Chapter 8.1. Synchrotron-radiation instrumentation, methods and scientific utilization

The so-called first generation of SR machines were those which were parasitic on high-energy physics operations, such as DESY in Hamburg, SPEAR in Stanford, NINA in Daresbury and VEPP in Novosibirsk. These machines had high fluxes into the X-ray range and enabled pioneering experiments. Parratt (1959) discussed the use of the CESR (Cornell Electron Storage Ring) for X-ray diffraction and spectroscopy in a very perceptive paper. Cauchois et al. (1963) conducted L-edge absorption spectroscopy at Frascati and were the first to diffract SR with a crystal (quartz). The opening experimental work in the area of biological diffraction was by Rosenbaum et al. (1971). In protein crystallography, multiple-wavelength anomalous-dispersion effects (Fig. 8.1.5.1) were used from the onset (Phillips et al., 1976, 1977; Phillips & Hodgson, 1980; Webb et al., 1977; Harmsen et al., 1976; Helliwell, 1977, 1979), and a reduction in radiation damage was seen (Wilson et al., 1983) for high-resolution data collection. Historical insights into the performances of those machines, from the current-day perspective, are described in detail, for example, by Huxley & Holmes (1997) at DESY, Munro (1997) at Daresbury, and Doniach et al. (1997) at Stanford. A principal limitation was the problem of source movements, which degraded the focusing of the source onto a small crystal or single fibre and thus degraded the intrinsic brilliance of the beam; see, for example, Haslegrove et al. (1977), who advocated machine shifts dedicated to SR as a working compromise with the high-energy physicists. Some possible applications discussed were unfulfilled until brighter sources became available. The two-wavelength crystallography phasing method of Okaya & Pepinsky (1956) (see also Hoppe & Jakubowski, 1975) and the three-wavelength method of Herzenberg & Lau (1967), as well as the implementation of the algebraic method of Karle (1967, 1980, 1989, 1994), awaited more stable beams, which had to be rapidly and easily tunable over a fine bandpass (<10⁻³). Experiments to define the anomalous-dispersion coefficients, including dichroism effects, at a large number of wavelengths at several example absorption edges in a variety of crystal structures were conducted at SPEAR (Phillips et al., 1978; Templeton et al., 1980, 1982; Templeton & Templeton, 1985). Values of f′ over a continuum of wavelengths in a real compound (i.e., not a metal in the gas phase) (Fig. 8.1.5.1b) were explored in a profile approach (now called DAFS, diffraction anomalous fine structure) by Arndt et al. (1982) at the newly commissioned SRS, the first dedicated second-generation SR source (see Section 8.1.5.2).

Figure 8.1.5.1| top | pdf | Anomalous dispersion. (a) f″ as represented by an absorption spectrum [Pt LIII edge for K2Pt(CN)4 as the example] (Helliwell, 1984). Reproduced with the permission of the Institute of Physics. (b) f′ as estimated by a continuous polychromatic profile method. Reproduced with permission from Nature (Arndt et al., 1982). Copyright (1982) MacMillan Magazines Limited.

The building of dedicated X-ray sources began with the SRS at Daresbury, which came online in 1980, having followed the NINA synchrotron (closed in 1976) and the associated Synchrotron Radiation Facility at Daresbury. Elsewhere in the world, LURE (Lemonnier et al., 1978) and CHESS at Cornell were building up their SR macromolecular crystallography operations in the late 1970s and early 1980s, and the NSLS in Brookhaven and the Photon Factory (PF) in Japan were both under construction. The NSLS and the PF came online in 1983 and 1984, respectively. Thus, there was a rapid increase in the number of operating machines and beamlines worldwide in the X-ray region for protein crystallography. There were teething problems with the SRS with the r.f. cavity window problem, interrupting operation for many months in 1983, and at the NSLS in its early period due to vacuum chamber problems. Pioneering experiments continued and blossomed. Seminal work ensued in virus crystallography [Rossmann & Erickson (1983) at Hamburg and Daresbury; Usha et al. (1984) at LURE], Laue diffraction for time-resolved protein crystallography [Moffat et al. (1984) at CHESS; Helliwell (1984, 1985) at the SRS; Cruickshank et al. (1987, 1991); Hajdu, Machin et al. (1987); Helliwell et al. (1989); Bourenkov et al. (1996); Neutze & Hajdu (1997)], enzyme catalysis in the crystal [Hajdu, Acharya et al. (1987) at the SRS], MAD [Phillips et al. (1977); Einspahr et al. (1985); Hendrickson (1985); Hendrickson et al. (1989) at SPEAR, the SRS and the PF; Guss et al. (1988) at SPEAR; Kahn et al. (1985) at LURE; Korszun (1987) at CHESS; Mukherjee et al. (1989) and Peterson et al. (1996) at the SRS; Hädener et al. (1999) at the SRS and the ESRF, to cite a few experiments], protein crystallography involving isomorphous replacement with optimized anomalous scattering [Baker et al. (1990) at the SRS; Dumas et al. (1995) at LURE], small crystals [Hedman et al. (1985) at the SRS] and diffuse scattering with SR [Doucet & Benoit (1987); Caspar et al. (1988); Glover et al. (1991)].

As early as 1979, there were discussions on planning a proposal for a high-brilliance, insertion-device-driven European synchrotron-radiation (ESR) source. A wide variety of discussion documents and workshops, and the ESR project led by B. Buras and based in Geneva at CERN, culminated in the so-called `Red Book' in 1987, the ESRF Foundation Phase Report (1987), totalling some 1000 pages of machine, beamline and experimental specifications and costs. This, then, was the progenitor of the third-generation sources, characterized by their high energy and high brilliance, tailored to optimized undulator emission in the 1 Å range. Actually, the ESRF machine energy was initially set at 5 GeV, but increased to 6 GeV to optimize the production of 14.4 keV photons to better match the nuclear scattering experiments proposed initially by Mossbauer in 1975. Proposals for the US machine, the Advanced Photon Source at 7 GeV, and the Japanese 8 GeV SPring-8 machine followed, with the higher machine energy enhancing the X-ray tuning range of undulators. Thus, MAD tuning-based techniques were facilitated with these machines and studies involving ultra-small samples (crystals, single fibres, or tiny liquid aliquots) or very large unit cells were enabled. As a result, micron-sized protein crystals as well as huge multi-macromolecular biological structures (of large viruses, for example) also became accessible.

Today a variety of enhanced national SR machines are being proposed and/or built. In the UK there is the DIAMOND 3 GeV machine and in France there is SOLEIL. The SLS in Switzerland, the country's first SR light source, is under construction. These machines are more tailored to the bulk of a country's user needs, distinct from the special provisions at the ESRF. The different countries' SR needs, of course, have many aspects in common, with some historical biases. The new sources are, in essence, characterized by high brilliance, i.e., low emittance. The 2 GeV high-brilliance SR source ELETTRA in Trieste, the MAXII machine in LUND and the Brazilian Light Source are already operational. In many ways, national sources like the SRS, LURE, DORIS and so on fuelled the case and specification for the ESRF. Now the developments at the ESRF, including high harmonic emission of undulators via magnet shimming (Elleaume, 1989) and narrow-gap undulator operation (Elleaume, 1998), are fuelling ideas and the specification of what is possible in these new national SR sources. Table 8.1.5.1 compares the parameters of the mature SRS of 1997 (from Munro, 1997) with the proposed design for DIAMOND (Suller, 1994). A shift of emphasis to high brilliance is again clear, as the applications of SR involving small samples dominate. Likewise, a 3 GeV machine energy is indicative of the need to include a provision of high photon energies for many applications, including, obviously, access to short-wavelength absorption edges. The extent to which undulators, for <3 GeV, will reach the hard X-ray region at high brilliance (e.g. around 1 Å wavelength) will depend on the minimum undulator magnet gaps realizable, along with magnet shimming to improve high harmonic emission. Moreover, longer wavelengths in protein crystallography are being explored on lower-energy SR machines (e.g. <3 GeV) at >1.5 Å, even 2.5 Å (Helliwell 1993, 1997a; Polikarpov et al., 1997; Teplyakov et al., 1998), and even softer wavelengths are under active development to utilize the S K edge for anomalous dispersion (Stuhrmann & Lehmann, 1994). Such developments interact closely with machine and beamline specifications. At very short (~0.5 Å) and ultra-short (~0.3 Å) wavelengths, a high machine energy yields copious flux output; pilot studies have been conducted in protein crystallography at CHESS (Helliwell et al., 1993) and at the ESRF (Schiltz et al., 1997).

Table 8.1.5.1| top | pdf | A comparison of the parameter list for the 2 GeV SRS, 1997, and the new higher-energy machine for the UK, DIAMOND S/C = superconducting magnet; MW = multipole wiggler (permanent magnet design).   SRS † DIAMOND ‡ Storage ring energy 2 GeV 3 GeV § Circumference 96 m 350 m ¶ Beam emittance 110 nm rad 15 nm rad Beam current after injection 300 mA 300 mA Typical dipole beam source sizes (σ)      horizontal 900 µm 400 µm  vertical 200 µm 150 µm Critical energy      dipole 3.2 keV 20 keV (S/C)  wiggler 13.3 keV (S/C) 10 keV (MW) †From Munro (1997). ‡From Suller (1994) and Suller (1998). §Up to 3.5 GeV. ¶A larger circumference is now proposed.

In terms of the evolution of X-ray sources, mention should be made of the X-ray free electron laser (XFEL); it now seems feasible that this will yield wavelength output well below the visible region of the electromagnetic spectrum. At DESY in Hamburg (Brinkmann et al., 1997) and at SLAC (Winick, 1995), such considerations and developments are being pursued. Compared to SR, one would obtain a transversely fully coherent beam, a larger average brilliance and, in particular, pulse lengths of ~200 fs full width at half-maximum with eight to ten orders of magnitude larger peak brilliance. Such a machine is based on a linear accelerator (linac)-driven XFEL utilizing a linear collider installation (e.g., for a high-energy physics centre-of-mass energy capability of 500 GeV). For this machine there is a `switchyard' distributing the electrons in a beam to different undulators from which the X-rays are generated in the range 0.1 to ~12 keV. The anticipated r.m.s. opening angle would be 1 mrad and the source diameter would be 20 µm. This source of X-rays would then compete in time resolution with laser-pulse-generated X-ray beams [see Helliwell & Rentzepis (1997) for a survey of that work and a comparison with synchrotron radiation] and would also have higher brilliance.

Laue geometry involves the use of the polychromatic SR spectrum as transmitted through the beryllium window that is used to separate the apparatus from the machine vacuum. There is useful intensity down to a wavelength minimum of , where is the critical wavelength of the magnet source. The maximum wavelength is typically ≥3 Å; however, if the crystal is mounted in a capillary, then the glass absorbs the wavelengths beyond ~2.5 Å.

The bandwidth can be limited somewhat under special circumstances. A reflecting mirror at grazing incidence can be used for two purposes. First, the minimum wavelength in the beam can be sharply defined to aid the accurate definition of the Laue spot multiplicity. Second, the mirror can be used to focus the beam at the sample. The maximum-wavelength limit can be truncated by use of aluminium absorbers of varying thickness or a transmission mirror (Lairson & Bilderback, 1982; Cassetta et al., 1993).

The measured intensity of individual Laue diffraction spots depends on the wavelength at which they are stimulated. The problem of wavelength normalization is treated by a variety of methods. These include:

Each of these methods produces a `λ curve' describing the relative strength of spots measured at various wavelengths. The methods rely on the incident spectrum being smooth and stable with time. The bromine and silver K absorption edges, in AgBr photographic film, lead to discontinuities in the λ-curve. Hence, the λ-curve is usually split up into wavelength regions, for example to 0.49 Å, 0.49 to 0.92 Å, and 0.92 Å to . Other detector types have different discontinuities, depending on the material making up the X-ray absorbing medium. Most Laue diffraction data now recorded on CCDs or IPs. The greater sensitivity of these detectors (expressed as the DQE), especially for weak signals, has greatly increased the number of Laue exposures recordable per crystal. Thus, multiplet deconvolution procedures, based on the recording of reflections stimulated at different wavelengths and with different relative intensities, have become possible (Campbell & Hao, 1993; Ren & Moffat, 1995b). Data quality and completeness have improved considerably.

The production and use of narrow-bandpass beams, e.g. , may be of interest for enhancing the signal-to-noise ratio. Such bandwidths can be produced by a combination of a reflection mirror used in tandem with a transmission mirror. Alternatively, an X-ray undulator of 10–100 periods should ideally yield a bandwidth behind a pinhole of 0.1–0.01. In these cases, wavelength normalization is more difficult, because the actual spectrum over which a reflection is integrated is rapidly varying in intensity; nevertheless, high-order Chebychev polynomials are successful (Ren & Moffat, 1995a).

The spot bandwidth is determined by the mosaic spread and horizontal beam divergence (since ) as where η is the sample mosaic spread, assumed to be isotropic, is the horizontal cross-fire angle, which in the absence of focusing is , where is the horizontal sample size, is the horizontal source size and P is the sample to the tangent-point distance. This is similar for in the vertical direction. Generally, at SR sources, is greater than . When a focusing-mirror element is used, and/or are convergence angles determined by the focusing distances and the mirror aperture.

The size and shape of the diffraction spots vary across the detector image plane. The radial spot length is given by convolution of Gaussians as and tangentially by where is the size of the X-ray beam (assumed to be circular) at the sample, is the detector point spread factor, and where ψ is the angle between the vertical direction and the radius vector to the spot (see Andrews et al., 1987). For a crystal that is not too mosaic, the spot size is dominated by and . For a mosaic or radiation-damaged crystal, the main effect is a radial streaking arising from η, the sample mosaic spread.

A wide variety of perfect-crystal monochromator configurations are possible and have been reviewed by various authors (Hart, 1971; Bonse et al., 1976; Hastings, 1977; Kohra et al., 1978). Since the reflectivity of perfect silicon and germanium is effectively 100%, multiple-reflection monochromators are feasible and permit the tailoring of the shape of the monochromator resolution function, harmonic rejection and manipulation of the polarization state of the beam. Two basic designs are in common use. These are the bent single-crystal monochromator of triangular shape (Fig. 8.1.4.1a) and the double-crystal monochromator (Fig. 8.1.4.1b).

8.1.7.2.1. Curved single-crystal monochromator

| top | pdf |

In the case of the single-crystal monchromator, the actual curvature employed is very important in the diffraction geometry. For a point source and a flat monochromator crystal, there is a gradual change in the photon wavelength selected from the white beam as the length of the monochromator is traversed (Fig. 8.1.7.1a). For a point source and a curved monochromator crystal, one specific curvature can compensate for this variation in incidence angle (Fig. 8.1.7.1b). The reflected spectral bandwidth is then at a minimum; this setting is known as the `Guinier position'. If the curvature of the monochromator crystal is increased further, a range of photon wavelengths, , is selected along its length so that the rays converging towards the focus have a correlation of photon wavelength and direction (Fig. 8.1.7.1c). The effect of a finite source is to cause a change in incidence angle at the monochromator crystal, so that at the focus there is a photon-wavelength gradient across the width of the focus (for all curvatures) (Fig. 8.1.7.1d). The use of a slit in the focal plane is akin to placing a slit at the tangent point to limit the source size.

Figure 8.1.7.1| top | pdf | Single-crystal monochromator illuminated by SR. (a) Flat crystal. (b) Guinier setting. (c) Overbent crystal. (d) Effect of source size (shown at the Guinier setting for clarity). From Helliwell (1984). Reproduced with the permission of the Institute of Physics.

8.1.7.2.2. Double-crystal monochromator

| top | pdf |

The double-crystal monochromator with two parallel or nearly parallel perfect crystals of germanium or silicon is a common configuration. The advantage of this is that the outgoing monochromatic beam is parallel to the incoming beam, although it is slightly displaced vertically by an amount , where d is the perpendicular distance between the crystals and θ is the monochromator Bragg angle (unless the second crystal is unconnected to the first, in which case it can be translated as well to compensate for that). The monochromator can be rapidly tuned, since the diffractometer or camera need not be re-aligned significantly in a scan across an absorption edge. Since the rocking width of the fundamental is broader than the harmonic reflections, the strict parallelism of the pair of the crystal planes can be relaxed or `detuned', so that the harmonic can be rejected with little loss of the fundamental intensity. The spectral spread in the reflected monochromatic beam is determined by the source divergence accepted by the monochromator, the angular size of the source and the monochromator rocking width (see Fig. 8.1.7.2). The double-crystal monochromator is often used with a toroidal focusing mirror; the functions of monochromatization are then separated from the focusing (Hastings et al., 1978).

Figure 8.1.7.2| top | pdf | Double-crystal monochromator illuminated by SR. The contributions of the source divergence, αV [less than or equal to γ−1, equation (8.1.2.4), depending on the monochromator vertical entrance slit aperture; see also Colapietro et al., 1992], and angular source size, , to the range of energies reflected by the monochromator are shown. From Helliwell (1984). Reproduced with the permission of the Institute of Physics.

8.1.7.2.3. Crystal sample rocking width

| top | pdf |

The rocking width of a reflection depends on the horizontal and vertical beam divergence or convergence (after due account for collimation is taken), and , the spectral spreads and , and the mosaic spread, η. We assume that the mosaic spread η is , the angular broadening of a reciprocal-lattice point (relp) due to a finite sample. In the case of synchrotron radiation, and are usually widely asymmetric. On a conventional source, usually . Two types of spectral spread occur with synchrotron (and neutron) sources. The term is the spread that is passed down each incident ray in a divergent or convergent incident beam; the subscript refers to the conventional source type. This is because it is similar to the line widths and separation. At the synchrotron, this component also exists and arises from the monochromator rocking width and finite-source-size effects. The term is special to the synchrotron or neutron case. The subscript `corr' refers to the fact that the ray direction can be correlated with the photon or neutron wavelength. In this most general case, and for one example of a arising from the horizontal ray direction correlation with photon energy and the case of a horizontal rotation axis, the rocking width of an individual reflection is given by where and L is the Lorentz factor, .

The Guinier setting of an instrument (curved crystal monochromator case, Fig. 8.1.7.1b) gives . The equation for then reduces to (from Greenhough & Helliwell, 1982). For example, for , (0.01°), , and (0.05°), then . But increases as ζ increases [see Greenhough & Helliwell (1982), Table 5]. In the rotation/oscillation method as applied to protein and virus crystals, a small angular range is used per exposure. For example, the maximum rotation range per image, , may be 1.5° for a protein and 0.4° or so for a virus. Many reflections will be only partially stimulated over the exposure. It is important, especially in the virus case, to predict the degree of penetration of the relp through the Ewald sphere. This is done by analysing the interaction of a spherical volume for a given relp with the Ewald sphere. The radius of this volume is given by (Greenhough & Helliwell, 1982).

In Fig. 8.1.7.3, the relevant parameters are shown. The diagram shows in a plane, usually horizontal with a perpendicular (vertical) rotation axis, whereas the formula for above is for a horizontal axis. This is purely for didactic reasons since the interrelationship of the components is then much clearer.

Figure 8.1.7.3| top | pdf | The rocking width of an individual reflection for the case of Fig. 8.1.7.1(c) and a vertical rotation axis. From Greenhough & Helliwell (1982).

The use of high SR intensity, cryo-freezing of a protein crystal to largely overcome radiation damage and sensitive, automatic area detectors (CCDs and/or image plates) is allowing diffraction data to be recorded at resolutions equivalent to smaller molecule (chemical) crystallography. In a growing number of protein crystal structure studies, atomic resolution (1.2 Å or better) is achievable (Dauter et al., 1997). The `X-ray data to parameter' ratio can be favourable enough for single and double bonds, e.g. in carboxyl side chains, to be resolved [Fig. 8.1.8.1; Deacon et al. (1997) for concanavalin A at 0.94 Å resolution]. Along with this bond distance precision, one can see the reactive proton directly. This approach complements H/D exchange neutron diffraction studies. Neutron studies have recently expanded in scope by employing Laue geometry in a synergistic development with SR Laue diffraction (Helliwell & Wilkinson, 1994; Helliwell, 1997b; Habash et al., 1997, 2000). The scope and accuracy of protein crystal structures has been transformed.

Figure 8.1.8.1| top | pdf | Determination of the protonation states of carboxylic acid side chains in proteins via hydrogen atoms and resolved single and double bond lengths. After Deacon et al. (1997) using CHESS. Reproduced by permission of The Royal Society of Chemistry.

Compensating for small crystal sample volume by increasing the intensity at the sample has been of major interest from the outset, and tests have shown that the use of micron-sized samples is feasible (Hedman et al., 1985). Third-generation high-brilliance sources are optimized for this application via micron-sized focal spot beams, as described in the ESRF Foundation Phase Report (1987). Applications of the ESRF microfocus beamline include the determination of the structure of the bacteriorhodopsin crystal at high resolution from micro-crystals (Pebay-Peyroula et al., 1997). Experiments using extremely thin plates involving only 1000 protein molecular layers are described by Mayans & Wilmanns (1999) on the BW7B wiggler beamline at DESY, Hamburg. A review of small crystals and SR, including tabulated sample scattering efficiencies, can be found in Helliwell (1992), pp. 410–414.

Time-resolved SR Laue diffraction of light-sensitive proteins, such as CO Mb studied with sub-nanosecond time resolution in pump-probe experiments (see Srajer et al., 1996), are showing direct structural changes as a function of time. Enzymes, likewise, are being studied directly by time-resolved methods via a variety of reaction initiation methods, including pH jump, substrate diffusion and light flash of caged compounds pre-equilibrated in the crystal. Flash freezing is used to trap molecular structures at optimal times in a reaction determined either by microspectrophotometry or repeated Laue `flash photography'. For overviews, see the books edited by Cruickshank et al. (1992) and Helliwell & Rentzepis (1997). Enzyme reaction rates can be altered through site-directed mutagenesis (e.g. see Niemann et al., 1994; Helliwell et al., 1998) and matched to diffraction-data acquisition times.

Multi-macromolecular complexes, such as viruses (Rossmann et al., 1985; Acharya et al., 1989; Liddington et al., 1991) (Fig. 8.1.8.2), the nucleosome (Luger et al., 1997), light-harvesting complex (McDermott et al., 1995) and the 13-subunit membrane-bound protein cytochrome c oxidase (Tsukihara et al., 1996), and large-scale molecular assemblies like muscle (Holmes, 1998) are very firmly recognizable as biological entities whose crystal structure determinations rely on SR. These single-crystal structure determinations involve extremely large unit cells and are now tractable despite very weak scattering strength. The crystals often show extreme sensitivity to radiation (hundreds, even a thousand, crystals have been used to constitute a single data set). Cryocrystallography radiation protection is now used extensively in crystallographic data collection on whole ribosome crystals (Hope et al., 1989); SR is essential for this structure determination (Yonath, 1992; Yonath et al., 1998; Ban et al., 1998). These large-scale molecular assemblies often combine electron-microscope and diffraction techniques with SR X-ray crystallography and diffraction for low-to-high resolution detail, respectively. A major surge in results has come from the ESRF, where the X-ray undulator radiation, of incredible intensity and collimation in a number of beamlines (Helliwell, 1987; Miller, 1994; Branden, 1994; Lindley, 1999), has been harnessed to yield atomic level crystal structures of the 780 Å diameter blue tongue virus (Grimes et al., 1997, 1998) and the nucleosome core particle (Luger et al., 1997). A very large multi-protein complex solved using data from the Daresbury SRS wiggler is the F₁ ATPase structure (Fig. 8.1.8.3), for which a share in the Nobel Prize for Chemistry in 1997 was awarded to John Walker in Cambridge. The structure (Abrahams et al., 1994; Abrahams & Leslie, 1996) and the amino-acid sequence data, along with fluorescence microscopy, show how biochemical energy is harnessed to drive the proton pump across biological membranes, thus corroborating hypotheses about this process made over many years. This study, made tractable by the SRS wiggler high-intensity protein crystallography station (Fig. 8.1.4.1), illustrates the considerable further scope possible with yet stronger, more brilliant SR undulator and multipole wiggler sources.

Figure 8.1.8.2| top | pdf | A view of SV40 virus (based on Liddington et al., 1991) determined using data recorded at the SRS wiggler station 9.6 (Fig. 8.1.4.1a).

Figure 8.1.8.3| top | pdf | The protein crystal structure of F1 ATPase, one of the largest non-symmetrical protein structure complexes, solved using SR data recorded at the SRS wiggler 9.6, Daresbury. The scale bar is 20 Å long. Reprinted with permission from Nature (Abrahams et al., 1994). Copyright (1994) MacMillan Magazines Limited.

Rapid protein structure determination via the MAD method of seleno protein variants (Hendrickson et al., 1990), as well as xenon pressure derivatives (Schiltz et al., 1997), and improved heavy-atom isomorphous replacement data are removing a major bottleneck in protein crystallography, that of phase determination. One example of a successful MAD study with an especially weak anomalous signal, from one selenium atom per 147 amino acids, undertaken at ESRF BM14, is that of van Montfort et al. (1998). At another extreme is the largest number of anomalous scatterer sites; for example, Turner et al. (1998), using the NSLS, reported the successful determination of 30 selenium atoms in 96 kDa of protein (one dimer) in the asymmetric unit using one-wavelength anomalous differences (at peak) as E values and `Shake n' Bake' (Miller et al., 1994), followed by MAD phasing from three-wavelength data and solvent flattening. Overall, as the number of protein structures in the Protein Data Bank doubles every few years (currently the number is 9000), the possibility of considering whole genome-level structure determinations arises (Chayen et al., 1996; Chayen & Helliwell, 1998). The human genome, the determination of the amino-acid sequence of which is currently underway, comprises some 100 000 proteins. Of these, some 40% are membrane bound and somewhat difficult to crystallize. A MAD protein crystal structure currently requires roughly 1 day of SR BM beamtime. A coordination of 20 SR instruments worldwide, or an SR machine devoted solely to the project, could make major progress in 20 years. This estimate assumes no further speeding up of the technique, such as would acrue with faster detectors like the pixel detector. The smaller yeast genome, comprising amino-acid sequences of 10 000 proteins, has recently been completed. The molecular-weight histogram peaks at 30 kDa. Assuming, on average, that one amino acid out of 56 is a methionine, it is clear that the MAD method, and six Se atoms on average for each protein, is a good match to the task. This approach, along with homology modelling and genetic alignment techniques, opens the immense potential for `structural genomics' as a basis for understanding and controlling disease (e.g. see Bugg et al., 1993). SR and crystallography are now intricately intertwined in their scientific futures and in facilities provision (Helliwell, 1998).