THE JOURNAL OF Btowtc.u. CXEMI~Y VoI. 269, No. 5, Issue of February 4, pp. 3616-3622, 1994 Q 1994 by The American Sociee for Biofhemietry and Molecular Biology, Inc. Printed in USA.

Structure of an I ~ ~ ~ n o ~ ~ o ~ ~ i n Fab Fragment Specific for ~p~emstranded DNA*

(Received for ~ublication, September 30, 1993)

Clifford D. Mol$, Alastair K. S, Muirg, Miroslaw Cyglerll, Jeremy S. L e e l l , and Wayne F. Amdemons ** From the wepartment of B i o c ~ e m i s t ~ , Vanderbilt U n ~ v e r s i ~ , N ~ h v ~ l ~ e , %nnessee, 37232-0146, the $Division of B ~ o ~ ~ ~ ~ Sciences, National Research Council, Ottawa, Ontarw KlA OR6, Canada, the ~ ~ o t e & h ~ l ~ Research Inst~tute, National Research Council of Canada, Montreal, Quebec H4P 2R2, Canada, and the /Department of Biochemistry, the University of Saska~chewan, Saskatoon, Saskatchewan S7N OWO, Canada

Triple-stranded DNA of the form poly(Pyr)*poly- (Purppoly(Pyr) (where Pyr represents a pyrimidine, and Pur represents a purine) has become the subject of in- tense research because of its potential use in the control of gene expression and in the development of sequence- specific reagents for cleaving DNA,. In a triplex of this type, the second pyrimidine strand ie in a parallel ori- entation to the purine strand and forms Hoogsteen base pairs with it via the major groove. We describe here the t ~ e e - ~ r n e ~ i o n ~ crystal structure d e ~ ~ t i o n of the ~t igen-bin~ng fragment (Fab) &.om the murine monoclonal antibody Jel 318. Jel 318 is specific for triple-stranded DNA, with a preference for the sequence p0lyld(T~~~C)].poly[d(G.A)1.poly[(d(~~C+~T)l. The stmc- ture has been solved by the molecular replacement method and refined by molecular dynamics to an R value of 0.20 at a resolution of 2.8 A The crystale have cell dimensions that are very s imi la r to those of the pre- viously determined structure of Fab Kol, but the Fab fragments pack within the unit cell in a completely dif- ferent manner. The protein is in an extended conforma- tion, with an elbow angle of 184". The shape and electro- static sdace potential of the antibody combining site suggest a possible model for the remgnition of triplex DNA in which residues of CDR-H2 (where CDR repre- sents complementarity-determining region) make spe- cific contact with the DNA bases in the minor groove of the tripiex.

Triple-stranded nucleic acid structures have become the sub- ject of intensive research and increased scrutiny. Although a structure for a nucleic acid triple helix ~ n s i s t i n g of homopoly- meric purine and pyrimidine pol~bonucleotides was first de- scribed over 30 years ago (1) and synthetic triplexes were ex- tensively studied in the 1960s (2-61, the possible involvement of triplex nucleic acids in biological processes has only recently become apparent. In a landmark study, it was shown that the binding of polypyrimidinic RNA homopolymers can effectively inhibit the template activity of double-stranded DNA in the presence of ~ s c h e r ~ c h ~ a coli RNA polymerase (2). More recent

* This work was supported by an Alberta Heritage Fund for Medical Research students~p (to C. D. M) and fellows~p (to A. K" S. M), Cana-

Research Council Group on Protein Structure and Function grant (to dian Medical Research Council grants (to J. S. L), a Canadian Medical

W. F. A), and Grant DK42502 from the National Institutes of Health. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "adwrtisenaent" in accordance with 18 U.S.C. Section 1734 solely to indicate ?his fact.

** To whom correspondence should be addressed. "el.: 615-322-4336 Fax: 615-343-0704.

work has demonstrated that triplex DNA can form in oligopu- rine-oligopyrimidine mirror repeats under superhelical stress (H-DNA), in which duplex DNA folds back on itself to form a pyrimi~e-p~ne-pyrim~dine triplex, with the extrusion of an associated s~ngle-stranded polypurine region (7-111. €3-DNA may play a role in the control of gene expression (31). Many potential homopolymeric t r i p l e x - f o ~ n g sequences are found in the 5'-flanking regions of eukaryotic structural genes (121, and these regions are sensitive to the action of single strand- specific nucleases both in vitro and in vivo (13-16).

Although a detailed single crystal x-ray structure of the triple helix has yet to be determined, the available chemical and physical evidence (1-2,17-25) indicates that triplex DNA is formed by the second pyrimidine strand occupying the major groove of a duplex DNA that is in the A helix form. This poly- pyrimidine strand is in a parallel orientation to the polypurine strand and forms Hoogsteen base pairs with it, giving rise to T-A-T and C.G*C+ base triads (Fig. 1). The specificity of the Hoogsteen hydrogen bonding interac-

tions has been exploited in the design of t r ip lex- fo~~ng se- quences that have been used as artificial restriction enzymes and repressors of ans script ion (26-30). Since the formation of the C.G*C+ triplet requires protonation of the cytosines on the third (Hoogsteen) strand, the generation of triplex is facilitated by low pH (8). Stable triplex can be formed at neutral pH, however, by incorporation of S-methyleytosine in the polypy- rimidine strands (31) or by micromolar concentrations of poly- amines (32).

In this report, we describe the t ~ - ~ e ~ s i o n a l s t r u c t ~ of the first triple-stranded DNA-binding protein, the antigen- binding fragment (Fab) of the murine monoclonal antibody Jel 318 (IgG2b, )o. Jel318 was prepared by i m m ~ i ~ a t i o n with the triplex formed by the sequence poly[d(T.m6C)l.poly[d- (G~A)l.poly[d(T.m6C>1 and has been shown by immuno~uores- cent microscopy to bind to triplex DNA in eukaryotic chromo- somes (33,371. The overall structure and charge ~ s t r i b u ~ o n of the antibody combining site differ from those observed in the structures of Fab fragments that bind single-stranded (34,35) and doubIe-stranded (36) DNAs. A model is proposed for the recognition of triplex DNA by Jel318. E ~ r i m e n ~ to test and improve this model can now be attempted.

~T~~ AND METHODS Data Collection-The production of Jel318 i ~ u n o g l o b ~ i n h m the

hybridoma cell line and the c h a ~ ~ r i z a t i o n of its specificity for triple- stranded DNAf33), as well as the purification and crystallization of the Fab fragments (38) and the dete-a~on of the amino acid sequence of the variable domains 139), are described elsewhere. The ~ q 5 t a l S were grown by the ~ a n ~ ~ - d ~ p vapor diffusion method (40) from solutions containing 12% polyethylene glycol 8O00, 25 nub NaCl, and 50 m~ Ma-HCl, pH 8.0,

3615

3616 Structure of a Fab Fragment Specific for Diple-stranded DNA

i I / *

H H

CH. O-.H-N y kp ? T H "

$J 'R

R O \

FIG. 1. Structures of base triads T-A-T and CGC' involved in triplex formation.

The data were collected from a single crystal using a multiwire area detector from San Diego Multiwire Systems (41). The data were col- lected and reduced using the University of California a t San Diego program package (42) to produce structure factor amplitudes (F values) from the raw detector images. The data included 97,798 observations of 12,656 unique reflections, Rmerge = 6.9% (R,,, = ZhL1( I (Z) - Iobs I )/ Zh)l ( I (Z) I )), representing 92% of the observable data to 2.6-A resolution with Z > 2uI. The crystals are sensitive to radiation damage, and the quality of the measured intensities decreases rapidly at resolutions beyond 2.8 A. The quality of the data in the 2.8 to 2.6-A shell was extremely poor, with an unacceptably high merging R value (R,,, = 26.4%). These data, as well as an additional 310 reflections for which FJF, or FJFo exceeded 2.5, were not included in the refinement. For data in the 3.0 to 2.8-A shell, R,, = 17.6%, and the average intensity (IluZ) is 2.8. A Wilson plot (43) of the data yielded an upper estimate of the overall temperature factor ( (B)) of 53 Az. The crystals belong to the space group P212121, with refined unit cell parameters of a = 82.89, b = 139.19, and c = 40.96 A.

Structure Solution-The structure of Jel318 was solved by the mo- lecular replacement method (44). Following our experience with the molecular replacement solutions of the Hed 10 Fab fragment (35) and Fab Jel72 (36), we have used either the VL-VH or CL-CH domain pairs of Fab fragments as separate search models. We have also noted before that, despite a great structural similarity, there is enough variation between individual domain pairs of different Fab fragments to warrant trials with all available models.

Prior to performing the cross-rotation searches, the variable and constant domain pairs of the search models were superimposed on the appropriate domain pairs of Fab Kol(45). The latter molecule was first oriented with its long axis along the x ax is and the elbow bend ax is approximately along they axis. Rotation searches were calculated with a variety of resolution ranges and Patterson integration radii as de- scribed for Jel 72 (36). Of the Fab domain pairs tried, the clearest rotation function results were obtained using the VL-VH domain pair of Fab 5539 (46) and the CL-CH domain pair of Fab Hed 10 (35). The translation vectors for these domain pairs were determined by a corre- lation coefficient search (BRUTE) (47) and confirmed by calculations of the S function (48). Because of continued difficulties in the refinement of the CL-CH domain pair, the rotation angles and translation vectors for both domain pairs were verified using the molecular replacement pro- tocols as implemented in the X-PLOR package (49). These results are listed in Table I.

The correctness of the solution was supported by the self-consistency of the two sets of numbers, and upon the transformation of both domain pairs, their association resembled that of other known Fab fragment structures. The molecule was then divided into the four individual domains (VH, VL, CH, and CL), and their orientations and positions were refined by an interactive local six-dimensional search (47). The corre- lation coefficient between IF, I and I F, I for a model consisting of 3159 atoms and for data in the 4-8-A resolution shell was, at this stage, 0.58, and the R value for data up to 3-A resolution was 0.43.

Structure Refinement-Prior to the determination of the amino acid sequences of the VL and VH domains, preliminary refinement of the structure was camed out using PROLSQ (501, resulting in a decrease in the R value to 0.32 for data within the resolution range of 6.0 to 2.8 A. At this point, the complete amino acid sequence for the variable domain was determined, and the structure of the antibody combining site was modeled (39). These model variable domains and the constant domain

The abbreviations used are: V, domain, variable light chain domain; VH domain, variable heavy chain domain; CL domain, constant light chain domain; CH domain, constant heavy chain domain; CDR, comple- mentarity-determining region.

TABLE I Results of molecular replacement of Jel318

Domain pair

C H G L VH-VL Model used Jel 72 Jel 72 Atoms 1588 1778 Rotation search

Resolution (A) 4.0-10.0 4.0-10.0 Integration radius (A) 5.0-24.0 Peaka

5.0-24.0 4.85 5.07

01 182.2 206.8 0 2 50.2 51.6 0 3 86.0 80.6

Translation search Resolution (A) 4.0-15.0 Peak

4.0-15.0 8.58 8.44

y (A) x (A) 50.4 49.8

z (A) 32.5 31.3 102.4 102.4

a Peak heights are given in standard deviations above background.

pair from Fab Jel72 (36) were used as the starting points for molecular dynamics refinement using the X-PLOR system of programs (49). Rigid body refinement of these four individual domains converged and indi- cated that they were correctly positioned. Calculation of the diffraction energy term (W, = 216,820) was followed by 160 cycles of Powell con- jugate gradient energy minimization and then molecular dynamics re- finement. The simulated annealing was carried out by heating to 3000 K and slow cooling to 100 K with a step size of 25 K.

At this point, the entire structure was inspected with omit maps (51) and manually readjusted with interactive molecular graphics using the program INSIGHT (Biosym Technologies, Inc.) on an Iris Silicon Graph- ics work station. The electron density was clear for most of the variable domain pairs, but was weak and discontinuous for long stretches of surface loops in the constant domain pairs. For the map calculations, contiguous stretches of -10-15 amino acids were omitted from the structure, and energy minimization was performed prior to the calcu- lation of structure fadors. The structure was also inspected with omit maps and W, - F, electron density maps calculated using map coeffi- cients to minimize model bias (52). The residues composing the anti- body complementarity-determining regions (CDRa) were each in turn omitted from the structure for a round of positional refinement and manual refitting. Iterative cycles of molecular dynamics refinement and manual rebuilding resulted in a further decrease in the R value to 0.27.

The CH domain was then rebuilt to correspond to the correct heavy chain isotype (IgGab), and the structure was subjected to another cycle of molecular dynamics refinement. Inspection of Ramachandran plots (53) indicated problem areas of the structure that were manually re- built if necessary. Further cycles of positional refinement and refine- ment of individual isotropic temperature fadors decreased the R value to 0.21 for data in the resolution range of 6.0 to 2.8 A. Since the quality of the electron density was insufficient in the poorly determined regions of the molecule and there was no further improvement in the R value, the refinement was halted at this stage.

The structure presented in this report consists of 3301 non-hydrogen atoms and has an R value of 0.207 for 9342 reflections, with Z > 2.5uZ, representing 86% of the observable reflections for data in the resolution shell of 6.0 to 2.8 A. The root mean square deviation from ideality for the bond lengths is 0.014 A, for the bond angles is 3.9", and for planarity of the peptide bond is 1.8". The error in the coordinates is between 0.25 and 0.35 A as estimated by the method of Luzzati (54). ARamachandran plot (53) of the main chain torsion angles for the heavy and light chains of this structure is presented in Fig. 2. The temperature factor distri- bution, averaged over all atoms for the residues of both chains, is shown in Fig. 3. Although there is some electron density in the final maps that could be interpreted as water molecules, no ordered solvent has been included in the final model. Further refinement of this structure would require the collection of better quality, higher resolution data.

Electrostatic surface potential calculations were performed on this structure using the program DELPHI as implemented in version 2.1.0 of INSIGHT 11. For these calculations, the protein dielectric was set to 2.0, and the solvent dielectric was set to 78.0, with an ionic strength of 0.15 M and a probe radius of 1.7 A. The amount of total buried surface area in the modeled complex with triplex was calculated based on the algorithm of Lee and Richards (631, with a probe radius for the solvent of 1.7 A.

Structure of a Fub Fragment Specific for Diple-stranded DNA 3617

-180 . .

.. ___ "" Y O

fl

-180 -120 -60 0 60 120 $80

CZI FIG. 2. Ramachandran plot of main chain torsion angles (@/W

for heavy and light chains of Fab Jel 318. The positions of the glycine residues are indicated by squares, and the positions of CDR residues that lie outside of the allowed regions of the plot are labeled.

RESULTS AND DISCUSSION Structure Determination-The structure of the antigen-bind-

ing fragment of the triplex-specific monoclonal antibody Jel318 has been determined by molecular replacement using indi- vidual Fab variable and constant domain pairs as models in the rotation and translation searches. When the crystals of Fab Jel 318 were first obtained (38), it was noted that the unit cell dimensions are nearly identical to those of the previously de- termined structure of Fab Kol (45). Both Fab fragments crys- tallize in the orthorhombic space group P212121 and have unit cell dimensions of a = 81.9, b = 139.2, and c = 41.0 A for Jel318 and a = 78.3, b = 138.9, and c = 40.0 A for Fab Kol. It was thus assumed that the two crystals were nearly isomorphous and that the elbow angle (defined as the angle between the pseu- dodyad axes that relate the CH-CL and V H - V ~ domain pairs) of the two Fab fragments would be very similar. For this reason, the model domain pairs used in the molecular replacement calculations were first superimposed on the appropriate do- main pairs of Fab Kol. Thus, all of the search models were related to each other by an elbow bend of - 170°, equivalent to that observed in Fab Kol. Jel318 is, however, in a less extended conformation than Fab Kol, with an elbow bend of only 154", and Jel 318 is not positioned within the crystal lattice in a manner similar to that of Fab Kol. The actual relationship between Fab Kol and Fab Jel318 and their respective crystal lattices is illustrated in Fig. 4. As is apparent in Fig. 4, the packing mode of Jel 318 differs significantly from that of Fab Kol in the relationship between the position of the Fab frag- ment relative to the crystallo~aphic origin. For Fab Kol, the crystallographic a axis passes between the variable and con- stant domain pairs of a single molecule, whereas for Jel 318, this axis passes between the variable domain pair of one Fab fragment and the constant domain pair of a symmetry-related molecule. Thus, the constant domain pair of Jel 318 has ap- proximately the same translation relative to the origin as the variable domain pair of Fab Kol.

The orientation of the two Fab fragments with respect to the crystal axes is also different. The more compact conformation of Jel318 allows the molecule to fit within the unit cell, with the long axis of the Fab fragment approximately parallel to the crystallographic b axis. The more extended conformation of Fab

Kol is accommodated within the crystal lattice by a rotation of the entire molecule such that the long axis of the Fab is inclined toward the crystallographic b axis.

These differences in the orientations and translations of the Fab domain pairs initially complicated the interpretation of the results from the molecular replacement calculations and illus- trate quite clearly the point that crystals with nearly identical unit cell dimensions need not necessarily be isomorphous. We have observed a similar situation in orthorhombic crystals of a lysozyme-peptide chimeric protein, in which the lysozyme also packs differently relative to the origin despite unit cell dimen- sions that are very similar to those of the orthorhombic crystal form of hen egg white lysozyme (55h2

Crystal Packing-The crystal packing of Jel318 and Fab Kol are generally similar in that both molecules pack in a head-to- tail arrangement, with residues in the variable region contact- ing residues of symmetry-related constant domains. For both molecules, these crystal contacts are primarily between resi- dues within the CDRs of the VH domain and residues near the COOH terminus of a CL domain of a symmet~-related mol- ecule. The crystal contacts made by Jel 318 involve residues H31 to H33 in CDR-H1 and residues H95 to H97 in CDR-H3, interacting with the residues near the COOH terminus of the light chain (residues L207 to L210). Residues within CDR-H2 also interact with this portion of the CL domain and with the region of the CH domain that is linked to it via the interchain disulfide between cysteine L214 and cysteine H128. Atoms be- longing to the side chains of residues H50, H52, H53, and H58 are within 4.0 A of residues H128 to H130, whereas CDR-H2 residues H52, H52a, and H53 are also within contact distance of residues L208, L210, and L212, near the COOH terminus of the light chain. There is a potential hydrogen bond between lysine H53 and serine L208, withothe terminal nitrogen atom of the lysine side chain within 2.5 A of the carbonyl oxygen atom of the serine. This interaction can be seen in the 2F,, - F, electron density map shown in Fig. 5. Interestingly, there are no crystal contacts made by the COOH terminus of the heavy chain, which perhaps explains why the crystals of Jel318 could be grown from different isoelectric species of Fab fragments and could also be grown when mixtures of Fab species were present in the crystallization experiments (38).

Structure Description-The structure of Fab Jel318 contains the immunoglobulin fold that has been observed in all other Fab fragment crystal structures (Fig. 6). The elbow angle re- lating the variable domain pair to the constant domain pair is 154" and is approximately in the middle of the range of elbow bends observed for Fab fragments. The rotation and translatitn operations that relate the VL to the VH domain (171", 0.25 A) and the CL to the CH domain (168", 1.8 A) are also similar to those of other known Fab structures.

There is good, general agreement between the model and the electron density in 2 F o - F, electron density maps contoured at the lo level (Fig. 5). As is commonly the case in Fab fragment structures, this correspondence is generally clearer for the light chain than the heavy chain and for the variable domains com- pared to the constant domains, a fact that is reflected in the temperature factor distributions for these domains (Fig. 3). The electron density is clear for the main chain and side chains of most of the residues within the variable domain CDRs. As can be expected with data at this resolution, the fine details of some of the side chain torsion angles are not as well determined, particularly in those regions of the structure that are charac- terized by large temperature factors. The variations of the resi- due-averaged temperature factors correlate closely with the alternating structure of estrands and connecting loops along

H. Patel, unpublished results.

Structure of a Fab Fragment Specific for Diple-stranded DNA

40

20

c

20 I 40 60 80 100 120 140 160 180 200 (27)

Light Chain Residue (Kabat numbering)

H1 HZ H3

20 40 60 80 100 120 140 150

111 111 II I Ill1 I 7 1 2 i 7 5zAEc 82ABc (131-2) (170)

(lsa-al) (181-2) ('074 (2 1 Yo?' Heavy Chain Residue (Kabat numbering}

(64) is used for both chains, with the positions of deletions shown inparentheses and insertions labeled A, B, etc. The positions ofthe antibody CDRS FIG. 3. Plots of mean temperature factor by residue for liiht (A) and heavy (B) chains of Fab &el 518. The numbering of Kabat et al.

are as indicated.

the polypeptide chain. There is weak electron density and, con- sequently, higher temperature factors for the @-strand-connect- ing loops of the constant domains (residues H160 to H165 and H195 to H200 and residues L150 to L155 and L180 to L185). Despite the fact that there are crystal packing interactions in the vicinity of the interchain disulfide, those residues of the heavy chain loop between positions 132 and 140 and of the extreme COOH terminus of the light chain (residues 211-214) that are not specifically involved in crystal contacts also pos- sess large temperature factors and poorly defined electron den- sity. Poor quality electron density for the interchain disulfide bond is commonly observed in Fab fragment crystal structures (56, 57) and is usually attributed to the fact that this disulfide is on the surface of the molecule and may exist in multiple conformations. A few residues possess main chain torsion angles that lie outside of the allowed regions of the Ramach- andran plot (Fig. 2). These residues are in solvent-exposed loops associated with regions of poorly defined electron density

and are, for the most part, confined to the constant domains, such as CL Asp-151 and CH Glu-150 and Am-162. Of the resi- dues that are part of the antibody CDRs, the outliers on the Ramachandran plot include only Leu-H97 and Phe-H105, which lie just outside of the right- and left-handed a-helical regions, respectively. These residues have well-defined density for their main chain and side chain atoms, which appears to confirm their conformations. Similar distortions are not un- common and have been observed in other Fab fragment S t N C -

tures (57-59). The other significant CDR residue outlier is Thr- L50, which is in a region that was initially considered forbidden, with @ = 60" and tlt = -30". A similar conformation for this residue has been observed in a high resolution struc- ture of a recombinant V, domain from the antibody McPC603 (58), and calculations indicate that this region of the CDrV plot is, in fact, favored if there is some flexibility allowed for the atomic bonds (60).

Antibody Combining Region-The surface that Jel 318

Structure of a Fab Fragment Specific for Diple-stranded DNA 3619 presents for interaction with triplex is remarkably flat, punc- tuated by the protrusion of a single hypervariable loop (CDR- H2) away from the main body of the molecule (Fig. 6). In this respect, the combining region of Jel318 is similar to that of the double-stranded DNA-specific Fab fragment Jel72 (36) and is unlike the combining regions of the single-stranded DNA-spe- cific Fab fragments Hed 10 (35) and BV04-01(34). Both of the single-stranded DNA-binding Fab fragments are characterized by combining regions that form deep clefts created by having

A

KOL

T B

A -->

JEL 318 FIG. 4. Crystal packing of Fab Jel318 versus Fab Kol. For both

crystals, the view is approximately down the c axis, with the crystallo- graphic a axis running from left to right. The CXCL domain pairs of both Fab fragments are shown by thick lines, whereas the V , V , do- main pairs are shown by thin lines.

long CDR-L1 and CDR-H2 hypervariable loops. Jel 72, how- ever, forms a much less pronounced cleft since it possesses a shorter CDR-H2 loop and an exceedingly long CDR-HI loop that folds back into the center of the combining region. The relatively flat surface of the Jel318 combining region is created by having a small CDR-L1 loop, shorter by 6 residues than the same loop observed in either Hed 10 or Jel72, and also a short CDR-HI loop. The resulting combining surface is enriched in amino acids with small polar side chains, such as serine knd asparagine. Considering the extremely anionic nature of triple- stranded DNA, it is somewhat surprising that the combining region does not contain more amino acids with basic side chains. The basic amino acids are mostly located in the pro- truding CDR-H2 hypervariable loop, which contains 3 basic residues (Arg-H50, Lys-H52b and Lys-H53), although there is also a single arginine residue (Arg-H98) in CDR-HI and a lysine residue at position L53 in CDR-L2 (Fig. 7).

The main chain conformation of five of the six CDRs con- forms to that previously observed in Fab fragment structures and to the canonical structure models described by Chothia and Lesk (61). Thus, CDR-L1 belongs to the canonical structure class 4, CDR-L3 to the canonical structure class 1, CDR-H2 to the canonical structure class 4, and CDR-L2 and CDR-H1 to the only canonical structure that has been observed for them. No canonical structures for the CDR-H3 region have been de- rived because the length of this loop and its amino acid se- quence are highly variable among Fab fragments of known structure. In Jel318, CDR-H3 occupies a central position in the combining region and interacts with the other hypervariable loops to form the flat combining surface. This loop contains 7 amino acids with the sequence Glu-Leu-Leu-Arg-Ser-Phe-Ala- Tyr from positions H95 to H102. These residues possess main chain conformations of the form Spu"a"Sa"&S.

Several of these residues in CDR-H3 interact with residues in the other CDRa. The side chain of glutamic acid at position H95 forms a salt link with the side chain of Arg-H50 (distance of 2.8 A) at the beginning of the protruding CDR-H2 loop. The side chain of the arginine residue at position H98 is exposed to solvent, but its enitrogen atom is within hydrogen bonding

residues H52c to Ha of CDRHB. The electron density for residues near the COOH terminue of a eymmatry-related CL domain is visible (upper FIG. 5. Stereo representation of - ffe electron deMity map contoured at 1au mustrating representative electron density for

left)

3620 Structure of a Fab Fragment Specific for Diple-stranded DNA

FIG. 6. Stereo diagram of Jel318 Fv fragment illustrating immunoglobulin fold and relative positions of antibody CDRs. The traces of the main chains are approximated by ribbons that are thick (V,) and thin (VL).

B

FIG. 7. Stereo views of antibody combining region. Only the CDRs are illustrated, with the traces of their main chains shown as ribbons. A, view looking down into the combining region; B , view from the L3 side of the combining region.

distance (2.7 A) of the side chain of the first residue in CDR-L2 (aspartic acid L50), whereas the side chain hydroxyl group of serine H99 can also hydrogen bond with the side chain of the framework residue Arg-L46. These interactions serve to pack the CDR-H3 loop within the center of the combining region and illustrate the point that at least some of the charged residues in the combining regions of antibodies may play a role in the stabilization of the structure of the CDRs, rather than a func- tional role in the recognition of antigen. For DNA-binding an- tibodies, an assumption is often made that arginine residues within the CDRs necessarily are involved in recognition of the sugar phosphate backbone of the DNA antigen, whereas they may, in fact, be essential for stabilizing the structure of the hypervariable loops.

Model for Recognition of Poly[d(T~m5C)]~Poly[d(G~A)JPoly- [d(m5C+.T)]-Whereas the possibility that the antibody CDRs will undergo conformational change upon binding triplex can- not be discounted, the shape and distribution of positive charge in the combining region suggest a possible model for triplex recognition by Jel 318. The binding of Jel 318 to triplex has

been investigated by solid-phase radioimmunoassay, and the results show that the antibody has a preference for AT-rich over GC-rich triplexe~.~ The binding model (presented below) pro- poses that this preference for AT-rich triplex is achieved by the antibody making base-specific contacts with the DNA via the minor groove of the DNA.

The triplex DNA used in the modeling of the Fab-DNA com- plex is derived from the structure of poly(dT)ply(dA)-poly(dT) that has been determined from fiber diffraction experiments (20). This triplex consists of one polypurine and one polypy- rimidine strand of opposite polarity, forming an A-type duplex, with a wide and shallow minor groove and a deep and narrow major groove. The second polypyrimidine strand is wound along the major groove of this duplex parallel to the polypurine strand, forming the Hoogsteen base triad T.A.T. This structure presents a problem when trying to visualize antibody binding to triple-stranded DNA. What remains of the major groove in the triplex is too narrow for the antibody loops to penetrate and

J. S. Lee, unpublished results.

Structure of a Fab Fragment Specific for Diple-stranded DNA 3621

RG. 8. Stereo view of Jel318 com- bining region and its electrostatic surPace potential. The main chain traces of the V, (green) and VL (yellow) domains are shown as ribbons. Contours are displayed at +1 (blue) and -1 (red) k d m o l . A, view looking directly down into the combining region; B, view from the H3 side of the combining region.

' 8 - . . ,

4

FIG. 9. Ribbon picture of Fab-tri- plex complex in s t e w . The V, and VL domains are colored as described for Fig. 7. The triplex is shown, with the purine strand in purple and the two pyrimidine strands in red. A, view from the H3 side of the combining region; B , view looking di- rectly down into the combining region.

to make contacts with the DNA bases. The presence of the portion of CDR-H2 that projects outward from the combining sugar phosphate backbone of the second pyrimidine strand in region. the major groove makes this an area of extremely negative The model for the Fab-triplex complex is based on the as- electrostatic potential. It is therefore likely that the contacts sumption that this CDR-H2 loop interacts with the minor the antibody makes with this region of the triplex will be domi- groove of the DNA. This portion of CDR-H2 possesses positive nated by interaction with the DNA backbone. The minor groove electrostatic potential, whereas there is another region of posi- of the triplex, however, is wide enough to accommodate the tive potential across the combining region centered around

3622 Structure of a Fab Fragment Specific for ll-iple-stranded DNA CDR-H3 and CDR-L2 (Fig. 8). The antibody would then bind the triplex at an angle, with the long axis of the DNA helix approximately parallel to the V,-V, domain interface. The pro- truding CDR-H2 loop can thus make base-specific contacts with the DNA via the minor groove, and the region of positive charge near CDR-L!2 and CDR-H3 (created by lysine L53 and arginine H98) interacts with the sugar phosphate backbone of the DNA (Fig. 9). Substituting lysine L53 with arginine would increase the potential for forming hydrogen bonds with the DNA backbone while conserving the positive charge in this area of the combining region. This substitution should increase the overall affinity of Jel318 for triplex DNA.

With the DNA triple helix bound in this orientation, approxi- mately six phosphate groups on the DNA become inaccessible to a 1.7-A radius solvent probe. These DNA phosphates are mostly located near CDR-H3 in the center of the combining region. Upon complex formation, a total surface area of -700 Az on the antibody would become buried. A similar area of total buried surface is observed in the crystal structures of Fab- protein complexes (57,621.

By positioning CDR-H2 in the minor groove of the triplex, there is the potential for forming base-specific hydrogen bonds that can explain the preference of Jel 318 for AT-rich triplex. The minor groove of AT-containing DNA is more acidic than GC-rich triplex owing to the presence of 0-2 of thymine bases, In the Fab-triplex model, Lys-H52b, Ser-H52c, Lys-H53, and Asn-H54 contact atoms of the DNA in the minor groove, with the side chain of Lys-H53 in position to donate a hydrogen bond to the 0-2 atom of thymine. In GC-containing triplex, the exo- cyclic amino group of the guanine bases hydrogen bonds with 0-2 of cytosines and is incapable of accepting a hydrogen bond from the lysine side chain. The presence of the guanine NH2 group in the minor groove may interfere with the lysine inter- acting with the 0-2 atom of cytosine. If this proposed interac- tion is a major determinant of sequence specificity, then Jel318 should bind equally well to triplex containing inosine in place of guanine. Replacing Lys-H53 with a residue that can accept a hydrogen bond from the NH2 group of guanine may result in an antibody with increased affinity for GC-rich triplex.

In summary, the three-dimensional structure of Jel318, the first triple-stranded DNA-binding antibody Fab fragment, has been determined. The combining region presents a relatively flat surface characterized by a single protruding loop that ex- tends away from the main body of the molecule. A model for the Fab-triplex complex is proposed that places this protruding loop in the minor groove of the DNA, where the residues in CDR-H2 of the antibody may be able to make base-specific contacts with the triplex. Site-specific mutagenesis experi- ments to test this model are proposed and will be facilitated by the fact that Jel 318 has been cloned as a single chain Fv fragment into E. coli. These experiments will help to delineate further the mechanism of triplex binding by Jel 318 and may aid in the design of triplex-specific antibodies with altered or enhanced specificity for different triple-stranded nucleic acids.

REFERENCES 1. Felsenfeld, G., Davies, D. R., and Rich, A. (1957) J. Am. Chem. Soc. 79,2023 2. Morgan, A. R., and Wells, R. D. (1968) J. Mol. Biol. 37,63 3. Felsenfeld, G., and Miles, H. T. (1967) Annu. Rev. Biochem. 96,407 4. Riley, M., Maling, B., and Chamberlain, M. J. (1966) J. Mol. Biol. 20,359 5. Lipsett, M. (1964fd. Bioi. Chem. 239,1256 6. Howard, F. B., Frazier, J., Lipsett, M. N., and Miles. H. T. (1964) Biochem.

7. Lyamichev, V. I., Mirkin, S. M., and Frank-Kamenetskii, M. D. (1986) J. Bio-

8. Mirkin, S. M., Lyamichev, V. I., Drushlyak, K. N., Dobrynin, V. N., Filippov, 5.

Biophys. Res. Commun. 17,93

mol. Struct. & Dyn. 3, 667

A,, and Frank-Kamenetskii, M. D. (1987) Nature 930,495

10. 9.

11. 12.

13. 14. 15. 16. 17. 18.

20. 19.

21.

22.

23.

24. 26. 26.

27.

28, 29, 30. 31.

32. 33.

34.

35.

36.

37.

38.

39.

40. 41.

42. 43. 44. 45.

46.

47. 48.

49. 50.

51. 52. 53.

55. 54.

56.

57.

58. 59. 60.

61. 62, 63. 64.

Htun, H., and Dahlberg, J. E. (1988) Science 241,1791 Wells, R. D., Collier, D. A, Hanvey, J. C., Shimizu, M., and Wohlrab, F. (1988)

Glover, J. N. M., and Pulleyblank, D. E. (1990) J. Mol. Biol. 211, 653 Bimboim, H. C., Sederoff, R. R., and Paterson, M. C. (1979) Eur: J. Biochem.

FASEB J. 2,2939

Schon, E., Evans, T., Welsh, J., and Efstratiadis, A. (1983) Cell 31, 837 Laraen, A,, and Weintraub, H. (1982) Cell 2S, 609 Elgin, S. C. R. (1982) Nature 300,402 Weintraub, H. (1983) Cell 32,1191

Praseuth, D., Perroualt, L., Le Doan, T., Chassignol, M., Thuong, N., and Lee, J. S., Johnson, D. A., and Morgan, A. R. (1979) Nucleic Acids Res. 6,3073

Johnston, B. H. (1988) Science 241, 1800 Amott, S., and Selsing, E. (1974) J. Mol. Biol. 88, 509 Amott, S., Bond, P. J., Selsing, E., and Smith, P. J. C. (1976) Nucleic Acids Res.

Mooren, M. M. W., Pulleyblank, D. E., Wijmenga, S. S., Bbmmers, M. J. J., and

Pilch, D. S., Levenson, C., and Shafer, R. H. (1990) Proc. Natl. A d . Sei.

Rajagopal, E, and Feigon, J. (1989) Biochemistry 28,7859 de 10s Santos, C.. Rosen, M., and Patel, D. (1989) Biochemistry 28,7282 Birg, F., Raseuth, D., &rial, A., Thuong, N. T., Asseline, U., Le Doan, T., and

Cooney, M., Czemuszewics. G., Postal, E. H., Flint, S. J., and Hogan, M. E.

Maher, L. J., 111, Wold, B., and Denran, P. B. (1990) &iem 246,725 Povsic, T. J., and Denran, I? B. (1990) J. Am Chem. Soc. 112,9428 Strobel, S. A., and Dervan, P. B. (1990) Science 249, 73 Lee, J. S., Wocdsworth, M. L., Latimer, L. J. P., and Morgan, A. R. (1984)

Hampel, K. J., Crosson, P., and Lee, J. S. (1991) Biochemistry SO, 4455 Lee, J. S., Burkholder, G. D., Latimer, L. J. P., Haug, B. L., and Braun, R. P. (1987) Nucleic Acids Res. 11, 1047

Hemn, J. N., He, X. M., Ballard, D. W., Btier, P. R., Pace, P. E., Bothwell, A. L. M., Voss, E. W., Jr., and E ~ u n d s o n , A. B. (1991) Proteins Struct. Funct. Genet. 11,159

Cygler, M., Boodhoo, A., Lee, J. S., and Anderson, W. F. (1987) J. Biol. Chem.

Mol, C. D., Muir, A. K. S., Lee, J. S., and Anderson, W. F. (1994) J. Bwl. Chem. as2, 643

Burkholder, G. D., Latimer, L. J. P., and Lee, J. S. (1991) Chromosoma (Bed. 289,3605

Boodhoo, A., Mol, C. D., Lee, J. S., andhderson, W. F. (1988) J. Bid . Chem. 101.11

Barry, M. M., Mol, C. D., Anderson, W. F., and Lee, J. S. (1994) J. Bid. Chem. 253,18578

asS, 3623 Davies, D. R., and Segal, D. M. (1971) Methods Enzymol. 22,266 Xuong, Ng. H., Sullivan, D., Nielsen, C., and Hamlin, R. (1985) Acta Crystal-

Howard,A., Nielsen, C., and Xuong, Ng. H. (1985) Methods Enzymol. 114,452 Wilson, A. J. C. (1949) Acta Crystallogr: 2,316 Rossman, M. G. (1990) Acta Crystallop Sect. A Found. Crystailogr: 46,73 M a ~ u ~ ~ , M., Marquart, M., Jones, T. A, Colman, P. M., -18, HI, Huber,

Suh, S. W., Bhat, T. N., Navia, M. A., Cohen, 0. H., Rao, D. H., Rudikoff, S., and R., and Palm, W. (1978) J. Md. Bioi. 121,441

Fujinaga, M., and Read, R. (1987) J. Appl. Crystallogz 20.517 Davies, D. R. (1986) Proteins Struct. finct. Genet. 1, 74

Cygler, M., and Desmhers, M. (1989) Acta Cryntallogr: Sect. A Found. Crys-

Briinger, A. T., Kuriyan, K, and Karplus, M. (1987) Science 2SS, 458 Hendricluwn, W. A., and Konnert, J. H. (1981) in Bio&ular Structure,

C o ~ ~ ~ n , Function and Eu~ution (Srinivasan, R., edf Vol. 1, pp. 43-57, Pergamon Press, Oxford

98,301

Helene, C. (1988) Proc. Natl. had. Sci. U. S. A. 86,1349

11,4141

~i lbers , c. w. (1990) ~ u c l e i ~ kids RS. 14 6523

U. S. A. 87,1942

Helene, C . (1990) Nucleic Acids Res. 18,2901

(1988) Science 241,456

Nucleic Acids Res. 12,6603

logr: Sect. B Struct. Crystallogc Cryst. Chem. 41,267

t a l l o g ~ 46,563

Bhat, T. N. (1988) J. Appl. Crystallogx 21,279 Read, R. J. (1986) Acta Crystallogr. Sect. A Found. Crystallop 42, 140 Ramachandran, G. N., Ramakrishnan, C., and Sasisekharan, V. (1963) J. Mol.

Luzzati, V. (1952) Acta Crystallogr: 5,802 Artymiuk, €! J., Blake, C. C. F., Rice, D. W., and Wilson, K. S. (1982) And

‘lbnno, J., Stadler, E., Skem, T.. Auer, H., Kanzler, O., Betzel, C., Blaas, D., and

Biol. 7, 95

Crystdogr: Sect. B Struct. CwstaUogr: Ctyst. Chem. 30,778

%lip, W. R., Varghese, J. N., Laver, W. G., Webster. R. G., and Colman, P. M. Fits, 1. (1992) Protein Sci. 1,1174

Steipe, B., Pliickthun, A., and Huber, R. (1992) J. Mol. Biol. 22S, 739 Saul, F. A., and Poljak, R. J. (1992) Proteins Struct. Funct. Genet. 14,363 Weiner, S. J., Singh. U. C., O’Donnell, T. J., and Kollman, D. A. (1984) J. Am.

Chothia, C., and Lesk, A. M. (1987) J. Mol. Biol. ISS, 901 Davies, D. R., Padlan, E. A, and Sheriff, S. (1990)Annu. Reu. Biochem. S8.439 Lee, B., and Richads, F. M. (1971) J. Mol. Biol. 65,379 Kabat, E. A., Wu, T. T., Reid-Miller, M., Perry, H. M., and Gotteaman, K. S. (1987) Sequences of Proteins of Immunological Interest, 4th Ed., United States Department of Health and Human Services, National Institutes of Health, Bethesda, MD

(1992) J. MoLBiol. 227, 122

Chem. Soc. 106,6243