Tables for
Volume F
Crystallography of biological macromolecules
Edited by E. Arnold, D. M. Himmel and M. G. Rossmann

International Tables for Crystallography (2012). Vol. F, ch. 25.1, pp. 845-872   | 1 | 2 |

Chapter 25.1. How the structure of lysozyme was actually determined

C. C. F. Blake,a R. H. Fenn,a§ L. N. Johnson,a* D. F. Koenig,a‡‡ G. A. Mair,a‡‡ A. C. T. North,a§§ J. W. H. Oldham,a¶¶ D. C. Phillips,a¶¶ R. J. Poljak,a‡‡‡ V. R. Sarmaa§§§ and C. A. Vernonb¶¶

aDavy Faraday Research Laboratory, The Royal Institution, London W1X 4BS, England, and bDepartment of Chemistry, University College London, Gower Street, London WC1E 6BT, England
Correspondence e-mail:

Lysozyme was the second protein structure and the first enzyme structure to be solved by X-ray diffraction methods. The structure was published in 1965. Later work published in 1967 led to an explanation of the catalytic function of the enzyme and gave a vision of the explanatory power of structure for biological function. This chapter describes the crystals, the search for heavy-atom derivatives and the careful X-ray measurements made possible by the development of the linear diffractometer. By optimizing crystal mounting, by correcting for absorption effects and with considerable attention to radiation-damage effects implemented in the measurements and in the data-processing software, a complete data set to 2 Å resolution was measured with remarkable precision. The precision of the measurements allowed anomalous-scattering effects to be introduced to help resolve the phase problem. The resulting electron-density map was immediately interpretable and the initial draft model building took about 1 month. The model was completed in time for it to form the centrepiece of Sir Lawrence Bragg's 75th birthday party at the Royal Institution on 31 March 1965. Many of the historical details of the data collection and structure determination are presented for the first time.

25.1.1. Introduction

| top | pdf |

For protein crystallographers, the year 1960 was the spring of hope. The determination of the three-dimensional structure of sperm-whale myoglobin at 2 Å resolution (Kendrew et al., 1960[link]) had shown that such analyses were possible, and the parallel study of horse haemoglobin at 5.5 Å resolution (Perutz et al., 1960[link]) had shown that even low-resolution studies could, under favourable circumstances, reveal important biological information. All seemed set for a dramatic expansion in protein studies.

At the Royal Institution in London, two of us (CCFB and DCP) had used the laboratory-prototype linear diffractometer (Arndt & Phillips, 1961[link]) to extend the myoglobin measurements to 1.4 Å resolution for use in refinement of the structure (Watson et al., 1963[link]), and we had begun a detailed study of irradiation damage in the myoglobin crystals (Blake & Phillips, 1962[link]). Meanwhile, David Green, an early contributor to the haemoglobin work (Green et al., 1954[link]), and ACTN had initiated a study of β-lactoglobulin (Green et al., 1956[link]) and worked together on oxyhaemoglobin before Green went to the Massachusetts Institute of Technology (MIT) in 1959 on leave for a year. At roughly this time, many of the participants in the myoglobin and haemoglobin work at Cambridge went off to other laboratories to initiate or reinforce other studies. Thus, Dick Dickerson went with Larry Steinrauf to the University of Illinois, Urbana, to start a study of the triclinic crystals of hen egg-white lysozyme.

RJP went to MIT from the Argentine as a post-doctoral fellow in 1958 and worked initially with Martin Buerger. In 1959 he transferred to Alex Rich's laboratory and there he soon came into contact with a number of veterans of the myoglobin and haemoglobin work. In addition to David Green were Howard Dintzis, who had discovered a number of the important heavy-atom derivatives of myoglobin (Bluhm et al., 1958[link]) and was now on the staff at MIT, and David Blow, who had first used multiple isomorphous replacement and anomalous scattering to determine haemoglobin phases (Blow, 1958[link]) and was on leave from Cambridge. The influence of these people, combined with lectures by John Kendrew and then by Max Perutz on visits to MIT, soon convinced RJP that working on the three-dimensional structures of proteins was the most challenging and fruitful research that a crystallographer could undertake. Dintzis, in particular, persuaded him that preparing heavy-atom derivatives was no great problem, and Blow urged him to look for commercially available proteins that were known to crystallize. This soon focused his attention also on hen egg-white lysozyme (Fleming, 1922[link]), but in the tetragonal rather than the triclinic crystal form. He quickly learned to grow crystals by the method described by Alderton et al. (1945)[link] and then found that precession photographs of crystals soaked in uranyl nitrate showed intensities that differed significantly from those given by the native crystals. Encouraged by these results, he asked Max Perutz whether he could join the Cambridge Laboratory, but Max, having no room in Cambridge, suggested that he write to Sir Lawrence Bragg about going to the Royal Institution. Bragg replied with an offer of a place to work on β-lactoglobulin with David Green, who had by then returned to London. RJP accepted the offer and left for London late in 1960 – after first discussing what was going on at the Royal Institution with ACTN, who had just arrived at MIT for a year's leave with Alex Rich.

Early in 1961, RJP showed Bragg his precession photographs of potential lysozyme derivatives, and Bragg enthusiastically encouraged him to continue the work, at the same time urging DCP to arrange as much support as possible. This was a characteristic response by Bragg, who was well aware that at least two other groups were already working on lysozyme, Dickerson and Steinrauf at Urbana and Pauling and Corey at Cal Tech (Corey et al., 1952[link]): competition with Pauling was a common feature of his career. In describing his reaction to Bragg's encouragement, RJP recalled Metchnikoff's view of Pasteur. `He transferred his enthusiasm and energy to his colleagues. He never discouraged anyone by the air of scepticism so common among scientists who had attained the height of their success … He combined with genius a vibrant soul, a profound goodness of heart.'

25.1.2. Structure analysis at 6 Å resolution

| top | pdf | Technical facilities

| top | pdf |

In 1961, the Davy Faraday Laboratory was well equipped with X-ray generators. They included both conventional X-ray tubes, operating at 40 kV and 20 mA to produce copper Kα radiation, and high-powered rotating-anode tubes that had been built in the laboratory to the design of D. A. G. Broad (patent 1956) under the direction of U. W. Arndt. We had a number of Buerger precession cameras and a Joyce–Loebl scanning densitometer, which had been used in the analysis of myoglobin (Kendrew et al., 1960[link]). In addition, we had a laboratory prototype linear diffractometer (Arndt & Phillips, 1961[link]), which had been made in the laboratory workshop by T. H. Faulkner, and the manually operated three-circle diffractometer that had been used to make some of the measurements in the 6 Å studies of myoglobin (Kendrew et al., 1958[link]) and haemoglobin (Cullis et al., 1961[link]). The diffractometers were used with sealed X-ray tubes, since the rotating anodes were not considered to be reliable or stable enough for this purpose.

At this stage, most of the computations were done by hand, but we did have access to the University of London Ferranti MERCURY computer, usually in the middle of the night. This machine was programmed in MERCURY Autocode. The development of the early computers, their control systems and compilers mentioned in this article have been described by Lavington (1980)[link]. Lysozyme crystallization

| top | pdf |

Tetragonal lysozyme crystals were first reported by Abraham & Robinson (1937)[link] and the standard method of preparation was developed by Alderton et al. (1945[link]); RJP used this method. Lyophilized lysozyme was obtained commercially and dissolved in distilled water at concentrations ranging from 50 to 100 mg ml−1. To a volume of the lysozyme solution, an equal volume of 10% (w/v) NaCl in 0.1 M sodium acetate (pH 4.7) was added. About 1 to 2 ml aliquots of this mixture were pipetted into glass vials and tightly capped. Large crystals, frequently with volumes in the range 0.5 to 1 mm3, grew overnight or during the course of a few days.

The crystals were modified bipyramids with well developed {011} faces and bounded by hexagonal {110} faces that were developed to differing extents in individual crystals (Fig.[link]. Many crystals grew in contact with the glass and were less regular in shape. Precession photographs confirmed that the unit-cell dimensions were a = b = 79.1 Å and c = 37.9 Å, and that the space group was [P4_{1}2_{1}2] or [P4_{3}2_{1}2] (Corey et al., 1952[link]). Each unit cell contained eight lysozyme molecules (one per asymmetric unit), molecular weight about 14 600, together with sodium chloride solution that made up about 33.5% of the weight of the crystal (Steinrauf, 1959[link]).


Figure | top | pdf |

Tetragonal lysozyme crystals with well developed {110} faces (left-hand crystal) and small {110} faces (right-hand crystal).

With respect to structure analysis by the method of isomorphous replacement, these two enantiomorphic space groups have the great advantage of exhibiting three independent centrosymmetric projections, on the (001), (010) and (110) planes, corresponding to the hk0, h0l and hhl reflections, respectively. As a result, 173 of the 393 reflections from planes with spacings ≥ 6 Å have heavy-atom contributions exactly in or exactly out of phase with the protein contributions to the structure factors. This property greatly facilitated the determination and refinement of heavy-atom positions in the isomorphous derivatives used in the work on lysozyme. Preparation of heavy-atom derivatives

| top | pdf |

For this work by RJP, lysozyme crystals were grown in glass vials in which a 1 ml solution contained between 20 and 50 mg of lysozyme. Attempts were then made to diffuse heavy-atom compounds into the pre-formed crystals. Stoichiometric amounts of heavy-atom compounds, such as K2PtCl4, UO2(NO3)2, p-chloromercuribenzene sulfonate (PCMBS), p-chloro­mercuribenzoate (PCMB) and K2HgI4, were added to these vials, and precession photographs were taken of the crystals from each vial. The precession photographs of the putative derivative crystals were compared visually with those from the native protein by superimposing them on a light box. This showed immediately whether the cell dimensions had changed and if there were any significant changes in intensity. If these photographs showed no changes in intensity, then the amount of heavy-atom reagent was increased, and this process was continued until the crystals either showed intensity changes or disintegrated. When intensity changes were detected, the effect of increasing the concentration of heavy atom was explored with the object of establishing the optimum conditions for the preparation of derivatives with high occupancy of a small number of sites. In this way it was sometimes possible to follow decreases and increases of the intensities of weak reflections that went through zero with increasing concentration, and this indicated a reversal of the signs of the reflections from the native and derivative crystals.

In accordance with the example provided by the work on myoglobin, 9° precession photographs, which provide data to a resolution just beyond 6 Å, were used for these trials, and attention was concentrated on the [001] and [100] zones of reflections.

When these exploratory studies had produced a promising derivative, further precession photographs were taken for use in intensity measurements. For this purpose, two Ilford X-ray films were placed one behind the other in the camera cassette and they were exposed for about 24 h to Cu Kα radiation from sealed X-ray tubes or for about 4 h to the same radiation from a rotating-anode tube. A second exposure with two films in the cassette was made for about 4 h with the sealed tube or 1 h with the rotating-anode tube in order to cover the full range of intensities that had to be measured. The intensity measurements were performed on the Joyce–Loebl recording densitometer, which scanned each row of reflections automatically but had to be moved manually from row to row. The heights above background of the peaks on the densitometer traces were measured with a ruler and these measurements provided the basic intensity data. Intensities recorded on the two films of each pair were brought to the same scale by calculation and application of a film transmission factor (usually between 2.5 and 3.0) and the corresponding factor relating films exposed for different lengths of time was obtained similarly. Weighted mean intensities were then calculated for each of the reflections. Lorentz–polarization (Lp) factor corrections were derived from a plot of the Lp factor against [(\sin^{2} \theta/\lambda^{2})], without consideration of asymmetric effects (Waser, 1951[link]), and these factors were manually applied to the observed intensities to provide structure-factor measurements on an arbitrary scale.

The structure factors of the heavy-atom derivative crystals, [|F_{HP}|], were scaled to those of the native protein, [|F_{P}|], by a factor K, derived from the equation [K^{2} \textstyle\sum\displaystyle |F_{HP}|^{2} = \textstyle\sum\displaystyle |F_{P}|^{2}S,]where S was a parameter, between 1.0 and 1.10, that depended on the assumed heavy-atom occupancy and was refined later in the process. Values of [\Delta F = K|F_{HP}| - |F_{P}|] were then calculated for use in the determination of the heavy-atom positions. Determination of heavy-atom positions

| top | pdf |

The high symmetry of the space group was greatly to our advantage, since the heavy-atom positions could be determined from difference-Patterson projection on the (100), (110) and (001) planes. In principle, all three coordinates of a heavy atom can be determined from projections on (100) or (110) alone. In practice, however, it was more straightforward to begin with the interpretation of the simpler projection on (001) before determining the z coordinate of the heavy atom from one of the other projections.

Apart from the effect of cross-over terms, which were sometimes detected as indicated above, these maps are effectively true Patterson maps of the heavy-atom structures of the derivatives. These difference-Patterson maps were calculated on the MERCURY computer and even from time to time by the use of Beevers–Lipson strips – a less demanding task than it might appear, since only about 80 hk0 and 60 0kl reflections are included within the 6 Å limit. The mercuri-iodide (K2HgI4) derivative

| top | pdf |

After trying several levels of substitution, RJP used the K2HgI4 salt at a molar concentration eight to ten times that of lysozyme. Based on the fact that lysozyme contains two methionine residues per molecule, and in keeping with a suggestion of Bluhm et al. (1958)[link], RJP was expecting to see two heavy-atom sites,1 but the hk0 difference-Patterson map was interpretable in terms of a single site of substitution. This site, however, was very close to the crystallographic twofold axis that runs along a diagonal in the [001] projection of the unit cell, and it proved necessary to correct the details of the first interpretation when phase information became available from other derivatives. It then appeared that there was one [\hbox{HgI}_{4}^{2-}] (or [\hbox{HgI}_{3}^{-}]) on the twofold axis between two protein molecules, but that it was best modelled by two closely spaced sites to allow for the elongated shape of the group (see Table[link]. Several other heavy-atom salts, including K2HgBr4, K2PtBr4 and K2AuCl4, gave derivatives in which the heavy atom was attached to the same site as K2HgI4, and consequently seemed not to provide useful additional phase information. The palladium chloride (K2PdCl4) derivative

| top | pdf |

An attempt to use K2PtCl4 to produce a useful derivative gave disordered crystals, but a substitute for it was found by soaking crystals in K2PdCl4 at a molar ratio of 3:1 relative to lysozyme. Despite the relatively light Pd atom, this derivative gave an easily interpretable difference-Patterson map (see Fig.[link] that yielded very good R factors, [R = \textstyle\sum\displaystyle \|F_{HP}| - |F_{P}\| - |F_{H}\hbox{(calc)}|/\textstyle\sum\displaystyle |F_{H}\hbox{(calc)}|,]where the summations are over centric reflections only.


Figure | top | pdf |

Difference-Patterson h0l projection map for the derivative obtained with K2PdCl4. The ends of heavy-atom double-weight (filled circles) and single-weight (open circles) vectors are shown (R. J. Poljak, unpublished material). The o-mercurihydroxytoluene p-sulfonate (MHTS) derivative

| top | pdf |

The hk0 difference-Patterson map of the p-mercuribenzene sulfonate (PCMBS) derivative was interpretable in terms of a single site of substitution at 8 Å resolution, but it was not useful beyond about 8 Å because of lack of isomorphism. RJP and RHF then explored the usefulness of MHTS as a derivative. This compound had been specially synthesized by JWHO in the hope that a small rearrangement of groups present in PCMBS would lead to an isomorphous derivative. Happily, this strategy worked, and MHTS gave a useful isomorphous derivative in which the major site overlapped that of PCMBS (Fig.[link].


Figure | top | pdf |

Difference-Patterson hk0 projection for the derivative obtained with MHTS. The large peak at [{1 \over 4}], [{1 \over 4}] is not explained by this solution (Fenn, 1964[link]). Other potential derivatives

| top | pdf |

As is usual in protein work, RJP tried many other heavy-atom compounds (Poljak, 1963[link]), but none gave useful results. In particular, a uranyl derivative, obtained by the use of UO2NO3, gave difference-Patterson maps that were difficult to interpret, and it was not taken further at this stage. Refinement of heavy-atom parameters

| top | pdf |

Refinement of the heavy-atom parameters was first performed by the use of Rollett's (1961)[link] least-squares program on the MERCURY computer, using the [|\Delta F|] values as structure amplitudes. This procedure gave satisfactory results for the K2HgI4, K2PdCl4 and MHTS derivatives described above, and they were used, therefore, in an attempt to determine the structure of the protein to 6 Å resolution in three dimensions. Analysis in three dimensions

| top | pdf | X-ray intensity measurements

| top | pdf |

We had three options for the collection of three-dimensional data. First, we could have used precession photographs and densitometry, as in the study of myoglobin (Kendrew et al., 1960[link]). Second, the manually controlled three-circle diffractometer used to make some of the measurements in the 6 Å study of myoglobin (Kendrew et al., 1958[link]) and haemoglobin (Cullis et al., 1962[link]) was available and, third, there was the prototype linear diffractometer (Arndt & Phillips, 1961[link]). We chose the last option, since CCFB, RHF and DCP were well experienced in using the instrument, and it offered the opportunity of measuring the protein reflections automatically and in a relatively short time compared with the other methods. Before he went on leave to MIT, ACTN had written a computer program for the University of London MERCURY computer to process the diffractometer data (North, 1964[link]) and, on his return in September 1961, he readily accepted an invitation to join the team to continue with this and other related aspects of the work.

The design of the linear diffractometer was based directly on the reciprocal-lattice representation of the genesis of X-ray reflections. The principle is illustrated in Figs.[link] and (b)[link], which show the familiar Ewald construction. YXO represents the direction of the incident X-ray beam with X the centre of the Ewald sphere and O the origin of the reciprocal lattice. A′OA, B′OB and C′OC are the principal axes of the reciprocal lattice, here assumed to be orthogonal. XP is the direction of the reflected X-ray beam corresponding to the reciprocal-lattice point P, which lies on the surface of the sphere of reflection. The reciprocal lattice can be rotated about the axis C′OC, and this axis can be inclined to the direction of the incident X-ray beam by rotation about the axis D′OD, which is perpendicular to the incident beam.


Figure | top | pdf |

Reciprocal-space diagrams showing the direction of the incident X-ray beam, the Ewald sphere and the genesis of a reflection (a) in an equatorial plane and (b) in the equi-inclination setting. Principal reciprocal-lattice directions are shown as thick lines. They also represent the slides in the diffractometer. The rotation of the diffractometer slide system about the axis C′OC is coupled to the rotation of the crystal about the axis R′XR by gears, pulleys and steel tapes. The counter arm of the diffractometer is represented by the fixed link XP = XO. Reproduced with permission from Arndt & Phillips (1961[link]).

The linear diffractometer was simply a mechanical version of this diagram. The reciprocal lattice was represented by three slides, A, B and C, which were parallel, respectively, to A′OA, B′OB and C′OC. They were mounted to rotate about the axis C′O and arranged so that the saddle P could be set to any position in space within the coordinate system that they defined. This saddle P was connected to the point X by means of a link of fixed length, XP = XO, corresponding to the radius of the sphere of reflection. The link XP always lay along the direction of the reflected X-ray beam and thus became the counter arm of the diffractometer. The crystal was mounted at X for rotation about the axis R′XR (independent of the link XP, which pivoted about an independent coaxial bearing at X). The rotation of the crystal about this axis was coupled by means of gears, pulleys and steel tapes to the rotation of the slide system about the axis C′OC. The axes R′XR and C′OC, held parallel by means of parallel linkages, could be tilted with respect to the incident X-ray beam by rotation about the axes D′OD, E′XE, as shown in Fig.[link].

The scale of the instrument clearly depended only on the length chosen for XP = XO. In the instruments used in the lysozyme work, this length, which is equivalent to one reciprocal-lattice unit, was five inches. The position of the saddle P on the three slides was controlled by means of lead screws, all of which were cut with 20 turns per inch. Hence the counters, which indicated revolutions and fractions of a revolution of the lead screws, read directly in decimal divisions of reciprocal-lattice units. The screws in slides A and B were driven by means of synchro-receiver motors, forming a synchro link with corresponding transmitters in the control panel. Slide C was set manually, together with the inclination angle μ, for the measurement of upper-level reflections in the Weissenberg equi-inclination mode, Fig.[link].

The coupling of the rotations of the crystal and reciprocal lattice about the axes R′XR and C′OC, respectively, was interrupted by two ancillary mechanisms. The first simply allowed for independent rotation of the crystal with respect to the slide system and was used for setting the reciprocal-lattice axes in the equatorial plane parallel to slides A and B, and for any fine adjustment of the crystal rotation that might be necessary during the measurement procedure. The second interruption consisted of a mechanism for oscillating the crystal about the position for any reflection while X-ray intensity measurements were made. This oscillation mechanism (Arndt & Phillips, 1961[link]) rotated with the crystal as the diffractometer was being set to a reflection position, and then controlled the independent motion of the crystal for the measurement of the integrated intensity of the reflection. The crystal remained stationary at a given angular setting for time t, was rotated at a uniform rate over a pre­determined angular range for a time 2t, remained stationary at the final angular setting for a further time t, and then returned quickly to its original setting. The correct setting for the reflection peak was at the midpoint of the rotation, which might be set to be through any angle from 1 to 5°. For initial adjustments, the motor could be arrested at this midpoint by means of a micro-switch operated by a switching disc rotating with the cam. This disc otherwise actuated contacts that started and stopped the intensity measurements.

The X-ray intensities were measured with a side-window xenon-filled proportional counter made by 20th Century Electronics, together with associated amplifiers and pulse-counting circuits (Arndt & Riley, 1952[link]). The proportional counter had a high quantum efficiency for the measurement of Cu Kα radiation (about 80%) and, when the operating potential and pulse-height discriminating circuits were carefully set, it provided useful discrimination against radiation of other wavelengths. The output from the proportional counter and its associated circuitry was fed directly to a teleprinter, which gave both a plain-language print out and a five-hole punched paper tape for input to the computer. Each count was provided with a check digit derived by a `ring-of-three' circuit, wired in parallel with the main electronic counter. During data processing, the check digit was compared with the count modulo 3: inequality of the two numbers was taken to indicate an error in the counting circuit.

Three counts were made: the first was a background count n1, made while the crystal was stationary on one side of the reflection position; the second was an integrated intensity count N, accumulated as the crystal rotated through the reflection; and the third was a further background count n2. The background-corrected integrated intensity of the reflection was taken to be [N_{o} = N - (n_{1} + n_{2}).]

At this stage of the work, measurements were made of one reciprocal-lattice level at a time in the equi-inclination mode that has no blind region at the centre. In each level, the diffractometer was driven to a reflection hkl (for example) at the limit of resolution of the data set to be collected. The diffractometer then moved in a series of equal steps along the scanning slide. At the end of each step, the oscillation mechanism took control for the measurement of intensity and background. After each measurement, a further step was taken on the scanning slide, and the process continued until a limit switch, set to the required resolution limit, was reached. The diffractometer then completed the current translation, measured the last reflection in that row, and then moved one step on the stepping slide to the next parallel row. This row was scanned, in the opposite direction, until the limit switch was reached again. In this way, the whole of a reciprocal-lattice level could be measured. In order to change to another level, the vertical slide C and the inclination angle μ had to be manually adjusted.

This account ignores two difficulties, one inherent in the design of the diffractometer, and the other specific to the lysozyme crystals. First, the instrument required a good deal of supervision, since it did not set itself very well for the measurement of low-angle reflections. Second, the crystals were not easily mounted so that the c axis, the most convenient axis for efficient data collection, since it is perpendicular to the most densely populated reciprocal-lattice planes, coincided with the crystal-rotation axis of the diffractometer.

The first problem was overcome by efficient teamwork and was much eased by the fact that RHF assumed responsibility for the MHTS derivative as part of her PhD work; the second was solved by making most of the measurements from crystals mounted to rotate about the [100] axis. These crystals were oriented so that the [b^{*}] and [c^{*}] axes were parallel to the horizontal slides of the diffractometer, and the measurements were made in levels of constant H by scanning along rows parallel to [b^{*}] and stepping to adjacent rows along [c^{*}]. A number of reflections could not be measured in this way, however, because reflections near the [a^{*}] axis were too broad to measure, particularly in the upper reciprocal-lattice levels. This difficulty was overcome by mounting some crystals with the c axis of the tetragonal crystals perpendicular to the length of the capillary tube, with the [110] axis parallel to the tube. These specimens were then mounted on a right-angled yoke so that the capillary tube was perpendicular rather than parallel to the goniometer axis (Fig.[link].


Figure | top | pdf |

Crystal mounting. (a) Rotation about the [a^{*}] axis; (b) rotation about [110], preliminary to [c^{*}] mounting; (c) rotation about the [c^{*}] axis (elevation); and (d) rotation about the [c^{*}] axis (plan).

Given the morphology of the crystals, with an essentially square habit bounded by {110} faces (Figs.[link] and[link]), all the reflections in the hkl octant could be measured in levels with constant L values without inclining the capillaries by more than about 40° to the X-ray beam. The horizontal slides of the diffractometer were set to be parallel to the [a^{*}] and [b^{*}] axes of the crystal. Some quadrants of hkL reciprocal-lattice levels were then scanned along rows parallel to the [a^{*}] axis and stepped along the [b^{*}] axis. Enough measurements were made in this mode to cover the `blind' region in the Hkl levels and provide an appropriate number of intersecting levels for scaling all the measurements into a consistent set.

Care was taken during all these measurements to index the reflections in a right-handed system of axes. Given the transparent relationship between the slide system of the diffractometer and the crystal geometry, this was easily accomplished, and it was necessary for the subsequent use of anomalous scattering (Bijvoet, 1954[link]) in the phase determination.

During the measurements from the native and derivative crystals mounted for rotation about [100], the variation in peak intensity of the 200 reflection with ϕ, the angle of rotation about the axis C′OC (Fig.[link], was also recorded (Fig.[link]. (200 is the lowest-order reflection available for this purpose in this space group.) These records were then used in the data-processing stage to correct the measurements for absorption by the method described by Furnas (1957[link]). According to this method, the absorption suffered by the incident and reflected beams for any reflection hkl is indicated by the relative intensity of 200 when the X-ray beams are parallel to the reflecting planes hkl. This is a fair approximation for reflections at low angles. This method could not be used, however, for the measurements made from crystals mounted to rotate about the [001] axes, since a full rotation about this axis was not possible. These measurements were not corrected for absorption errors.


Figure | top | pdf |

Absorption curve. Variation of relative transmission [T(hkl) =] [ I(\varphi_{hkl})/I_{max}(\varphi )] [ [= 1/A(hkl)]] with rotation angle ϕ for the 200 reflection and the crystal rotating about the normal to (200). Solid line: measured curve; broken line: calculated curve, neglecting effect of mother liquor and capillary. Reproduced with permission from North et al. (1968[link]).

Finally, the diffractometer was reset manually at regular intervals during data collection to measure the intensities of a number of reference reflections. These measurements were used to monitor the stability of the system and the extent of irradiation damage to the crystal, and they were also recorded on the paper-tape output for analysis by the data-processing program. An attempt was made to minimize irradiation damage by using a shutter to expose the crystal only during the measuring cycle. Data processing

| top | pdf |

The measurements N, n1 and n2, together with the indices of the reflections, hkl, were all printed out in plain language on a teleprinter and punched in paper tape for direct transfer to a computer (Fig.[link]. The plain-language record was important during measurement of the low-angle reflections, when the diffractometer had to be adjusted by hand. Not all imperfections in the measurements were easily spotted at this stage, however, and ACTN's data-processing program (North, 1964[link]) therefore incorporated systematic checks on the quality of the measurements.


Figure | top | pdf |

Typical output from the linear diffractometer. (a) Indices h, k, l followed by background (n1), peak (N), background (n2) counts. (b) Listing ready for the next stage in data processing with indices * h k followed by l, background corrected peak and standard deviation. Reproduced with permission from North (1964[link]). Copyright (1964) Institute of Physics.

The program checked for the following contingencies:

  • (1) malfunction of the diffractometer-output mechanism leading to the paper tape being an inaccurate record of the measurements, generally because the tape punch had failed to perforate the tape or had `stuttered';

  • (2) errors by the pulse counters, detected by the `ring-of-three' circuit;

  • (3) peak counting rate so high that counting-loss errors were appreciable;

  • (4) count on reflection not significantly above background;

  • (5) failure of diffractometer to set crystal or counter correctly; and

  • (6) gradual drift in the experimental parameters, including movement of the crystal within its mounting and irradiation damage to the crystal.

These checks were made while the diffractometer tape was being read into the computer, and a monitor output was produced simultaneously, as shown in Fig.[link]. The checks depended in large part on the fact that the significance of an intensity measurement may be assessed in terms of counting statistics. The standard deviation of a background-corrected count, [N_{o}] [(= N - n_{1} - n_{2})], is given by [\sigma^{2}(N_{o}) = N + n_{1} + n_{2}], and the ratio [\sigma (N_{o})/N_{o}] may be taken as an indication of the significance of the measurement. Measurements were rejected when this ratio exceeded unity. [N_{o}] might then have been taken as zero but, following Hamilton (1955)[link], we considered it preferable to replace [N_{o}] by a fraction (0.33 for centric and 0.5 for acentric reflections) of the minimum background-corrected count that we should have considered acceptable. Reflections were treated in the same way whether the net count [N_{o}] was positive or negative, but measurements were rejected if [N_{o}] was negative and [|N_{o}|\, \lt\, \sigma (N_{o})].


Figure | top | pdf |

Format of monitor output in which the computer lists reflections that fail the tests for format or significance. PE1 signifies punching error, indices; PE2, punching error, measurements; PE3, failure of electronic check on counting circuits; SD, standard deviation greater than set limit; N-, net count negative; BG, backgrounds significantly different; N > H, gross counts exceed counting-loss limit. This output was from the version of the program designed to be used with the diffractometer fitted with three counters. The symbols: −, =, + refer, respectively, to reflections measured by the lower, central and upper counters. Reproduced with permission from Arndt et al. (1964[link]). Copyright (1964) Institute of Physics.

Mis-setting of the crystal was frequently revealed by marked inequality of the background counts. Measurements were therefore rejected if the difference between the two backgrounds exceeded three or four standard deviations, that is if [(n_{1} - n_{2})^{2} \,\gt] [ b^{2}(n_{1} + n_{2})], where b is the appropriate constant.

After monitoring the quality of the data in this way, the program proceeded: (i) to extract background-corrected counts; (ii) to apply a correction for irradiation damage derived from any systematic variation in the intensities of the reference reflections; (iii) to sort the reflections into a specified sequence of indices; (iv) to apply Lorentz–polarization factors; and (v) to apply absorption corrections (the data for which were read separately from a specially prepared punched tape, Fig.[link]). The outputs from this program comprised data sets from a number of individual crystals of the native protein and the three derivatives. The scale factors needed to bring the measurements from the individual crystals of each crystal species to a common scale were determined from the intensities in common rows by the use of a program written by Rollett (Rollett & Sparks, 1960[link]). The scaling factors were applied and the data merged, a weighted-average intensity being determined when more than one estimation was available. The absolute scale of the intensities

| top | pdf |

An attempt was made to determine the absolute scale of the measured intensities by comparison with the intensities diffracted by anthracene, a small organic crystal of known structure. This method had worked well in a determination of the absolute scale for seal myoglobin (Scouloudi, 1960[link]), but it did not give a satisfactory result with lysozyme, mainly because of the difficulty of measuring the crystal volumes precisely enough. Accordingly, we used Wilson's (1942)[link] method to provide an estimate of the absolute scale of the intensities, knowing very well that it does not give an accurate estimate for protein data, especially at low resolution. Nevertheless, this scale gave reasonable values for the occupancies of the heavy-atom sites. Re-assessment of heavy-atom derivatives

| top | pdf |

Given the three-dimensional data to 6 Å resolution for the native crystals and the three derivatives, it was next possible to calculate three-dimensional difference Patterson maps for the derivatives using the terms [|\Delta F|^{2} = ||F_{PH}| - |F_{P}||^{2}]as coefficients in the Fourier series. This synthesis, which is now well known in protein-structure analysis, gives a modified Patterson of the heavy-atom structure in which the heavy-atom vectors appear at reduced weight in a complex background (Blow, 1958[link]; Phillips, 1966[link]). Nevertheless, the [|\Delta F|^{2}] maps for PdCl4 and MHTS were readily interpreted in terms of single heavy-atom substitutions, particularly because the vectors involved were all confined to defined Harker sections (Fig.[link].


Figure | top | pdf |

Three-dimensional [|\Delta F|^{2}] syntheses for the PdCl4 and MHTS derivatives. The positions of peaks in the Harker sections are marked and numbered (Fenn, 1964[link]).

The map of the [\hbox{HgI}_{3}^{-}] derivative was not so satisfactory, and this led to the discovery, during refinement of the heavy-atom parameters, that the mercury occupancy declined as a function of irradiation time. Consequently, the [\hbox{HgI}_{3}^{-}] data from two individual crystals were treated separately during the remaining stages of the analysis.

The least-squares refinement mentioned in Section[link] was not wholly satisfactory in that it included no provision for refining heavy-atom occupancy. Accordingly, DCP – with some help from ACTN – wrote a computer program for the MERCURY computer based on Hart's (1961[link]) method, in which the heavy-atom positions, occupancies (O) and temperature factors (B) were refined simultaneously together with the scale factors (S) between heavy-atom and native structure factors. All the centric reflections from the [100], [110] and [001] zones were included in this refinement.

The quantity minimized for each derivative was [R' = \textstyle\sum\displaystyle (||F_{PH}| - |F_{P}|| - |F_{H}|)^{2},]where [F_{H}] is the calculated heavy-atom contribution to the derivative structure factor. The method involves calculating R′ for each parameter at its current value pn and at values [p_{n} \pm \Delta p_{n}] and [p_{n} \pm 2\Delta p_{n}], where each of the parameters is shifted in turn, the shifts having been specified, while the other parameters are at the unshifted value. Thus for each of the parameters (x, y, z, O, B and S), four values of R′ are obtained for the shifted values plus the value of R′ for the unshifted parameters, the latter being denoted [\varepsilon_{u}].

Let the minimum value of R′ from all the calculations be [\varepsilon_{\rm min}] and for the parameter pn, the minimum of the list of five values of R′, be [\varepsilon_{qn}], corresponding to the value qn of pn. Then, according to the method of steepest descents, the shift to be applied to the parameter pn is [(q_{n} - p_{n})(\varepsilon_{u} - \varepsilon_{qn})/(\varepsilon_{u} - \varepsilon_{\rm min}).]

If [\varepsilon_{qn} = \varepsilon_{u}], that is, if the unshifted value of the parameter gave the minimum value of R′, then the shift was divided by 4 for the next cycle. Otherwise the shift was kept constant. Thus the new parameters and shifts were determined for the next cycle of refinement, and the process was repeated until convergence.

This program worked well, and RJP, who was reading Candide at the time, named it Pangloss – it gave the best possible values for the heavy-atom parameters. These values, which include two separate sets for the [\hbox{HgI}_{3}^{-}] derivative, are shown in Table[link]. At this stage, an important check was carried out. The coordinates of the heavy-atom site in each derivative were referred to an origin at the junction of a twofold axis and a twofold screw axis. However, there are four such intersections in the unit cell and, in order to ensure that the same origin had been chosen for each derivative, the sign predictions for the centric reflections from each derivative – which were checked by hand throughout this exploratory stage – were compared. They agreed well, thus establishing that the choice of origin was the same for each derivative.

Table| top | pdf |
Heavy-atom parameters used in the final phase calculation for the lysozyme structure

E is the average difference between observed and calculated heavy-atom changes of centric reflections (electrons); R is the reliability index for observed and calculated heavy-atom changes of centric reflections and R′ is the Kraut (Kraut et al., 1962[link]) agreement index for all reflections.

Crystal 1Crystal 2
x 0.147 0.218 0.131 0.178 0.125 0.178
y 0.841 0.620 0.869 0.822 0.875 0.822
z 0.963 0.054 0.250 0.250 0.250 0.250
Occupancy (e) 72 47 74 74 72 72
B2) 52 −18 38 38 111 111
E (e) 57 66 66 55
R (%) 35 49 42 39
R′ (%) 10.7 11.6 11.6 12.2
Centric 0, 1, 2kl reflections only.
hk0 reflections only. Phase determination at 6 Å resolution

| top | pdf |

These parameters were used to determine the phases of the protein reflections. A proportion of these phases were first determined by the graphical method suggested by Harker (1956)[link], which had been used in the 6 Å stage of sperm-whale myoglobin (Kendrew et al., 1958[link]). We treated the process as a group activity in which different individuals took responsibility for reading out the various Harker components [[F_{P}], [F_{PH}], [F_{H}](calc) etc.] for each reflection in turn. This was a useful bonding exercise and improved our familiarity with the rather cosmopolitan accents in use within the group. Scientifically, it was also an encouraging experience since it showed that reasonably consistent results could be obtained from the three different derivatives, and that the anomalous-scattering measurements were capable of making a significant contribution to the phase determination. It soon became clear that the most significant anomalous contributions had to be included in such a way as to retard the phase of the derivative structure factor if these indications were to agree with the phase predictions derived from the isomorphous differences alone (Fig.[link].


Figure | top | pdf |

Phase determination for reflection 424. In this Harker diagram, the heavily traced protein circle (with radius [|F_{P}|]) is labelled P. The circles with radii [|F_{HP}|] obtained with PdCl4, MHTS and HgI3 are shown. Values of the anomalous-scattering pairs hkl and khl were used for PdCl4 and MHTS. The position of the protein vector, weighted by its figure of merit, is shown as a small open circle (R. J. Poljak, unpublished results).

The implication of this observation was that the space group is [P4_{3}2_{1}2] rather than [P4_{1}2_{1}2], that is, the fourfold screw axis is left handed.

These results encouraged us to go ahead with the computer calculation of phases by the phase probability method applied to the isomorphous differences (Blow & Crick, 1959[link]; Dickerson et al., 1961[link]) and to the anomalous-scattering differences (Blow & Rossmann, 1961[link]). This program was written by ACTN and it was used first to confirm the space-group identification. Overall, the mean figure of merit obtained with the anomalous contribution consistent with [P4_{3}2_{1}2] was somewhat higher than the alternative, though to a lesser extent than we had anticipated. This observation led at a later stage to reconsideration of the way in which the anomalous scattering was incorporated in the phase determination (see Section[link]).

The quality of the phase determination was indicated by the figure of merit. For the acentric reflections, this was 0.79, an encouraging result since it compared favourably with that obtained in the low-resolution study of haemoglobin (Perutz et al., 1960[link]). The mean figure of merit for the centric reflections, on the other hand, was 0.95, so that the overall value was 0.86. A check on the sign predictions for centric reflections showed that the three derivatives gave satisfyingly similar results.

These phases were also used to calculate difference-Fourier maps, showing the heavy atoms in the derivatives by the use of coefficients [|\Delta F| = ||F_{PH}| - |F_{P}||]associated with the protein phases and weighted by the figures of merit of each phase determination. These difference maps in three dimensions are shown in Fig.[link].


Figure | top | pdf |

Electron-density difference syntheses showing the main sites of substitution and subsidiary sites with low occupancy in the PdCl4 and MHTS derivatives (Fenn, 1964).

They revealed the presence of small subsidiary sites in these two derivatives, but these minor sites were not taken into account. The electron-density map of lysozyme at 6 Å resolution

| top | pdf |

The electron-density distribution was calculated on the MERCURY computer, by means of a program written by Owen Mills that was in general use at the time, with structure amplitudes weighted by the figures of merit so as to give the `best' Fourier (Blow & Crick, 1959[link]). The map was contoured by hand and plotted on clear plastic (Perspex; Plexiglass in the USA) sheets (Fig.[link].


Figure | top | pdf |

Electron-density distribution in lysozyme at 6 Å resolution viewed parallel to the c axis. The horizontal and vertical lines represent the twofold rotation axes and intersect the twofold screw axis, upper right of centre. The fourfold screw axis is at the lower left of centre. The contour interval is 0.07 e Å−3, the lowest heavy contour being at 0.6 e Å−3. The absolute scale is approximate. Reproduced with permission from Nature (Blake et al., 1962[link]). Copyright (1962) Macmillan Magazines Limited.

The first objective in studying this map was to determine the boundary of a single molecule. Comparison of the unit cells of various crystal forms of lysozyme (Steinrauf, 1959[link]) suggested that in tetragonal lysozyme, the molecule occupies the full length of the c axis and, on average, one-eighth of the ab plane. On this basis, the map of Fig.[link] includes the whole of the c axis and a sufficient area of each section to ensure that one whole molecule is included in addition to parts of neighbouring molecules. Only contours indicating where the electron density is greater than average are included.

Some featureless regions of average electron density (0.4 e Å−3) were immediately apparent. The most marked of these was around the twofold screw axis parallel to c. This axis is intersected at intervals of 9.5 Å by twofold rotation axes, and packing considerations, therefore, made it impossible for substantial parts of the molecule to penetrate into the neighbourhood. Similarly, the immediate vicinity of the fourfold screw axis was also without significant features and was clearly a region of intermolecular space. The twofold rotation axes also helped to determine the boundary, particularly where relatively high density approached or intersected them, as it did in two places. It was clear that such regions must represent close contacts or bridges between adjacent molecules.

Following the example of the haemoglobin study (Perutz et al., 1960[link]), we decided at this stage to make a balsa-wood model of the electron density to help visualize the molecule. Instead of producing a stack of sections through the molecule, however, CCFB devised a way of shaping the sections to make a smooth model of the volume occupied by electron density greater than about 0.53 e Å−3. The result is shown in Fig.[link].


Figure | top | pdf |

Views of a 6 Å resolution model of the regions in which the electron density exceeds about 0.53 e Å−3. The vertical rods indicate the twofold and fourfold screw axes parallel to c. The thinner horizontal rods are twofold rotation axes. At two places, shown hatched, rotation axes pass through continuous regions of density. The white regions alone make up the most compact asymmetric unit of the structure. In part (b), the grey section (B) is alternative to the region (A), to which it is related by the twofold axis T. In part (a), the grey section (D) is alternative to the white (C), related by the fourfold screw axis. The grey piece (F) in part (c) is alternative to the white (E), to which it is related by a unit-cell translation along c. The scale is indicated by the framework of symmetry elements, adjacent parallel twofold axes being 18.95 Å apart. Reproduced with permission from Nature (Blake et al., 1962[link]). Copyright (1962) Macmillan Magazines Limited.

One asymmetric unit is shown in white, with additional pieces in grey to show alternative shapes. The white pieces together make up the most compact asymmetric unit, which is roughly ellipsoidal in shape with axes 52 × 32 × 26 Å. Later work showed that this asymmetric unit represented a single molecule but, at this stage, we were scrupulous in detailing the alternative interpretations. There is a region of low density that divides the model roughly into two halves, and although we speculated about this we refrained from making any suggestions about its possible significance in our description of the structure (Blake et al., 1962[link]). Instead, we noted that the two halves of the model could be assembled differently, following the crystal symmetry, so as to form a dumb-bell shaped molecule connected at PP′ (Fig.[link].

Our second objective was to determine as far as possible the course of the polypeptide chain and the positions of the disulfide bridges. This proved to be impossible. In comparison with the maps of myoglobin (Kendrew et al., 1958[link]) and haemoglobin (Perutz et al., 1960[link]) at this resolution, it was immediately apparent that this map of lysozyme had a much smaller proportion of clear-cut rod-like features representing α-helices. This was not a surprise, since optical rotatory dispersion measurements (Yang & Doty, 1957[link]) suggested that only 30–40% of the polypeptide chain in lysozyme is in the form of α-helix, as compared with 77% in myoglobin. In addition, Hamaguchi & Imahori (1964[link]) had distinguished the presence of a region of β-sheet in lysozyme before completion of the X-ray analysis. The task of tracing the polypeptide chain, which was difficult with myoglobin, was impossible with lysozyme, since the connectivity of the non-helical regions was often not discernible. The existence of four disulfide bridges, which were expected to have about the same electron density as helices at this resolution, complicated the problem further.

Accordingly, we concluded that defining the shape of the molecule and its tertiary structure would have to await further studies at higher resolution. Meanwhile, Corey and his colleagues (Stanford et al., 1962[link]) and Dickerson et al. (1962)[link] published interim accounts of their work at the same time as our work was published (Blake et al., 1962[link]).

At this stage, in the autumn of 1962, RJP left for three months for the MRC Laboratory in Cambridge and then joined Howard Dintzis at Johns Hopkins. At about the same time, DFK joined the team to continue the analysis to high resolution, and LNJ joined DCP as a graduate student and began work related to the activity of the enzyme.

25.1.3. Analysis of the structure at 2 Å resolution

| top | pdf |

The structure of chymotrypsinogen (Kraut et al., 1962[link]) at 6 Å resolution was published a few months before the corresponding work on lysozyme. Compared with the work on the globins, however, neither analysis yielded much information on the structures of these proteins, and protein crystallographers generally found them discouraging. Indeed, some went so far as to suggest that only the structures of proteins with a high α-helical content would be amenable to study by X-ray methods. Our reaction, however, as we turned our attention to extending the study of lysozyme to high resolution, was to resolve that each step in the analysis must be conducted as well as possible. We considered carefully, therefore, what improvements might be made to the methods employed hitherto. In particular, we recognized that further work would be needed to identify heavy-atom derivatives suitable for use at high resolution, and we sought improvements in the methods used for data collection, the correction of absorption errors and the use of anomalous scattering in phase determination.

At a purely practical level, one of our major concerns was that our limited access to the London University computer would lead to serious delays in the next stage of the work, which was bound to be even more dependent on computing than the work at 6 Å. Accordingly, we sought support from the Medical Research Council (MRC) for the acquisition of a laboratory-based computer that would be able to handle the computations up to but not including the calculation of a high-resolution electron-density map. Happily, the MRC provided a grant for an Elliott 803B computer, which was installed in the laboratory in March 1963. At the same time, the MRC provided a grant to purchase the commercial version of the linear diffractometer, which was manufactured by Hilger & Watts, Ltd. Since the Elliott 803B had not previously been used for crystallographic computing, this change in our computer involved many members of the laboratory in new programming. Heavy-atom derivatives at 2 Å resolution

| top | pdf |

The potential usefulness of the three derivatives used at 6 Å resolution for phasing a higher-resolution map was analysed by CCFB and DFK. As the MHTS derivative had much the highest R factor at 6 Å resolution, and the K2HgI4 derivative had problems of stability and structure, only the K2PdCl4 (PD) derivative seemed likely to be useful for phasing at higher resolutions. An immediate search for additional heavy-atom derivatives was therefore undertaken, which included a re-examination of uranyl nitrate, UO2(NO3)2 (UN). Together with other compounds, DFK obtained samples of the then novel [\hbox{UO}_{2}\hbox{F}_{5}^{3-}] ion (UF) from Reuben Leberman at Cambridge, which generated a different pattern of changes in the lysozyme diffraction pattern to any of the previous heavy atoms. When the native phases from the 6 Å map were applied to the UF changes in the centric hk0 and h0l zones, they showed a novel two-site binding pattern with a low R factor.

The UF and PD derivatives were examined at 2 Å resolution. Photographs of the centric hk0 and h0l zones were taken with a 23° precession angle, the intensities were measured on the Joyce–Loebl densitometer and corrected for Lorentz and polarization effects by a program written by CCFB for the Elliott 803B computer. The heavy-atom parameters obtained in the refinement of these derivatives at 6 Å were used as a starting set for the refinement at higher resolutions. This refinement, like that at 6 Å resolution, was carried out with the program (Pangloss) based on Hart's (1961[link]) method, but improved and re-written for the Elliott 803B by DFK.

Initially only the PD and UF derivatives were refined: the mercuri-iodide derivative was not seriously considered to be a potentially useful derivative at high resolution because of the loss of heavy atom with irradiation observed during collection of the 6 Å data. However, it appeared probable that a mean occupancy would be suitable for photographic data, and that the signs predicted by the derivative might be very useful when the other two derivatives gave weak or ambiguous predictions. In fact, the PD and UF derivatives on refinement at 2 Å resolution predicted sets of signs that disagreed to an unacceptable extent, with the PD derivative having a much higher R factor. Refinement of parameters defining the central mercury atom of the mercuri-iodide derivative, followed by calculation of a high-resolution difference-Fourier map using only the most clearly predicted signs, revealed a planar trigonal [\hbox{HgI}_{3}^{-}] ion, whose shape and size were similar to those found by Scouloudi (1965)[link] and Fenn (1964)[link].

Unfortunately, refinement of the mercury and iodine parameters produced a set of protein signs at variance with both previous sets. On the basis of the agreement between [F_{H}](calc) and [F_{PH} - F_{P}], both the PD and [\hbox{HgI}_{3}^{-}] derivatives appeared to be non-isomorphous with the native structure. The apparent inability of all three derivatives used at 6 Å to be of use in phasing reflections at higher resolution was disappointing. One possible explanation that was explored for the poor performance of the PD derivative at high resolution, when it showed large changes and a low R factor at 6 Å resolution, was the question of its structure. As with mercuri-iodide, the number of electrons in the halogen substituents exceeded the number in the central palladium atom. Much time was expended in trying to resolve the structure of the square-planar ion without success. This included examining all PdX4 and PtX4 complexes, where X is Cl, Br, or I, and calculating double-difference maps between PdCl4 and PdBr4, but without revealing any real indication of structure in the heavy-atom peak.

An extended search for alternative heavy-atom derivatives was therefore begun. Crystals for this purpose were grown as described above (Section[link]), except that, on the initiative of DFK, small polyethylene bottles were used instead of glass vials. The polyethylene bottles did not affect crystal size or quality, but allowed crystals to be detached from their surfaces without the damage that occurred when glass vials were used. The search was carried out by the diffusion method. A small number of crystals were isolated together with a known volume of mother liquor, and a solution of the heavy-atom compound, usually at a concentration to make 5 mM in the final solution, was put into a dialysis bag and suspended in the liquid above the crystals. After one or two days, one of the crystals was mounted and an offset 5° screenless precession photograph of the hk0 zone was taken. (In these photographs there is no overlap of the zero and first upper levels.) If the approximately 30 unique reflections in this zone showed little or no change in intensities, the crystal was usually not examined further, while those that showed some changes were set aside until a higher-angle precession photograph could be taken. By the use of this technique, a rapid survey of potential derivatives could be made, and those heavy-atom compounds that did not bind to the protein were rapidly dismissed.

Crystals that showed substantial differences in intensity with respect to the native enzyme were used to take hk0 and h0l precession photographs. In these studies, the precession angle was reduced to 15° in order to reduce the amount of data to be collected to more manageable proportions, but otherwise the data were collected as for the initial three derivatives. The heavy-atom parameters were also refined in the same way. However, to cope with the disagreements observed between the initial derivatives, a procedure was introduced to combine the refinement of individual derivatives with a refinement of the protein signs in the two centric zones, using all current derivatives together.

When a potential new heavy-atom derivative was identified, the procedure worked as follows. Difference-Fourier maps were calculated, initially at 6 Å and later, when most of the 2 Å signs had been established, at the maximum resolution of the derivative data. When a heavy-atom binding site had been located, its parameters were refined by Hart's method, and the consequences of the refinement, [F_{P}] and [F_{PH}] with their signs and the calculated [F_{H}] values, were compared with the corresponding results from other derivatives. The signs of each [F_{P}] predicted by the current derivatives were noted, and the most probable sign was determined for each reflection. Initially this was done by inspection, and later by the method of Blow & Crick (1959)[link], with E values provided by the Pangloss refinement. The `best' signs given by this procedure were used to calculate a double-difference map for each derivative with coefficients [[(F_{PH} - F_{P}) - F_{H} (\hbox{calc})]]. When these maps showed either new sites or some new feature at the sites previously included, appropriate alterations were made to the current model of that derivative and further refinement was carried out. If, on the other hand, a derivative showed a high background without interpretable features, it was omitted from further cycles and replaced by another derivative so that a group of four to six derivatives was always in use.

After each derivative had been examined in this way, the set of `best' signs was updated with the new models of the derivatives, and the procedure was repeated. This procedure worked very efficiently, rapidly indicated a non-isomorphous derivative – or rather one that was significantly less isomorphous than the best – and clearly showed features such as new sites, structure around previously included sites, incorrect or anisotropic temperature factors, and incorrect positional parameters or occupancies. Finally, when the five best derivatives were left in the list, it was immediately apparent, from inspection of the sign predictions in comparison with the E values, which derivatives should be used to phase the high-resolution map.

This method confirmed that MHTS, one of the derivatives used in the 6 Å map but then discarded, was useful at 2 Å resolution and revealed that two new derivatives, UN and UF, were also satisfactory. The poor performance of the MHTS derivative at 6 Å resolution was due to anomalously large changes in the very low resolution reflections, which are sensitive to salt concentrations, but it exhibited good iso­morphism at higher resolutions. At about this time, in the summer of 1963, DFK left the laboratory to take up an appointment at Brookhaven National Laboratory.

It became apparent during the course of the heavy-atom search and refinement that the number of suitable derivatives was severely restricted by a feature of the protein. This was the existence of a close pair of very strong binding sites, whose occupation was always accompanied by non-isomorphism. These sites, referred to by the initial derivatives that they bound as the [\hbox{HgI}_{3}^{-}] and PD sites, were found to have a strong affinity for all the complex halogen anions of PdII, PtII, PtIV, AuIII, HgII, OsIV, IrIII and IrIV that were tried. The sites were mutually exclusive, probably because they shared a protein side chain that acted as one of the important metal-binding groups. The [\hbox{HgI}_{3}] site bound [\hbox{HgI}_{3}^{-}], [\hbox{PtCl}_{4}^{2-}], [\hbox{PtBr}_{4}^{2-}], [\hbox{PdI}_{4}^{2-}], [\hbox{OsCl}_{6}^{2-}], [\hbox{IrCl}_{6}^{3-}] and [\hbox{AuCl}_{4}^{-}]. All these compounds caused disordering of the protein structure, as indicated by a decrease in intensities at high resolution, and at least two of the compounds were sensitive to X-irradiation. The PD site also bound HgCl2, [\hbox{PdBr}_{4}^{2-}], [\hbox{PtCl}_{6}^{2-}], [\hbox{PtBr}_{6}^{2-}] and [\hbox{PtI}_{6}^{2-}]. All these derivatives were seriously non-isomorphous with the native structure at medium to high resolution, but careful analysis suggested that [\hbox{PtCl}_{6}^{2-}] would be useful at low resolution. In contrast, the two uranium compounds, UN [which most probably gave rise to a bound UO2(OH)n cation] and UF, gave a substitution pattern entirely different from other derivatives and, in particular, avoided the two sites that gave so much trouble with other complex ions. This finding is in accord with their known tendency to complex with protein oxygens, as opposed to the protein nitrogens that form the binding sites of most other heavy metals.

The two mercury benzene sulfonates that were investigated, MHTS and PCMBS, had a common sulfonate site, but the mercury atoms in the two derivatives were found to be about 3 Å apart. This suggested that they were bound to the protein by their charged sulfonate groups. A similar orientation and position of the benzene ring was implied by the observations that the Hg–SO3 distance was 5.76 Å in the MHTS derivative, compared with 5.67 Å in the crystal structure of MHTS (Fenn, 1964[link]), and 7.00 Å in the PCMBS derivative, compared with 7.25 Å in the structure of PCMBS. Moreover, the angle Hg(MHTS)–SO3–Hg(PCMBS) was found in the complexes with lysozyme to be 32°, the same as that calculated from the mercurial structures. This indicates that the benzene sulfonate groups of these two derivatives were fixed in the same position in the lysozyme crystals, and the location of the mercury in the protein crystal depended solely on its position on the benzene ring.

It is interesting to notice that PCMBS, which was investigated first, caused the c axis of lysozyme to lengthen by 1.5% and consequently could not be used at high resolution, while the slightly different MHTS reduced the lengthening to only 0.25% and could be used. This was an early example of an engineered isomorphous derivative. Nevertheless, PCMBS seemed potentially useful at low resolution.

A total of about 50 compounds were used to prepare derivatives (Blake, 1968[link]). Many of these compounds were available commercially but some of the most important, including MHTS and [\hbox{UO}_{2}\hbox{F}_{5}^{3-}], were not. MHTS was designed and synthesised by JWHO, as described above, while the initial sample of [\hbox{UO}_{2}\hbox{F}_{5}^{3-}] was provided by Reuben Leberman, and further supplies were synthesized by CCFB.

In order to prove these new heavy atoms in three dimensions and also to eliminate any `cyclic' effect due to the use of Fourier methods, 6 Å three-dimensional data were collected early in 1964 for the UF, UN, MHTS, PCMBS and [\hbox{PtCl}_{6}^{2-}] derivatives. Two- and three-dimensional Patterson functions were calculated and interpreted afresh. These interpretations were wholly consistent with the results from the high-resolution projections, with the exception of the UN complex. At 3 Å, the hk0 and h0l projections showed only two sites, but at 6 Å there appeared to be three additional sites. This point was cleared up when refinement of the 2 Å data collected for phase determination showed that the three extra sites had very high temperature factors, which would result in these peaks being very low at 3 Å resolution. These five compounds were used to phase a second 6 Å electron-density map of the enzyme, as described below, in advance of the analysis at 2 Å, which was based on the only three derivatives that had been found to be reasonably isomorphous at high resolution: MHTS and the two uranyl derivatives, UF and UN.

The discovery of the two uranyl derivatives was not entirely accidental. In the preliminary study at 6 Å resolution, we had been impressed by the potential utility of anomalous scattering in phase determination and had noted the very high value of the imaginary component [(\Delta f'' = 16)] of the anomalous scattering by uranium (Dauben & Templeton, 1955[link]). Intensity measurements

| top | pdf |

A further advance in diffractometry arose from the observation by DCP that more than one reflection can be measured at the same time in the flat-cone setting of a diffractometer (Phillips, 1964[link]). In the flat-cone setting, the crystal axis is inclined to the X-ray beam as in the equi-inclination setting (Fig.[link], but the motion of the counter is confined to the plane perpendicular to the crystal axis. This flat-cone plane of the reciprocal lattice is midway between the zero-level and equi-inclination levels. When measurements are made in the flat-cone setting from crystals rotating about a reciprocal-lattice axis perpendicular to reciprocal-lattice planes, the crystal and counter settings for reflections in levels adjacent to the flat-cone level are identical to one another and closely similar to those for reflections in the flat cone. This setting is illustrated in Fig.[link].


Figure | top | pdf |

Perspective drawing of the sphere of reflection showing inclination geometry with simultaneous reflections (reciprocal-lattice points P and Q) from levels symmetrically related by the flat-cone setting. Reproduced with permission from Phillips (1964[link]). Copyright (1964) Institute of Physics.

This property of flat-cone geometry made it possible to modify the linear diffractometer so as to measure three reflections quasi-simultaneously (Arndt et al., 1964[link]). Even more reflections could be measured in this way from crystals with large unit cells but, at this stage, three 20th Century Electronics side-window proportional counters were mounted in echelon on the counter arm, with appropriate entrance slits and with their windows 0.75 cm apart. The distance of this array from the crystal could be varied between 22 and 50 cm so that the angular separation of adjacent counters lay between 0.034 and 0.015 rad. The separation of reciprocal-lattice levels in which quasi-simultaneous measurements could be made also varied, therefore, from 0.034 to 0.015 reciprocal-lattice units (r.l.u.'s), corresponding to crystal-lattice dimensions of 45 to 100 Å with copper Kα radiation.

The array of proportional counters could be set in position on the counter arm by means of horizontal and vertical fine controls. The reflected X-rays passed through adjustable slits before entering the counter windows, and the left and right sides of these slits, or their top and bottom halves, could be blocked to facilitate precise setting.

A disadvantage of using flat-cone as opposed to equi-inclination geometry is that reciprocal-lattice levels measured in a flat cone, and adjacent levels, have a blind region at their centre in which reflections are not accessible. With the linear diffractometer, this blind region falls within the region in which the diffractometer does not automatically set the crystal and counter precisely enough for reliable measurements. The problem is not severe for measurements at low resolution, and it was avoided during the measurement of high-resolution data by the operation of low-angle limit switches that controlled the operation of the scanning and stepping slides. Fortunately, the high symmetry of the lysozyme crystals greatly reduced the seriousness of this problem, since reflections in the blind region close to the rotation axis could usually be measured as symmetry-equivalent reflections in the measurable region.

The first measurements made by this method were 6 Å data for the native protein and the five derivative crystals that had been identified as giving useful phase information at low resolution. Crystals were first mounted to rotate about their a axes, and the crystal-to-counter distance was 38.5 cm. Unfortunately, this long crystal-to-counter distance gave rise to weak measured intensities, because of the significant absorption of the reflected X-rays by the air in the counter arm. For this reason we considered filling the counter arm with hydrogen, but came to the conclusion that a simpler approach would be to make the bulk of the measurements from crystals mounted to rotate about their [[\bar{1}10]] axes.

In this setting, the pyramidal end of the crystals fitted against the wall of the capillary tube in which they were mounted (Fig.[link]. The roughly square cross section of the crystals perpendicular to the rotation axis tended to minimize the absorption variation, which was significant with crystals of linear dimensions of about 0.5 mm. The use of large crystals was necessary with a relatively weak X-ray source running at 800 W and with a foreshortened focal spot of 0.4 × 0.4 mm.

It was convenient to index the reflections in a monoclinic cell, as shown in Fig.[link]. In this orientation, the reciprocal lattice presented a diamond pattern to the triple-counter array, whose windows were set parallel to the rotation axis. The reflections therefore occurred in two intersecting sets of levels, the odd and even levels, which had to be measured separately. However, the need to collect alternate levels in this way conferred the advantage that the counters could be positioned closer to the crystal (27.1 cm) than in the a-axis mounting so that >70% of the reflected X-rays were transmitted to the counters. Despite the complexity that this geometry introduced at the data processing and reduction stages, the significant advantages that it offered at the experimental stage ensured its use. The use of a [c^{*}] mounting would have been even more advantageous, but it was ruled out both by the difficulty of mounting the crystals in this orientation and by the fact that the counter arm could not be set to the required length of 18.5 cm.


Figure | top | pdf |

The [a^{*}b^{*}] plane of the reciprocal lattice oriented to rotate about the [[\bar{1}10]] axis. The indexing of reflections in a monoclinic unit cell with axes [A^{*}] and [B^{*}] is also shown. [A^{*}] coincides with [110] in the tetragonal lattice and [B^{*}] coincides with [b^{*}]. The rotation axis is the B axis in the monoclinic cell. The reciprocal-lattice dimensions with Cu Kα radiation are [a^{*} = b^{*} = 0.0195]; [c^{*} = 0.0406] r.l.u., while the cell diagonal in the [a^{*}b^{*}] plane is 0.0276 r.l.u.

Native data were collected both from crystals rotated about the a axis and from crystals rotated about [[\bar{1}]10] so that they could be scaled together to form a consistent set of three-dimensional data. For the derivatives, however, data were collected only from crystals rotating about their [[\bar{1}10]] axes and these data were scaled together to form two sets, one comprising levels with odd indices corresponding to the rotation axis and the other with even indices. These measurements, comprising some 1200 reflections, could be made from a single crystal exposed to the X-ray beam for about 20 h. The odd and even data sets for the derivatives were then scaled separately to the native data to give complete sets of three-dimensional data. These low-resolution measurements were made during the first half of 1964 and, after processing by the methods described below, they were used in September 1964 to calculate a new image of the structure at 6 Å resolution. The second low-resolution map at 6 Å

| top | pdf |

Our purpose in calculating a new electron-density map at 6 Å was fourfold. First to ascertain whether the procedures used to identify the five derivatives thought to be satisfactory at this level of resolution had worked satisfactorily. Second, to judge the quality of the measurements made by the triple-counter diffractometer. Third, to explore the effects of the modified method of applying absorption corrections to the intensities that are described below, although these were not expected to have a very great effect at low resolution. Fourth, to examine the effectiveness of the new procedure for incorporating anomalous-scattering information in the phase determination, which is also described below.

Comparison of the two sets of structure amplitudes gave a conventional R value of 0.075, which is not particularly good – perhaps because of the comparatively large background values associated with these low-angle measurements. However, the mean figure-of-merit obtained in the new phase calculations was 0.97 as compared with the 0.86 obtained originally. The root-mean-square difference in electron density between the two maps was 0.012 e Å−3, from which it may be judged that the two maps were very similar. Nevertheless, the outline of the molecule was certainly clearer in the new map, and within the molecule there was improved continuity, suggesting the course of a folded polypeptide chain, and there were a number of stronger rod-like features suggestive of α-helices. Two of these were prominent, running upwards from right to left, in the view of the new model shown in Fig.[link].


Figure | top | pdf |

Solid model of the electron density greater than about 0.5 e Å−3 in the second study of lysozyme at 6 Å resolution. This view of the model is equivalent to a view of the original model seen horizontally from the right of Fig.[link]. (a) The new model has a marked cleft running roughly vertically down the other side of the model, corresponding to the one that can be seen in Fig.[link]. (b) The cleft was shown to bind inhibitor molecules. The black density is that observed for the lysozyme–GlcNAc complex at 6 Å resolution.

The result was very encouraging, and we therefore went ahead immediately with data collection at 2 Å resolution, using essentially the same methods. At the same time, we began to plan low-resolution studies of inhibitor binding to lysozyme, from which we hoped to derive information about the nature of the enzyme–substrate complex. Intensity measurements at high resolution

| top | pdf |

At high resolution, measurements were made from only six levels from native crystals set to rotate about the a axis while a complete set of data was collected from crystals rotating about [[\bar{1}11]]. As at low resolution, the a-axis levels were important for scaling together the odd and even sets of levels measured from crystals rotated about the tetragonal [[\bar{1}11]] axis. No high-resolution measurements were made from derivative crystals rotated about the a axis, the intention being to scale the odd and even subsets of derivative data directly to the native. Experimental methods

| top | pdf |

The crystals can be thought of as roughly flattened on [(00\bar{1})] with liquid between [(00\bar{1})] and the capillary tube in which they were mounted. In order to minimize the effect of the liquid on absorption, we therefore measured the reciprocal-lattice hemisphere with C and [c^{*}] positive. Reflections were scanned along A, and the origin of the scanning slide was offset for upper levels so that, for example, the centre of the scanning slide was the H = −4 for the level K = 8.

Crystals could be set fairly well by eye. Viewed through the diffractometer microscope, they looked roughly square in the [001] direction, and the arcs of the goniometer head were set so that the edges were horizontal and vertical. When the crystal was turned through 90° from this position, the reflection of a light held level with the microscope could often be seen in the true (110) face. Setting the crystals on the goniometer head with [001] parallel to one of the arcs facilitated subsequent adjustments.

Fine adjustment of crystal orientation was achieved by setting on the [\bar{4}40] reflection. For this purpose, C – the vertical slide – was set to 0.1104 and the inclination angle to [\mu = 3^{\circ}\;5']. Then, if one arc was set fairly well by eye, the other could be oriented parallel to the incident X-ray beam and adjusted to locate the reflection. The two arcs were then adjusted in turn to give the optimum setting, in which the crystal was rotated about the normal to the [(\bar{4}40)] plane.

The orientation in the AC reciprocal-lattice plane was determined by returning the vertical slide to zero and setting the upper (scanning) slide to 0.1104 with the lower (stepping) slide at zero. The crystal was then rotated until the monoclinic 400 reflection appeared. At this stage, checks were made with the top/bottom and left/right slits to make sure that the crystal was well centred in the X-ray beam and that the counter apertures were well positioned. Similar checks were made with the crystal rotated about 180° to the monoclinic [\bar{4}00] reflection.

The final check on crystal orientation was to locate the 008 reflection near 0.3248 on the lower (stepping) slide, with the other slides set at zero. There was some variation in the value of [c^{*}] for different crystals, and [c^{*}] was often closer to 0.0404 than to 0.0406 r.l.u.

At this stage, the first measurements were made of the intensities of the reflections (monoclinic) 400, [16_{\prime}0_{\prime}0] and 008. These reference reflections were remeasured at intervals during the measurement of each triplet of reciprocal-lattice levels as a check on the stability of the whole system and irradiation damage to the crystal. The measurements were manually entered on the diffractometer output tape and monitored by the data-processing program.

In order to set the diffractometer for a particular triplet of levels, we found it convenient to leave the lower slide set for the 008 reflection, to adjust the vertical slide and the tilt angle (μ) for the levels in question, and then to run out along the stepping slide until a suitable high-angle reflection was found. The crystal rotation angle was then optimized for this reflection.

Finally, careful checks were made to ensure that the reflections to be measured fell in the reciprocal-lattice hemisphere with L positive, and that they were indexed in a right-handed axial system. The automatic run was then begun at the 2 Å limit on the stepping [(C^{*})] slide, and all reflections were scanned from Hmax to −Hmax on the scanning slide. Virtually all reflections within (and in some directions a little beyond) the 2 Å limit were measured in this way in fourteen triplet levels, seven even and seven odd. About 2000 reflections were measured in a typical overnight run, and data collection for each species of crystal took slightly more than two weeks.

The derivative crystals required for these measurements were prepared by the diffusion method described above, and about 20 of them were mounted at the same time at the beginning of the data-collection process to ensure that they were all the same. At the end of a run, careful measurements were made of the peak intensity of the [\bar{4}40] reflection as the crystal was rotated through steps of 15° about the crystal-mounting axis (ϕ). These measurements were used during data processing in an improved method of absorption correction devised by North et al. (1968)[link]. Diffractometer output

| top | pdf |

The output from the triple-counter diffractometer again consisted of a plain-language output for immediate checking of the results and output at this stage on eight-hole-punched paper tape for immediate transfer to the computer. An example of this record is shown in Fig.[link].


Figure | top | pdf |

Paper-tape output from the triple-counter linear diffractometer, showing the indices of the central reflection and the background, peak, background counts for the three reflections. Reproduced with permission from Arndt et al. (1964[link]). Copyright (1964) Institute of Physics. Data processing

| top | pdf |

Although the in-house Elliott 803B computer provided much improved facilities over those previously available, its limited store capacity (4096 words of magnetic core memory, but no magnetic drum, disk, or tape facilities) resulted in data processing being carried out in a series of stages, each of limited complexity. Input and output was by means of eight-hole paper tape, which was used as the medium for intermediate storage. The output tape from each stage was used for input to the next stage, together with appropriate parameters. Manual input was restricted to such parameters as unit-cell dimensions and ordinates of the absorption curves, with all other input being in the form of computer-generated output from a previous stage, starting with the diffractometer output tapes. This approach resulted in a great reduction in manual labour and intervention compared with the low-resolution work on lysozyme and the high-resolution stage of myoglobin, and it was probably the first crystal-structure determination that was fully computerized.

As for the initial low-resolution work, the first stage of data processing was input and checking of the diffractometer output, with the programs modified appropriately to deal with the sets of three reflections measured simultaneously. The background-corrected intensities were then output together with a table of the reference reflections used to estimate the extent of any radiation damage. Corrections were then applied for radiation damage if required, followed by application of Lorentz–polarization corrections, followed by absorption corrections. Absorption corrections

| top | pdf |

Although absorption of X-rays by protein crystals is low compared with crystals having a preponderance of heavier atoms, corrections for absorption are required in order to give F values that are sufficiently precise for calculation of the relatively small changes due to the introduction of heavy atoms or anomalous dispersion. The mounting of protein crystals within a glass capillary, normally with a small amount of mother liquor between the crystal and the capillary wall, presents a complicated situation for absorption calculations. Although Wells (1960)[link] wrote a computer program to deal with the situation, a severe impediment to the use of theoretical methods of correcting for absorption results from the very great difficulty in obtaining precise measurements of the mounted crystal, the liquid meniscus and the capillary-tube walls. An alternative approach was to use a semi-empirical method and, for low-resolution lysozyme studies, Furnas' (1957) method had been employed, as described above.

Despite the fact that the method of Furnas had been successful in improving the agreement between symmetry-related reflections in the earlier studies, the implicit assumption that the absorption depended upon the mean direction of the incident and reflected X-ray beams became clearly less valid as the Bragg angle increased. We therefore implemented a development of the method in which the absorption correction applied to any reflection was given by the mean of the two values for the directions of the incident and reflected beams (North et al., 1968)[link].

Although the method was easy to apply and was of significant value, as judged by the improved agreement between symmetry-related reflections, it nevertheless provided only a partial correction for absorption, because of the assumption that absorption is dependent solely on the directions of the incident and reflected beams. The limitations of this assumption are particularly important where precise values are required for Friedel pairs of reflections in order to make use of anomalous-scattering differences in phase determination. Fig.[link] shows two contrasting situations that can arise when the environment of a crystal is asymmetrical because of its mounting. The Friedel pair of reflections shown in Fig.[link] would suffer similar absorptions, whereas the pair shown in Fig.[link] would have significant differences because of the location of the mother liquor. In the 2 Å structure determination of lysozyme, an approximate correction was made for this effect. From Fig.[link], it is clear that the absorption error arising from the asymmetric distribution of the mother liquor is 0 for reflections with h = 0 and becomes increasingly great as h increases. The assumption was made, therefore, that the required correction was a function only of h and the reflections hkL with constant L were divided into groups with constant h and −h. The ratios [\sum_{k}I(h)/\sum_{k}I(-h)] were then plotted against h, as shown in Fig.[link].


Figure | top | pdf |

Asymmetric mounting of a protein crystal with its mother liquor in a capillary tube. Anomalous-scattering differences would be seriously affected in (b) but not in (a). Reproduced with permission from North et al. (1968[link]).


Figure | top | pdf |

Plot of the ratio [\textstyle\sum_{h}\displaystyle I(h)/ \textstyle\sum_{h}\displaystyle I(-h)] against h. In this example, a linear correction may be safely applied to equalize the average intensities in opposite rows (North et al., 1968[link]).

Such plots were frequently found to be linear, and the corresponding linear correction was then applied to each row on the more highly absorbed side in order to bring its mean intensity up to that of the other. Where the plots were not linear, so that a simple form of correction was not applicable, the entire set of measurements was usually rejected. Further stages of data processing

| top | pdf |

Many of the programs for subsequent stages of data processing were written by VRS, who had discussed the work with DCP at a meeting in Madras in January 1963 and who joined the team in October 1963.

As described in Section[link], the native data comprised three sets, one of six reciprocal-lattice levels measured from crystals rotated about the a axis, and the other two comprising the odd and even subsets of levels collected about the [[\bar{1}11]] axes. Within these three sets, the odd and even [[\bar{1}11]] levels had no rows in common with one another, but each had rows in common with the a-axis data. Extraction of the related rows permitted the calculation of scale factors by the method of Hamilton, Rollett & Sparks (Hamilton et al., 1965[link]). Application of these scale factors produced a complete set of self-consistent native data.

The derivative data, which consisted of the odd and even levels collected about the [[\bar{1}11]] crystal axes, were scaled to the native data by the application of two scale factors derived by comparing the totals of the intensities in the odd and even subsets. Because of the high symmetry of the tetragonal space group, up to eight measurements were available for many symmetry-equivalent reflections; for the heavy-atom derivative crystals, these formed four sets of Friedel-related pairs. At this stage, the X-ray amplitude data were on 40 paper tapes for each of the native and three derivative crystals, one for each reciprocal-lattice level. The native and derivative data were then compared level by level, but it was found that the ratio was very nearly a constant, independent of sin θ. A further numerical factor was applied to all data to bring them approximately to an absolute scale. In the final stage of data processing, the data from the native and the three derivative sets were brought together into a single list containing the native F's and the Friedel pairs for each of the three derivatives. In addition, all of the centric reflections were extracted in order to provide the data for refinement of the heavy-atom parameters.

At this late stage in the data processing, we noticed that some levels agreed significantly less well than the norm with intersecting levels. On 23 October 1963, we had to recognize that the high-resolution measurements had been made from two different types of crystals with essentially identical unit-cell dimensions, but with subtly different diffraction patterns. We designated these two crystal types type I and type II, and set about the task of producing sets of data of one type to use in the structure analysis. The crystal-type problem

| top | pdf |

This discovery was particularly galling because, although at the time variations in diffraction patterns had been reported for some protein crystals, we had failed to notice that this phenomenon had been mentioned years earlier in a study of lysozyme (Corey et al., 1952[link]).

A preliminary analysis of the differences in the diffraction patterns suggested two important characteristics:

  • (1) the differences tended to increase with resolution and

  • (2) the differences appeared to be more consistent with two discrete diffraction patterns than a continuum of patterns lying between two extremes.

The two diffraction patterns were characterized operationally by specific patterns of intensities in the 3–4 Å resolution range (where the differences appeared to be maximal), and particularly by a few adjacent pairs of reflections whose relative intensities interchanged in the two types. The principal diagnostic reflections were [11_{\prime}11_{\prime}4] and [11_{\prime}11_{\prime}5]. In data associated with crystal type I, [I(11_{\prime}11_{\prime}4) \,\gt\, I(11_{\prime}11_{\prime}5)], while in crystal type II, [I(11_{\prime}11_{\prime}4)\,\lt] [ I(11_{\prime}11_{\prime}5)]. The [11_{\prime}11_{\prime}]L rows of reflections from the two types of crystals had the structure amplitudes shown in Table[link].

Table| top | pdf |
Structure amplitudes of the [11_{\prime}11_{\prime}L] reflections from crystal types I and II

Crystal typeReflection
I 29.2 21.8 10.6 18.2 23.1 17.0
II 36.9 25.7 19.7 27.7 26.4 36.8

In order to stand any chance of successfully calculating the lysozyme map by isomorphous replacement, we had to sort all the data so far collected, both native and derivative, into the two types and then recollect `rogue' data sets in order to assemble complete data sets of one particular type. The alternative was to recollect the whole data on a sounder basis, which we were loath to do, especially as other teams seemed likely to be well advanced in their solution of the lysozyme structure.

It appeared that the bulk of the data that had already been collected using the diffractometer was what we had called type II. We also observed that nearly all the photographic data collected in the heavy-atom proving stage was of type I. This observation was of great importance because it gave us a sound basis for defining the differences between the type I and II diffraction patterns, and it also provided a vital clue in identifying from their shapes the crystals that gave the two types of diffraction pattern. This was very important to us in the selection of crystals to replace the rogue data sets. The crystals that gave the best results for the photographic work tended to be relatively small and flattened along the tetragonal fourfold axis direction, while those that gave the best diffractometer data were the larger isometric crystals, which were more extended along the crystal fourfold axis (Fig.[link]. These crystals could be definitely associated with the type I and type II diffraction patterns, respectively. Batches of lysozyme crystals grown according to the procedures defined earlier usually contained both types of crystal. This suggested that the two crystal forms might have originated when the pH of crystallization was on the borderline between two crystal forms of lysozyme. This hypothesis was supported by the observation that crystals grown at a somewhat higher pH had diffraction patterns closely similar if not identical to type I.

The more thorough analysis of the differences between the diffraction patterns for types I and II that these findings permitted showed that the differences in the intensities of equivalent reflections were resolution-dependent. They were very small in the 6 Å region (which probably accounts for the differences not being observed earlier), increased to a maximum at the position of the normal 4 Å peak in the protein diffraction pattern, and fell off at higher resolutions. This pattern is consistent with a lack of isomorphism between the two types of crystal of the kind that may be caused by a slightly different orientation of the lysozyme molecule in the unit cells of the two crystals (Crick & Magdoff, 1956[link]). Such effects may be brought about, for example, by slightly different charge distributions in the protein molecules, rather than by the presence of additional diffracting material in one crystal type or the other. [This conclusion was later confirmed by a detailed analysis of the two structures by Helen Handoll (1985)[link].]

These observations suggested it was safe to go ahead with trying to determine the structure of lysozyme in either type of crystal, and the decision to proceed with the type II crystals was solely on the basis that the bulk of the data already collected were of this type. Knowing the characteristic pairs of reflections that distinguished the two crystal types, we found it relatively straightforward to ascribe each data set to a particular crystal type, even for the heavy-atom derivatives, because the type differences were much larger than the heavy-atom changes at approximately 4 Å resolution. All triplet levels of data that belonged to the type I diffraction pattern were extracted from the total data sets and replaced by equivalent data sets recollected from confirmed type II crystals. This process, whose success was carefully tested and confirmed during the data processing and reduction stage, resulted in consistent data sets at 2 Å resolution for the native lysozyme and all three isomorphous derivatives that were derived from type II crystals. Final refinement of heavy-atom parameters

| top | pdf |

At this stage, we were able to review the refinement of the heavy-atom parameters in relation to the type II crystal data that were now available. Prompted in part by the observation that the heavy-atom positions in myoglobin appeared in regions of high negative density in the final electron-density map, we considered the need to adopt an improved way of modelling the heavy atoms, particularly for the complex uranium compounds, in which the uranium atoms were coordinated to several oxygen or nitrogen atoms. Calculation of the R factors and occupancies of the heavy atoms as a function of sin θ showed that the heavy-atom compounds, particularly the uranyl complexes, could not be modelled well by a single atom, but that it was necessary to take into account both the O and the N atoms in the complex and the water that had been displaced by the heavy-atom cluster. The difference electron-density maps did not show the orientation of the O and N atoms, and our first thought was to model the complex in terms of a central U atom, surrounded by a spherical shell of electron density representing the O and N atoms minus a sphere of electron density representing the displaced water. Trial calculations suggested that this model would give improved agreement between observed and calculated [|\Delta F|] values, most significantly for the uranyl derivatives. On second thought, it occurred to us that a more satisfactory approach might be to derive empirical scattering-factor curves by fitting the curves representing the variation of heavy-atom occupancies with sin θ to polynomial functions of [\sin^{2} \theta]. This was easily accomplished by the use of a standard curve-fitting program and, for ease of use in the phase program, these scattering curves were fitted to curves of the type [f(H) = a + b\sin^{2} \theta + c\sin^{3} \theta + d\sin^{4} \theta]. With this modification, Pangloss was used with the type II crystal centric data to obtain the heavy-atom parameters shown in Table[link].

Table| top | pdf |
Heavy-atom parameters for the 2 Å structure

O is the occupancy of the heavy-atom site (electrons); B is the isotropic temperature-factor constant; E is the root-mean-square difference between observed and calculated heavy-atom differences for centric reflections (electrons); N is the number of centric reflections in the range [0.01 \,\lt\, \sin^{2} \theta \,\lt\, 0.15] used in the refinement and R is the reliability index for observed and calculated heavy-atom changes of centric reflections.

MHTS I 0.2068 0.6138 0.0507 39.2 17.8 58 1247 0.60
  II 0.2415 0.6393 0.9326 8.8 14.9      
UF I 0.1783 0.5849 0.7204 55.5 21.0 74 1277 0.52
  II 0.0974 0.8976 0.4650 29.3 24.3      
UN I 0.0961 0.8938 0.2664 47.1 19.2 80 1140 0.57
  II 0.1898 0.5901 0.7168 42.1 124.8      
  III 0.0446 0.7266 0.5150 9.0 190.2      
  IV 0.0869 0.8976 0.4866 11.4 68.4      
  V 0.2024 0.6388 0.6781 28.6 42.8      

The temperature factors shown in Table[link] most deserve comment. All those obtained for the significant sites (of which there are no more than two for any derivative) are comparable with the overall value for the protein crystals themselves. The very large values obtained for other sites show that these sites are of little importance at high angles and may not represent real sites of heavy-atom attachment. The minor site of MHTS is clearly the [\hbox{SO}_{3}^{-}] group of this molecule (see Section[link]). Calculation of phase values

| top | pdf |

Blow (1958)[link], in his determination of the phase angles of the noncentrosymmetric [100] zone of horse haemoglobin, and, later, Cullis et al. (1961[link], 1962[link]), in their determination of the three-dimensional structure of horse haemoglobin, used anomalous-scattering data to supplement the information available from the isomorphous-replacement differences. In each of these studies, phase determination had been carried out by constructing probability curves from the multiple-isomorphous-replacement data and, when the most probable phase angle had been deduced, the anomalous-scattering data were examined. For many reflections, they allowed a choice to be made between two apparently equally probable values of phase angle given by the isomorphous data. This procedure was clearly rather arbitrary and subjective, and a method of combining anomalous scattering with isomorphous replacement in a more rigorous way was described by Blow & Rossmann (1961)[link]. In their method, which was subsequently employed for the low-resolution work on lysozyme, use was made of the fact that the mirror image of the Argand diagram for a [\bar{h}\bar{k}\bar{l}] reflection is similar to the Argand diagram for the hkl reflection, but for the reversal of the sense of the imaginary part of the heavy-atom contribution. The data for the [\bar{h}\bar{k}\bar{l}] reflections may therefore be treated as though they came from a separate isomorphous compound, with parameters identical to those of the original compound, but with the opposite sign for the imaginary component of the atomic scattering factor.

In the low-resolution lysozyme phase determination (Section[link]), intensities of the Friedel pairs of reflections were measured for each of the three heavy-atom compounds, and the problem was treated as if there had been a total of six heavy-atom compounds. Although the method had been found helpful to some extent, analysis of the phases showed that the anomalous-scattering data had played comparatively little part in determining the positions of the centroids of the phase probability distributions, even for reflections with apparently significant anomalous differences.

ACTN observed that this apparent contradiction is because of the fact that the anomalous differences between Bijvoet pairs of reflections measured from the same crystal are inherently more accurate than the isomorphous differences that are measured from different crystals and subject to different systematic errors (North, 1965[link]; Phillips, 1966[link]). Indeed, analysis of the equivalent reflections from native and derivative crystals (Section[link]) showed that the r.m.s. error E′, corresponding to the anomalous differences, was about one-third of E, the error in the isomorphous differences. The result of incorporating this distinction in the phase program is illustrated in Fig.[link]. Phase calculations for the new 6 Å and 2 Å maps of lysozyme were therefore carried out by using ACTN's method, with E′ set at one third of E.


Figure | top | pdf |

(a) Phase probability curve for a Bijvoet pair of reflections (broken lines) with the joint probability curve (full line) derived by the method of Blow & Rossmann (1961)[link]. (b) Isomorphous-replacement phase probability curve derived from the mean of [F_{PH^{+}}] and [F_{PH^{-}}] (broken line); anomalous-scattering probability curve (chain line); and joint probability curve (full line) derived by the method of North (1965)[link], using [E'(\hbox{anomalous}) = E(\hbox{isomorphous})]. (c) As (b), but with [E' = (1/3)E]. Reproduced with permission from North (1965[link]).

The data tapes containing the F values for the native and the Friedel pairs of F values for the three derivatives were used as input to a phase program written by ACTN. For acentric reflections, phase probabilities were calculated as described in the previous section, and the centroids of the distributions were determined in order to derive a `figure of merit', which was applied to the structure amplitudes, as first proposed by Blow & Crick (1959)[link], so as to produce a `best' Fourier map. For the quite high proportion of centric reflections in the lysozyme diffraction pattern, phase probabilities were calculated by the formula appropriate to the case in which the native and derivative F's are collinear with each other and with the vector due to the heavy atom.

The phases of the 9040 reflections were calculated on the Elliott 803B computer and had a mean figure of merit of 0.60. The variation with angle was very similar to that obtained with sperm-whale myoglobin and is shown for the centric and acentric reflections separately in Fig.[link].


Figure | top | pdf |

Variation of the mean figure of [\langle m\rangle] with [\sin^{2}\theta / \lambda^{2}] (crosses represent acentric reflections, open circles represent centric reflections). The electron-density map at 2 Å resolution

| top | pdf |

GAM joined the team in early 1964 with the specific task of writing a program for calculating the 2 Å electron-density map. Fortunately, at this stage, the University of London Computing Centre was in the course of acquiring a Ferranti ATLAS computer, which was then one of the most powerful computers available for this kind of work, and GAM eagerly set about taking advantage of its power. Whereas we ourselves had operated the previous MERCURY computer at night, there were machine operators for the ATLAS with whom we left our input tapes; these were very large reels of paper tape comprising all of the amplitude and phase data for the 2 Å set. Following two or three unsuccessful attempts to load the whole of the data without tearing the tapes, GAM modified his program so that the data could be loaded in several sections, which was achieved satisfactorily.

The electron density was calculated at 1/120ths of the cell edge along a and b, and 1/60ths along c. The output from the computer was on punched paper tape, arranged with appropriate insertions of carriage returns and line feeds so that the teleprinter output was in a form suitable for immediate contouring to a scale of 0.75 inches equal to 1 Å. Each x, y section of the map was printed out in five strips, which had to be glued together to cover the whole area. The contours were drawn initially in pencil on these paper sheets and were then copied in drawing ink to thin Mylar sheets, which were supported for interpretation on Perspex sheets that were stacked and bolted together, with spacers of appropriate dimensions to maintain the scale in the c direction, in groups of five for ease of handling. The whole map was drawn on 60 sections perpendicular to the z axis, and the bolts holding the blocks of five sheets together were designed to fit into one another to keep successive blocks in register. The maps were viewed on large light boxes, specially constructed for the purpose, though they were not transparent enough for more than fifteen sheets, three blocks of five, to be studied in detail at one time. The contours were drawn on Mylar sheets to avoid waste of the more expensive Perspex that would have arisen from errors. A grid was drawn on a Mylar sheet, which could be superimposed on the contour stacks and used to read the atomic coordinates directly in ångstroms.

The electron density had been calculated using a scale factor such that it was convenient to draw contours at intervals of 0.25 e Å−3; as no F(000) term had been included in the calculation, electron-density values were relative to the mean value for the unit cell. Contours were drawn only for electron densities above the mean, the two lowest levels being drawn in orange ink, and the higher ones in black ink. This proved to be a satisfactory form of representation, as it very clearly revealed regions of continuous high electron density while also showing the positions of significant features of lower density. All the members of the lysozyme group participated in drawing the maps.

To illustrate the result that was obtained, sections z = 35/60 to 44/60 of the electron-density map are shown in Fig.[link].


Figure | top | pdf |

Photograph of sections z = 35/60 to 44/60 of the three-dimensional electron-density map of hen egg-white lysozyme at 2 Å resolution. AA′ shows the axis of a length of α-helix lying in the plane of the sections. B indicates an α-helix more nearly normal to the sections. C indicates the disulfide bridge between residues 30 and 115, the sulfur atoms of which lie one above the other. The side chain of a phenylalanine residue is located four residues along the helix from the disulfide, towards the lower sections. Reproduced with permission from Nature (Blake et al., 1965[link]). Copyright (1965) Macmillan Magazines Limited. Map interpretation and model building

| top | pdf |

We were fortunate that by the time the map was ready for interpretation, two independent groups of protein chemists, led by Pierre Jollès in Paris and R. E. Canfield in New York, had studied the amino-acid sequence of hen egg-white lysozyme in detail and published their results. Two slightly different amino-acid sequences, complete with the arrangement of the four disulfide bonds, were published by Jollès et al. (1964)[link] and by Canfield & Liu (1965)[link], and this information was used intensively in the interpretation of the electron-density map of the protein. The sequence published by Canfield & Liu (1965)[link] is shown in Fig.[link].


Figure | top | pdf |

The amino-acid sequence of hen egg-white lysozyme (Canfield & Liu, 1965[link]).

Many features were immediately identifiable in the map. These included the side chains of many of the amino-acid residues, especially the disulfide bridges and the aromatic side chains of tryptophan, tyrosine and phenylalanine. Nevertheless, no attempt was made to interpret the map in detail without recourse to the amino-acid sequence. Interpretation began in the part of the map shown in Fig.[link]. The strongest feature in this part of the map corresponded to a disulfide bridge (C), in which the two sulfur atoms lie one above the other in the direction of the c axis. The first challenge was to identify this bridge. It is clearly connected to a helical region of the molecule (A), which ran in the direction from top right to bottom left of the diagram, with the main-chain carbonyl groups pointing in this direction. Consequently, this helix ran from its amino terminus on the right to its carboxyl terminus in the centre of the map. The map was quite clear enough to count the α-carbons from the cysteine residue that forms part of the disulfide bridge, and it was immediately apparent that the fourth residue from the cysteine towards the carboxyl terminus is an aromatic residue, probably phenylalanine. Inspection of the amino-acid sequence in Fig.[link] showed that only one pair of residues satisfied this condition, Cys30 and Phe34.

Given this start, interpretation of the map and the construction of a molecular model were relatively straightforward. The model was constructed in a metal frame, the top and bottom of which consisted of sheets of blockboard. The a and b axes were drawn parallel to the diagonals of these boards to cover the coordinate ranges, respectively, x = −1/4 to +1/4 and y = 0 to +1/2 to a scale of 2 cm to 1 Å. This was the scale of the brass models, constructed by Cambridge Repetition Engineers Ltd, which were used to build the model. The height of the frame covered the full extent of the c axis. The heavy-atom coordinates and the computer programs were both based on the wrong-handed space group [P4_{1}2_{1}2]. It was not until the anomalous scattering from the heavy atoms was incorporated that the correct space group [P4_{3}2_{1}2] was assigned. The Fourier-map sheets were actually stacked the opposite way round and a left-handed system of axes was used for the model. In retrospect, this should have been put right at once, but the system was not easy to change.

Holes were drilled in the top and the base boards on the grid defined by the a and b axes, and these were used to support an array of brass rods parallel to the c axis to which the model components could be attached. The model building was carried out by two subgroups, CCFB and VRS in one and ACTN and DCP in the other, so that work could go on continuously throughout each day. The method employed was to examine the map density corresponding to the next amino-acid residue to be located and to mark the positions of the constituent atoms with small washers or nuts. The coordinates of these atoms were then read from the map (making use of the superimposed grid and estimating the z coordinates from the extent to which adjacent z sections contributed to the density). These coordinates were then located in the model by means of the coordinate grids drawn on the base and top boards and by the use of a plumb line marked with the z coordinates. At this stage it was usually possible to fix a model of the amino-acid residue in place in the model frame with remarkably little trouble, though, of course, fine adjustment was necessary as the model grew.

This bout of model building began towards the end of February 1965 and proceeded quite rapidly. The main difficulties arose from the fact that the two amino-acid sequences that were available did not agree in every respect. Eleven amino-acid residues were identified differently by Jollès (Jollès et al., 1964[link]) and by Canfield (Canfield & Liu, 1965[link]). Four of the discrepancies involved Asn and Asp, which cannot be distinguished in the electron-density map with any degree of certainty. The remaining discrepancies are shown in Table[link].

Table| top | pdf |
Discrepancies in amino-acid sequences (excluding Asp/Asn)

Canfield & Liu (1965)[link] Thr Gln Ala Ile Asn Val Asn
Jollès et al. (1964)[link] Gln Ala Thr Asn Ile Asn Val

Inspection of the residues 40, 41, 42, 92 and 93 showed quite clearly that the shapes in the electron-density map fitted the Canfield side chains. Our initial conclusion, however, was that the electron densities corresponding to residues 58 and 59 were more consistent with the sequence proposed by Jollès than that published by Canfield. Accordingly, in our first detailed description of the structure (Blake et al., 1965[link]), we accepted the identification of these residues by Jollès and his colleagues. As will be seen, this was a mistake.

The model was completed in time for it to form the centrepiece at Bragg's 75th birthday party at the Royal Institution on 31 March 1965, and a description was submitted to Nature at about the same time (Blake et al., 1965[link]). One of the difficulties at this stage of development of the subject was that computer graphics had not yet been developed to the stage at which illustrations of protein structures could be produced with any degree of facility. Consequently, we were most grateful that Sir Lawrence Bragg enthusiastically drew the model freehand. His drawing provided the main illustration for the published paper: the original sketch is reproduced in Fig.[link].


Figure | top | pdf |

Schematic drawing of the main-chain conformation of lysozyme. The drawing was made from observations of the molecular model by Sir Lawrence Bragg and later prepared for publication by Mrs S. J. Cole.

During the early summer of 1965, we rebuilt the model, taking care to ensure that the components fitted the electron density as well as could be judged by eye, and that the contacts between them were fully consistent with current understanding of van der Waals forces and hydrogen bonds. The model components were fixed firmly in position with small brass clamps, designed and constructed by Bruce Morris in the Royal Institution workshop, so that at the end of the process we had an unrealistically rigid model that incorporated all the information available to us at the time. At this stage it became clear that the Canfield sequence (Fig.[link] was to be preferred in every respect.

A stereo-photograph of this model (Blake et al., 1967[link]) is shown in Fig.[link] in a view that shows the most striking feature of the molecule. The molecular face in the foreground of this view is crossed by a deep cleft, roughly parallel to the c axis of the crystal unit cell.


Figure | top | pdf |

Stereo-photographs of a model of the lysozyme molecule to a scale of 2 cm to 1 Å. The main polypeptide chain is painted white, and nitrogen, oxygen and sulfur atoms are indicated by blue, red and green sleeving, respectively. Some hydrogen bonds are shown by red connections. Oxygen atoms of the acid side chains near the cleft, Glu 35, Asp 52, Asp 101 and Asp103, are shown by red hemispheres (Blake et al., 1967[link]).

25.1.4. Structural studies on the biological function of lysozyme

| top | pdf | Lysozyme substrates

| top | pdf |

By the summer of 1964, the results for the structure determination of lysozyme were sufficiently promising for us to consider diffraction studies on the biological function of lysozyme. At about the same time, the Royal Society invited Max Perutz to organise a Discussion Meeting on lysozyme at The Royal Institution in early 1966, to which the leading workers worldwide would be invited. This provided an additional spur to our efforts.

Simultaneously with his discovery of lysozyme in 1922, Fleming had found a gram-positive species of bacteria, Micrococcus lysodeikticus, that was particularly susceptible to lysis by lysozyme (Fleming, 1922[link]). During the 1940s and 1950s, further work by a number of authors had shown that lysozyme exerted its biological effects through hydrolysis of the bacterial cell wall. These studies led to a definitive description of the polysaccharide component of the bacterial cell wall, as shown in Fig.[link].


Figure | top | pdf |

The cell-wall tetrasaccharide with the [\beta (1\!\rightarrow \!4)] glycosidic bond that is hydrolysed by lysozyme indicated (Blake et al., 1967[link]).

The structure is composed of alternating sugars, linked by [{\beta(1\!\rightarrow\! 4)}] glycosidic links, of N-acetylglucosamine (GlcNAc, abbreviated by us at the time as NAG) and N-acetylmuramic acid (MurNAc, previously abbreviated as NAM). The 3-hydroxyls of the MurNAc residues are attached to short peptides, and the peptides themselves are cross-linked to provide an extensive and rigid two-dimensional proteoglycan network. Lysozyme cleaves the [\beta(1\!\rightarrow \!4)] glycosidic bonds between MurNAc and GlcNAc residues, thus leading to the dissolution of the bacterial coat and lysis of the bacterium.

Kinetic studies on the activity of lysozyme were hampered by the lack of a suitable small-molecular-weight substrate. The turbidometric assay in use at the time, and still in use today, followed the change in optical density of a suspension of Micrococcus lysodeikticus cells as the cells were lysed by lysozyme. The assay could work reliably but it was sensitive to physical parameters such as ionic strength, the method by which the cells were suspended and product inhibition. In 1962, Wenzel et al., in an effort to obtain a small-molecular-weight substrate, had reported that lysozyme promoted the cleavage of the trimer of GlcNAc, tri-N-acetylchitotriose, releasing dimer and monomer sugars, and that the monomer, GlcNAc, was an inhibitor of lysozyme (Wenzel et al., 1962[link]). Compounds such as glucose or cellobiose that lacked the N-acetyl group did not inhibit.

John Rupley extended this work. Rupley, a chemist from the University of Arizona, Tucson, played a crucial role in developing the mechanisms for lysozyme catalysis. In results that were summarized in 1964 and 1967 (Rupley, 1964[link], 1967[link]) and made available to workers at the Royal Institution in 1965, he showed that the rate of hydrolysis of GlcNAc homopolymers increased by 30 000 as the chain length increased from trimer to hexamer, and that there was no further increase in rate for substrates of greater length than the hexamer. Furthermore, the cleavage patterns for the smaller polymers were complex, but the optimal substrate, the hexamer, was cleaved to a tetrasaccharide and a disaccharide.

Thus by the time we began structural studies, the nature of the substrate and the specificity of lysozyme were established. GlcNAc was the only compound that was available commercially. This seemed a good starting point. It was anticipated that as a competitive inhibitor it would bind to and allow the identification of the catalytic site. The crystal structure of GlcNAc

| top | pdf |

During the first two years of her graduate work (1962–1964), LNJ determined the crystal structure of GlcNAc. β-D-N-Acetylglucosamine was crystallized from a methanol–water mixture, and diffraction data were recorded on film with a modified Weissenberg camera. Intensities were estimated visually from comparison with a scale of fixed-time exposures of the attenuated main beam. The structure was solved manually using a sharpened Patterson function and application of the minimum function of Buerger (1959)[link]. The structure revealed a standard glucopyranose ring in the chair conformation with the plane of the N-acetyl group normal to the ring (Johnson & Phillips, 1964[link]; Johnson, 1966[link]). The structure of glucosamine hydrochloride had been solved by Cox & Jeffrey in 1939 (Cox & Jeffrey, 1939[link]), a remarkable early achievement.

By 1963, about ten glucopyranose structures were available, including that of cellobiose (Jacobson et al., 1961[link]). An analysis by Ramachandran et al. (1963)[link] showed that the glucopyranose ring is remarkably uniform in its conformation and may be regarded as a rigid structure.

The final difference-Fourier synthesis of β-GlcNAc revealed an additional peak adjacent to the C1 carbon atom in the position of a hydroxyl group in the α configuration. Peak heights suggested that there could be a mixture in the crystal of approximately 80% β-N-acetylglucosamine and 20% α-N-acetylglucosamine. Refinement indicated that a mixture of the two anomers could be accommodated in the crystal lattice.

Tests on the optical rotation of the crystalline sample after dissolution, compared with the starting material, also added support to the notion that the crystal contained a mixture of α and β anomers, probably as a result of mutarotation during crystallization. Although a later and more precise structure determination of GlcNAc indicated a lower proportion of the α anomer (Mo & Jensen, 1975[link]), mixtures of α and β sugars in crystals have been observed for other compounds (Jeffrey, 1990[link]). The consideration of the α and β anomers of GlcNAc meant that we were keenly aware of the importance of configuration at the C1 atom. This turned out to be essential when we were interpreting the results with lysozyme. Low-resolution binding studies of lysozyme with GlcNAc and other sugars

| top | pdf |

In the first binding studies of lysozyme with GlcNAc, crystals were soaked overnight in solutions of 0.15 M GlcNAc in the standard crystallization medium, and a 15° precession photograph was recorded by LNJ. The changes in intensities of reflections compared to a native lysozyme photograph were extremely small, much smaller than those observed with the heavy-atom derivatives, but were sufficiently promising to encourage us to collect three-dimensional data to 6 Å resolution. Data were collected on the linear diffractometer using the three counters in the flat-cone setting over a period of about 20 h, as described for the heavy-atom derivatives. The data were processed using the programs of ACTN and VRS on the Elliott 803B computer. The 6 Å difference-Fourier map, using the phases from the improved set of heavy-atom derivatives, was obtained at the end of October 1964. It showed a single rather elongated peak which, when superimposed onto the 6 Å model, was located in the cleft in the enzyme surface between the two domains (Johnson & Phillips, 1965[link]).

The power of the difference-Fourier technique in protein crystallography was immediately demonstrated by this first 6 Å electron-density difference map for the lysozyme–inhibitor complex, as had also been demonstrated earlier in the work (Stryer et al., 1964[link]) on the binding of azide to sperm-whale myoglobin. Once a protein structure had been solved, it was apparent that ligand-binding sites could be established with ease. Following the GlcNAc result, 6 Å binding studies were repeated with a number of other compounds (Blake et al., 1967[link]). Kinetic studies using the turbidometric assay were carried out with each of the compounds in order to establish the mode of inhibition of lysozyme activity. We were fortunate in being able to use the skills of JWHO, a carbohydrate chemist who had been at the Royal Institution for many years. JWHO had synthesized 6-iodo-6-deoxy-N-acetyl β-methylglucosaminide, a compound which was found to inhibit more powerfully than GlcNAc itself. The low-resolution binding study showed a stronger and more compact peak at the catalytic site than that observed with GlcNAc, but it was not possible to resolve the iodine and hence identify the six positions of the sugar.

By January 1965, we had been given a sample of the disaccharide di-N-acetylchitobiose, (GlcNAc)2, sent by John Rupley. Efficiency of data collection and map production had increased. Data collection was started on 19 January and the map was obtained by 28 January. Other results followed with N-acetylmuramic acid (a gift from R. W. Jeanloz) and the disaccharide N-acetylglucosamine β-(1,4)-N-acetylmuramic acid (GlcNAc-MurNAc – a gift from N. Sharon) (Fig.[link].


Figure | top | pdf |

Inhibitor molecules of lysozyme (Blake et al., 1967[link]). (a) N-acetylglucosamine; (b) N-acetylmuramic acid; (c) 6-iodo-α-methyl-N-acetylglucosaminide; (d) α-benzyl-N-acetylmuramic acid; (e) di-N-acetylchitobiose; (f) N-acetylglucosaminyl-N-acetylmuramic acid; (g) tri-N-acetylchitotriose.

As part of these studies, penicillin V and p-iodophenoxymethyl penicillin potassium salt were also investigated. They were observed to inhibit lysozyme. Crystallographic studies showed that the penicillins did indeed bind to lysozyme in the catalytic cleft, but at a site remote from the GlcNAc binding site (Johnson, 1967[link]). As established shortly after this result was obtained in 1965, penicillin exerts its potent antibiotic activity by inhibition of the enzymes responsible for the biosynthesis of the peptide cross-linking component of the bacterial cell wall (Tipper & Strominger, 1965[link]; Wise & Park, 1965[link]). An original suggestion that penicillin might resemble MurNAc turned out to be incorrect (Collins & Richmond, 1962[link]). The interactions of penicillin with lysozyme are probably fortuitous, but were not fully investigated.

By the end of May 1965, we were ready to move on to 2 Å data collection, a formidable task that required fourteen crystals and more than two weeks continuous data collection on the multiple-counter linear diffractometer. Data for the lysozyme–GlcNAc complex were completed first, and the map was available around October 1965. The electron density was puzzling, and an interpretation was not possible until the high-resolution results with the trisaccharide were available. Binding studies of lysozyme with tri-N-acetyl-chitotriose, (GlcNAc)3, at 2 Å resolution

| top | pdf |

John Rupley visited the Royal Institution in July 1965. He brought with him a sample of tri-N-acetylchitotriose (Fig.[link], previously abbreviated as tri-NAG and now as (GlcNAc)3. The lysozyme–(GlcNAc)3 complex turned out to be the crucial structure for understanding activity. When lysozyme crystals were soaked in 0.05 M (GlcNAc)3 solution, they turned opaque and became difficult to mount for X-ray studies. The problem was overcome by soaking the crystals already mounted in a capillary tube. By 17 July 1965, a 6 Å map of the lysozyme–(GlcNAc)3 complex was obtained and showed a peak occupying the whole top part of the cleft on the enzyme surface.

Rupley had also succeeded in co-crystallization of lysozyme with (GlcNAc)3. Data collected from these crystals gave an identical difference-Fourier synthesis to that obtained with crystals soaked in (GlcNAc)3. Data from another co-crystallized crystal, which were collected ten days later, also gave an identical result, indicating that there had been no hydrolysis of the trisaccharide in the crystal during this time. The way was open for high-resolution data collection with (GlcNAc)3. Rupley set up co-crystallizations of lysozyme with (GlcNAc)3 (with concentrations 1:1.1 molar ratio under standard conditions) at the Royal Institution during the hot August of 1965. The crystals grew well and provided us with a plentiful supply of crystals for the 2 Å data collection, which was carried out by LNJ following the same procedures as for the native crystals described above.

The 2 Å difference-electron-density map arrived on Thursday 6 January 1966, but there was a mistake. After correction and the laborious task of contouring by hand and transferring the contours to transparent sheets had been completed, serious model building by DCP, LNJ, ACTN and other members of the team began on 19 January 1966.

The interpretation of the difference map for (GlcNAc)3 was straightforward. Three sugars could readily be identified in the density. The first occupied a site at the centre of the cleft at a position similar to that occupied by GlcNAc by itself and labelled site C. The acetamido group was visible and fitted neatly into a pocket where there were complementary hydrogen bonds between the NH and CO of the acetamido group and the main-chain carbonyl oxygen of residue 107 and the amido group of residue 59. Lysozyme has three tryptophan residues at the catalytic site. One of these, Trp108, was at the bottom of the acetamido pocket and made van der Waals contacts to the methyl group of the acetamido. The precise and extensive contacts to the acetamido group explained immediately the specificity of lysozyme for N-acetyl sugars. The free reducing group of the sugar in site C pointed down (towards lower z). The second sugar could be fitted to the density above the first (in the direction of increasing z), linked to the O4 of the sugar in site C. It was clear that a second tryptophan, Trp62, stacked against the glucopyranose ring for this sugar in site B, and that there was a shift of this tryptophan towards the sugar. This and other conformational changes could explain the tendency of the lysozyme crystals to become disordered when soaked in (GlcNAc)3 solutions.

In cellobiose, whose crystal structure had been determined in 1961 by Lipscomb and colleagues (Jacobson et al., 1961[link]), the conformation about the [\beta (1 \!\rightarrow \!4)] glycosidic link rotates the second sugar about 180° with respect to the first, and there is an intramolecular hydrogen bond between the O3 hydroxyl of one sugar and the O5 of the adjacent sugar. This relative orientation of the two adjacent sugars was found to fit the density for sugars C and B and also to agree with models proposed by Carlstrom (1962)[link] for chitin and the analysis of Ramachandran et al. (1963)[link]. The lobes representing the acetamido and C6 alcohol groups for site B provided further confirmation for this orientation. The third sugar in site A appeared less well located, but nevertheless could be placed in density and was observed to make satisfactory contacts with the protein.

The interpretation of the difference-Fourier map was made by one worker without reference to the protein structure, while a second worker fitted the model into the protein structure. It was most exciting and satisfying to observe the contacts when the skeletal models of the sugars were fitted into the skeletal model of the protein. Shifts in atomic coordinates of selected protein groups were estimated from the gradient of electron density in a difference map divided by the curvature of the electron density at that point, as described by Lipson & Cochran (1968)[link]. Using this formula, the shift in Trp62 was calculated to be 0.75 Å, a value that turned out to be accurate when the lysozyme–(GlcNAc)3 complex was refined by least-squares methods many years later (Cheetham et al., 1992[link]). It was also noted that there were two acidic residues on either side of the catalytic cleft, Glu35 and Asp52, which were some distance from the (GlcNAc)3 position. There was a further acidic residue, Asp101, near the top of the cleft.

The structure of the (GlcNAc)3–lysozyme complex allowed an interpretation of the GlcNAc result. It was apparent that GlcNAc bound to site C in one or other of two distinct but closely related ways, depending upon whether it is in the α or β enantiomeric form (Fig.[link].


Figure | top | pdf |

Contemporary drawings of the binding to lysozyme of: (a) β-N-acetylglucosamine and (b) α-N-acetylglucosamine (Blake et al., 1967[link]).

The GlcNAc molecule had evidently undergone mutarotation in solution to produce an equilibrium mixture of α and β forms, both of which bound to lysozyme, although of course the specificity of the enzyme for longer oligosaccharides is for β-linked sugars. Both binding modes exploited the specificity of the N-acetyl group and its interactions with the enzyme.

One mode was characteristic of β-GlcNAc, as observed for the terminal sugar of (GlcNAc)3. The other mode was characteristic of α-GlcNAc and involved a rotation about the hydrogen bonds of the acetamido group to the protein. This finding could explain the result for 6-iodo-6-deoxy-N-acetyl α-methylglucosaminide, which could not fit into the α-GlcNAc site because of the additional methyl group but could only be accommodated in the `β-site'. The dual mode of binding for the terminal residue in the α or β configuration could also explain the bifurcated peak observed for the disaccharides GlcNAc-MurNAc and (GlcNAc)2. When the terminal sugar is in the α configuration, the second sugar is placed out of the main run of the cleft, while in the β configuration the second residue is as in site B of the trisaccharide complex. Proposals for the catalytic mechanism of lysozyme

| top | pdf |

A Royal Society Discussion Meeting had been arranged for 3 February 1966 under the organization of Max Perutz. CAV, from University College, London, who had worked on mechanisms for the non-enzymic hydrolysis of glycosides in solution, had been invited to contribute to this meeting. In mid-January (the exact date is not recorded) he visited the Royal Institution to discuss the possible reaction mechanism of lysozyme in the light of the crystallographic evidence. At that stage, the crystallography had shown, from low-resolution studies, the inhibitor binding sites within a surface cleft lined with several acidic residues that included Glu35 and Asp52. CAV's work had shown that the hydrolysis of methyl-α-D-glucopyranoside proceeded via a ring-closed carbonium-ion intermediate (i.e. a carbocation) and that the rate-determining step was the heterolysis of the C1 carbon–oxygen bond (reviewed in Vernon, 1967[link]). Rupley's studies with 18O water had shown that hydrolysis of oligosaccharide substrates by lysozyme also proceeded through cleavage of the C1—O bond (Rupley, 1967). Whereas in free solution the carbocation intermediate is presumed to be stabilized by interaction with solvent, in the catalytic site of lysozyme it was suggested that the carbonium-ion intermediate might be stabilized by the nearby ionized carboxylate of one of the acidic residues in the catalytic site.

The interpretations of the (GlcNAc)3 and the GlcNAc complexes were finished by Monday 31 January 1966. The date is significant. The Royal Society Discussion Meeting had been arranged for the following Thursday. The results with the non-hydrolysable complex with (GlcNAc)3 had identified the catalytic site and had provided immediate explanations for the specificity requirement for the enzyme. However, information on how a true substrate might bind was missing. This problem was solved in one day. DCP, noting that Rupley's work had shown the optimal substrate to be a hexasaccharide, constructed a hexasaccharide substrate in sites A–F, using the position of the experimentally determined trisaccharide in sites A–C and model building those sugars in sites D–F. Noting also the specificity of lysozyme for bacterial-cell-wall substrates, where the polysaccharide is composed of alternating [\beta(1\!\rightarrow\! 4)] linked GlcNAc and MurNAc residues, and the bond cleaved is that between MurNAc and GlcNAc residues, but not that between GlcNAc and MurNAc residues, he also examined sites likely to be specific for MurNAc. It was apparent that, for steric reasons, sites A, C and E could not accommodate MurNAc residues, but that sites B, D and F could do so. Thus the site of cleavage must be between sites D and E (or sites A and B, but this could be discarded because the trisaccharide complex was stable).

Close to the bond between sites D and E were two acid residues (Fig.[link]. As noted previously by members of the team, one of these, Glu35, was in a non-polar environment and shielded by Trp108, Val109 and Ala110. In this environment, the carboxylic acid was likely to have a raised pK and to be protonated at pH 5, the optimal pH for activity against homopolymers of GlcNAc. The other acid group, Asp52, was also buried, but in an environment that was largely polar and at the centre of a hydrogen-bonding network. Hence it might be predicted that its pK was likely to be similar to that of a carboxyl group in an aqueous environment, and the group would therefore be mostly ionized at pH 5. There was one other important factor. The sugar in site D could not be accommodated in the usual chair conformation of a glucopyranose ring. In order to avoid overcrowding between the C6O6 atoms of the alcohol group and the protein atoms at site D, it was necessary to distort the sugar ring from the conventional chair conformation to a sofa conformation. This relieved the overcrowding by bringing the C5—C6 bond axial and retained the links from C4—O4 and C1—O4′ equatorial, so that the adjacent sites of C and E were not perturbed by the distortion of the sugar in site D (Fig.[link]. CAV visited the Royal Institution on a second occasion between the completion of the model and the Royal Society meeting, probably on 1 February. When shown the hexasaccharide substrate complex with two acid residues on either side of the susceptible bond and the distortion of the sugar in site D, CAV explained that this was exactly what he would expect from his own work for a carbonium-ion mechanism with transition-state distortion of the substrate.


Figure | top | pdf |

Stereo-photographs of a model of the lysozyme molecule showing how a hexasaccharide substrate may bind to the enzyme. The yellow trisaccharide model shown at the top of the cleft is in the position occupied by tri-N-acetylchitotriose in the crystals of its complex with lysozyme. The orange trisaccharide in the lower part of the cleft has been added by model building alone. The fourth sugar residue from the top, the uppermost orange one, is distorted from the chair conformation (Blake et al., 1967[link]). The glycosidic linkage that is broken is the one between the fourth and fifth sugar residues from the top. It lies between the side chains of Glu35 (to the right) and Asp52 (to the left), the oxygen atoms of which are marked by red hemispheres. Asp101 and Asp103, near the top of the cleft, also have their oxygen atoms marked.

The mechanism was presented for the first time at the Royal Society meeting on 3 February 1966. It was noted that the nearest oxygen of the protonated Glu35 was about 3 Å from the glycosidic oxygen between residues D and E (the O1 atom of residue D and the equivalent O4 atom of residue E). On the other side of the cleft, the nearest oxygen of the ionized Asp52 was about 3 Å from the C1 atom of residue D and about the same distance from the O5 of that residue. The distances were, of course, subject to some error from experimental model building. It was suggested (Blake et al., 1967[link]; Phillips, 1967[link]) that:

  • (1) Asp52 carries a negative charge that promotes the formation of a carbonium ion at C1 of residue D and stabilizes it when formed;

  • (2) distortion of residue D from the chair conformation into the sofa conformation would contribute to stabilization of a carbonium ion at C1 by favouring a conformation in which the charge at C1 could be shared with the ring oxygen atom (Lemieux & Huber, 1955[link]), and hence contribute to the consequent weakening of the C1—O1 bond; and

  • (3) Glu35 could act as a proton donor, facilitating the formation of a hydroxyl group with the bridge oxygen atom (O1), and release of residues E and F.

Fig.[link] shows a sketch of the proposed lysozyme–hexasaccharide substrate complex prepared by Irving Geis for a Scientific American article published later that year (Phillips, 1966[link]).


Figure | top | pdf |

Draft sketch of the lysozyme–hexasaccharide substrate complex prepared by Irving Geis and annotated by David Phillips for an article published in Scientific American in 1966 (Phillips, 1966[link]).

As Perutz commented in his closing remarks at the Royal Society meeting: `For the first time we have been able to interpret the catalytic activity of an enzyme in stereochemical terms' (Perutz, 1967[link]). The scheme for catalysis has been subjected to numerous experimental tests and has been investigated further in crystallographic experiments. The mechanism provided a satisfactory explanation for the body of evidence already in existence at the time of the proposals and has been substantiated by the results of new experiments designed to test it (see, for example, Imoto et al., 1972[link]; Ford et al., 1974[link]; Strynadka & James, 1991[link]; Hadfield et al., 1994[link]; Jolles, 1996[link]).


This work would not have been possible without the continuous support and encouragement of Sir Lawrence Bragg and the Medical Research Council, UK, for which we were and are most grateful. We are also grateful to our colleagues Robert Canfield, Pierre Jolles, Gordon Lowe, John A. Rupley, Nathan Sharon and others who shared their results on the catalytic mechanism with us before publication. The successful outcome also depended greatly upon the skilled work of the craftsmen in the Royal Institution's mechanical and electronic workshops, Messrs T. H. Faulkner, S. B. Morris, J. E. T. Thirkell and A. R. Knott. Finally, we acknowledge gratefully the unstinting contributions made by our team of research assistants, Mrs W. J. Browne, Mrs A. Hartley, Mrs K. Sarma, Miss D. Glass, Mrs R. Arthanari, Mrs S. J. Cole, Mrs J. A. Conisbee and Miss M. Hibbs. Clearly, thirty years ago protein-structure analysis was still very labour-intensive, despite the advances in computing and diffractometry described above. This was largely because we were still designing and building our own apparatus and drawing maps and building models by hand. This situation has been transformed chiefly by the availability of more powerful computers and X-ray detectors, by the developments of computer graphics, and by the exploitation of synchrotron radiation. The review has used the original figures prepared during the 1960s with a few exceptions. We acknowledge with special thanks the skills of Stephen Lee in scanning and reproducing these figures.


Abraham, E. P. & Robinson, R. (1937). The crystallization of lysozyme. Nature (London), 140, 24.
Alderton, G., Ward, W. H. & Fevold, J. (1945). Crystallization of lysozyme chloride. J. Biol. Chem. 157, 43–58.
Arndt, U. W., North, A. C. T. & Phillips, D. C. (1964). Adaptation of a linear diffractometer to measure three X-ray reflections quasi-simultaneously. J. Sci. Instrum. 41, 421–425.
Arndt, U. W. & Phillips, D. C. (1961). The linear diffractometer. Acta Cryst. 14, 807–818.
Arndt, U. W. & Riley, D. P. (1952). Side-window proportional counters. Proc. Phys. Soc. London A, 65, 74–84.
Bijvoet, J. M. (1954). Structure of optically active compounds in the solid state. Nature (London), 173, 888–891.
Blake, C. C. F. (1968). The preparation of isomorphous derivatives. Adv. Protein Chem. 23, 59–120.
Blake, C. C. F., Fenn, R. H., North, A. C. T., Phillips, D. C. & Poljak, R. J. (1962). Structure of lysozyme. A Fourier map of the electron density at 6 Å resolution obtained by X-ray diffraction. Nature (London), 196, 1173–1178.
Blake, C. C. F., Johnson, L. N., Mair, G. A., North, A. C. T., Phillips, D. C. & Sarma, V. R. (1967). Crystallographic studies of the activity of hen egg-white lysozyme. Proc. R. Soc. London Ser. B, 167, 378–388.
Blake, C. C. F., Koenig, D. F., Mair, G. A., North, A. C. T., Phillips, D. C. & Sarma, V. R. (1965). Structure of hen egg-white lysozyme. A three-dimensional Fourier synthesis at 2 Å resolution. Nature (London), 206, 757–761.
Blake, C. C. F. & Phillips, D. C. (1962). Effects of X-irradiation on single crystals of myoglobin. In Biological Effects of Ionizing Radiation at the Molecular Level. Brno, Czechoslovakia: International Atomic Energy Agency, Vienna.
Blow, D. M. (1958). The structure of haemoblobin. VII. Determination of phase angles in the non-centrosymmetric [100] zone. Proc. R. Soc. London Ser. A, 247, 302–336.
Blow, D. M. & Crick, F. H. C. (1959). The treatment of errors in the isomorphous replacement method. Acta Cryst. 12, 794–802.
Blow, D. M. & Rossmann, M. G. (1961). The single isomorphous replacement method. Acta Cryst. 14, 1195–1202.
Bluhm, M. M., Bodo, G., Dintzis, H. M. & Kendrew, J. C. (1958). The crystal structure of myoglobin IV. A Fourier projection of sperm-whale myoglobin by the method of isomorphous replacement. Proc. R. Soc. London Ser. A, 246, 369–389.
Buerger, M. J. (1959). Vector Space. New York: Wiley.
Canfield, R. E. & Liu, A. K. (1965). The disulfide bonds of egg white lysozyme (muramidase). J. Biol. Chem. 240, 1997–2002.
Carlstrom, D. (1962). The polysaccharide chain of chitin. Biochim. Biophys. Acta, 59, 361–364.
Cheetham, J. C., Artymiuk, P. J. & Phillips, D. C. (1992). Refinement of an enzyme complex with inhibitor bound at partial occupancy. J. Mol. Biol. 224, 613–628.
Collins, J. F. & Richmond, M. (1962). A structural similarity between N-acetyl muramic acid and penicillin as a basis for antibiotic action. Nature (London), 195, 142–143.
Corey, R. B., Donohue, J., Trueblood, K. N. & Palmer, K. J. (1952). An X-ray investigation of air-dried lysozyme chloride crystals: the three-dimensional Patterson function. Acta Cryst. 5, 701–710.
Cox, E. G. & Jeffrey, G. A. (1939). Crystal structure of glucosamine hydrobromide. Nature (London), 143, 894–895.
Crick, F. H. C. & Magdoff, B. S. (1956). The theory of the method of isomorphous replacement for protein crystals. 1. Acta Cryst. 9, 901–908.
Cullis, A. F., Muirhead, H., Perutz, M. F., Rossmann, M. G. & North, A. C. T. (1961). The structure of haemoglobin VIII. A three-dimensional Fourier synthesis at 5.5 Å resolution: determination of the phase angles. Proc. R. Soc. London Ser. A, 265, 15–38.
Cullis, A. F., Muirhead, H., Perutz, M. F., Rossmann, M. G. & North, A. C. T. (1962). The structure of haemoglobin IX. A three-dimensional Fourier synthesis at 5.5 Å resolution: description of the structure. Proc. R. Soc. London Ser. A, 265, 161–187.
Dauben, C. H. & Templeton, D. H. (1955). A table of dispersion corrections for X-ray scattering of atoms. Acta Cryst. 8, 841–842.
Dickerson, R. E., Kendrew, J. C. & Strandberg, B. E. (1961). The crystal structure of myoglobin: phase determination to a resolution of 2 Å by the method of isomorphous replacement. Acta Cryst. 14, 1188–1195.
Dickerson, R. E., Reddy, J. M., Pinkerton, M. & Steinrauf, L. K. (1962). A 6 Å model of triclinic lysozyme. Nature (London), 196, 1178.
Fenn, R. H. (1964). An X-ray Crystallographic Study of Some Mercury Compounds and Their Use in Protein Structure Analysis. Davy Faraday Research Laboratory, The Royal Institution, London 182.
Fenn, R. H., Phillips, D. C. & Oldham, J. W. H. (1963). Crystal structure of (CH3)3S.HgI3 and the configuration of the [HgI_{\it 3}^{-}] ion. Nature (London), 198, 381–382.
Fleming, A. (1922). On a remarkable bacteriolytic element found in tissues and secretions. Proc. R. Soc. London Ser. B, 93, 306–317.
Ford, L. O., Johnson, L. N., Machin, P. A., Phillips, D. C. & Tjian, R. (1974). Crystal structure of a lysozyme–tetrasaccharide lactone complex. J. Mol. Biol. 88, 349–371.
Furnas, T. C. (1957). Single Crystal Orienter Instruction Manual. General Electric Company, Milwaukee, USA.
Green, D. W., Ingram, V. M. & Perutz, M. F. (1954). The structure of haemoglobin IV. Sign determination by the isomorphous replacement method. Proc. R. Soc. London Ser. A, 225, 287–307.
Green, D. W., North, A. C. T. & Aschaffenburg, R. (1956). Crystallography of the β-lactoglobulins of cows' milk. Biochim. Biophys. Acta, 21, 583–585.
Hadfield, A. T., Harvey, D. J., Archer, D. B., MacKenzie, D. A., Jeenes, D. J., Radford, S. E., Lowe, G., Dobson, C. M. & Johnson, L. N. (1994). Crystal structure of the mutant D52S hen egg white lysozyme with an oligosaccharide product. J. Mol. Biol. 243, 856–872.
Hamaguchi, K. & Imahori, K. (1964). Optical measurements of the secondary structure of lysozyme. J. Biochem (Tokyo), 55, 388–391.
Hamilton, W. C. (1955). On the treatment of unobserved reflexions in the least-squares adjustment of crystal structures. Acta Cryst. 8, 185–186.
Hamilton, W. C., Rollett, J. S. & Sparks, R. A. (1965). On the relative scaling of X-ray photographs. Acta Cryst. 18, 129–130.
Handoll, H. H. G. (1985). Crystallographic Studies of Proteins. DPhil thesis, University of Oxford.
Harker, D. (1956). The determination of the phases of the structure factors of non-centrosymmetric crystals by the method of double isomorphous replacement. Acta Cryst. 9, 1–9.
Hart, R. G. (1961). Refinement of heavy atom parameters other than relative y's. Acta Cryst. 14, 1194–1195.
Imoto, T., Johnson, L. N., North, A. C. T., Phillips, D. C. & Rupley, J. A. (1972). Vertebrate Lysozymes. New York and London: Academic Press.
Jacobson, R. A., Wunderlich, J. A. & Lipscomb, W. N. (1961). The crystal and molecular structure of cellobiose. Acta Cryst. 14, 598–607.
Jeffrey, G. A. (1990). Crystallographic studies of carbohydrates. Acta Cryst. B46, 89–103.
Johnson, L. N. (1966). The crystal structure of N-acetyl-α-D-glucosamine. Acta Cryst. 21, 885–891.
Johnson, L. N. (1967). An interaction between lysozyme and penicillin. Proc. R. Soc. London Ser. B, 167, 439–440.
Johnson, L. N. & Phillips, D. C. (1964). Crystal structure of N-acetyl­glucosamine. Nature (London), 202, 588–589.
Johnson, L. N. & Phillips, D. C. (1965). Structure of some crystalline lysozyme–inhibitor complexes determined by X-ray analysis at 6 Å resolution. Nature (London), 206, 761–763.
Jollès, J., Jaurrgui-Adell, J. & Jollès, P. (1964). Amino-acid sequence & S–S bridges of HEWL. C. R. Acad. Sci. Paris, 258, 3926–3928.
Jolles, P. (1996). Editor. Lysozymes: Model Enzymes in Biochemistry and Biology. Basel, Boston, Berlin: Birkhauser Verlag.
Kendrew, J. C., Bodo, G., Dintzis, H. M., Parrish, R. G., Wyckoff, H. & Phillips, D. C. (1958). A three-dimensional model of the myoglobin molecule obtained by X-ray analysis. Nature (London), 181, 662–666.
Kendrew, J. C., Dickerson, R. E., Strandberg, B. E., Hart, R. G., Davies, D. R., Phillips, D. C. & Shore, V. C. (1960). Structure of myoglobin. A three-dimensional Fourier synthesis at 2 Å resolution. Nature (London), 185, 422–427.
Kraut, J., Sieker, L. C., High, D. F. & Freer, S. T. (1962). Electron density map of chymotrypsinogen at 6 Å resolution. Proc. Natl Acad. Sci. USA, 48, 1417.
Lavington, S. (1980). Early British Computers: The Story of Vintage Computers and the People Who Built Them. Manchester University Press.
Lemieux, R. U. & Huber, G. (1955). Can. J. Res. 33, 128–133.
Lipson, H. & Cochran, W. (1968). The Determination of Crystal Structures. London: Bell.
Mo, F. & Jensen, L. H. (1975). A refined model for N-acetyl-α-D-glucosamine. Acta Cryst. B31, 2867–2873.
North, A. C. T. (1964). Computer processing of automatic diffractometer data. J. Sci. Instrum. 41, 42–45.
North, A. C. T. (1965). The combination of isomorphous replacement and anomalous scattering data in phase determination of non-centrosymmetric reflexions. Acta Cryst. 18, 212–216.
North, A. C. T., Phillips, D. C. & Mathews, F. S. (1968). A semi-empirical method of absorption correction. Acta Cryst. A24, 351–359.
Perutz, M. F. (1967). Concluding remarks. Proc. R. Soc. London Ser. B, 167, 448.
Perutz, M. F., Rossmann, M. G., Cullis, A. F., Muirhead, H., Will, G. & North, A. C. T. (1960). Structure of haemoglobin. A three-dimensional Fourier synthesis at 5.5 Å resolution, obtained by X-ray analysis. Nature (London), 185, 416–422.
Phillips, D. C. (1964). On the design of single crystal diffractometers to measure a number of reflections simultaneously. J. Sci. Instrum. 41, 123–129.
Phillips, D. C. (1966). The three-dimensional structure of an enzyme molecule. Sci. Am. pp. 78–90.
Phillips, D. C. (1967). The hen egg white lysozyme molecule. Proc. Natl Acad. Sci USA, 57, 484–495.
Poljak, R. J. (1963). Heavy atom attachment to crystalline lysozyme. J. Mol. Biol. 6, 244–246.
Ramachandran, G. N., Ramakrishnan, G. & Sasisekharan, V. (1963). Aspects of Protein Structure. London: Academic Press.
Rollett, J. S. (1961). Least-squares refinement in crystal-structure analysis. In Computing Methods and the Phase Problem in X-ray Crystal Analysis, p. 122. London: Pergamon Press.
Rollett, J. S. & Sparks, R. A. (1960). The correlation of intersecting layers of X-ray intensity data. Acta Cryst. 13, 273–274.
Rupley, J. A. (1964). The hydrolysis of chitin by concentrated hydrochloric acid and the preparation of low molecular weight substrates for lysozyme. Biochim. Biophys. Acta, 83, 245–255.
Rupley, J. A. (1967). The binding and cleavage by lysozyme of N-acetylglucosamine oligosaccharides. Proc. R. Soc. London Ser. B, 167, 416–428.
Scouloudi, H. (1960). Structure of seal myoglobin in projection. Proc. R. Soc. London Ser. A, 258, 181.
Scouloudi, H. (1965). The nature and configuration of the mercuri-iodide ion in the seal myoglobin derivative. J. Mol. Biol. 12, 17–26.
Stanford, R. H., Marsh, R. E. & Corey, R. B. (1962). An X-ray investigation of lysozyme chloride crystals containing complex ions of niobium and tantalum: three-dimensional Fourier plot obtained from data extending to a minimum spacing of 5 Å. Nature (London), 196, 1172–1177.
Steinrauf, L. K. (1959). Preliminary X-ray data for some new crystalline forms of β-lactoglobulin and hen-egg-white lysozyme. Acta Cryst. 12, 77–79.
Stryer, L., Kendrew, J. C. & Watson, H. C. (1964). The mode of attachment of the azide ion to sperm whale myoglobin. J. Mol. Biol. 8, 96–104.
Strynadka, N. C. J. & James, M. G. (1991). Lysozyme revisited: crystallographic evidence for distortion of an N-acetyl muramic acid residue bound in site D. J. Mol. Biol. 220, 401–424.
Tipper, D. J. & Strominger, J. L. (1965). Mechanism of action of penicillins: a proposal based on their structural similarity to acyl-D-alanyl-D-alanine. Proc. Natl Acad Sci. USA, 54, 1133–1141.
Vernon, C. A. (1967). The mechanisms of hydrolysis of glycosides and their relevance to enzyme-catalysed reactions. Proc. R. Soc. London Ser. B, 167, 389–401.
Waser, J. (1951). Lorentz factor for precession photographs. Rev. Sci. Instrum. 22, 563.
Watson, H. C., Kendrew, J. C., Coulter, C. L. & Brändën, C.-I. (1963). Progress with the 1.4 Å resolution myoglobin-structure determination. Acta Cryst. 16, A81.
Wells, M. (1960). Computation of absorption corrections on EDSAC II. Acta Cryst. 13, 722–736.
Wenzel, M., Lenk, H. P. & Schutte, E. (1962). Z. Physiol. Chem. 327, 13–20.
Wilson, A. J. C. (1942). Determination of absolute from relative X-ray intensity data. Nature (London), 150, 151–152.
Wise, E. M. & Park, J. T. (1965). Penicillin: its basis site of action as an inhibitor of a peptide cross-linking reaction in cell-wall mucopeptide synthesis. Proc. Natl Acad. Sci. USA, 54, 75–81.
Yang, J. T. & Doty, P. (1957). ORD measurements on lysozyme. J. Am. Chem. Soc. 79, 761.

to end of page
to top of page