Determination of protein secondary structure has long been a major application of optical spectroscopic studies of biopolymers . In most cases these efforts have been aimed at evaluating the average fractional amount of helix and sheet contributions to the overall secondary structure in a protein or peptide. In some cases further interpretations in terms of turns and specific helix and sheet segment types have developed. This focus on average secondary structure is a consequence of the interactions of primary importance to these techniques and their relatively low resolution, which ordinarily does not provide site-specific information without technologically challenging selective isotopic substitution. Such a limit is in contrast to x-ray crystallography or nmr spectroscopy which naturally yield site-specific structural information, due to their very high resolution (at least for nmr), requiring at most uniform labeling. However, these invaluable structural biology techniques are very slow both in terms of data acquisition and in completion of the complex interpretive process. Furthermore, the intrinsic time scales of these measurements are slow, thereby not permitting reliable analysis of dynamic structures or of conformations undergoing fast changing events. Only more limited applications of optical spectra to determination of the tertiary structure or fold of the secondary structural elements have appeared, and these have typically used fluorescence or near-uv electronic circular dichroism (ECD) of aromatic residues to sense a change in the fold rather than to determine its nature .

IR and Raman analyses of secondary structure historically took a different approach , due to the natural resolution of the vibrational region of the spectrum into contributions from modes characteristic of different bond types in the molecule. Initial focus was on assigning component frequencies to various secondary structural component types as has been discussed in several reviews . Most effort focussed on the mid IR amide I (C=O stretch) band with additional use of the amide II (N-H deformation plus C-N stretch) in IR spectra and amide III (oppositely phased N-H deformation plus C-N stretch) with Raman spectral methods. In non-aqueous media, the near IR amide A (N-H stretch) can also be useful. Since both IR and Raman techniques give rise to single-signed spectral bandshapes that are effectively just the dispersed sum of the contributions from all the component transitions, and those components differ by only relatively small amount in frequency, the bandshapes for different proteins are very similar. However, due to the high signal to noise ratio (S/N) of Fourier transform IR (FTIR), this approach can be pushed further with resolution enhancement using second derivative or Fourier self deconvolution (FSD) techniques . These latter methods partially compensate for the relatively low level of bandshape variation available in the IR and Raman spectra, but are subject to abuse by the unwary user and submit the analysis to several assumptions that may well be unwarranted . Most notably, the frequencies assigned to specific secondary structural types are assumed to be unique, whereas any given type will shift significantly under influence of different solvents and residues . Additionally non-uniformity and end effects for these segments of a given structural type will have significant impact on the frequencies resulting in some dispersion of the contribution of a given secondary structure segment over the spectrum . Finally most methods assume that the dipole strengths (extinction coefficients) of all the residues, regardless of their conformation, is the same, while they in fact vary . None-the-less such FTIR methods have proven to be very useful for sorting out differences between proteins with similar IR spectra. Bandshape-based analyses similar to those used for ECD have also been applied to FTIR and Raman spectra with reasonable success as seen with ECD .

Of course this benefit comes at a cost, which arises from significantly reduced S/N and some theoretical interpretive difficulty as compared to IR. Developments on the latter front are fast bringing the theoretical capability for prediction of VCD spectra for small molecules to a level that is demonstrably superior to that for ECD spectra . However, these ab initio quantum mechanical methods are severely limited if one were to apply them to large molecules, such as proteins. Thus most biomolecular applications of VCD use empirically based analyses . Experimentally, instrumentation has reached a stage where VCD spectra for most molecular systems of interest can be measured under at least some sampling conditions . It is true that most VCD studies of biomolecules in aqueous solutions, naturally the conditions of prime interest, are restricted to relatively high concentration samples, much as is characteristic of nmr. (By comparison, ROA measurements demand even higher concentrations .)

On the other hand, qualitative analyses can be done with facility utilizing the VCD bandshape and its frequency position to predict the dominant secondary structural type in a peptide or protein. This has become the standard approach for most peptide studies . In such smaller fully solvated, biopolymers, the frequency shifts due to solvent effects and the inhomogeneity of the peptide secondary structure (as well as end fraying) are severe problems for frequency based analyses. In globular proteins, qualitative estimations of structure remain of interest for determining the dominant fold type, eg. highly helical, highly sheet or mixed helix and sheet type, but quantitative estimations of secondary structure content based on empirical spectral analyses are usually of more interest. Quantitative methods for analysis of protein VCD spectra in terms of structure follow the lead of most ECD analyses and employ bandshape techniques referenced to a training set of protein spectra . In this respect, globular proteins in aqueous solution are assumed, on average, to have their peptide segments in similar environments and of similar lengths, so that the solvent and length effects on the peptide modes will be relatively consistent for the training set and any unknowns studied.

In summary, unlike ECD, VCD can be used to correlate data for several different spectrally resolved features; and, unlike IR and Raman spectroscopies, each of these features will have a physical dependence on stereochemistry. But from another point of view, the combination of these techniques can compensate for each other, providing a balance between accuracy and reliability. The prime questions remaining in the VCD field now relate to application and interpretation of the method. It is clear that, despite claims of fundamental advantages of any one technique, progress in understanding of biomolecular structures will come from synthesizing all the data gathered from various techniques. In our biomolecular work, different types of spectral data are used to place bounds on the reliability of structural inferences that might be drawn from any one technique. Furthermore, such insight into average secondary structure may provide constraints on site-specific structure analyses using nmr and computational methods.

This new dimension in optical activity comes at some cost in that the rotational strengths of vibrational transitions as detected in VCD are much weaker than are those of electronic transitions detected in ECD. Similarly since VCD is a differential IR technique, its S/N can never approach that of FTIR which represents a summed response. Several research groups have developed instrumentation that makes the measurement of VCD reasonably routine over much of the IR region commercial FTIR vendors are now providing VCD accessories or, in one instance (Bomem-Biotools), a stand-alone VCD instrument that has now been shown by its users to have an exceptional S/N and baseline characteristics . In this section, instrument designs are briefly summarized and compared.

2.1 FT-VCD vs. DISPERSIVE VCD.

Available instrumentation makes routine measurement of VCD possible over much of the IR region down to ~700 cm^-1 on many samples. Development of a VCD instrument is normally accomplished by extending a dispersive IR or an FTIR spectrometer to accommodate, in terms of optics, time-varying modulation of the polarization state of the light and, in terms of electronics, detection of the modulated intensity that results from a sample with non-zero VCD. Our instruments and those of others are described in the literature in detail as referenced in recent reviews . A detailed review contrasting these designs and detailing components needed to construct either type of instrument has been published by this author . Here only a brief survey of the important components is given.

VCD instruments share several generic elements with "normal" CD instruments. All current instruments use a broad band source of light, typically utilizing black-body radiation from something like a ceramic or graphite-based glower (or tungsten in the near IR), to allow sampling of a spectrum over the IR region. The method chosen for encoding the optical frequencies divides VCD instruments into two styles. Dispersive VCD instruments use a monochromator, based on grating technology, which scans only through the wavelength spectrum of interest, recording the response sequentially. Such an instrument must be optimized for efficient light collection. On the other hand, Fourier transform (FT) VCD instruments use a Michelson interferometer that encodes the optical frequencies as an interferogram and gains efficiency through the multiplex advantage. With FT-VCD, spectral responses at all wavenumbers of interest are obtained simultaneously, with longer scans serving to improve the resolution of spectral features. Both styles of instrument then are modified to provide for linear polarization of the light beam, normally with a wire grid polarizer, and modulation of it, with a photo-elastic modulator (PEM), between (elliptically) right- and left-hand polarization states. The beam then passes through the sample and onto the detector, typically a liquid N₂ cooled Hg_1-x(Cd)_xTe (MCT) photoconducting diode. After preamplification, the electrical signal developed in the MCT is separated to measure the overall transmission spectrum (I_trans) of the instrument and sample via one channel and the polarization modulation intensity (I_mod), which is related to the VCD intensity, via the other.

Our original dispersive instrument is configured around a 1.0 m focal length, ~f/7 monochromator (Jobin-Yvon, ISA) that is illuminated with a home built carbon rod source . A mechanical chopper provides the modulation necessary for detecting the instrument transmission with an MCT detector. The monochromator output is filtered with a long-wave pass interference filter (OCLI) to eliminate light due to higher order diffraction from the grating and uses mirrors (achromatic) to focus the beam on the sample. A more recent, compact design has been shown to have advantages in terms of S/N and baseline stability .

The light is linearly polarized by means of a wire grid polarizer on a BaF₂ substrate (Cambridge Physical Sciences, Molectron Detectors in the USA) and modulated between left and right circularly polarized states with an AR coated ZnSe PEM (Hinds Instruments). Following the sample a ZnSe lens focuses the light onto an MCT detector chosen in terms of size and shape to match the slit image (we have an array of 3 2mm x 2mm elements) for optimal operation down to ~800 cm^-1. Alternatively, very high sensitivity in the near-IR (~5000-1900 cm^-1) is possible with an InSb photovoltaic detector, CaF₂ modulator and lens, and a grating optimized for that region. Lower frequencies can be accessed with different detectors but with loss of S/N .

To process the signal, a lock-in amplifier is used to detect the transmission intensity of the instrument and sample as evidenced by the signal developed in phase with the chopping frequency. The polarization modulation intensity is measured with a separate lock-in amplifier as that component of the detector signal in-phase with the PEM frequency. Since the VCD is also modulated by the chopper, the signal can be demodulated again by using another lock-in referenced to the chopper, but now measuring the output of the one referenced to the PEM. Dynamic normalization varies the amplification gain such that the transmission signal is constant. Applying the same gain to the polarization modulated signal assures a normalization much as is accomplished in "normal" CD instruments (which vary the high voltage applied to the photomultiplier to vary its gain and yield a constant average, or dc, current). In an alternate design, both signals can be A-to-D converted and normalization effected by digital division in the data computer .

Our FTIR-VCD spectrometer uses a Digilab (BIORAD) FTS-60A FTIR as its core but the choice of FTIR is fully open to the user, since these optics seem to impart little limitation for practical VCD operation and very successful instruments in other laboratories have been configured around a number of FTIRs . In a separate compartment, an external beam goes through the same type of linear polarizer, stress optic modulator, lens and relatively large area MCT detector as described above, with only weak focussing at the sample which leads to flatter baselines. MCT detectors can easily saturate to a non-linear response due to the high light levels in an FTIR. This can be controlled by using optical filters (e.g. 1900 cm^-1 cut-off low-pass) to isolate the spectral region of interest and by controlling the preamp gain. For aqueous, biological samples, the spectral band pass is limited by the solvent, so high light level is not a major problem.

The raw VCD is obtained by ratioing the spectrum of the polarization modulated signal with the normally developed transmission single beam spectrum using the FTIR computer processing software. In a rapid scan instrument, the detector signal is processed by a lock-in amplifier referenced to the modulator to form an ordinary interferogram of the modulated signal which the FTIR electronics can process. Most rapid scan FTIR VCD spectra have concentrated on mid-IR bands since the near-IR corresponds to higher frequency sidebands, which the lock-in attenuates. Slow- or step-scan operation yields better response for higher frequency, near IR, components of the spectrum . The optical frequencies are also encoded through correlation to the mirror position, as measured by use of laser fringe counting, but the time element is removed.

While the instrument throughput (equivalent to I_trans) is an ordinary IR intensity measurement for which all instruments are adequately programmed to process, the polarization modulation signal (I_mod) is not so simply processed. Often the integral of the modulated spectrum is very small having a rough balance between positive and negative VCD bands. This results in their being only a very weak center burst in the interferogram. For purposes of interferometer alignment and phase correction, this can pose difficulties . Normally, specialized software is required to overcome these limits, provide for simultaneous or sequential measurement from two independent detector (I_trans and I_mod) inputs, and permit a variety of arithmetic manipulations of the I_trans and I_mod spectra. Unlike the case for FTIR hardware, software is a central consideration in choosing an instrument for VCD use.

While FT-VCD has many advantages, the restriction to measurement only in the spectral windows of water and the relatively broad bands seen in biopolymer IR spectra can nullify the multiplex and throughput advantages of FTIR and, all other things being equal, favor use of dispersive VCD. Until now, the expected FTIR advantages have not been experimentally realized in terms of the S/N for low resolution biomolecular (aqueous) FTIR-VCD spectra as compared to what can be measured with the dispersive instrument over a similar time span FT-VCD measurement, which always encompasses the full spectrum, can be quite inefficient as compared to concentrating oneís effort and maximizing S/N in just one or two spectral bands, such as are accessible for a protein or peptide in D₂O.

To get adequate signal to noise ratio (S/N) and determine scan-to-scan reproducibility, the dispersive spectra are averaged for several scans, often using time constants of the order of 10 sec and resolutions of ~10 cm^-1. This means a typical IR band can take about 1/2 hour for a single scan. FTIR measured VCD spectra can sample a much wider spectral region and take ~1/2 hour (on our instruments) to collect an adequate number of scans for detecting the features of interest at higher resolution for a rigid, chiral organic molecule, but would require extensive averaging over much longer times to match the S/N available using the dispersive instrument for single bands in aqueous phase biopolymers. If, in the end, only one or two adjacent bands are needed for the analysis, much time can be lost with the FTIR-based technique; but if multiple bands are to be studied, FTIR-VCD retains its advantage, even for biological samples . In both cases, these inherently single-beam, though corrected for light-beam intensity by the I_mod/I_trans ratioing step, VCD scans must be coupled with equally long collections of baseline spectra to correct for instrument and sample induced spectral response. Finally in this comparison of FTIR and dispersive based VCD instruments, it might be noted that the more direct dispersive measurement is intuitively easier to interpret in case something goes wrong, such as noise or baseline artifact.

Due to its weak signal size, VCD is subject to artifacts which must be corrected by careful baseline subtraction. The best baseline is determined using racemic material, which is impractical for most biological materials. However, satisfactory baselines for spectral corrections can often be acquired with carefully aligned instruments by measuring VCD spectra of the same sample cell filled with just solvent. There exists no satisfactory theory of these artifacts that can be used to control baselines. Rather there is an empirical body of evidence that parallel or slowly converging beams, few reflections and uniform detector surfaces give the best results. (More recently, Nafie, in Applied Spectr. 2000, has shown that use of a double modulator approach can largely correct baseline offset and eliminate much of the worst artifacts.) Finally, most artifacts can also be minimized by careful optical adjustments which, at least in our instruments, are very stable, not requiring corrections for months. In this case, baseline subtraction also subtracts the bulk of the artifacts.

2.2 SAMPLING TECHNIQUES.

Most biomolecular systems are best studied in an aqueous environment. This poses difficulties for IR techniques due to water being a very strong absorber whose fundamental transitions strongly overlap regions of interest in biomolecules such as the N-H and C=O stretches. Consequently, the peptide amide I' (primes for N-deuterated amides) band at ~1650 cm^-1, which is dominated by the C=O stretch, has normally been measured in D₂O based solution. On the other hand, the amide II at ~1550 cm^-1 and amide III at ~1300 cm^-1 are best studied in H₂O.

In our laboratory, ECD spectra are additionally measured for the samples studied using a commercial instrument (now a JASCO J-810). These spectra are usually obtained under more dilute conditions using strain-free quartz cells (NSG Precision Cells or Helma Cells) obtained with various sample path lengths from 0.1 to 10 mm, the shorter path length cells being somewhat difficult to clean. Since relatively small amounts of biopolymer can give rise to significant ECD signals, it is very important to thoroughly clean sample cells between uses. Concentrations used in our laboratory for ECD are often of the order of magnitude of 0.1-1 mg/ml. For comparison of data obtained under comparable conditions, it is possible to measure ECD on the same samples used for VCD (> 10 mg/ml) by employing 6-15 µm path cells constructed with quartz windows and a teflon spacer .