Crystallization and Structure Determination of an Autoimmune Anti-poly(dT) Immunoglobulin Fab Fragment at 3.0 A Resolution*

HEDlO is an autoimmune antibody (IgG) which shows considerable specificity for the single-stranded DNA poly(dT). Production of Fab fragments by papain digestion resulted in heterogeneity as judged by isoe- lectric focusing gels, which had a marked negative effect on crystallization. However, a single species of Fab with a PI of 7.6 could be isolated in good yield by DEAE-cellulose chromatography, and good crystals were produced by the hanging drop vapor diffusion method. The space group was P2, with cell dimensions, a = 64.2, b = 90.0, c = 42.3 A, and fl = 9 6 . 7 O . These crystals diffract to about 2.2 A resolution. The struc- ture of Fab HEDlO was solved by the molecular replacement method using the known struqture of MC- PC603 and is refined to R = 27.2% at 3.0 A resolution. is extended than and an elbow angle (between the variable and constant domains) of 162", very similar to in Fab KOL. The majority of the hypervariable regions are in the

Antibodies provide a family of proteins with very similar overall structure. The similarity of their architecture is accompanied by variations in their binding sites which result in their specificity for a large variety of molecules. Anti-DNA antibodies are the subject of intensive studies (for a review see Ref. 1). They provide the opportunity to investigate how proteins can recognize different DNA structures and/or sequences and the range of interactions that can provide the capability for such recognition. Crystallographic studies of these proteins and their complexes with DNA will provide an understanding of the protein-nucleic acid interactions that give rise to their specificity, whether it is for the configuration of the sugar-phosphate backbone or for particular bases. The properties of these proteins make them very interesting models for studying protein-nucleic acid interactions.
The determination of the structures of DNA-binding immunoglobulins will allow the comparison of their structures with those of proteins that have evolved to bind particular DNA structures or base sequences. This is particularly of interest in view o f the observed similarities of three-dimensional structure and amino acid sequence among proteins which are involved in repression and activation of transcription (2)(3)(4)(5)(6)(7)(8).
*This work was supported in part by grants from the Medical Research Council of Canada to the MRC Group on Protein Structure and Function and to J. S. L. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. § To whom correspondence should be sent.
Anti-DNA antibodies are also of interest because of their involvement in the disease systemic lupus erythematosus (9). The sera of patients with this autoimmune disorder (as well as animal models of the disease) frequently contain high titers of anti-DNA antibodies which have been implicated in the pathogenicity of systemic lupus erythematosus. It is thought that these antibodies form immune complexes which, for as yet unknown reasons, are not cleared from the system, resulting in tissue damage.
For these reasons HEDlO is a particularly suitable antibody for structural studies. Not only was it derived from autoimmune mice (IO) which develop a disease resembling human systemic lupus erythematosus but also it shows considerable specificity in its interactions with single-stranded DNA (11). Detailed binding studies revealed a strong preference for poly(dT) and poly[d(BrU)] with the recognition of four consecutive bases. Its binding constant is strongly dependent upon ionic strength suggesting that two phosphates are involved in the interaction. It therefore recognizes features of both the base and the sugar-phosphate backbone of nucleic acid.
Preliminary crystallographic studies of another Fab fragment specific toward single-stranded DNA (BV04-01 antibody) have recently been reported (12). This antibody also recognizes thymine residues in single-stranded DNA but unlike HED10, phosphate groups play a minor role in the interaction. Comparison of these two Fab fragments will be most instructive.
The structure of a number of Fab fragments (13)(14)(15)(16)(17), Bence-Jones proteins (18)(19)(20), and variable domain dimers, F, (21)(22)(23)(24) have been reported and are known at atomic resolution. In addition, the low resolution structure of an intact immunoglobulin molecule is also known (15,(25)(26)(27). Comparison of the structures of different Fab fragments (28) has revealed a common folding pattern for the variable and constant domains of the light (VL and C L ) and heavy (V, and CH1) polypeptide chains and a similar association of Vt-VH and CL-CH1 into the variable (V) and constant (C) domain pairs.
There are only two polypeptide connections between the domain pairs, and there are relatively few interatomic contacts between them. For these reasons Fab molecules display considerable flexibility. The high purity of the protein used for crystallization is very often an imperative in growing crystals suitable for xray diffraction experiments (29). We present here the methods used for the preparation of a homogeneous HEDlO Fab fragment, which was essential for the crystallizatioq of the protein, and the determination of its structure at 3.0 A resolution.

MATERIALS AND METHODS
The preparation of the HEDlO IgG from the hybridoma cell line was as described by Lee et al. (10). The immunoglobulin isotype was 643 determined by solid-phase radioimmunoassays with the aid of isotypespecific rabbit anti-mouse IgGs (Zymed Laboratories). The HEDlO Fab fragment was produced by incubating IgG with papain (Sigma), typically for 4 h at 37 "C in 0.25 M NaCl, 25 mM Tris-HC1, pH 7.5, 1 mM EDTA, 25 mM 2-mercaptoethanol (11). The papain was inactivated with iodoacetamide, and the digest was dialysed versus 10 mM Tris-HCI, pH 8.0, before being loaded on isoelectric focusing (IEF)' gels for analysis, or applied to a DEAE-Sephacel column equilibrated with the same Tris buffer for purification. The Fc portion was retained by the DEAE-Sephacel and eluted late in a 0-0.5 M NaCl gradient. The Fab fragments eluted in the flow through or early in the gradient.
Isoelectric focusing in polyacrylamide gels was carried out according to the Pharmacia product manual (31).
Data Collection-The intensities were collected on a SYNTEX P21 diffractometer equipped with a long detector arm (crystal to detector distance of 55 cm). Since the crystals were sensitive to radiation and decayed relatively quickly in the x-ray beam, the data were collected by scanning only the tops of the peaks (32,33). Each reflection was scanned in seven steps over a narrow w-range (0.10-0.15") and checked to ensure that the top of the peak was scanned. Five consecutive steps within the top of the peak were used to derive the net intensity. For each reflection the background was measured for 2 s on either side of the peak with 0.9" offset in the w angle. Ten reflections, distributed over the reciprocal space and covering the 28 range investigated with each crystal, were measured every 100 reflections and used later for proper scaling and correction for decay. Approximately 100 reflections were measured per hour. Data collection continued until the intensities of the control reflections declined on average to 50% of their original intensities.
To estimate the background for each reflection Fletterick et al. (33) made the assumption that the background is a slowly varying function independent of x and that the dependence of 28 and 6 could be separated. They measured the background at the end of data collection for one value of 6 as a function of 28 and for one value of 28 as a function of 6, and the background at any position was calculated based on these observations. Application of this procedure to our data sometimes led to large negative intensities which showed a systematic pattern, indicating that for some 6 angles the background was overestimated. This is due to the oversimplified description of the angular dependence of the background. To obtain a better estimate of the background around each reflection, the (6,28) plane was divided into a grid with 6(6) = 5" and 6(28) determined empirically from the 28 dependence of the background. An average background was calculated for each grid point from the actual short (4 s/reflection) background measurements of reflections with 6 and 28 within & (6(6),6(28)) from that grid point.
Measurements differing by more than 3u from the mean were discarded and the mean value recalculated. For each reflection the average background from the nearest grid point was taken as an estimate. Most of those were based on a total measurement time between 50 and 100 s. With the background estimated in this way only a few reflections had negative intensities below the -3u(I) level. The absorption correction was estimated according to the method of North et al. (34), and the anisotropic decay correction was derived from control reflections after Fletterick et al. (33).
Data to 3A resolution were collected from two crystals ( Table I). The symmetry equivalent reflections from each crystal were averaged (R-= 0.04 and 0.08, respectively) and then merged together (Rmea, = 0.10). Data for two heavy atom derivatives, K2Hg14 and HgC12, were collected and processed in a similar way. Some statistical information is given in Table I.

RESULTS
Protein Purification and Crystallization-An examination of the microheterogeneity of Fab fragments that were used for crystallization experiments was initially undertaken bel The abbreviations used are: IEF, isoelectric focusing; PEG, polyethylene glycol; MOPS, 4-morpholinepropanesulfonic acid. cause one preparation yielded crystals while subsequent preparations did not. Analysis of the different preparations of Fab by IEF indicated that they were, in fact, a mixture of several species with PI values between 7.0 and 8.5. The preparation that crystallized contained the same PI species as the preparations that did not crystallize but the distributions differed. Analysis of the products of the papain digestion of IgG HEDlO by IEF revealed that the conditions of the digestion affected the distribution of PI species produced. By altering the IgG/papain ratio, it was found that the distribution of the different species could be altered (Fig. 1). Digestions at a constant ratio of IgG/papain for different periods of time were not as effective in altering the distribution of products (data not shown). Presumably the papain became inhibited, although this was not verified. In the case of HED10, the major initial product of papain digestion is clearly subjected to further proteolysis to yield a species with a PI of 7.6. At still higher levels of papain further degradation occurs to smaller molecular weight material.
Although the distribution of the PI species could be changed by altering the conditions of the papain digestion, this was not in itself sufficient for obtaining high quality crystals for x-ray diffraction measurements. IEF analysis of single crystals (see Fig. 1) revealed that they contained predominantly the PI 7.6 species and very little, if any, of the others. This information, combined with the analysis of the products of the papain digestion and their dependence on the conditions of the digestion, allowed the adaptation of the digestion to provide for maximal production of the PI species that occurred in the crystals. Although this species could have been purified by preparative IEF, it was found that it could be purified from the other Fab fragments by DEAE-cellulose chromatography (Fig. 2). This allowed a simpler purification with better overall yield.
The HEDlO PI 7.6 Fab fragment was crystallized as described above. The protein crystallizes in the monoclinic space group P21 with cell dimensions a = 64.2, b = 90.0, c = 42.3A, and / 3 = 96.7". One Fab fragment of 50, !00 daltons per asymmetric unit would result in a V,,, of 2.43 A3/dalton, which is in the range generally observed for protein crystals (35). The x-ray diffraction pattern of these crystals extends to about 2.2 A resolution. The purity of the protein dramatically affects the crystal growth process (29). Having a mixture of PI species prevented the nucleation of crystal growth. In some cases seeding with microcrystals promoted crystallization in preparations containing a distribution of species such as those in Fig. 1, but the crystals obtained were not suitable for x-ray diffraction data collection. Seeding was not required when the purified PI 7.6 Fab was used although large crystals were most efficiently obtained using a macro-seeding technique. Clearly the papain digestion of IgG to produce Fab fragments introduces heterogeneity which can affect crystallization. Production of crystals of monoclonal antibody Fab fragments that were suitable for x-ray diffraction analysis required. careful control of the conditions of the papain digestion and purification of the species of interest. The best results were obtained when dealing with as conformationally and structurally pure a sample as possible (29). Structure Solution-The structures of three immunoglobulin Fab fragments, NEW (36), McPC603 (14), and KOL (15) have been determined at atomic resolution and are well refined. The comparison of their structures (28) shows that the conformation of each of the domains of the heavy and light chains are similar among the three proteins. The largest differences are within the hypervariable loops. The V and C domain pairs of the Fab molecule are connected by short polypeptide links, one within the heavy chain and one within the light chain, providing substantial flexibility in the so called "switch" or "elbow" region. The elbow angle, defined as the angle between the pseudo 2-fold axes of symmetry of the VL-VH and CL-CHl domain pairs, varies between 132" (McPC603, (14)) and 166" (KOL, (15)). This variation is even greater when the light chain dimers Mcg (113", (18)) and LOC (97", (19)) are considered. Because three good models of an Fab molecule were available, molecular replacement seemed to be the method of choice for the structure determination. Although the amino acid sequence of HEDlO is at present only partially known, the homology to McPC603 is expected to be higher than that of KOL or NEW. McPC603 was obtained from the same species (mouse) as HEDlO and its light chain was also a K-type. Since only one gene codes for the CL(%) region in mouse (371, the sequence of that part of both proteins had to be nearly identical. The heavy chain of McPC603 belongs to class (Y while that of HEDlO belongs to class y2A and the similarities of their amino acid sequences are less pronounced (38).
Determination of Orientation-A series of rotation functions were calculated, employing the fast rotation function algorithm (39) with the McPC603 Fab fragment as a model. Parameters such as the resolution shell, the integration radii, and the number of reflections were varied. Two peaks, corresponding approximately to a 30" difference in the orientation of the model, appeared consistently in all calculations. Since a difference in the elbow angle between McPC603 and HEDlO was expected, these results were interpreted as an indication of a different spatial arrangement of V and C domain pairs in HEDlO as compared to McPC603. According to that presumption one peak corresponded to the correct orientation of the V domain pair (and the incorrect orientation of C domains) while the other indicated the orientation of the C domain pair. To test this hypothesis, and to determine which peak was related to which pair, a series of rotation functions were calculated using only the V or the C domain pair as a model. In each case only one dominant peak was observed and it was at one of the two positions obtained previously (Table 11). A detailed account of these calculations and comparison with the results obtained by using other Fab structures as starting models will be presented elsewhere?
Determination of the Position of the Molecule within the Unit Cell-In the polar space group P21, the dimensionality of the problem of finding a proper position within the asymmetric unit of the correctly oriented molecule is decreased from three to two. Only translations along the x-and z-axes have to be determined while translation along the y-axis may be arbitrarily set to zero. To determine the x-and z-translations, the model was first rotated into the orientation indicated by the rotation function peak and then moved in small steps (1A or less) within the part of the x,z-plane corresponding to one permissible origin. The correlation coefficient * M. Cygler and W. F. Anderson, manuscript in preparation.   Table   11). To determine the relative translation along the y-axis between the two pairs, they were positioned according to their respective peaks in the z,z-plane. The correlation coefficient was then calculated as a function of the translation of the C domain pair along the y-axis. The maximum correlation coefficient was r = 0.45. At this stage the Fab fragment was divided into four rigid domains, VL, CL, VH, CHl, and the orientation and position of each of them was adjusted stepwise (within a narrow angular and translational range) to maximize the correlation coefficient. The final results are shown in Table 11. As an independent test of the correctness of this model, phases determined by it were used to calculate difference maps for the two available derivatives. The highest peaks in these maps corresponded to the positions of heavy atoms derived independently from the analysis of the difference Patterson function. The K2Hg14 derivative was also used independently to determine the translations of correctly oriented domains? Model Building and Partial Refinement-One of the major concerns in the process of refining a protein at medium resolution on the basis of molecular replacement results is the bias introduced by the starting model (41). To minimize this effect a rather conservative approach was used in building the first model of the HEDlO Fab molecule. Since the amino acid sequence of the HEDlO Fab fragment is at present only partially known (about 23 NHZ-terminal residues of each chain plus the CL domain), the amino acid sequence of Mc-PC603 was used. The C, backbones of three Fab fragments, McPC603, KOL, and NEW were visually compared on a graphic system. Residues forming the part of the chain that adopted different conformations in these molecules, were removed from the starting model. This included, of course, the hypervariable loops and the switch region. A total of about 2400 atoms out of 3399 of McPC603 were then used to phase the reflections. An electron density map was calculated using the coefficients ZF, -DF, (42), which has the property of minimizing the model bias. The map was displayed on the MMS-X graphics system with the help of the program M3 (43). The parts of the chain that were not used for the phasing of this map and were clearly recognizable, were then added to the model. In the places where no sequence information was available, the side chains were altered according to the shape of the electron density. This model contained 2822 atoms. The R-factor calculated at this point was 38.0%. Eight cycles of restrained least squares refinement (44) with relatively rigid restraints reduced the R-factor to 32.7%. The model was then refit on the graphics system to a new electron density map calculated with 2F, -DF, coefficients and phases calculated from the improved model. It was possible at this point to follow the polypeptide chain in the switch region, as well as parts of the hypervariable regions. The next eight cycles of refinement further reduced the R-factor to 27.2%. The model will be refined further when the resolution of the diffraction data is extended and the amino acid sequence is determined.

DISCUSSION
Analysis of Papain Digestion for Producing Fab Fragments-Papain digestion of the HEDlO IgG was carried out with the aim of reducing the molecular weight and the conformational heterogeneity due to the flexibility of the intact IgG. This increases the probability of obtaining suitable crystals. One side effect is that the papain digestion introduces chemical heterogeneity due, presumably, to variability in the position or number of proteolytic cleavages. Our results show that, in the case of HED10, it is important to purify the Fab fragments produced to obtain crystals suitable for x-ray diffraction analysis. IEF provides a convenient and sensitive method for analyzing the products of the papain treatment and optimizing t.he conditions that are used for the digestion. In this case we were fortunate to obtain the initial crystals which allowed us to determine what PI species was incorporated in the crystal and accordingly adapt our procedures to purify that particular species. In most cases it will be useful to carry out crystallization trials on each of the purified PI species in order to obtain the most satisfactory crystals (29)3 whether or not the initial mixture of Fab fragments crystallize. When there are problems with the reproducibility of crystallization, or the size or the quality of crystals, careful purification to yield a chemically and conformationally pure sample is often a good first step towards alleviating the problem.
Crystal Structure of HEDIO-Th? structure of the HEDlO Fab fragment has been solved at 3  that is observed in the crystal exhibits some differences from McPC603, which was used as a model for the rotation/translation searches. The most significant difference is in the magnitude of the elbow angle. The HEDlO molecule is much more extended than McPC603 with an elbow angle of 162", compared to 132" for the latter. In this respect HEDlO resembles Fab KOL (15) which has an elbow angle of 166" and is the most extended of the Fab molecules and light chain dimers that have been investigated to date by x-ray crystallography. There are also differences between HEDlO and McPC603 at the level of the association of the light and heavy chains into the V and C domain pairs. These differences are immediately visible from the somewhat different values of the final rotation angles and translation parameters (Table 11) Fig. 3.
In the recently determined structure of the light chain dimer LOC a somewhat different association of the two VL domains was observed than is found in other immunoglobulin fragments (19). While the transformation relating the VL and VH domains usually has a very small transla$onal component along the pseudo 2-fold axis (less than 0.5 A') in LOC this has a value of 3. 5 A. The relative translation of the two VL domains in the direction perpendicular to the axis of the 6barrel results in the closing of the central cavity in the antigen-binding region. Rather than forming a binding pocket or cavity this creates a site that protrudes from the protein surface. They speculate that such a different mode of association may, if present in other immunoglobulins, contribute at yet another level to the diversity of the immune response system. It is, however, equally possible that the association mode observed in LOC is a result of having two identical domains forced to form a complex. Up to now, the mode of association within the V domains that is observed in Fab fragments does not show such differences.
The electron density corresponding to the first complementarity determining regions of both the heavy and light chains is low and difficult to interpret at this stage and is not represented in the current model. At this moment, we attribute the weak density to the relatively large errors in our current phases rather than to disorder or large thermal motion of that segment of the polypeptide chain. The density in the other hypervariable regions and the switch region was stronger, and it was possible to trace the chain in these locations.
Packing of Molecules in the Crystal-The molecules in the crystal form layers parallel to the x,%-plane. Two contacting layers are related by the 2-fold screw symmetry. Molecules in one layer interact in a head-to-tail fashion, with the V domains of one molecule packed against the C domain of the neighboring one, forming infinite lines in the [ 1101 direction. This mode of packing, where the antigen-binding region of a Fab molecule packs tightly against symmetry-related molecules, is often observed in crystals of Fab fragments (28). The detailed nature of these interactions in HEDlO crystals is not yet known since the electron density in the antigen-binding region is only partly interpreted. The contacts between symmetry related molecules along the y-axis is shown in Fig. 3 and the arrangement of molecules in one layer is shown in Fig. 4. The side-by-side packing of C domains forming an extended /?-sheet observed in Mcg and LOC light chain dimers (46) does not take place here. This is the first detailed structure of an autoimmune Fab fragment to be reported. The overall architecture would appear to be similar to conventional antibodies. Thus, the pathogenic potential of autoimmune antibodies probably resides in the antigen-binding site rather than in the structure itself.