J. Mol. Biol. (1991) 220, 723-737

Structures of Met atd Azidomet Hemerythrin at 166 A Resolution Margaret A. Holmes and Ronald E. Stenkamp University

Department of Biological Structure of Washington, Seattle, WA 98195, U.S.A.

(Received 27 August 1990; accepted 26 March 1991) The crystallographic refinement of met and azidomet hemerythrin has been carried out at 1.66 A resolution in an attempt to characterize precisely the binuclear iron center in this protein. Restrained least-squares refinement has produced molecular models giving R-values of l&9% for met (65,683 reflections from 10 A to I.66 A) and 17.6% for azidomet hemerythrin (68,747 reflections from l@O A to 1.66 A). The protein structure in each derivative is very similar to that of myohemerythrin. The ~-0x0 bridged iron center differs between the two forms. The complex in met hemerythrin is asymmetric with the bridging oxygen closer to one of the iron atoms while the complex in azidomet hemerythrin is symmetric. After investigations of the effects of correlation in the refinement, we believe this difference between the two complexes is associated with chemical differences and is not a refinement artefact. Keywords:

hemerythrin;

refinement;

crystallography;

1. Introduction

Raman studies (Zhang et al., 1988; Gomez-Romero et al., 1989) have not supported the asymmetric complex so we have carried out higher resolution

Hemerythrin and myohemerythrin, the oxygenbinding proteins found in sipunculids and other marine invertebrate phyla, have been the objects of several crystallographic investigations aimed at determining the structure of the protein and its nonheme-iron active site. Myohemerythrin and the hemerythrin subunits consist of four roughly parallel a-helices; this structure provides the aminoacid

side-chain

ligands

for

the

binuclear

(1.66 A) crystallographic cules to obtain

justification

EXAFST

2.

iron

could

of the mole-

structural

models for

Methods

(a) Data collection The crystals of T. dyscrita hemerythrin are the same as those studied earlier (Loehr et al., 1975). (For a discussion of the composition of met hemerythrin including pK and pH values, see Stenkamp et al., 1983a,b.) The space group is P4, with unit cell parameters of a = b = 866 A and c = 898 A. The unit cell contains 2 octameric molecules and the asymmetric unit contains 4 subunits, 2 from each octamer. We have made extensive use of this structural redundancy in the model-building and refinement. High resolution (1.66 A) data were measured for met and azidomet hemerythrin using the multiwire area detector facility at the University of California, San Diego (Xuong et al., 1985). Table 1 provides a statistical summary for the data sets. While the R-values between replicate reflections are not good for the highestresolution shell, we believe this is due to the low intensities of these reflections and is not unique to these data sets or the data collection method. In the outermost shell, 1.77 A to 166 K resolution, about 28% of the reflections for met and 20% of the reflections for azidomet have intensities larger than 28(l).

be

and resonance

t Abbreviations used: EXAFS, extended X-ray absorption fine structure; c.p.u., central processing

refinement

more accurate

them.

p-oxygen-bridged metal center (Fig. 1). The protein, in addition to operating between deoxy and oxy states, can bind other exogenous small molecule ligands. A detailed view of the differences between unliganded and liganded states of the metal center was obtained in the 20 A (1 A = 01 nm) study of met and azidomet hemerythrin from Them&e dyscrita (Stenkamp et al., 1982, 1983u,b, 1984). One unusual feature of those refined models was an asymmetric ~-0x0 bridge in which the bridging oxygen atom was closer to one iron atom than the other. The Fe-O bond distances observed were 1.92 A and 1.68 A for the met complex and 1.89 A and 164 A for the azidomet complex. No reasonable experimental or chemical found for this asymmetry.

non-heme iron; structure

unit.

723 0022-2836/91/K&723-15

$03.00/O

0

1991 Academic

Press Limited

724

M. A. Holmes and R. E. Stenkamp His73

refined in any given cycle, the usual procedure being to run 5 or more cycles of positional refinement, then 2 or 3 cycles of temperature factor refinement, followed by more cycles of positional refinement. This procedure was interrupted 4 times to examine the structure using the program FRODO (Jones, 1978, 1982) and to make changes indicated by the F, and F,- F, electron density maps. Amino acid side-chains were moved and adjusted, alternative conformations were added for side-chains that showed static disorder, stereochemistry of the model was improved, and water molecules involved in bad contacts were deleted. Several times, difference electron density maps were calculated where parts of the model requiring attention were omitted. Comparison of equivalent residues in the 4 different subunits in met hemerythrin (and occasionally with the 4 azidomet hemerythrin subunits) led to changes in atom names (for example, 061 and 062 in an Asp residue might need to be interchanged to agree with the other subunits). When supported by the Fourier maps, changes in the conformation of the sidechains were also made consistent. Initially, the chiral centers at the p-carbon atoms of some of the isoleucine and threonine residues inverted during refinement. TNT does not normally restrain chiral centers, so starting with cycle 24, those 2 types of chiral centers were restrained using non-standard torsion angle restraints such as C”-CBCy’-Cyz for isoleucine. (We first became aware of this technique in J. Hermans’ REFIKE program (Hermans & McQueen, 1974).)

(a)

Figure 1. Azidomet hemerythrin. complex of subunit IA.

(a) Active-site

metal

(b)

Figure

1. (b) C” plot of subunit IA.

(b) Met hemerythrin rejinement The starting model for the high resolution refinement of met hemerythrin consisted of 4296 atoms, 508 of which were water molecules. The program used for the 1st portion of the refinement was Release 3 of TNT (Tronrud et al., 1987). A program written by L. Andrews to adjust the derivatives for the thermal parameters (B-values) and occupancy by the curvature was inserted into the flow of TNT programs and wss found to improve greatly the behavior of the B-values of the active-site Fe atoms (L. Andrews, personal communication). One cycle of TNT took 25 to 3 h of c.p.u. time on our VAX 11/799. As recommended, only positional or thermal parameters were

From the beginning, the thermal parameters did not behave well in refinement. B-values for a few protein atoms refined to below 1.0 A’, and a number of water molecules had B-values higher than 199 A’. The atoms with low B-values were located in regions of good electron density and showed no sign of being incorrectly positioned in Fourier and difference Fourier maps. We ran TNT without using the option to correlate the B-values of bonded atoms (the program BCORRELS); later experimentation including BCORRELS in the TNT run showed that it made no difference. The problem with the water molecules was solved by removing from the structure all water molecules with B-values greater than 75 A’; this

Met and Azidomet Hemerythrin had to be done only once. The low B-values were replaced by the average of the thermal parameters for the atom in the other 3 equivalent subunits. This had to be done several times. Overall, however, the major problem was that the atoms as a whole were drifting towards ever higher B-values. After 33 cycles of refinement, the average thermal parameter for the protein atoms was 23.6 A2 and the average for the active site Fe atoms was 183 A2, values not compatible with a 1.66 A resolution diffraction pattern. We had refined the scale factor between F, and Fc, with the result that the scale factor continually decreased while the B-values increased. The 1st attempt to control refinement of the scale factor involved changing the inner resolution cut-off from 5.0 A to 190 A. The 1st 4 cycles (cycles 34 to 37) run this way lowered the B-values by over 2 A*, but subsequent cycles started to reverse that. At that point, we decided to fix the scale to a value that we obtained from a Wilson (1942) plot. The remainder of the TNT refinement (cycles 38 to 64) was run this way. Initially, the geometry restraints included those for the active-site geometry that were used in the 2-O A refinement (Stenkamp et al., 1983b). To obtain a more bias-free structural model, beginning with cycle 48, the active-site restraints were removed and were not applied during the remainder of the refinement. Occupancies were not refined. although partial occupancies were assigned to atoms that were involved in alternate conformations. They were occasionally adjusted to make the B-values of the at,om similar in the different conformations. In all, 64 cycles of TNT refinement were carried out. At that point, in an attempt to obtain more reasonable refinement. of the B-values and the active-site atoms, we decided to use the program PROFFT (Hendrickson & Konnert, 1980). The starting co-ordinates were the positional co-ordinates from cycle 64 of the TET refinement, and the B-values were those from the 2.0 A PROLSQ refinement. The structure factor derivatives were calculated analytically with a subroutine specific for the space group P4. Initially with PROFFT, the scale factor was refined. After 6 cycles, it was obvious that PROFFT suffered from the same problem as TNT; namely, if allowed to change the scale, it would continually lower the scale and raise the overall R-value. At that point, we scaled the 1.66 A data to the 2.00 A diffractometer data (which had been scaled manually to the calculated structure factors in earlier refinement), and fixed the scale factor for the rest of the PROFFT refinement. Four rounds of FRODO intervention were needed during the PROFFT refinement. The 1st was after cycle 12. All water molecules were deleted from the structure, an F,- F, map was calculated, and water molecules were placed in likely looking peaks in the map. More water molecules were added in subsequent FRODO sessions. Beginning with cycle 43, the contribution of anomalous scattering from the iron atoms was included in the calculations; the number of structure factors and therefore the computing time nearly doubled. In all, 57 cycles of PROFFT refinement were run. The high resolution refinement was completed by refining the structure of the active-site complex using the program CRYLSQ from the XRAY package (Stewart et al., 1976). Beginning with the cycle 42 PROFFT met hemerythrin co-ordinates, 12 cycles of TNT refinement were performed using TNT Release 4. During these 12 cycles, the Fe-O-Fe complex was frozen in a symmetric state with 1.8 A Fe-O distances and an Fe-O-Fe angle of 152” (EXAFS values available at that time (Elam et al.,

at 1.66 A Resolution

725

1982)). This was done so that the rest of the struct,ure would be unbiased towards an asymmetric complex. Next, structure factors were calculated for this protein model, excluding the iron and bridging oxygen atoms. Using these calculated structure factors as partial contributions, CRYLSQ was used to perform full-matrix leastsquares refinement on the iron and bridging oxygen atoms. The starting model of the complex was the symmetric structure used in generating the partial structure factors. Positional parameters for all 12 atoms, isotropic B-values for the oxygen atoms. and anisotropic temperature factors for the iron atoms were refined for 5 cycles. Our final model for the structure of met hemerythrin consists of the final PROFFT protein model plus the CRYLSQ model of the Fe-O-Fe complex. A summary of refinement statistics for the final model is given in Table 2 and a plot of R-value versus resolution is shown in Fig. 2(a). (c) Azidomet hemerythrin

refinement

The starting model for the high resolution refinement of azidomet hemerythrin consisted of 4304 atoms, 504 of which were water molecules. Refinement of azidomet hemerythrin proceeded much as for met hemerythrin; however, far fewer cycles were run because the experimentation that went on with met hemerythrin preceded the same stage of refinement of azidomet hemerythrin. A total of 36 cycles of TNT Release 3 was run with 4 rounds of FRODO intervention. After cycle 12, all water molecules with R-values greater than 75 A2 were removed from the structure, and chiral center torsion angle restraints were implemented for Thr and Ile side-chains. Problems with very low and very high B-values appeared just as they had for met hemerythrin. Beginning with cycle 20, the inner resolution limit was changed from 5 A to 10 A, and the structure factor scale was fixed at a value determined from a Wilson plot. Beginning with cycle 31 active-site geometry restraints, except for azide bond lengths, were removed. PROFFT refinement began with positional parameters from the TNT refinement and B-values from the 2.9 A PROLSQ refinement for the protein atoms. The water molecules from met hemerythrin PROFFT cycle 20 were added to the protein (a few were deleted due to bad contacts). As with met hemerythrin, a scale factor was obtained by scaling the 1.66 A data to the 2.0 A diffractometer data and was not allowed to vary. Twenty cycles of refinement were run and there were 2 rounds of modelbuilding, followed by 8 cycles of refinement with the anomalous scattering from the iron atoms included. CRYLSQ refinement of the active-site complex completed the high resolution refinement. Eleven cycles of TNT Release 4 refinement were carried out with a frozen symmetric Fe-O-Fe complex, and partial F, values were calculated. Eight cycles of CRYLSQ full-matrix refinement were performed on the symmetric starting model. Atomic parameters for the iron and bridging oxygen atoms, including anisotropic R-values for the iron atoms, were refined. Our final model for the structure of azidomet hemerythrin consists of the final PROFFT protein model pius the CRYLSQ model of the Fe-O-Fe complex. A summary of refinement statistics for the final model is given in Table 2 and a plot of R-value ?)ersusresolution is shown in Fig. 2(b). The co-ordinates for the high resolution met and azidomet hemerythrin structures have been deposited in the Protein Data Bank (Bernstein et al., 1977) with identifiers 2HMQ and ZHMZ, respectively.

M. A. Holmes and R. E. Stenkamp

726

Table 1A Reflection 8tfztistics

of met hemeythrin

I. Summary of obsetvation intensities by resolution della Shell lower limit (A)

Number of observations Average resol.

Average obs. F=

F=/a(F=)

520 4.13 361 3.28 3.04 2.86 2.72 2.60 2.50 2.41 2.34 2.27 2.21 2.16 2.11 2.06 2.02 1.98 1.95 1.92 1.88 1.86 1.83 1.80 1.78 1.76 1.73 1.71 1.69 1.67 1.66

673 457 384 341 315 295 279 2.66 255 246 238 230 224 218 213 2.09 204 2430 1.97 1.93 190 1.87 1.84 1.82 1.79 1.77 1.74 1.72 1.70 1.69 1.67

353596 5004.71 396278 2981.31 210753 147732 1115-65 101978 89858 831.41 741-62 65865 59777 537.08 45048 39420 35319 28891 255.03 21692 182-07 14702 13301 128.01 118-92 10044 9293 9309 7827 7875 67.02

32-324 31.241 23303 24959 21.359 17.661 14-637 13241 11.979 11.080 10090 s-999 8231 7379 6333 5655 5124 4251 3814 3307 2807 2-292 2081 2002 1.859 1.584 1.460 1454 1.224 1.222 1.065

Totals

239

142360

13582

with Fz/o(F2)

in given range

Average

0