger of the three zinc- finger protein makes sequence-specific contacts to three DNA bases by three positionally defined residues of the recognition helix, with a predilection for guanines by the arginines. The orientation of the zinc fingers relative to the DNA is such that the N-terminal finger binds to the triplet base pairs near the 3' end of the DNA. The zinc fingers interact with the DNA in the major groove, and their spatial orientation relative to the DNA is conserved so that the position of each finger is related to the next by simple transformations and translations (see Fig. 3A). Thus the consistency of zinc finger-DNA interactions observed from the co-crystal enables one to formulate simple guidelines of sequence-specific recognition, so that inspection of novel, three CC/HH zinc finger sequences may predict their cognate DNA binding site.

The amino acid sequences of the three EKLF zinc fingers indicate that they belong to the classical CC/HH family of zinc

fingers (Fig. 4; also see Ref. 10). Except for the first residue of finger 1 (histidine instead of an aromatic residue), all three fingers contain the consensus sequence YiFXCX,,CX,F/ YX&X2HX3H (50). Data from a variety of zinc finger struc- tures that contain this motif are fully consistent with the "clas- sical" PPa zinc finger folding properties of xfin31 (511, mKr2

A comparison of the Zif268 and EKLF primary structures indicates that they share similar features (Fig. 4). Both Zif268 and EKLF contain three zinc fingers, the binding regions of which comprise 9 base pairs, 3'-GCG GGT GCG-5' and 3'-GGT GTG GGA-5', respectively. In addition, each zinc finger is sepa- rated from the successive zinc finger by the consensus linker sequence TGxR/B: (57). Thus, if the conserved interactions for sequence-specific contacts between each zinc finger and its re- spective triplet base pairs observed in the Zit268 structure (arginine and histidine for guanine bases) are applied to the novel EKLF protein, then a predicted binding site for EKLF would be 3'-nGn GnG GGn-5' on the G-rich strand or 5'-nCn CnC CCn-3' on the C-rich strand (Fig. 4). The predicted binding sequence of EKLF based on the Zif268 co-crystal data and the known binding sequence, 5'-CCA CAC CCT-3', are remarkably consistent, suggesting that the type of sequence-specific con- tacts found in Zit268 may also be applied to EKLF. Thus, the similarity in amino acid sequence and apparent conservation of base contacts between Zit268 and EKLF gave us confidence for construction of an initial EKLF-DNAcomplex predicated on the high resolution co-crystal data of Zif268.

The orientation of the three EKLF zinc fingers relative to the DNA in the initial molecular structure is described in Fig. 4 with the XYZ residues in the recognition helix of each zinc finger facing the major groove of the DNA. Because the detailed analysis of the sequence-specific arginine and histidine con- tacts with the guanine bases in Zit268 indicated a striking consistency in the molecular geometry in which the gua- nidinium or imidazole side chain interacts with the guanine bases among all the Zif268 zinc fingers, we modified the XYZ histidinelarginine side-chain dihedral angles of EKLF in order to form Zif268-like contacts. Although both aspartates in fin- gers 1 and 3 do not make any significant base contacts in Zif268, they were proposed to potentiate the arginine-guanine contacts. Therefore, the aspartates in fmgers 2 and 3 of EKLF were positioned in a manner analogous to the Zit268 aspar- tates.

Results of the Energy Minimization Procedur+The EKLF- DNA model was constructed as described under "Materials and Methods." Strict molecular replacement of Zif268 amino acids and binding site nucleotides with EKLF amino acids and CAC site nucleotides resulted in unfavorable contacts and steric clashes between overlapping atoms because of their dissimilar- ity. Therefore, the EKLF-CAC DNA model was further refined with an energy minimization procedure (see below) to relieve energetically unfavorable contacts and to form potentially fa- vorable (though not necessarily observed in the Zif268 complex) contacts between the EKLF zinc fingers and CAC DNA, result- ing in the structure shown in Fig. 3B. The similarity of the structures shown in Fig. 3 (A and B ) reveals that, even after energy minimization, there are no low resolution changes in the EKLF zinc finger/CAC DNA structure compared to the Zif268 zinc fingermif DNA structure.

However, a high resolution analysis reveals significant dif- ferences in these structures. Table I quantitates the effects of the energy minimization process. As expected, the root mean square deviation of the energetically minimized Zif268-DNA structure (mzi . f zIF; m = minimized, DNA component is in ital- ics, protein component is in bold) compared to the co-crystal structure ( c z i m ; c = crystal) had small protein and DNA

(52), ADRl (531, SWI5 (541, ZFY (55), and MBP-1 (56).

1497

DNA site specificity of Zif268 and FIG. 4. Similarity of sequence and

EKLF zinc fingers. A, the amino acid sequences of each of the three zinc fingers from Zif268 and EKLJ? are aligned. Bold residues indicate the CCMH canonical se- quence motie boxed residues indicate the XYZ amino acids important for base se- quence recognition (59). X is +2 residues from the conserved aromatic residue in the second f3 strand, Y is -1 residue from the conserved leucine in the a helix, and 2 is -1 residue from the first invariant his- tidine. B , DNA site recognition as deter- mined from the XYZ residues and the ac- tual (Zif268) or predicted (EKLF) target sequence.

A m G E R 1

A H T C G H E G C G K S Y S E L R T E T G E K (EKLF)

P Y A C P V E S C D R R P S E I R I E T G Q K ( Z I F 2 6 8 )

FINGER 2 P Y A C S W D G C D W R T A E Y R K E T G H R (EKLF)

P T Q C R I C M R N ~ S H I R T E T G E K ( Z I F ~ ~ ~ )

FINGERS P ~ C C G L C G L C P R A P S ~ S D~ L A~ E M K R E (EKLF)

P t A C D I C G R K P A S D E R K E T K I H f Z I F 2 6 8 )

I3 ZIF268

Finger 1 2 3 P o s i t i o n s XYZ XYZ XYZ Residues RER RHT RER DNA GCG GGT GCG (3'" 5 ' )

TASLE I Comparison of crystallographic fc) and energetically minimized (m)

DNAlprotein structures Energy minimization was performed as described under "Materials

and Methods." The root mean square deviation of the protein backbone or the DNA backbone is compared before and after the refinement procedure. The DNA component is italicized, and the protein component is in bold. Analogously, the protein backbone numerical comparison in the table is in bold, and the DNA phosphate backbone numerical com- parison in the table is italicized.

Root mean square com arison of protein and DNA ba&ne (A)

aim mzZ#ZIF mzzflELF mCACEKLF

cziflIF 0.30 1.98 mzifAF 0.49 1.98

1.92 188

mziflEawF 1.78 0.97 mCACEKLF 1.66

1.65 1.57 1.29

structural deviations (0.30 and 0.49 A, respectively), demon- strating that the minimization procedure did not appreciably alter the known favorable structure. However, the protein and DNA of both mziflEKLF and mCACEKLF underwent signifi- cant structural changes during the refinement procedure (both complexes should be compared to czifZJP, since the ZiE268 co- crystal was the initial conformation of the modeled structures). Refinement of the molecular models through energy minimiza- tion maintained an energetically favorable complex yet allowed molecularly significant structural deviations from an initially unfavorable complex.

To further quantitate these structural changes and relate them to the binding studies, we examined the number and source (specific and nonspecific) of hydrogen bonds found be- tween the zinc-fingers and DNA in each complex. Hydrogen bonds are ideal sources for sequence-specific recognition, be- cause they form strong non-covalent intermolecular interac- tions in addition to being distance and vectorially dependent (58). Fig. 5 describes the number and source of hydrogen bonds determined &om the structural analysis of the three com- plexes: mzifZJP, mzimKLF', and mCACEKLF'. Hydrogen bonds were defined by two criteria: the distance separating the hydrogen atom (H) to the acceptor (A) heavy atom (1.7-2.2 A), and the donor-H---A angle (+60°, 0" being colinear). Hydrogen

EKLF Finger 1 2 3 P o s i t i o n s XYZ XYZ XYZ Residues KHA RER RHL P r e d i c t i o n nGn GnG GGn (3'" 5 ' )

bonds from the side chains of the proteins to the DNA base pairs were designated "specific," while other hydrogen bonds to the DNA (e.g. phosphates) were designated as "nonspecific." While all three complexes had similar numbers of specific hy- drogen bonds, only mziflIF and mCACEKLF had the largest number of total hydrogen bonds (21 and 24, respectively). The greater number of total hydrogen bonds correlate with the ex- perimental observations, which indicate that Zif268 and EKLF bind to their respective cognate sequences with high amnity, while the "hybrid" complex fmziflWL&" did not produce a sig- nificant gel shift (10). Thus, the analyses of number and type of hydrogen bonds derived from the modeling provided a basis for evaluation of the mutant mut6, mut7, and mutt4 promoters.

Analysis of the Thalassemia Mutants-Fig. 6 compares the hydrogen bond analysis of mutants mmutGEKLF, mnutREKLF, and mmut8EKLF complexes to that of the com- plexes previously described. Each of the mutant complexes were constructed and refined in a manner identical to that used with the previous models. The total number of hydrogen bonds in each of the EKLFDNA mutants was less than in the m z i - mKLF complex. Thus an initial prediction of the mutant pro- moters' affiity for EKLF based on this data suggests that it would be more similar to Zif268 DNA than to CAC DNA. Fur- thermore, mmut6EKI.F' formed twice as many specific hydro- gen bonds as did either mmut~EKLF' or ~ u t 8 E K L F , sug- gesting that its binding f f i i t y may be slightly higher (although still quite deficient) than the other two mutant pro- moters. DNA affinities based solely on an analyses of the hy- drogen bonds between EKLF and various DNA sites correlate well with the experimental evidence shown in Figs. 1 and 2 (i.e. binding by EKLF to all the non-wild type sites tested is pre- dicted to be significantly inferior).

EKLF Forms Close Contacts to Nucleotides in the Natural CACCC Element-The preceding functional and modeling studies clearly implicate the guanine residues on the "G-rich" strand of the CAC site as being critical for high affinity inter- action with EKLF'. We directly assessed this by using a meth- ylation interference assay to probe EKLFs interaction with a DNA fragment encompassing the murine $3-globin promoter. EKLF formed a doublet complex with the partially methylated

1498 EKLFI p-Thalassemia Mutant DNA Interactions

bonds formed after molecular model- FIG. 5. Quantitation of hydrogen

ing of protein-DNA complexes. Spe- cific, nonspecific, and total hydrogen bonds were estimated from the molecular modeling (energy minimization) of Zif268 and EKLF protein andor Zif268 and CAC DNAcomplexes (m = minimized, the DNA component is italicized, and the protein component is boldface).

fn P z 0 m

CE W m 5 z

301 25-

20-

15-

10-

5-

0"

Speclflc

"1 Total

4

no. 6. Quantitation of hydrogen bonds formed after molecu- lar modeling of E M F protein-mutant CAC DNA complexes. Spe- cific, nonspecific, and total hydrogen bonds were estimated from the molecular modeling (energy minimization) of Zif268 and EKLF protein andor Zif268, CAC, or mutant DNA complexes (m = minimized, the DNA component is italicized, and the protein component is boldface).

promoter fragment as expected from the gel shift data with the CAC oligonucleotide (data not shown). The results from cleav- age reactions of the separated free and bound fragments (Fig. 7) indicate that methylation of the G residues solely within the CACCC element interfere with EKLF proteinDNA interac- tions. In addition to directly verifying the importance of these residues for optimal binding, the results are concordant with the in vivo footprinting data, which indicate these same resi- dues in the murine P-globin promoter are protected from meth- ylation (22).

DISCUSSION

EKLF is an erythroid-specific DNA-binding protein that ac- tivates transcription through its interaction with the CACCC element. EKLF is proposed to be intimately involved in gen- eration of erythroid-gene specificity (lo), as the CACCC se- quence and its derivatives are required for optimal transcrip- tion of a large number of erythroid-specific promoters (see Introduction). The CACCC element in the mammalian adult P-globin gene is the site of naturally occurring point mutations that give rise to P-thalassemias ( 4 2 4 ) . In addition to attest-

NonSpecific

Total

I.@@ *eb *" 4 so

GC AT TA TA GC GC CG CG CG TA j e GC

TA 0 GC

f TA AT CG GC GC AT

TA AT

CG AT GC GC *

Frc. 7. Methylation interference m y of EgLF binding to the murine p-globin promoter. A gel shift assay was used to separate the EKLFDNA complex (bound) from free DNA (free) after incubation of EKLF with a partially methylated murine P-globin promoter fragment that was radiolabeled on the non-coding strand (*). The autoradiograph of the sequencing gel resulting from chemical cleavage at guanine resi- dues of bound and free DNA is shown. The marker lane was from chemical cleavage of the promoter fragment at guanine and adenosine. Part of the P-globin promoter sequence is shown and indicates the CAC site (boxed) and the guanine sites which, when methylated, interfere with protein binding (0).

ing to the importance of this site for adult P-globin gene tran- scription, these results imply that the phenotype results from a DNA binding site with reduced affinity for an essential, acti- vating transcription factor. We used this prediction as a strin- gent test of EKLF's proposed role in activation of this gene in vivo. In addition, we used molecular models and methylation

EKLFI P-Thalassemia Mutant DNA Interactions 1499

interference studies to predict and explain how such appar- ently subtle point mutations have a deleterious effect on EKLF DNA binding.

Our in vivo and in vitro analyses demonstrate that EKLF is unable to transactivate the mutated CAC sites and that this is due to their drastically lowered affinity for EKLF. These re- sults, together with the low levels of adult P-globin seen in naturally occurring P-thalassemias that map to the same site, imply that lack of binding by human EKLF may play a deter- mining role in the phenotype of a subset of nondeletion p-thal- assemias.

Although each of the point mutations appear subtle, they disrupt specific contacts between guanine (on the G-rich strand of the CAC site) and arginine or histidine (located at the critical "XYZ" positions (59)) in fingers 2 and 3 of EKLF (10). None- theless, one might not have anticipated such a drastic effect on binding affinity; for example, the Zit268 sequence (CGC- CCACGC) is more divergent and thus potentially more disrup- tive. However, our molecular modeling of the EKLF-CAC site revealed that there are other interactions besides hydrogen bonding at crucial XYZ positions that play a large role in the overall affinity of EKLF for its target site. These energetically refined models queried either the possibility of maintaining Zif268-like protein-DNA interactions for the EKLF complex or the possibility that the EKLF complex may contain novel pro- tein-DNA interactions. Although these two zinc fingerDNA structures are similar at a relatively low resolution, our results suggest that, at a higher resolution, the EKLFDNA complex may be distinct from the Zif268-like structure and yet can still acquire a greater number of hydrogen bonds. Alteration of a particular zinc finger-DNA contact resulted in the side chain of the disrupted residue seeking andor interfering with other sites of interaction. Thus, the entire hydrogen bonding profile between the protein and the DNA, and hence its binding affin- ity, was altered from a single base pair change, while the com- plex as a whole maintained the overall zinc finger and B-con- formation structure. Unlike mCACEKLF', the mutant complexes were unable to attain the critical number of hydro- gen bonds required for high affinity interaction with DNA.

The molecular model used in this study of the EKLF-DNA complexes was constructed with assumptions derived from fin- ger swapping experiments and from structural data of zinc finger-DNA complexes. This model accurately predicted that each of the CAC site mutants would be more drastically im- paired than Zif268 in their binding to EKLF. More importantly, these models provided a means to precisely investigate the nature of DNA specificity in EKLF.

With one exception, the CAC site mutants described in this work do not overlap those previously tested for transcription of a linked sequence in transfected erythroid (12, 45,60, 61) and non-erythroid (45, 62, 63) cells. These earlier studies utilized an otherwise intact P-globin promoter in the absence of any cotransfected activator, although the analyses in non-erythroid cells also required the presence of a viral enhancer in cis to boost transcription levels (45,62,63). These experiments dem- onstrated that the CAC site was required for maximal tran- scription of its linked P-globin gene. A mutant corresponding to our mut6 decreased linked transcription to -30% of wild type levels in MEL cells (61). However, the activity of the P-globin promoter in non-erythroid cells (45, 62, 63), in addition to in vitro analyses that demonstrate multiple proteins, including Spl (and possibly related proteins (64)), are able to bind the CAC site (4,5,61,65,66), have clouded the question of whether the CAC site binds an erythroid-specific factor (see Ref. 10).

We have previously argued that EKLF is likely to be the CACCC-binding protein responsible for transcription from that element in erythroid cells (10). The present studies reveal that

the EKLF DNA binding module exhibits a strong discrimina- tory ability among potential binding sites consistent with, and as required by, the p-thalassemia data. Since the CAC sites do not form a homogeneous group, especially when related GTG sites are included, it will be important to evaluate these other sites for their ability to bind EKLF. The availability of models that accurately gauge the affinity of these interactions will clearly aid this process. These molecular analyses of EKLF, along with direct tests of its biological function (e.g. by gene disruption), will further resolve whether EKLF is the protein specifically responsible for red cell-specific transcriptional ac- tivation through erythroid CACCC elements.

Acknowledgments-We thank Tina Lopingco, Yasodha Natkunam, and Ira Miller for a critical reading of the manuscript.

REFERENCES

2. Evans, T., Reitman, M., and Felsenfeld, G. (1988) Proc. Natl. Acad. Sci. 1. Orkin, S . H. (1990) Cell 63,665-672

U. S. A.S,59765980 3. Mignotte, V., Wall, L., deBoer, E., Grosveld, F., and Romeo, P. H. (1989) Nucleic

Acids Res. 17,37-54 4. Mantovani, R., Malgaretti, N., Nicolis, S. , Gigloni, B., Comi, P., Cappellini, N.,

Bertero, M. T., Calgaris-Cappio, F., and Ottolenghi, S . (1988) Nucleic Acids Res. 16,429%4313

5. Jackson, P. D., Evans, T., Nickol, J. M., and Felsenfeld, G. (1989) Genes & Deu. 3,1860-1873

6. Evans, T., and Felsenfeld, G. (1989) Cell 68, 877-885 7. Perkins, N. D., Nicolas, R. H., Plumb, M. A,, and Goodwin, G. H. (1989) Nucleic

8. %ai, S.-F., Martin, D. I. K., Zen, L. I., DAndrea, A. D., Wong, G. G., and Orkin,

9. Andrews, N. C., Erdjument-Bromage, H., Davidson, M. B., Tempst, P., and

A c i d s Res. 17, 1299-1314

S . H. (1989) Nature 339,44fi451

Orkin, S . H. (1993) Nature 362.722-728 10. Miller, I . J., and Bieker, J. J. (1993) Mol. Cell. Biol. 13, 27762786 11. Gumucio, D. L., Heilstedt-Williamson, H., Gray, T. A., Tarle, S . A,, Shelton, D.

A., Tagle, D. A., Slightom, J. L., Goodman, M., and Collins, F. S . (1992) Mol. Cell. Biol. 12, 4919-4929

12. Cowie, A,, and Myers, R. M. (1988) Mol. Cell. Biol. 8, 312243128 13. Watt, P., Lamb, P., Squire, L., and Proudfoot, N. (1990) Nucleic Acids Res. 18,

14. Gong, Q.-H., Stem, J., and Dean, A. (1991) Mol. Cell. Biol. 11, 2558-2566 15. Zhang, Q., Reddy, P. M., Yu, C. Y., Bastiani, C., Higgs, D., Stamatoyannopoulos,

G., Papayannopoulou, T., and Shen, C. K. J. (1993) Mol. Cell. Biol. 13,

1339-1350

16. Philipsen, S., Pruzina, S. , and Gmsveld, F. (1993) EMBO J. 12, 1077-1085 17. Sun, J. M., Penner, C. G., and Davie, J. R. (1992) Nucleic Acids Res. 20,

18. Frampton, J., Walker, M., Plumb, M., and Hamson, P. R. (1990) Mol. Cell. Biol.

19. %ai, S.-F., Strauss, E., and Orkin, S . H. (1991) Genes & Deu. 6,919-931 20. Strauss, E. C., and Orkin, S . H. (1992) Proc. Natl. Acad. Sci. U. S. A. 89,

21. Ikuta, T., and Kan,Y. W. (1991)Proc. Natl. Acad. Sci. U. S. A. 88,10188-10192 22. Reddy, P. M. S. , and Shen, C. K. J. (1993) Mol. Cell. Biol. 13, 1093-1103 23. Strauss, E. C., Andrews, N. C., Higgs, D. R., and Orkin, S. H. (1992) Mol. Cell.

24. Miller, J., McLaughlin, A. D., and Klug, A. (1985) EMBO J. 4, 1609-1614 25. Rhodes, D., and Klug, A. (1986) Cell 46, 12S132 26. Hanas, J. S., Littell, R. M., Gaskins, C. J., and Zebrowski, R. (1989) Nucleic

27. Klug, A. (1987) I)pnds Biochem Sci. 12,464-469 28. Kaptein, R. (1991) Cum Opin. Struct. Biol. 1,6%70 29. Jacobs, G. (1992) EMBO J. 11, 4507-4517 30. Vallee, B. L., Coleman, J. E., and Auld, D. S . (1991) Proc. Natl. Acad. Sci.

31. Diakun, G. P., Fairall, L., and Hug, A. (1986) Nature 324,698-703

33. Parraga, G., Horvath, S. , Hood, L., Young, E. T., and Klevit, R. (1990) Proc. 32. Berg, J. M. (1989)J. Am. Chem. Soc. 111,3759-3761

34. Mortishire-Smith, R. J., Lee, M. S. , Bolinger, L., and Wright, P. E. (1992) FEBS

35. Nardelli, J., Gibson, T. J., Vesque, C., and Chamay, P. (1991) Nature 349,

36. Thiesen, H.J., and Bach, C. (1991) FEBS Lett. 289, 23-26 37. Thiesen, H.J., and Schroder, B. (1991) Biochem. Biophys. Res. Commun. 176,

38. Desjarlais, J. R., and Berg, J. M. (1992) Proc. Natl. Acad. Sci. U. S. A, 89,

39. Desjarlais, J. R., and Berg, J. (1992)Proteins Struct. Fumt. Genet. 12,101-104 40. Berg, J. M. (1992) Proc. Natl. Acad. Sei. U. S. A. 89, 11109-11110 41. Pavletich, N. P., and Pabo, C. 0. (1991) Science 262,80%816 42. Orkin, S . H., Kazazian, H. H., Jr., Antonarakis, S . E., Goff, S . C., Boehm, C. D.,

Sexton, J. P., Waber, P. G., andGiardina, P. J. V. (1982)Nature296,627431 43. Orkin, S . H., Antonarakis, S . E., and Kazazian, H. H., Jr. (1984) J. Biol. Chem.

269,8679-8681

229a2308

6385-6392

10,3838-3842

58Ok5813

Biol. 12, 2135-2142

Acids Res. 17,9861-9870

U. S. A. 88, 999-1003

Natl. Acad. Sci. U. S. A. 87. 137-141

Lett. 296, 11-15

175-178

33-38

7345-7349

1500 EKLFI P-Thalassemia Mutant DNA Interactions 44.

45. 46.

47. 48.

49. 50. 51.

52.

53.

54.

Kulozik, A. E., Bellan-Koch, A,, Bail, S., & h e , E., and Kleihauer, E. (1991) 55. Kochoyan, M., H a d , T., Nguyen, D,, D&I, c,, ~eutmann, H,, and ~ e j ~ , M, Blood 77,2054-2058

Charnay, P., Mellon, P., and Maniatis, T. (1985) Mol. Cell. Bwl. 5, 1498-1511 56. Omichinski, J. G., Clore, G. M., Robien, M.. Sakaguchi, K, Appella, E., and Berger, S. L., and Kimmel, A R. (1987) Guide to Molecular Cloning lbchnques, Gronenborn, A. (1992) Biochemistry 31, 39073917

Academic Press, San Diego 57. Knochel, W., Poting, A,, Koster, M., Baradi, R. E., Nietfeld, W., Bouwmeester, M m m , A M., and Gilbert, W. (1980) Methods Enzymol. 66,499660 Pearlman. D.A.. Case. D.A.. Caldwell. J. C.. Siebel. G. L.. Sin&. U. c.. weher. 58. Baker, E. N., and Hubbmd, R. E. (1984) Prog. Biophys. Mol. Biol. 44,

(1991) Eioehemistzy SO, 33714386

T., and Pieler, T. (1989) h. Natl. Acad. Sci. U. S. A. 86, 60974100 " "_ , ,

P., and'Kollman, P.' (1991) University of California, San Francisco Pabo, C. O., and Sauer, R. T. (1992)Annu. Reu. Biochem. 61, 1053-1088 Berg, J. (1988) Proc. Natl. Acad. Sei. U. S. A. 86.99-102 Lee, M. S., Gippert, G. P., Soman, K. V., Case, D. A, and Wright, I? E. (1989)

Carr, M. D., Pastore, A,, Gausepohl, H., Frank, R., and Roesch, P. (1990) Eur.

Klevit, R. E., Herriott. J. R., and Horvath, S. J. (1990) P r o t e i n s Stnrct. Funct.

- I .

Science 246,63&637

J. B k h e m . 1&3,45&461

59. Klevit, R. E. (1991) Science 259, 1367, 1395

61. Hartzog, G. A., and Myera, R. M. (1993) Mol. Cell Biol. 1% 44-56 60. Stuve, L. L., and Meyers, R. M. (1990) Mol. Cell. Biol. 10, 972-981

62. Dierks, P., van Ooyen, A,, Cochran, M. D., Dobkin, C., Reiser, J., and Weiss-

63. Myers, R. M., Tilly, K, and Maniatis, T. (1986) Science 232, 613418 64. Kingsley, C., and Whoto, A (1992) Mol. Cell. Bwl. 12,42514261 65. Philiosen. S.. Talbot. D.. Fraser. P.. and Grosveld. E (1990) EMBO J. 9.2159-

Y8-17Y

mann. C. (1983) Cell 32,695-706

Genet. 7,215-226

179-184 Chem. 266,89074915

. . . I

2:67 . .

Neuhaus, D.. Nakaseko, Y.. Nagai, K., and Klug, A (1990) FEES Lett. 262, 66. Yu, C. Y., Motamed, K, Chen, J., Bailey,A. D., and Chen, C. K. J. (1991) J. Bwl.