2.1 Instrumentation

2.1.1 Synchrotron light sources

SRCD stations use SR light in the UV and VUV regions; to date nearly all SRCD beamlines (Table 1) are installed on bending magnets. The flux off of bending magnets in this wavelength range is more than sufficient for the requirements of CD experiments (indeed higher flux densities are contraindicated for protein studies due to radiation heating effects (see Section 2.4). Using bending magnets instead of insertion devices has been a major advantage in obtaining approval for inclusion of SRCD beamlines as part of the portfolio of beamlines at many synchrotrons: bending magnets are not only less costly, but such ports tend to be in considerably less demand for physics, materials and crystallography beamlines than are ones on insertion devices such as undulators and wigglers.

Table 1.

Characteristics of SRCD Beamlines

• Key:

• SF=Stopped Flow

• TJ=Temperature Jump

• CF=Continuous Flow

• LD=Linear Dichroism

• F=Fluorescence

• (p)=planned

• Websites:

• U11: http://www.nsls.bnl.gov/beamlines/beamline.asp?blid=U11

• U9b: http://www.nsls.bnl.gov/beamlines/beamline.asp?blid=U9b

• 3.1: http://www.srs.ac.uk/srs/stations/station3.1.htm

• CD12: http://www.srs.dl.ac.uk/VUV/CD/new12.html

• UV1: http://www.isa.au.dk/facilities/astrid/beamlines/uv1/uv1OpticalSpecs.asp

• CD1: http://www.isa.au.dk/facilities/astrid/beamlines/cd1/CD1OpticalSpecs.asp

• BL15: http://www.hsrc.hiroshima-u.ac.jp/english/bl15.htm

• 4B8: http://www.ihep.ac.cn/bsrf/english/facility/html/VUV.htm

• 3m_NIM-C: http://www.bessy.de/bit/bit_show_object.html.php?i_bit_id_object=168

• 04B1: http://140.110.203.42/manage/fck_fileimage/file/bldoc/04BSNM.htm

• B23: http://www.dls.ac.uk/Home/Beamlines/B23.html

• DISCO: http://www.synchrotron-soleil.fr/Recherche/LignesLumiere/DISCO/DescriptionLigne#equipe

• CD12: http://ankaweb.fzk.de/website.php?page=instrumentation_beam&id=21&field=1

Because of the wavelength ranges covered and the lower flux requirements, many very successful SRCD beamlines have been installed on low energy synchrotrons such as ISA (0·58 GeV) and HiSOR (0·7 GeV). Indeed, SRCD beamlines that are being developed on higher-energy third-generation synchrotron sources such as Soleil (in France – 2·75 GeV) and Diamond Light Source (in the United Kingdom – 3 GeV) have encountered additional design issues due to the high flux of photons produced. These include the materials used for the windows (they can rapidly develop colour centres), the requirement for a cold finger on the first mirror to dissipate the heat generated by the white light (including hard X-rays) impinging on the mirror, the requirement to spread the beam on the sample thereby reducing the flux density and, potentially, the need to include filters or baffles or additional reflecting surfaces to decrease the total flux onto the sample.

2.1.3 Synchrotron radiation circular dichroism sample cells

The sample cells ordinarily used for cCD spectroscopy are made from Suprasil quartz, a material that has good transparency properties in the near- and far-UV wavelength range. Due to the limited light flux in cCD instruments, the optical properties of the material used in the sample cell windows are generally not the factors that limit the lowest wavelength achievable. However, when using SR as the light source, the light flux is not the limiting factor; instead the absorbance properties of the materials that lie in the beam limit the transmission of the low-wavelength light. These can include the material used for windows in the beamline itself, but also include the materials used in the manufacture of the sample cells, as well as the contents of those cells, especially the solvent present. In the case of water, used for most biological experiments, the molar excess of water to protein and the absorbance of the water (peak ~168 nm) means that if there is a large amount of water in the beam (i.e. the sample cell has a long pathlength) this will be an important factor limiting the minimum wavelength measurable. The material used to make the sample cells is also critical, since even very thin quartz cells tend to result in wavelength cut-offs in the range of 190 nm (the exact value depending on the thickness and composition of the quartz). Because of these two factors, several attempts have been made to produce new sample cells for use in SRCD measurements that employ either different materials that are more optically transparent or have shorter pathlengths so there is less solvent in the beam or both.

The first such cell specifically designed for SRCD measurements, designated the Gray cell, was developed at the NSLS (Gray et al. Reference Gray, Lang, Kuner, Vaughan and Sutherland1984). It was constructed from thin quartz plates that were demountable (which also aided in cleaning) and could be made to have different pathlengths depending on the thickness of Teflon spacer used between the windows. They had the advantages of ease of use, very thin pathlengths, and compatibility with existing circular cell holders. They permitted measurements down to ~185 nm, a significant achievement at the time. A disadvantage was the physical properties of the Telfon spacers when they became wet and stretched, which made manipulation and reproducible loading of the cells challenging.

The second type of cell produced some years later (Matsuo et al. Reference Matsuo, Sakai, Matsushima, Fukuyama and Gekko2003) extended the idea of short pathlengths, but made the cells out of magnesium fluoride. The design made them suitable for use in a vacuum sample chamber, such as is present at the HiSOR beamline. They used adjustable spacers in the range from 1·3 to 50 μm to achieve the different pathlengths, which reportedly enabled measurements down to 160 nm in aqueous solutions (although the data realistically seemed to cut-off at around 170 nm) and showed good thermal stability over a wide temperature range from −30 to 70°C. Disadvantages include the fragility of the material and the time required for their loading and assembly and the issue of calibration and reproducibility after adjustment of the pathlength.

The third type of cell specially designed for SRCD measurements (the ‘Birkbeck’ cell) was one made from calcium fluoride (Wien & Wallace, Reference Wien and Wallace2005). These cells, with fixed pathlengths ranging from ~4 to ~20 μm, had the advantages of being made from an optically transparent material enabling data collection to low wavelengths (Fig. 4a), plus the short pathlengths needed to limit the amount of solvent present. In addition, when placed in their specially designed cell holder, they could be loaded quickly and reproducibly and could be unloaded and cleaned easily, even allowing recovery of some of the solution afterwards. Measurements in aqueous solution were possible to ~167 nm. Disadvantages included the birefringence present in the calcium fluoride crystals, which meant that the two plates had to have very consistent alignments relative to each other for the corresponding sample and baseline loading. This was overcome by the design of a cell holder that produced not only reproducible pressure, resulting in identical pathlengths on each loading, but also reproducible orientations of the plates (Wien & Wallace, Reference Wien and Wallace2005). An additional advantage of the custom-made cell holder was that it could be used on both conventional CD instruments and at all existing SRCD beamlines, thereby enabling strictly comparable measurements and cross calibration studies. A commercial version of these cells is now available (F. Wien, personal communication; Tao & Wallace, Reference Tao and Wallace2009).

Fig. 4.

High-tension plots of different sample cells with and without different solvents, showing their relative absorbance characteristics and their low-wavelength cut-off limits. (a) Comparison of empty calcium fluoride ‘Birkbeck’ cells (purple) and empty quartz Suprasil cells (black). The peaks between ~130 and 140 nm that can be seen in the empty calcium fluoride cells are due to the nitrogen used to flush the sample chamber and can be used for wavelength calibrations of the beamline. (b) Calcium fluoride cells without solvent (purple), with trifluorethanol (green), acetonitrile (red) and water (blue).

An added benefit of all of the custom-made short pathlength cells, in addition to the lower wavelength achievable, is that they tend to have a much smaller volume and therefore require proportionately less material than do the longer pathlength standard CD cells. For example, the Birkbeck cells require only ~2 μl of solution, as opposed to ~50–200 μl for standard demountable cells. A disadvantage of all short pathlength cells, however, is that usually they require reasonably high protein concentrations (~1–10 mg/ml); however, conditions of pathlength and buffers can be chosen which require concentrations of only ~0·1 mg/ml solutions (1 μm for a protein of MW 100 000), which can usually be attainable. Conversely, the requirement for higher protein concentrations means that samples can be examined under conditions that more closely approximate the conditions used for crystallisation of proteins (~5–20 mg/ml) or the protein concentrations present in ‘crowded’ cells.

In general, sample cells used in SRCD instruments tend to be circular rather than the more common rectangular cells used in many cCD instruments. This is because circular cells are isotropic in shape and thus tend to be more compatible with the more isotropic beam shapes produced by beamlines; they further have the advantage that for scattering samples such as membranes, the light is symmetrically transmitted in all directions before impinging on the detector face, thereby giving rise to less apparent differential light scattering artefacts for these type of samples (see Section 4.1.2). In addition, circular cells exhibit less strain-induced dichroism.

Hence, the developments of new sample cells have also greatly aided the quality of the data obtainable by SRCD.

2.2 Cross-calibrations

An essential consideration if protein spectra are to be compared from different beamlines and cCD instruments is that all are properly calibrated and cross-calibrated so that the same spectrum is obtained for each protein no matter where it is measured. This is especially important if standard reference data sets are to be used in empirical analyses of the data (see Section 3.2). The need for cross-calibration may seem fairly obvious, however, several years ago, two independent studies (Jones et al. Reference Jones, Schiffmann, Knight and Windsor2004; Miles et al. Reference Miles, Wien, Lees, Rodger, Janes and Wallace2003) examining the comparabilities of spectra collected both on different types of conventional CD instruments and on different instruments of the same type in different labs, showed that there was a massive variation between the spectra obtained, even though all examined the exact same protein sample provided by a single laboratory (Fig. 5a). This may have been because, in general, most laboratory-based CD instruments rarely tend to have calibration checks done on them, and even when they do, the calibration tests tend to use a single peak (290·5 nm) in the spectrum of one standard material, usually camphor sulphonic acid (CSA) or the related compound ammonium camphorsulphonate (ACS). Single calibration points, however, are not sufficient to ensure proper calibration across the entire UV region used in CD studies. As a result, a series of standards and protocols (Miles et al. Reference Miles, Wien, Lees, Rodger, Janes and Wallace2003, Reference Miles, Wien, Lees and Wallace2005a; Miles & Wallace, Reference Miles, Wallace, Wallace and Janes2009a; Ravi et al. Reference Ravi, Hills, Knight, Wallace and Janes2009) were established to cover the magnitudes, optical rotations and peak positions across the spectral ranges of interest for both cCD and SRCD instruments. In addition, software was produced to facilitate such cross-calibrations (Lees et al. Reference Lees, Smith, Wien, Miles and Wallace2004) (Fig. 5b) and the molar elliptical extinction coefficient of CSA was accurately re-determined (Miles et al. Reference Miles, Wien and Wallace2004). These standards and procedures have since been used to test and cross-calibrate many SRCD beamlines (Miles et al. Reference Miles, Wien, Lees, Rodger, Janes and Wallace2003, Reference Miles, Wien, Lees and Wallace2005a, Reference Miles and Wallace2007; Janes & Wallace, unpublished). The results obtained for all the SRCDs were remarkably consistent, and the spectra of a number of test proteins such as myoglobin, lysozyme, concanavalin A and collagen showed excellent agreement between the beamlines at SRS, NSLS, the National Synchrotron Radiation Research Centre (NSRRC), ISA, Soleil, and the Beijing Synchrotron Research Facility (BSRF). This is likely because SRCD beamlines are regularly (sometimes as often as every beam fill) calibrated. As a result the data they produce are in general much better cross-correlated than conventional CD spectra. Furthermore, at a recent international SRCD meeting (Tao & Wallace, Reference Tao and Wallace2009), beamline scientists of both new and old beamlines agreed on standards and values for calibration of their SRCD beamlines, which should maintain this high correspondence between data obtained at SRCD beamlines in the future. This will be very important for empirical analyses using the new reference data sets collected on such well-calibrated beamlines (Lees et al. Reference Lees, Miles, Wien and Wallace2006a).

Fig. 5.

Cross-calibration: (a) Comparison of CD spectra of myoglobin obtained on two different instruments that have not been cross-calibrated. (b) Cross-calibration correction curves produced using CDtool software (Lees et al. Reference Lees, Smith, Wien, Miles and Wallace2004). (c) CD spectra (from a) after application of the cross-calibration procedure. (Adapted from Miles et al. Reference Miles, Wien, Lees, Rodger, Janes and Wallace2003.)

2.3 Practical considerations

Although CD is an apparently simple technique suitable for employment by non-expert users, its apparent simplicity means it is also simple for errors to be introduced by the way data are collected, resulting in problems in the final spectra produced. An excellent review of good practice procedures for CD by Kelly et al. (Reference Kelly, Jess and Price2005) includes a discussion of common errors and their solutions. These issues are all also relevant to SRCD, but in addition there also are a number of further considerations for good practice for SRCD data collection that are discussed by Miles & Wallace (Reference Miles, Wallace, Wallace and Janes2009b) and reviewed by Miles & Wallace (Reference Miles and Wallace2006).

The most common problem in CD and SRCD is caused by using samples that have too high an absorbance, out of the range of the detection limits. This can be monitored by simultaneously recording the dynode voltage or detector high-tension (HT) or high-voltage readings, all of which are effectively pseudoabsorbance measurements. This problem can also be detected by comparisons of spectra of the same sample obtained in different pathlengths or at different concentrations (Miles & Wallace, Reference Miles and Wallace2006). In many cases, the high absorbance is not the result of the protein component but may arise due other components included in protein samples to maintain their structural integrity (such as buffers, salts and other additives) that have significant absorbance in the UV wavelength range. Although such components tend to be non-chiral and thus do not produce a CD signal, they do absorb each of the left- and right-circularly polarised light beams, resulting in a greatly diminished transmission of light through the sample and impinging on the detector. A commonly-used example of a component producing high absorbance is high concentrations of sodium chloride, as the chloride absorbs significantly in the far-UV and VUV regions. Replacement of sodium chloride with sodium fluoride can often improve the ability to measure the protein components, as can the replacement of certain amide-based buffers with other non-absorbing buffers (Kelly & Price, Reference Kelly, Price, Wallace and Janes2009; Miles & Wallace, Reference Miles and Wallace2006).

The limitation to the wavelength limit achievable also arises in biological solutions due to the solvent water, which is present in vast molar abundance relative to the protein and which also absorbs in the VUV region (with a peak at ~168 nm and significant absorbance as high as 190–200 nm). Indeed, in the absence of water, spectra can be measured down to at least 130 nm (Fig. 6), revealing even more peptide transitions in the wavelength range between 130 and 170 nm. By using short pathlength cells, the water absorbance can be minimised, but not eliminated. All of these factors together conspire to limit the light flux into and out of a protein sample, in many cases making measurements below even 200 nm difficult in a conventional CD instrument and practically limiting SRCD measurements to above ~170 nm. Although some conventional CD instruments attempt to overcome the limited light flux by opening their slits at low wavelengths, but changes in the spectral bandwidth effectively distorts the low-wavelength data (Lees & Wallace, Reference Lees and Wallace2002) producing low-wavelength peaks that appear broader than they actually are (Fig. 7). However, because of the high light penetration of SR, even if absorbing salts and/or buffers are present, it may be possible to measure SRCD spectra for samples which are not possible to examine by CD spectroscopy.

Fig. 6.

SRCD spectrum of a dry horse myoglobin film showing that additional transitions are detectable in the wavelength region between 120 and 170 nm, when measurements are made in the absence of water (Wallace & Janes, unpublished results).

Fig. 7.

Comparisons of the low-wavelength regions of SRCD (blue) and CD (green) spectra of myoglobin. Whilst the CD instrument software permitted collection of data down to around ~170 nm, it is distorted in the low-wavelength region below 190 nm, because in order to have sufficient light flux on the sample, the slits are wide open. The resulting spectrum appears to have a broader peak but this is an artefact of the bandwidth which results in a range of wavelengths input onto the sample (Miles & Wallace, unpublished results).

A second good practice issue is the ability to obtain accurate determinations of the spectral magnitude, which requires knowledge of the correct protein concentration and the sample cell pathlength. Both the reviews by Kelly et al. (Reference Kelly, Jess and Price2005) and by Miles & Wallace (Reference Miles and Wallace2006), discuss good and bad methods for precise protein concentration determination (for instance, quantitative amino acids and denatured extinction coefficients can be sufficiently accurate, but colorimetric assays or gravimetric determinations generally are not). The short pathlength cells used in SRCD are particularly susceptible to erroneous pathlength values (varying by as much as 50% from the values reported by manufacturers) and for SRCD measurements especially, users need to independently determine their cell pathlengths. Methods for determining pathlengths using interferometry or dilution methods are discussed by Miles et al. (Reference Miles, Wien, Lees and Wallace2005a). The consequence of using inaccurate spectral magnitude is that the resulting protein secondary structure determinations can be very wrong (Miles et al. Reference Miles, Whitmore and Wallace2005b). Although correct spectral magnitudes are important in all cases, a significant advantage of SRCD is that if the very-low-wavelength data down to 170 nm is included in the analyses, knowledge of the correct magnitude is not nearly as critical for obtaining a correct analysis (Lees et al. Reference Lees, Miles, Janes and Wallace2006b).

2.4 Thermal radiation effects

Because of the high light flux on the samples in an SRCD cell, one early concern was whether this would cause any damage to the protein during the time of data collection (Orry et al. Reference Orry, Janes, Sarra, Hanlon and Wallace2001). Although the energy of the VUV light is much less than that of X-rays and so was not expected to break covalent bonds, one suggestion (Clarke & Jones, Reference Clarke and Jones2004) had been that it might produce free radicals that would damage the chemical integrity of protein samples. Thus, to test this was an essential proof-of-principle study needed before the SRCD method could find wide acceptance and use. The first such study was conducted on two of the low-flux beamlines (SRS 3.1 and ISA UV1). Using mass spectrometry and gel electrophoresis it was shown that neither chemical fragmentation nor modification, nor any kind of cross-linking occurred after many hours of irradiation in the beam (Orry et al. Reference Orry, Janes, Sarra, Hanlon and Wallace2001). Furthermore, no change in the protein conformation could be detected after this treatment, as evidenced from the SRCD spectra which were not altered after 20 or more repeat scans (Fig. 8). However, the situation changed when high-flux beamlines such as SRS CD12 were developed. In this case, the flux on the sample was several orders of magnitude higher (Clarke & Jones, Reference Clarke and Jones2004) than on the two beamlines originally tested, and the initial observations were that, for most proteins, the SRCD signal changed dramatically in the course of collecting even one or two repeated scans (Fig. 8). Using the highly-sensitive protein human serum albumin for testing, it became clear that the protein was not chemically modified, but did undergo a substantial conformational change (as reflected in sequential SRCD spectra) as it was irradiated in the beam (Wien et al. Reference Wien, Miles, Lees, Hoffmann and Wallace2005). For some of the proteins tested, after the sample was removed from the beam and allowed to re-equilibrate and then re-examined, the protein was able to refold to its native conformation. Hence, the suggestion was made that the irradiation was heating the sample, causing the protein to unfold or denature. Although there was no detectable change in the macroscopic temperature of the protein solution, it was postulated that the high flux of the beam, especially at low wavelengths near the water absorption peak, was causing the heating of internal water molecules in the proteins, waters that were integral to the proteins maintaining their native structures (Wien et al. Reference Wien, Miles, Lees, Hoffmann and Wallace2005). Comparison of the behaviour of a number of other proteins indicated that the denaturation was not limited to a particular type of protein (i.e. with/without disulphides, with high or low aromatic amino acid contents, specific types of secondary structures or folds, single or multiple domain or single or multiple subunit proteins). The only class of protein that appeared to be less sensitive was that of membrane proteins. This, too, would be consistent with the idea of heating bound waters as membrane protein structures tend to have fewer internal water molecules. One study (Janes & Cuff, Reference Janes and Cuff2005) showed that if the total flux onto the sample was physically decreased by inserting baffles upstream in the beam, the denaturation could be diminished or even eliminated.

Fig. 8.

Comparisons of the radiation induced thermal denaturation of a sensitive protein (human serum albumin) in lower flux density beamlines at (a) ISA (CD1), (b) BSRF and (c) NSRRC, versus (d) the CD12 high-flux-density beamline at the SRS. Each plot represents 20 raw consecutive scans on the protein, collected at a set temperature of 4°C. The solid black lines indicate the first scan and the dashed black lines the last scan. The intermediate scans are plotted in grey. (Adapted and augmented from Miles et al. Reference Miles, Hoffmann, Tao, Janes and Wallace2007.)

Subsequently, tests on a number of other beamlines (CD1, 04B1, U11 and 4B8 – see Table 1) were undertaken (Miles et al. Reference Miles, Hoffmann, Tao, Janes and Wallace2007, Reference Miles, Janes, Brown, Clarke, Sutherland, Tao, Wallace and Hoffmann2008a; Wallace & Janes, unpublished results) (Fig. 8). They showed that the critical parameter was flux density and that at beamlines where the total flux was low or spread across a large spot size (usually by moving the sample relative to the focal point of the beam), no such degradation occurred. At very-high-flux beamlines, especially on third-generation sources, such as B23 at Diamond, initially measurements showed degradation was very rapid and significant and design modifications were needed in order to allow multiple scans on the same protein sample. Using comparisons from several beamlines with several different flux levels and spot sizes, the approximate flux density threshhold for denaturation was determined to be ~0·4×10¹¹ photons/sec-mm², which now is a guideline for future beamline designs (Miles et al. Reference Miles, Janes, Brown, Clarke, Sutherland, Tao, Wallace and Hoffmann2008a). Hence, unlike many other SR beamlines, such as those for macromolecular crystallography, where higher flux is usually desirable, for SRCD beamlines high flux is not necessarily optimal, and new beamlines are now being designed in ways to attenuate the flux and thereby minimise the denaturation effects.

These thermal radiation effects have important implications for different types of beamline usages: For standard spectral measurements, obtaining more than one repetitive scan is important for good practice, in order to be able to determine the reproducibility (error bar) levels; this is essential for studies comparing samples that, for example, have been modified or for which ligands have been bound. So these need to be done under conditions where the flux density does not produce differences between repeated scans. Thermal melt experiments (see Section 4.2.3) cannot be done in existing high-flux density beamlines because there will be competing/augmenting denaturation effects from the beam-induced local heating and the external macroscopic heating of the whole sample, although future development plans for many beamlines include providing means for attenuating the flux to enable such experiments. On the other hand, high-flux density will be important for stopped flow experiments (see Section 6.1.3); in those cases the dwell time within the beam will be so short that irradiation effects should be negligible during the course of the measurements.

3.1 Algorithms

Empirical analyses of protein secondary structures from CD and SRCD data are based on the premise that the n→π* and π→π* transitions arising from different types of secondary structures (i.e. different peptide backbone φ, φ angles) produce spectral peaks of different magnitudes and centred at different wavelengths. These techniques use CD or SRCD spectra derived from proteins with known (crystal) structures in reference data sets. Most methods are based on using proportional combinations of the component spectra in the reference data sets to reconstruct the spectrum of an unknown (query) protein; the calculated secondary structure is then derived from the secondary structure of the components used to reconstruct the unknown spectrum in the relative proportions that they contributed to the reconstructed spectra. All empirical analyses assume that the reference data sets include representative structures of the type found in the query (unknown) protein (Janes, Reference Janes2005). If that is not the case, then the analyses will fail. As an example, if there is no polyproline II (PPII)-type structure in the reference data set, and an attempt is made to analyse a collagen-type protein, then the results will be meaningless. Hence, the breadth of coverage of the reference data sets will often be more critical than the choice of algorithm in the ability to produce a correct analysis.

Over the years, a number of algorithms have been developed to undertake secondary structure calculations based on CD data. These include SELCON3 (Sreerama & Woody, Reference Sreerama and Woody2000), a self-consistent method combined with a singular value deconvolution procedure, CONTINLL, a modified version of the original ridge regression algorithm of Provencher & Glockner (1982) that incorporates the locally linearised model of Van Stokkum et al. (Reference Van Stokkum, Spoelder, Bloemendal, Van Grondelle and Groen1990), and CDSSTR, which uses a variable selection procedure and is a modified version of the original VARSLC method (Compton & Johnson, Reference Compton and Johnson1986). In addition, a number of types of neural network procedures have been developed (Andrade et al. Reference Andrade, Chacón, Merelo and Morán1993; Lees et al. Reference Lees, Miles, Wien and Wallace2006a), which are based on training derived from protein reference data sets. All of the empirical methods have been shown to produce reasonably good estimates of secondary structure composition for proteins containing standard types of secondary structures. In general, when using CD data they produce much more accurate secondary structure calculations for proteins with high helical contents. When SRCD data are included (Lees et al. Reference Lees, Miles, Wien and Wallace2006a), there is a significant improvement in the analyses of beta sheet and other types of secondary structure. A number of new algorithms (Lees et al. Reference Lees, Miles, Janes and Wallace2006b) have been developed to optimally use or be trained on SRCD data. These have been shown to produce more accurate analyses for all types of secondary structures. Besides the improved accuracy in the analyses, an additional advantage of using SRCD data in the analyses is that although knowledge of the correct spectral magnitudes has been shown to be very important for obtaining correct secondary structure analyses using any of the algorithms (Miles et al. 2005) if the very-low-wavelength data down to 170 nm is included in the analyses, knowledge of the correct magnitude is not nearly as critical for obtaining a correct result (Lees et al. Reference Lees, Miles, Janes and Wallace2006b).

To assess how successful an analysis has been, several goodness-of-fit parameters have been introduced, the most common of these being the normalised root mean square deviation (NRMSD), which is the spectroscopic equivalent of a crystallographic R-factor (Mao et al. Reference Mao, Wachter and Wallace1982). It is defined as the square root of (the squared sum of the differences between the experimental values and the back-calculated values for the best solution at all wavelengths divided by the squared sum of the experimental values). The smaller the NRMSD value, the closer the calculated spectrum is to the experimental data; generally values less that 0·100 are considered good fits (Fig. 9). However, whilst a small NRMSD is a necessary condition for a correct solution, a low value does not necessarily mean the analysis is accurate, because multiple types of structures could produce similar solutions. Large values of NRMSD suggest that the reference data set used in the analysis may not contain the same sort of structures present in the protein under study.

Fig. 9.

Comparisons of the SRCD spectrum and spectral fits of lentil lectin analysed using the CONTINLL (Van Stokkum et al. Reference Van Stokkum, Spoelder, Bloemendal, Van Grondelle and Groen1990) algorithm at the DichroWeb server (Whitmore & Wallace, Reference Whitmore and Wallace2008) analysed with (a) the new broad-based SP175 reference data set (Lees et al. Reference Lees, Miles, Wien and Wallace2006a) designed for SRCD data and (b) reference set 1 from Sreerama & Woody (Reference Sreerama and Woody2000) originally developed for CD studies. In each case, the plot compares the experimental spectrum (solid line with crosses) with the back-calculated spectrum from the best fit analysis (solid line with squares) and includes the difference spectrum (vertical lines) to emphasize where the experimental and calculated spectra do not correspond well. The calculated values for the NRMSD parameters for (a) and (b) were 0·024 and 0·163, respectively.

3.2 Reference data sets

Until recently, the only reference data sets for empirical analyses of secondary structure were compiled (Sreerama & Woody, Reference Sreerama and Woody2000) from data collected many years ago on cCD instruments in a number of different laboratories. In general, they do not include data that extend to low wavelengths. With the advent of SRCD, it was clear that the increased information in low-wavelength data could enhance the accuracy of secondary structure determinations (Wallace & Janes, Reference Wallace and Janes2001), so new reference data sets were created independently at several beamlines to take advantage of the newly available data. At the SRS, a reference data set containing an unspecified number of proteins (their identities were also not stated) was created and used locally with the programme SELCON (Sreerama & Woody, Reference Sreerama and Woody2000) to analyse clathrin and its component proteins (Clarke & Jones, Reference Clarke and Jones1999). The improvement in the analyses reported was dramatic, but the reference data set was not made publicly available. At HiSOR, a new reference data set with first 15 proteins (Matsuo et al. Reference Matsuo, Yonehara and Gekko2004), and later with 31 proteins (Matsuo et al. Reference Matsuo, Yonehara and Gekko2005), was produced for use with SRCD measurements at that beamline. For the most part, the HiSOR data set included spectra of proteins that had also been part of the existing cCD data sets. Thus, it significantly improved their wavelength range and extended their information content, but did not greatly expand their breath of coverage of protein components. This data set was used in early ab initio calculations (Oakley & Hirst, Reference Oakley and Hirst2006). It was clear that a publicly-accessible reference data set that was applicable for data produced on all beamlines was needed to generally benefit this emerging technique. Hence, a new reference data set, which included spectral data on 71 proteins with high quality protein crystal structures and which was cross-calibrated at several SRCD beamlines (SRS, ISA, NSLS and BESSYII) was produced (Lees et al. Reference Lees, Miles, Wien and Wallace2006a). It had the specific aim of not only including high-quality low-wavelength data, but was also based on bioinformatics analyses of protein structures (Janes, Reference Janes2005) to broadly cover secondary structure and fold space. It included representatives of the types of structures seen in most proteins and included at least one member of each of the CATH-defined (Orengo et al. Reference Orengo, Michie, Jones, Jones, Swindells and Thornton1997) protein superfamilies. That data set is now available publicly on the DichroWeb calculation website (Whitmore & Wallace, Reference Whitmore and Wallace2008) (see Appendix 8.2) for use with five different analysis algorithms and has been the basis of new ab initio calculations (Bulheller et al. Reference Bulheller, Miles, Wallace and Hirst2008). Other specialist reference data sets, for denatured proteins (Sreerama et al. Reference Sreerama, Venyaminov and Woody2000), membrane proteins (Wallace et al. Reference Wallace, Lees, Orry, Lobley and Janes2003, in preparation) and proteins with crystallin-type folds (Evans et al. Reference Evans, Bateman, Slingsby and Wallace2007) have been created in recent years; the crystallins reference data set is also available on the DichroWeb site, and the membrane protein data set will be made publicly available there soon. As these SRCD-defined reference data sets are augmented in the future, they should enable increasingly more accurate secondary structure analyses and include an even wider range of protein architectures and folds.

3.3 Cluster analyses

Other important developments in bioinformatics are driving new and extended analyses of SRCD data. The additional data available in the low VUV wavelength region have been shown to contain additional information, based on principal component analyses (Lees et al. Reference Lees, Miles, Janes and Wallace2006b; Toumadje et al. Reference Toumadje, Alcorn and Johnson1992). This additional information has been proposed to include information regarding the intramolecular interactions within a protein arising specifically from aspects of their supersecondary structures and their different types of folds due to the summed charge transfer transitions arising between different adjacent peptide bonds (Wallace & Janes, Reference Wallace and Janes2001). Preliminary studies using cluster analyses have identified spectral neighbours and compared them with proteins with related folds; these studies have shown that if the low-wavelength SRCD spectra are used (data down to 170 nm has been shown to be required), they cluster proteins effectively not only into groupings similar to the architecture level of CATH fold classifications (Lees et al. Reference Lees, Miles, Wien and Wallace2006a; Miles & Wallace, in preparation) (Fig. 10) but also can further discriminate between topological classes in some cases. This classification is a type of fold recognition based on spectral properties alone (no information input about sequence or crystal structure). Information on spectral nearest neighbours could have applications in molecular modelling and even might be able to help identify related structures that could be used for molecular replacement in protein crystallographic studies.

Fig. 10.

Cluster analysis showing proteins can be grouped into similar CATH classes/architectures based on their SRCD spectra alone (without inclusion of any sequence or crystal structure data). (After Lees et al. Reference Lees, Miles, Wien and Wallace2006a.)

3.4 Ab initio calculations

To date, most calculations of secondary structure based on CD spectra have used empirical methods based on reference data sets derived from spectra of proteins of known structures. However, a number of studies have undertaken theoretical calculations based on first principles in attempts to produce CD, and now SRCD, spectra for proteins of known structure (see a review by Woody, Reference Woody and Fasman1996). Over the years, a number of ab initio methods have been developed to compute the far-UV CD spectra of proteins for the peptide n→π* and π→π* transitions. Experimental helical spectra are particularly well reproduced based on such methods. Other ab initio calculations aimed to match only the magnitude at 220 nm (Besley & Hirst, Reference Besley and Hirst1999). However, such single point fits can be misleading given that the magnitude at a single point is one of the least well determined aspects of a spectrum (because of errors in protein concentrations and pathlengths). More recently, the contribution of charge-transfer transitions to the calculated VUV spectra have been explored (Bulheller et al. Reference Bulheller, Miles, Wallace and Hirst2008; Bulheller & Hirst, Reference Bulheller, Hirst, Wallace and Janes2009; Oakley & Hirst, Reference Oakley and Hirst2006). The full SRCD spectra of first 31 (Oakley & Hirst, Reference Oakley and Hirst2006) and then 71 proteins (Bulheller et al. Reference Bulheller, Miles, Wallace and Hirst2008) were calculated based on their atomic coordinates and compared with the corresponding experimental spectra (Lees et al. Reference Lees, Miles, Wien and Wallace2006a; Matsuo et al. Reference Matsuo, Yonehara and Gekko2005). The inclusion of inter-peptide charge transfer information appears to improve the correlation with experimental data at 175 nm, although there are still significant differences both at high and low (~160 nm) wavelengths for many of the proteins (Fig. 11). A webserver, DichroCalc, based on these methods is available at http://comp.chem.nottingham.ac.uk/dichrocalc/ for the back-calculation of CD spectra derived from ab initio calculations.

Fig. 11.

Comparisons of experimental spectra (solid) and ab initio-calculated SRCD spectra for (a) haemoglobin and (b) lentil lectin, showing the effects at low wavelengths when charge transfer contributions are included (dotted) and not included (dashed). (Reprinted from Bulheller et al. Reference Bulheller, Miles, Wallace and Hirst2008.)

4.1 Sample types

4.1.1 Soluble proteins

CD spectra of soluble proteins exhibit well-known characteristics associated with the principal types of secondary structure present (Fig. 2). In the far-UV wavelength range, mostly helical proteins produce spectra with two negative peaks around 208 and 222 nm and a large positive peak around 190 nm. The spectra for mostly beta sheet proteins are much more diverse (Sreerama & Woody, Reference Sreerama and Woody2003; Wallace et al. Reference Wallace, Wien, Miles, Lees, Hoffmann, Evans, Wistow and Slingsby2004) (Fig. 12), but generally exhibit a single negative peak around 215 nm and a positive peak around 195 nm that is roughly of the same magnitude as the negative peak. The peaks arising from beta sheet structures are much less intense than those from helical structures, and this is one of the reasons that analyses of protein sheet content is generally less accurate than analyses of their helical contents (especially if there is a small sheet content in the presence of helix, which results in the sheet signals being ‘swamped’). The spectra of PPII-rich proteins are generally relatively featureless in the far-UV region above 210 nm (although they often have a small positive peak around 220 nm); they do, however, have a large negative peak around 200 nm (Miles & Wallace, Reference Miles and Wallace2006; Sreerama & Woody, Reference Sreerama and Woody1994). Spectra of proteins without large amounts of ordered structure (often called natively disordered or unfolded proteins or previously referred to as being ‘random coils’) tend to have no positive peaks and a single broad negative peak at around 195–198 nm and often with a small positive peak (or negative minimum) at around 180–185 nm. They can have a somewhat similar appearance to a PPII spectrum except the negative peak may be slightly blue-shifted (Fig. 13). Because the peaks due to PPII and disordered or ‘other’ structures are similar and have minima at wavelengths near the low-wavelength data collection limit for CD spectra, they too are often more poorly defined by secondary structure calculations than are helical structures.

Fig. 12.

SRCD spectra of beta sheet-rich proteins showing the diversity of spectral characteristics associated with different folds. The proteins shown are: concanavalin A (open squares), avidin (open triangles), elastase (filled circles), beta-galactosidase (filled diamonds), soybean trypsin inhibitor (open circles) and βB2-crystallin (filled squares). (Adapted from Wallace et al. Reference Wallace, Wien, Miles, Lees, Hoffmann, Evans, Wistow and Slingsby2004.)

Fig. 13.

Comparison of the SRCD spectra of a disordered protein (red) with that of a PPII-rich protein (blue). (Miles, Richards & Wallace, unpublished data). It should be noted that the spectra of unfolded disordered proteins vary significantly depending on the conditions used for unfolding, and the reader is refereed to Matsuo et al. (Reference Matsuo, Sakurada, Yonehara, Kataoka and Gekko2007) for examples of different unfolded proteins.

Because there is no such single entity as a ‘random coil’ type of secondary structure, it is unlikely that there will be a single exemplar of a disordered spectrum that can be used in reference data sets employed in empirical analyses. Using SRCD, Matsuo et al. (Reference Matsuo, Sakurada, Yonehara, Kataoka and Gekko2007) examined the different types of structures and spectra produced under different types of denaturation conditions (acid, cold, chaotropic agents and heat). The spectra produced under these conditions were distinctly different, confirming that there are different end points in the processes of unfolding resulting in different ‘denatured’ structures, depending on the conditions used for denaturation. Indeed, when the SELCON3 analysis algorithm was applied to the different spectra, it identified that in a number of cases there was still a significant amount of helix or sheet present. For this study, SRCD was particularly valuable for two reasons: (1) the low-wavelength peaks around 190–200 nm were easily detectable and distinguishable and (2) even in the highly absorbing chaotropic agent guanidine hydrochloride, measurements were possible to wavelengths below 200 nm. At present several reference data sets (Sreerama et al. Reference Sreerama, Venyaminov and Woody2000) include ‘denatured’ protein spectra, although inclusion of proteins treated in only one of the above manners to represent ‘unordered’ or ‘other’ types of secondary structure may not be sufficiently complete for reference data sets used in secondary structure analyses.

SRCD extends the spectral wavelength range to VUV wavelengths where not only are the existing transitions more complete, but also additional transitions can be detected (Fig. 2). This enables an improved distinction of beta sheet contributions from helical ones, since at low wavelengths the sheet and helix signals have opposite signs (Miles & Wallace, Reference Miles and Wallace2006). The low-wavelength peaks also show a rich diversity between proteins with different types of beta sheet structures, including parallel and antiparallel sheets, sheets with different twists, strands with different sheer, and barrels with different staggers (Wallace et al. Reference Wallace, Wien, Miles, Lees, Hoffmann, Evans, Wistow and Slingsby2004) (Fig. 12). In addition, SRCD spectra are also particularly good for discriminating between examining natively disordered proteins (Guerra-Giraldez et al. Reference Guerra-Giraldez, Moore, Neves, Wallace, Svergun, Brown and Smith2005) and ones with significant amounts of PPII, such as collagen (Miles & Wallace, Reference Miles and Wallace2006). Not only can this be useful for qualitative visual assessments of the spectra for secondary structure, the additional data result in a higher information content (Lees et al. Reference Lees, Miles, Wien and Wallace2006a; Toumadje et al. Reference Toumadje, Alcorn and Johnson1992; Wallace & Janes, Reference Wallace and Janes2001) than present in conventional CD spectra, which in turn means that additional types of protein secondary structural components (such as 3₁₀ helices and turns) can be computationally-derived from the data (Lees et al. 2006b). Hence, SRCD can enhance the utility and accuracy of secondary structure determinations of soluble proteins, especially those with beta sheet and PPII compositions.

4.1.2 Membrane proteins

Membrane proteins are notoriously difficult to examine by CD spectroscopy due to artefacts arising from absorption flattening (Wallace & Teeters, Reference Wallace and Teeters1987), differential scattering (Wallace & Mao, Reference Wallace and Mao1984) and wavelength shifts (Cascio & Wallace, Reference Cascio and Wallace1995; Wallace et al. Reference Wallace, Lees, Orry, Lobley and Janes2003). These effects result because membrane proteins are not surrounded by a uniform and isotropic aqueous solution, but are rather embedded in hydrophobic and anisotropic phospholipid bilayers or detergent micelles. SRCD has proven to be particularly useful in obviating each of these potential issues for several reasons.

4.1.2.1 Absorption flattening

Absorption flattening arises from the non-uniform distribution of chromophores in a membrane sample. This distribution is because although the proteins may (or may not) be randomly distributed within the membrane particles (vesicles, liposomes or bicelles) and the membrane particles may be randomly distributed throughout the sample volume, the proteins are not uniformly distributed throughout the sample, being sequestered into areas occupied by the particles. This means that the major tenet of Beer's law which requires the random distribution of absorbers throughout the sample is not met. The consequence is that each of the absorbance peaks will be depressed with respect to what it would have been for the equivalent protein in an isotropic solution. Furthermore, the more intense the absorbance, the more depressed in magnitude is the peak (Teeters et al. Reference Teeters, Eccles and Wallace1987).

This effect can be eliminated in principle if the lipid-to-protein molar ratio is sufficiently high such that there is only a single protein present in each particle. To achieve this, lipid-to-protein molar ratios of ~2000 or more may be needed, depending on the size of the particle. However, such samples are normally difficult to measure by cCD because whilst lipids generally do not produce significant CD signals, they do absorb to a major extent at wavelengths in the far-UV due to the electronic transitions of the carbonyl groups linking their fatty acid chains and head groups. As a result, the total amount of light that penetrates a liposome sample containing high concentrations of lipids will be very small, so the signal-to-noise levels in both the left- and right-handed circularly polarised signals will be very low, producing a very noisy net CD spectrum. When coupled with a need to keep the protein concentration low (again to maintain the high lipid-to-protein ratio), the polypeptide CD signal tends to be unmeasureable. However, because the absolute amount of penetration of light is higher when using a SRCD beamline, membrane protein samples can often be measured with good fidelity and signal-to-noise levels. A demonstration of this was the study of opsin in bicelles (McKibbin et al. Reference Mckibbin, Farmer, Jeans, Reeves, Khorana, Wallace, Edwards, Villa and Booth2007) where SRCD permitted such measurements and samples with lipid-to-protein molar ratios of ⩾1000. Indeed such studies enabled detection of subtle differences in protein structure associated with the reconsititution of the protein in different lipids. Such a study would not have been viable using cCD.

4.1.2.2 Differential scattering

Differential scattering is an artefact that arises when the size of the object under examination is large (>1/20) relative to the wavelength of light being used for the observation (Gitter-Amir et al. Reference Gitter-Amir, Rosenheck and Schneider1976; Wallace & Mao, Reference Wallace and Mao1984). As a result, the particles (in the case of membrane proteins these are usually vesicles or liposomes) tend to scatter light other than in the forward direction, which then falls outside the acceptance angle of the detector. Because the standard geometry of most cCD instruments places the sample relatively far from the detector and the detector face is relatively small, this can result in an acceptance angle as small as 6°, which, even for small lipid vesicles means that much of the light never reaches the detector. If the scattering is from a chiral object (i.e. the vesicle includes a protein), then the scattering will be different for the left- and right-handed circularly polarised light because the refractive indices for the two light hands are different, giving rise to a distorted CD spectrum. This can have severe consequences for the spectral shape and the subsequent analyses (Wallace & Mao, Reference Wallace and Mao1984). Non-chiral objects in the solution (i.e. vesicles not containing protein) will also scatter light, but in this case equally for both handed types. Although this does not change the spectral shape it will also produce a lower signal-to-noise level. To overcome the scattering effects, either samples with smaller particle sizes (i.e. in detergent micelles instead of lipid bilayers) or changes to the instrument geometry are required. The former solution can be problematic if the protein does not have the same conformation in lipids as it does in detergents (quite often the case). However, the latter solution to the problem can be attained by changing the instrument geometry by modification of the sample chamber and detector; this is possible in some commercial instruments (i.e. those made by Aviv Biomedical). However, as most SRCDs already have a large detection angle automatically incorporated into their designs and tend to have larger detector faces, SRCD beamlines greatly reduce the differential scattering problem for membrane protein samples.

4.1.2.3 Wavelength shifts

Because the dielectric constants of the hydrophobic fatty acid chains of lipids or the hydrophobic tails of amphipathic detergent molecules in which membrane protein transmembrane segments are embedded, are very different than the dielectric constant of the aqueous milleux of soluble proteins, the energy difference between the ground and excited states of the transitions, and therefore the peak positions in the spectra, of membrane proteins and soluble proteins can differ considerably (Cascio & Wallace, Reference Cascio and Wallace1995; Chen & Wallace, Reference Chen and Wallace1997; Park et al. Reference Park, Perczel and Fasman1992). This can result in significant differences in the spectra of membrane proteins and soluble proteins, even when they have arisen from proteins with the same secondary structure content (Fig. 14). Furthermore, when membrane protein spectra are analysed using reference data sets derived from soluble proteins (Wallace et al. Reference Wallace, Lees, Orry, Lobley and Janes2003), the results are often incorrect. Because the peak shifts for the different transitions are of different magnitudes and signs, the problem cannot simply be alleviated by shifting the whole spectrum by the same amount in nm. Neither can it be reduced by converting the spectra to energy units and shifting all transitions by the same amount in energy terms (Cascio & Wallace, Reference Cascio and Wallace1995).

Fig. 14.

Comparisons of the calculated fits (blue) of (a) a soluble protein (concanavalin A) and (b) a membrane protein (FepA), with their experimental SRCD spectra (red), both analysed using a reference data set derived from soluble proteins. These two proteins were chosen for comparison because they had very similar net secondary structures. (After Wallace et al. Reference Wallace, Lees, Orry, Lobley and Janes2003.)

SRCD has been advantageous with respect to studies assessing the extent of the wavelength shift for several reasons: (1) The SRCD spectra can be measured in the detergents and lipids that membrane proteins require for solubility, (2) the spectra have higher signal-to-noise levels so subtle shifts can be more easily detected, and (3) the lower wavelength data in the VUV region enables the observation of the complete peaks for all three n→π* and π→π* peptide transitions.

This has led not only to the creation of a new membrane protein reference data set using SRCD data, with significant improvements for the analysis of membrane protein spectra, but also to the development of methods for the separate identification of transmembrane and extramembrane secondary structural elements (Abdul-Gader, Miles & Wallace, unpublished results). The latter significantly increases the information content derivable from a spectrum as it, for instance, could identify that of the 50% helix content of a protein, the transmembrane region may be nearly 100% helical whilst the extramembranous region may be only 30% helical. This can also enable identification of whether conformational changes occur in the transmembrane or extramembrane domains, again an increase in information with respect to that obtainable by CD spectroscopy.

4.1.2.4 Lipid/protein ratios

In order to detect the signals due to the polypeptide, in cCD, the lipid-to-protein molar ratios are normally maintained at reasonably low levels (i.e. 50:1 – or in some cases even less). This can however, be problematic (in addition to the absorption flattening effect described above), since proteins and peptides at such high levels can either tend to lyse the membranes or form aggregates within the bilayers. SRCD can mitigate against these problems by enabling the use of much higher (up to 10² times as high) lipid ratios. This is because of the high flux, which even when the transmission is attenuated by the absorbance of the lipid component, enables sufficient light to reach the detector to result in a reasonable signal-to-noise ratio. Hence, SRCD spectra can readily be measured down to wavelengths of 180 nm in small unilamellar lipid vesicles with lipid-to-protein mole ratios even as high as 2000:1 (Miles & Wallace, unpublished).

As an example of how this has been useful, it was possible to examine the eukaryotic cytolysin equinatoxin II in liposomes of various compositions, enabling the identification of the lipid features essential for binding interactions (Miles et al. Reference Cowieson, Miles, Robin, Forwood, Kobe, Martin and Wallace2008). This protein lyses membranes when present at high concentrations. However, the high lipid-to-polypeptide ratios achievable for SRCD measurements (250:1) enabled the examination the protein in intact (not lysed) liposomes. These studies were all only possible using SRCD instrumentation (indeed the study had been attempted unsuccessfully in advance using conventional CD, in case it would not have been necessary to use valuable SRCD beamtime for the experiment, but the signal was too small and noisy to permit any conclusions). However, the SRCD study produced high quality data that enabled the comparison of membrane-bound and soluble forms of the polypeptide and the identification of the importance of sphingomyelin in the lipid mixture to enable the toxin to adopt a folded, active conformation.

4.1.6 Non-aqueous solutions

A major limitation on the ability to obtain data in the very low VUV wavelength range when using SRCD spectroscopy is due to the presence of water, which has a strong absorption peak near 168 nm. Even though the water absorption is not chiral, it does severely limit both the left- and right-circularly polarised light being transmitted because of the relatively large molar excesses of water relative to protein in any aqueous solution. To circumvent this, several methods have been employed to prepare samples for SRCD studies with reduced or no water content. They include decreasing the pathlength of the sample cell whilst increasing the protein concentration (see Section 2.1.3) or the use of dehydrated or partially hydrated films (described in the next section). An alternative has been to use organic solvents that do not absorb strongly in this wavelength range instead of water. Solvents such as acetonitrile and 2,2,2-trifluoroethanol (TFE) have been shown to have improved absorbance profiles in the VUV range relative to water (Wien & Wallace, Reference Wien and Wallace2005) (Fig. 4b) and permit data collection to lower wavelengths. However, the principal issue with the use of non-aqueous solvents to examine proteins and peptides is whether or not they cause the protein to adopt a non-native conformation because of their different solvation properties and dielectric constants. SRCD spectra of myoglobin in both TFE (Thulstrup et al. Reference Thulstrup, Brask, Jensen and Larsen2005) and hexafluoroisopropanol (Clarke & Jones, Reference Clarke and Jones2004) have been collected to wavelengths below 169 nm. However, whilst the protein retains a significant alpha-helical content in these media, the spectra clearly differ significantly from that of native myoglobin in water (Wallace & Janes, unpublished results) (Fig. 15), with changes to both peak positions and peak magnitudes suggesting that the protein is at least partially unfolded in these solvents. This suggests an important caution: it is tempting to look at the spectrum of a protein or peptide in a non-aqueous helix-inducing solvent such as TFE and see that the polypeptide is helical and therefore assume it is folded. However, as the authors in the two studies cited above did, before deriving any conclusions from SRCD spectra in non-aqueous media it is important to apply the critical test of comparison of the protein in water and the other solvent in the far-UV (or other spectral) regions where measurements in both solvent can be undertaken, to ascertain if there are any significant differences.

Fig. 15.

Comparison of SRCD spectra of myoglobin in aqueous solution (blue) and in 2,2,2,-trifluorethanol (red).

Hence, the use of organic solvents enables VUV data collection, but often at the price of altering the protein conformation. So unless there is a specific question about the conformation of a protein in a particular solvent, even though enabling for data collection, the biological relevance of such SRCD studies need to be interpreted with caution.

4.2 Examples of SRCD-enabled applications

4.2.1 Detecting macromolecular complex formation

One contribution SRCD can make in Systems Biology is the investigation of interactions that occur between macromolecules. How proteins interact and form complexes with other proteins, nucleic acids and carbohydrates provide important clues as to their roles in the cascade of functional and developmental events in the cell. Several studies have indicated new ways in which SRCD can add to the cadre of methods which detect complex formation and the definition of the structural consequences that occur upon binding their macromolecular partners.

4.2.1.1 Complex formation involving natively disordered structures

Bioinformatics analyses on the human genome (Edwards et al. Reference Edwards, Lobley, Pentony and Jones2009) suggest there is a significant population of proteins that contain disordered regions. These elements may be critical to some types of complex formation, and that when they interact with binding partners (either macromolecules or smaller biological partner molecules such as lipids or carbohydrates), they will partially or completely refold to form a stable complex. Sometimes the changes involve relatively small regions of their structures. Often, the changes are subtle, and because they involve comparisons of differences of differences [i.e. the difference between the spectrum of protein A plus the spectrum of protein B (scaled by the appropriate proportions due to their relative molecular weights) and the spectrum of the protein A plus protein B complex (in stoichometric ratios)], their accurate detection requires a good signal-to-noise ratio, as well as a reasonably high binding constant. Hence, SRCD can be particularly valuable due to its high sensitivity.

One example of such an interaction detected by SRCD is the structural consequences of binding the C-terminal unordered peptide of rhodopsin to its binding partner arrestin (Wallace & Hargraves, unpublished data; Wallace & Janes, Reference Wallace and Janes2010). The difference of the difference in this case shows that upon complex formation, there is a large gain in helical content. Because the arrestin component is already folded before the interaction, the result obtained could be interpreted as corresponding to the equivalent of nearly 100% of the peptide component changing from disordered to helical structure. A complication can sometimes come in interpreting the results if both partners are proteins; then the mutual changes that occur are overlapping, resulting in a net change, so it is often difficult or impossible to ascribe which partner underwent the change. If, however, one partner starts as completely unordered and the other is substantially ordered and the spectrum of the complex is more ordered than the other partner is originally, as in the arrestin/rhodopsin case, assumptions can be made about the changes to the individual partners. In cases where the binding partner is non-peptidyl, the interpretation is much simpler (assuming there is no induced CD in the partner, a relatively rare occurrence) because all changes seen at these wavelengths will be due to the protein component. Another example of detecting such a complex formation by SRCD is the SHERP protein, a cell-surface protein involved in the disease leishmaniasis (Guerra-Giraldez et al. Reference Guerra-Giraldez, Moore, Neves, Wallace, Svergun, Brown and Smith2005). In this case, the protein is completely unfolded, but in the presence certain lipid partners folds to an almost entirely helical structure. Because the lipids and detergents do not produce any chiral absorbance in the UV wavelength range, all the change is attributable to the protein.

These examples suggest that SRCD can be a valuable technique for monitoring disorder-to-ordered transitions, and the enhanced speed of data collection means that reasonably high-throughput screening of potential binding partners may be possible.

4.2.1.2 Complex formation involving tertiary changes

Traditionally, far-UV CD spectroscopy has only been useful for detecting conformational changes involving net changes in secondary structures, as the wavelengths monitored tend to be dominated by peptide backbone transitions. Sometimes tertiary interactions can be detected in the near-UV region when they involve changing the environment of aromatic amino acids, but these measurements are difficult due to the very low extinction coefficients of the aromatic absorbances and the small number of aromatic residues in proteins, which mean that the spectral magnitudes tend to be two or more orders of magnitude lower than the backbone transitions. Measurements can be made in this wavelength region using either cCD or SRCD; however, they require the use of long pathlength cells and large volumes and amounts of protein and they are often difficult to interpret in terms of structural changes. Recently, however, it has been shown that SRCD can be used to detect complex formations involving only tertiary interactions without changes in secondary structures, due to changes detectable in the VUV range (Cowieson et al. Reference Cowieson, Miles, Robin, Forwood, Kobe, Martin and Wallace2008). In this study, the two proteins examined had known crystal structures as separate proteins and as components of a crystal structure of the binary complex. They were chosen because the proteins in isolation and in the complex had the same secondary structures, so complex formation involved rigid-body interactions. It was observed that although there were no differences detectable in the far-UV region of the spectrum of the complex relative to the spectrum derived from summed spectra of the individual proteins, in the VUV region below 190 nm, distinct and significant changes could be measured by SRCD (Fig. 16) but not by cCD.

Fig. 16.

Detection of rigid-body complex formation by SRCD spectroscopy. Experimental spectrum (red) of the complex of latexin and carboxypeptidase compared with the summed spectra (blue) of the individual proteins. (Data from Cowieson et al. Reference Cowieson, Miles, Robin, Forwood, Kobe, Martin and Wallace2008.)

From the above examples it can be seen that SRCD can enhance the ability to detect complex formation not only by detecting smaller conformational changes as a result of its enhanced sensitivity, but also because the additional information at lower wavelengths can be accessed and provide novel information on intermolecular interactions. It thus enables the detection of complex formation involving both induced fit mechanisms (involving secondary structural changes of one or more of the components), plus rigid body complex formation (involving tertiary changes and intermolecular interactions such as aromatic associations and in some cases hydrogen bond formation), thereby extending the types of complex formation that can be monitored by CD spectroscopy.

4.2.3 Monitoring protein unfolding and folding

Understanding the processes of unfolding/refolding of proteins in vitro is important not only for discerning their thermodynamic properties, but also has important applications for the small and large scale production of correctly folded proteins. On a secondary structure level, unfolding generally involves changes from helical or sheet structures to random coil or disordered structures. SRCD can be particularly valuable for examining disorder-to-order transitions because disordered structures have as their largest spectral feature a negative peak at low wavelengths (~190 nm), a region easily detected with SRCD. This peak is often very noisy in cCD due to the components present in the solution necessary for either keeping the protein in an unfolded state or for mimicking physiological conditions; this has meant that, in general, refolding studies that aim to examine the transition between folded and unfolded states (in either direction) at a single wavelength as a function of time often must concentrate only on the high-wavelength peak associated with helix formation. SRCD expands the information obtainable by enabling the monitoring of both the disappearance/appearance of the disordered peak and the appearance of the ordered peaks for helix and sheet. This permits detection of whether or not the folding of different elements occurs in a concerted manner and whether they appear/disappear in parallel or not.

4.2.3.1 Chemical stability

Chemically-induced unfolding studies usually employ chaotropic agents, such as urea or guanidinium HCl, or hydrophobic additives, such as the detergent sodium dodecyl sulphate (SDS), to interrupt hydrogen bonds and hydrophobic interactions in the interior of proteins. The usefulness of SRCD in folding studies was exemplified in a study of the alpha/beta protein shikimate kinase (Cerasoli et al. Reference Cerasoli, Kelly, Coggins, Boam, Clarke and Price2002). This study used a combination of stopped flow studies on a cCD instrument (at the time there were no working SRCD stopped flow setups, see Section 6.1.3) to follow the refolding process, but also used SRCD scans to examine the whole spectrum of the protein in its unfolded state. SRCD proved of benefit because of the high concentrations of urea used to produce the unfolded state, which absorbs strongly in the far-UV region. The refolding was monitored at 225 nm, but as the authors noted, in the future it should be possible to monitor the refolding of the lower wavelength peaks when a stopped flow SRCD becomes available. An advantage in this case could arise because that the protein contains a considerable amount of beta sheet structure, which is not monitored by the 225 peak changes.

With the recent development of a continuous flow facility using microfluidic mixer on an SRCD beamline (see Section 6.1.3), it has been possible to temporally monitor the rapid time course of the refolding (Hoffmann et al. Reference Hoffmann, Kane, Nettels, Hertzog, Baumgärtel, Lengefeld, Reichardt, Horsley, Seckler, Bakajin and Schuler2007) of a small cold shock protein from 4 M guanidinium hydrochloride at 205 nm, where the change in beta sheet content can be monitored. The high flux of the SRCD beam was critical for being able to do these measurements.

4.2.3.2 Thermal stability

The thermal stability of a protein may be monitored by either following the change in magnitude of a single peak or the change in the whole spectrum as a function of temperature. In this way, the melting temperature (T_M) of the unfolding transition can be determined. However, to accurately determine this value it is essential that for each temperature increment, the sample has come to equilibrium (i.e. multiple scans exhibit no systematic change with time). As a result measurements of whole spectra have an advantage of being able to confirm if the sample has equilibrated. However, in cCD, often only high wavelengths can be monitored with accuracy because of the noise levels in measurements at lower wavelengths. The ability to measure the thermal characteristics at many wavelengths using SRCD provides the additional advantage that the different types of secondary structures can be separately monitored during unfolding or refolding to ascertain if the process is concerted and cooperative or sequential (Fig. 17). The 222-nm peak is a good indication of the helix content and the 215- and 208-nm peaks are due to a mixture of the sheet and helix content, whilst the 190-nm peak represents a mixture of helix, sheet and unfolded conformations and the 170-nm peak is a different mixture of helix, sheet and unfolded. By monitoring all of these peaks (Lees et al. Reference Lees, Smith, Wien, Miles and Wallace2004), it is possible to deconvolute the changes due to each type of secondary structure and separately define the differential stabilities of each of the secondary structural components. In addition, folding intermediates can be identified from difference spectra. Thus, SRCD has the potential for defining the full of path of an unfolding process.

Fig. 17.

Plot of a three-dimensional stack of SRCD thermal melt spectra for lysozyme, a protein which contains both helix and sheet elements. This is plotted using CDtool software (Lees et al. Reference Lees, Smith, Wien, Miles and Wallace2004) with the z axis being the temperature and the projection contours on the xz plane highlighting changes at each wavelength with temperature.

As an early example, SRCD was used to follow the unfolding of the protein lactoglobulin. Because the helical content of this protein is very small, rather than monitoring the changes at a single wavelength, Qi et al. (Reference Qi, Holt, McNulty, Clarke, Brownlow and Jones1997) obtained full SRCD spectra (down to 170 nm) at 12 temperatures between 20 and 80°C, and then calculated the secondary structure content present at each temperature and found a trend associated with the thermal denaturation. The use of SRCD was essential because the changes directly observable in the spectra at ~222 nm were effectively undetectable. In addition, secondary structure differences calculated from using the spectra only down to 200 nm did not show differences that were significant; however, when the whole spectrum was analysed using data down to 170 nm, changes in the helical content were apparent. This illustrates the power of including information from the lower wavelengths in such analyses.

More recently (Maytum & Janes, Reference Maytum and Janes2007), SRCD was used to identify a new intermediate in the folding pathway for tropomyosin. Although cCD had been used for many years to study this protein, because spectra were only measured to ~200 nm, the transition had not been previously seen. The new thermal transition, which was detectable by SRCD in the spectra region between 170 and 190 nm, and at a physiologically relevant temperature (30–37°C) was capable of distinguishing between wildtype and mutant protein.

Hence, SRCD provides added value to unfolding/refolding studies by enabling the examination of additional peaks, which in turn provide insight into the types of secondary structures involved in thermal transitions, the presence of intermediates and the cooperative nature of the processes.

5.1 Synchrotron radiation circular dichroism and infrared spectroscopy

The ability to combine two spectroscopic techniques for examining a single protein has a number of advantages. The pairing of SRCD and Fourier transform infrared spectroscopy (FTIR) is a particularly good example of this synergy, as the methods measure different wavelength ranges which provide information on different features of a protein and have different sample requirements. CD detects electronic transitions due to the peptide backbone, whereas FTIR measures vibrational transitions due to stretching and bending of, for instance, NH, CO and CN bonds. In both types of spectra, distinct features arise depending on what type of secondary structure is present, but the two methods have different accuracies in their ability to quantify different types of structures and different potential limitations. For example, scattering can be problematic for membrane proteins in CD measurements, but not so for FTIR (due to the longer wavelengths used); however, aqueous solvent peaks can interfere with the protein peaks in FTIR, making quantitation, especially for helical structures difficult; however water does not produce a CD signal and therefore does not interfere with protein spectra. CD generally requires less material than FTIR, and in this regard, SRCD provides an even greater advantage with respect to the amount of material and time needed for measurements.

Beyond just providing different separate measurements that can be compared, recent studies (Lees & Janes, Reference Lees and Janes2008; Oberg et al. Reference Oberg, Ruysschaert and Goormaghtigh2004) have shown that CD and FTIR data can actually be combined in secondary structure analytical procedures to produce more accurate determinations than either method alone. This combination may be successful in part because CD is particularly good for helical structure determinations whilst FTIR is especially good for defining beta sheet structures. Lees & Janes (Reference Lees and Janes2008) have further shown that if SRCD data rather than conventional CD is used, the improvement is even more notable.

Future developments in beamline designs could exploit this synergy if IR and CD endstations could be juxtaposed on the same beamline to enable measurement on the same sample. Alternatively, simple coordination of beamtime on two different beamlines could enable both measurements during the same synchrotron visit, especially if sample cells are made mutually compatable. The later will be more facile now that a number of synchrotrons are developing ‘life science villages’ where beamlines (and their associated laboratory spaces) used for biological samples are located in close proximity around the synchrotron ring.

5.2 Synchrotron radiation circular dichroism and sequence-based secondary structure predictions

Combinations of SRCD measurements and sequence-based neural network modelling methods have been used to improve the empirical predictions of protein secondary structures (Lees & Janes, Reference Lees and Janes2008; Matsuo et al. Reference Matsuo, Watanabe and Gekko2008). By providing accurate information on the net amounts of each type of secondary structures as calculated from SRCD spectra to augment the training of a standard neural network for the position-specific prediction of the secondary structure type of each amino acid in a protein, Matsuo et al. (Reference Matsuo, Watanabe and Gekko2008) found a significant improvement in the prediction of the individual amino acids' secondary structures, when compared the known crystal structure. A similar approach was taken by Lees & Janes (Reference Lees and Janes2008), except in the latter case experimental information from both SRCD and FTIR measurements were used to enhance the predictions. In both studies, SRCD-derived net secondary structures rather than CD-derived values were used because they produced more accurate experimental values. SRCD had another distinct advantage: measurements could be made when only small amounts of the protein of interest were available. These studies suggest that combining the methods of SRCD and bioinformatics can have useful applications in homology modelling using sequence-based structure predictions to produce improved modelled structures.

Conversely, SRCD experimental data have been used to test the accuracy of modelled structures, either by comparing back-calculated spectra from a theoretical model with the experimentally-measured SRCD spectrum of the protein of interest (Richards et al. Reference Richards, Hanlon, Berrow, Butcher, Dolphin and Wallace2002) or by comparing the SRCD-derived secondary structure determined for a modelled homologue with the secondary structure derived from crystal structures of other members of the same protein family (Miron et al. Reference Miron, Réfregiers, Gilles and Maurizot2005).

5.3 Synchrotron radiation circular dichroism and protein crystallography

SRCD spectroscopy can be a valuable adjunct to protein crystallographic studies, both as an aid to crystallisation conditions and as a means of relating a protein crystal structure to the biological function of the protein.

As a crystallisation aid, CD can be used to monitor (i.e. by thermal melt studies) the effects of various additives, detergents or other conditions on the stability of the protein; because it has been shown using other techniques that optimising protein stability can lead to identifying initial conditions for crystallisation and, later, improved crystals (Warne et al. Reference Warne, Serrano-Vega, Tate and Schertler2009), CD studies of thermal stability could provide useful information for crystallisation. Alternatively, ligand binding studies can monitor the titration of complex formation, thereby assuring a single species is present under the crystallisation conditions (as opposed to having a mixture of proteins with and without ligand bound). Both of these types of studies can be undertaken successfully with conventional CD spectroscopy, but they can be improved with SRCD because of the wider range of conditions possible, including monitoring additives that may absorb in the UV.

Other applications of CD/SRCD are the ability to demonstrate that constructs produced for crystallisation have well-ordered structures and as a guide to dissecting the protein (deletion of termini, loops), removing disordered regions that can prevent crystallisation (Richards et al. Reference Richards, Hanlon, Berrow, Butcher, Dolphin and Wallace2002). SRCD has also been used to identify whether domain constructs designed for crystallisation are properly folded, and if not, to use the SRCD measurements to help identify the location of domain boundaries (Richards et al. Reference Richards, Hanlon, Berrow, Butcher, Dolphin and Wallace2002). SRCD is preferable to CD for these studies because of the requirement for small amounts of protein (it would not be sensible to use a method that could require much larger amounts of protein than needed for the crystallisation setups), because the unordered structures that are being assessed in these studies have their major spectral features at wavelengths below 200 nm and because domain constructs often require relatively high salt concentrations in order to stabilise them in solution (as neighbouring domains that can contribute to solubility may have been removed in the process). Furthermore, CD/SRCD can be used to monitor the refolding into viable preparations of constructs initially expressed as inclusion bodies or in denatured forms. SRCD has the advantage in this case of enabling such measurements in a range of denaturants such as urea, detergents and guanidine hydrochloride.

One of the potentially most useful applications of SRCD with respect to crystallography is to be able to relate crystal structures to structures present under ‘physiological’ conditions. SRCD can measure samples under both types of conditions (which are often inaccessible to CD measurements due to the high concentrations of salts, buffers and precipitants present in crystallisation mixtures and high ionic strength and concentrated solutions representing cellular conditions). In addition, whilst a crystal structure is a ‘static’ representation, SRCD can monitor conformational changes in the protein associated with ligand binding and altering conditions such as pH and temperature, thereby relating its structure to function.

Another example of where SRCD can provide important information that is complementary to protein crystallography is when a crystal structure has been determined for a protein (for example the wildtype), but a mutant that is biologically important cannot be made to form high resolution crystals. Often the mutant can be compared spectroscopically with the wildtype protein to determine subtle conformational differences. This was the strategy adopted for examining a mutant eye lens γD-crystallin protein (P23T) that causes a form of cataract disease but did not produce suitable crystals, where there was a good quality wildtype protein crystal structure (Evans et al. Reference Evans, Wyatt, Wistow, Bateman, Wallace and Slingsby2004). SRCD showed small but significant differences (Fig. 18 a) that could be interpreted structurally based on the native crystal structure and molecular modelling and dynamics studies as resulting in an extension to one beta strand which changed the protein's solubility; this could then make structural sense of the disease phenotype of opaque (insoluble) protein in the eye.

In a similar type of study, SRCD spectra were used to compare wildtype and mutant forms of the enzyme phosphoribosylpyrophosphate synthetase (Liu et al. Reference Liu, Peng, Zhao, Zhang, Tao, Luo, Li, Teng, Li and Wei2009). The mutant version of this enzyme is super active relative to the wild type, leading to increased synthesis of phosphoribosylpyrophosphate, which is associated with the overproduction of uric acid and the disease gout. The crystal structure of the wild type had been determined, but not of the mutant. Small but significant differences in the SRCD spectra of the two enzymes were evident in both the peaks at ~190 and ~220 nm. These led to calculated differences in helix contents of around 1% between native and mutant. The differences could be rationalised based on the crystal structure of the wildtype in terms of potential structural changes around the single amino acid that was changed. If only one amino acid conformation had changed in the protein, this would have produced much less change than the 1% of the total helix content seen. However, examination of the crystal structure suggested that the asparagine to serine substitution in the mutant would alter the local hydrogen bonding pattern in a way that would change the phi/psi angles of a segment of polypeptide backbone, resulting in the change from a disordered helix to an ordered one. This is an excellent example of the synergy possible between protein crystallography and SRCD spectroscopy: when no crystal structure of the mutant was available, the spectra provided data that were interpretable on a molecular level based on the wildtype crystal structure.

It is significant that in both these examples, the differences between the mutant, disease-causing proteins and the wildtype native proteins were visible in the SRCD spectra. However, they were so small (on the order of 1% differences in the magnitudes of specific peaks) that they would have been undetectable using cCD instruments, where a noise level better than 2%–3% can rarely be achieved, even when averaging many repeats and under optimal conditions. Indeed, in the case of the crystallins study, a direct comparison between the wildtype and mutant proteins with cCD (Fig. 18 b) showed that the data quality was insufficient to make any significant comparisons.

SRCD can also have a role in structural genomics programmes which aim to crystallise examples of all possible protein fold types. Fold recognition studies based on SRCD spectra, using the enhanced information at low wavelengths, may be useful in identifying if a potential target protein may represent a new fold. This and other applications of SRCD in structural and functional genomics are discussed by Miles & Wallace (Reference Miles and Wallace2006).

Finally, a future application of SRCD could be as an aid in identifying a potential model structure for molecular replacement phasing of crystal data. Although this has yet to be realised, with improved amounts of (low wavelength) data, a broader reference data set of spectra of known proteins and further developments in spectral matching and cluster analyses, SRCD spectra of proteins with closely related folds may prove to be useful for identifying a close structural (although not necessarily a close sequence) neighbour for such purposes.

5.4 Synchrotron radiation circular dichroism and other biophysical methods

Isothermal calorimetry (ITC) is a valuable method in structural biology for monitoring binding of various compounds to proteins; this can be combined with SRCD to examine the structural effects of such binding on the protein. For example, Bagger et al. (Reference Bagger, Hoffmann, Fuglsang and Westh2007) used SRCD in conjunction with ITC to compare two forms of the protein phytase from P. lycii (one heavily glycoslyated and the other enzymatically deglycosylated) and their interactions with the denaturing detergent SDS. The aim was to investigate that nature/role of the glycans in maintaining structural integrity. They showed that both the native and deglycosylated versions of the protein produced similar SRCD spectra and that the denatured versions of both variants were also similar, although there was a small additional requirement for denaturant to reach the same level of unfolding in the glycosylated protein. This study also showed, as other SRCD studies of denatured proteins have found (Matsuo et al. Reference Matsuo, Sakurada, Yonehara, Kataoka and Gekko2007), that the ‘denatured’ proteins were not fully unfolded.

SRCD and small angle X-ray scattering (SAXS) are examples of two synchrotron techniques that can produce useful complementary information if applied to a single sample type. This is because SRCD is particularly good at providing secondary structure information and SAXS can provide valuable information on the tertiary or quaternary structure of a protein. Grossmann et al. (Reference Grossmann, Hall, Kanbi and Hasnain2002) exemplified this by using SRCD to examine the conformational differences in the protein rusticyanin with and without its N-terminal extension. SRCD was important for the studies as the differences involved small amounts of beta sheet structures and SAXS was essential for showing that the changes did not involve changes to the association into hexamers.

The combined information from SRCD and SAXS can also provide powerful information against which to test molecular models. As an early example (Scott et al. Reference Scott, Grossmann, Tames, Byron, Wilson and Otto2002), the SRCD spectrum of protein haemoglobinase, a principal component of an iron acquisition system in pathogenic Escherichia coli, was used to confirm the predicted secondary structure of the protein, and then its size and shape were probed using analytical ultracentrifugation and SAXS. The experimental results from all these methods were compared with models of the structure as an extended crescent-shaped conformation and found to be in good agreement. A later example of this combination of methods was as a test for the predicted structure of a designed carboprotein (Hoiberg-Neilsen et al. 2008).

SAXS and SRCD have also been combined to identify the nature of complexes formed between two proteins. Stanley et al. (Reference Stanley, Sokolova, Brown, Clarke, Wilmanns and Svergun2004) combined SRCD and SAXS for studies of the C-terminal domain of Pex5p and the protein SCP2. They showed that the complex formation required little secondary structure rearrangement and comparisons of the irregular ‘shoe’-shaped particle of the complex with the known crystal structures of the individual proteins suggested that complexation required a conformational change of the C-terminus of SCP2.

In another study (Guerra-Giraldez et al. Reference Guerra-Giraldez, Moore, Neves, Wallace, Svergun, Brown and Smith2005), three complementary methods (SRCD, SAXS, NMR) were used to study the SHERP protein associated with the disease leishmaniasis. SRCD showed that in the absence of binding partners, the protein was ‘natively disordered’, a conformation not suitable for NMR studies, but very amenable to SAXS measurements, which confirmed it had an extended conformation under these conditions. SRCD was then used to identify specific binding partners which induced the protein to fold to a regular helical structure, whose detailed conformation was then determined by solution NMR spectroscopy.

6.1 Instrumentation

6.2 Bioinformatics: the Protein Circular Dichroism Data Bank

The Protein Circular Dichroism Data Bank (PCDDB) has recently (December 2009) been inaugurated. It has been developed for the deposition of and access to, CD and SRCD spectral and meta data (Wallace et al. Reference Wallace, Whitmore and Janes2006, 2009a; Whitmore et al. Reference Whitmore, Janes and Wallace2006), in a manner that parallels the information currently available in the Protein Data Bank (PDB) for crystallographic and NMR spectroscopic studies (Berman et al. Reference Berman, Westbrook, Feng, Gilliland, Bhat, Weissig, Shindyalov and Bourne2000). The goal of the PCDDB is to enable public availability of data collected on SRCD and CD instruments worldwide, for analysis, comparisons and data mining. It will also fulfil data sharing (NIH, 2003) and tracability requirements for drug development regulations (Guideline Q6B, 2001). The creation of the data bank (Wallace et al. Reference Wallace, Whitmore and Janes2006) and software for deposition and validation [the latter being comparable to the data checking programs WHATIF (Vriend, Reference Vriend1990) and PROCHECK (Laskowski et al. Reference Laskowski, Macarthur, Moss and Thornton1993)] associated with the PDB are a joint effort of Birkbeck College and Queen Mary College, University of London. Once fully established, the sustainability of the project will become a collaboration with SRCD beamlines, with mirror sites already agreed at the NSRRC and Soleil and expressions of interest from many other beamlines (Tao & Wallace, Reference Tao and Wallace2009).

The data bank is available at http://pcddb.cryst.bbk.ac.uk/. One of the aims of this is as a repository for validated spectra from a wide range of proteins, which will be as a resource for the future development of new SRCD reference data sets to further improve secondary structure analyses and new methods for de novo calculation of spectra from first principles. The access to spectra in this data bank is expected to enable or facilitate a wide range of other new studies, as outlined by Wallace et al. (Reference Wallace and Janes2009).

8.1 CDtool software

CDtool (Lees et al. Reference Lees, Smith, Wien, Miles and Wallace2004) is a downloadable (http://cdtools.cryst.bbk.ac.uk) software package for the processing and analysis of SRCD data. It was developed in order to enable comparisons of data obtained on different SRCD beamlines and on different conventional CD instruments that were all output in different formats. It includes functions for calibrating, displaying and archiving data.

CDtools is capable of accepting as input all data formats currently in use at SRCD beamlines, including spectra with different wavelength ranges, wavelength intervals, wavelength orders and including both CD and HT signals. Data processing functions include averaging, smoothing, baseline subtraction, zeroing, conversion between various units, calculation of error bars and scaling. Multiple data sets can be processed at one time. CDtools includes software to calibrate/cross-calibrate the data according to the procedure by Miles et al. (Reference Miles, Wien, Lees, Rodger, Janes and Wallace2003) if a standard reference spectrum has been obtained on the same beamline (Fig. 5). A header, which includes a record of all the processing functions that were performed, is created and appended to the output file. The headers can later be consulted for details of how the final spectrum was obtained. Output is in ASCII format files that are compatible with standard spreadsheet software as well as as input to the DichroWeb analysis server (Whitmore & Wallace, Reference Whitmore and Wallace2008) (see below). It can produce plots in a range of different graphics formats and styles.

Probably one of its most useful functions is the interface to a user-supplied MYSQL database, which enables the archiving of spectra for later retrieval and downloading, as well as creation of specialist reference data sets that can be used for spectral analyses.

Other functions include the ability to perform principal component and cluster analyses on a series of related spectra. Additional display functions include the ability to display protein structures from Protein Data Bank (Berman et al. Reference Berman, Westbrook, Feng, Gilliland, Bhat, Weissig, Shindyalov and Bourne2000) files, plus Ramachandran plots in order to identify which residues in a structure adopt specified phi, psi angles. It also includes a facility to display three-dimensional plots when a series of related spectra have been obtained (for instance, a thermal melt series, see Fig. 17), which enable detection of unfolding intermediates and clearly visualise the order in which different types of secondary structures unfold.

Fig. 18.

Comparison of the (a) SRCD spectra and (b) cCD spectra of a wildtype (black) and single point mutation (P23T) (red) human γD-crystallin. The error bars indicate one standard deviation in repeated measurements. It is clear that the two cCD spectra are not significantly different, but that the SRCD spectra are different in the region around 208 nm. (Data are from Evans et al. Reference Evans, Wyatt, Wistow, Bateman, Wallace and Slingsby2004.)

The website also includes an extensive downloadable users' manual.

8.2 DichroWeb server

The DichroWeb analysis server (Lobley et al. 2004; Whitmore & Wallace, Reference Whitmore and Wallace2004, Reference Whitmore and Wallace2008) is an online facility for calculating protein secondary structures based on SRCD and CD spectra. It is located at http://dichroweb.cryst.bbk.ac.uk. The server is capable of accepting a wide range of specific data formats from various beamlines and commercial CD instruments, as well as output files from a number of processing programs (including CDtool, described above) and free-format data in columns of wavelength versus CD signal. The input data can be in a wide range of data units (CD machine units or millidegrees, delta epsilon and mean residue ellipticity values), with different wavelength ranges, wavelength order and wavelength intervals. The user is also provided with the option of reading in a full wavelength data set, but specifying a cut-off wavelength to truncate it at a higher wavelength than the data was collected at, if the user later determines that the data below a certain wavelength are not valid for analysis. If the data are read in as millidegrees, they can be converted and output to more standard units such as delta epsilon or mean residue ellipticity by the server.

The server enables analyses using the five most popular analysis algorithms, SELCON3 (Sreerama & Woody, Reference Sreerama and Woody2000), CONTINLL (Van Stokum et al. Reference Van Stokkum, Spoelder, Bloemendal, Van Grondelle and Groen1990), CDSSTR (Sreerama & Woody, Reference Sreerama and Woody2000), VARSLC (Compton & Johnson, Reference Compton and Johnson1986) and K2D (Andrade et al. Reference Andrade, Chacón, Merelo and Morán1993). These can be combined with any of ten reference data sets (Whitmore & Wallace, Reference Whitmore and Wallace2008) including seven described by Sreerama & Woody (Reference Sreerama and Woody2000), the SP175 database (Lee et al. Reference Lees, Miles, Wien and Wallace2006a), a truncated version of SP175 (intended for use with conventional CD data) and CRYST175 (Evans et al. Reference Evans, Bateman, Slingsby and Wallace2007), which is meant specifically for proteins with a double Greek key fold (as is found in the βγ crystallins family). A new membrane protein reference data set (MP180) and the combination data set MSP180 (containing the SP175 and MP180 proteins and 25 additional soluble proteins) will also be made available as options in the near future. As a further analysis option, the user can specify a scale factor between 0·5 and 1·5, which can be used to test if the spectral magnitude provided is correct, as described in Miles et al. (Reference Miles, Whitmore and Wallace2005b).

The output produced includes the calculated secondary structure, the goodness-of-fit parameter (NRMSD) (Mao et al. 1981) for the calculated best fit, an overlaid plot of the experimental spectrum with the spectrum back-calculated from the best fit solution (Fig. 9) and several types of ASCII files of the experimental and calculated spectra that can be input into standard spreadsheet programs. A difference spectrum between the experimental and calculated spectrum is also provided, which can be useful in detecting features of the spectra that are not well-represented in the standard reference spectra or which result from systematic changes such as peak shifts produced by membrane proteins (Wallace et al. Reference Wallace, Lees, Orry, Lobley and Janes2003).

The website includes extensive help pages and a FAQ section; an extended users' guide, including examples of potential errors in data input and their solutions is included in the work of Whitmore & Wallace (Reference Whitmore, Wallace, Wallace and Janes2009).

8.3 Other analysis algorithms

The original analysis algorithms (Sreerama & Woody, Reference Sreerama and Woody2000) designed for use with CD data are available for download as part of the CDPRO package, which can be obtained at http://lamar.colostate.edu/~sreeram/CDPro/main.html. With the download are a number of reference data sets, including ones in which denatured proteins have been included (Sreerama et al. Reference Sreerama, Venyaminov and Woody2000). Several of the reference data sets have low-wavelength cut-off limits (178 nm) which would be suitable for use with SRCD data. The algorithms include SELCON3, a self-consistent method combined with a singular value deconvolution procedure, CONTINLL, a modified version of the original ridge regression algorithm of Provencher & Glockner (1982) method that incorporates the locally linearised model of Van Stokkum et al. (Reference Van Stokkum, Spoelder, Bloemendal, Van Grondelle and Groen1990), and CDSSTR, which uses a variable selection procedure and is a modified version of the original VARSLC method (Compton & Johnson, Reference Compton and Johnson1986). The input files require a specific data format, data in delta epsilon units, defined wavelength ranges and specific header information. The output includes an ASCII file that can be incorporated into a spreadsheet for graphics display. A short users' guide, including information on the input and output formats is available at the download site.

Other analysis algorithms (Lees et al. Reference Lees, Miles, Janes and Wallace2006b) designed specifically to take advantage of the extra information content in the low-wavelength data present in SRCD spectra, are available for download from http://webspace.qmul.ac.uk/rwjanes/algorithms.htm. These have been shown to be particularly useful for analysing the beta sheet components of proteins. One method included in this package is SELMAT, a version of the popular SELCON3 (Sreerama & Woody, Reference Sreerama and Woody2000) algorithm, but without the constraint that limits the number of reference spectra that can be used in the analyses. Also, by relaxing the sum, fraction, spectral and helix rules at each stage of the algorithm, it ensures that at least one solution can be obtained, something that is not always possible with the original algorithm. This software package also includes chemometric methods such as partial least squares (PLS), simultaneous partial least squares (SIMPLS) and principal component regression (PCR), as well as neural network (NN) and support vector machine (SVM) techniques. It outputs the calculated secondary structures from each of the methods in a number of different secondary structure classification schemes, including the most common one of regular and distorted helix and regular and distorted sheet (as used in CDPRO). Other schemes include the simple division of the structures into helix, sheet and ‘other’ categories or alternative classifications that separately identify the amounts of 3₁₀ and PPII helices, parallel and anti-parallel sheets and two types of beta turns. This separation into additional classes is possible only because of the additional information content in the low-wavelength SRCD data (Wallace & Janes, Reference Wallace and Janes2001).