Using Circular Dichroism to Determine Secondary Structure Elements of Proteins

Circular Dichroism (CD) spectroscopy was traditionally used to help establish the absolute stereo-configuration of small molecules. Naturally, once its potential application in the world of proteins had been realized, the CD technique was considered suitable for the determination of the relative fractions of a-helix and b-sheets — which are proteins’ most common secondary structure elements — using the deconvolution of CD spectra. However, once it became apparent that many other structural elements existed, it became important to approach the interpretation of a deconvolution analysis with caution, as the outcome depends largely on the choices that scientists make during analysis:

1. Reference spectra are required in deconvolution for the different secondary structure elements. In an ideal scenario, only those reference spectra that correspond to pure structural elements should be selected. In reality, however, proteins containing only a single element do not exist. Moreover, there is a bias in the selection of reference data — for instance, how does one decide how many reference spectra are chosen, and which ones? At the same time, what are the experimental conditions under which they have been acquired? Moreover, how well do they represent a structural element? Unfortunately, questions like these are often difficult to answer.

2. The structural elements included in the model dictate the results of deconvolution. An incorrect result is likely when a protein contains less common elements (as depicted in human thioredoxin above) that are not considered in the analytical model. On the other hand, if the model erroneously includes elements that are not present in the protein, the experimental data poses a risk of overfitting, and may be described perfectly by arbitrary combinations of secondary structure elements.

3. There are a variety of algorithms available to deconvolute CD spectra: for instance, least-squares, ridge regression, singular-value decomposition, variable selection, neural networks, principle component analysis, convex constraint analysis, matrix descriptors and the self-consistent method, among many more. Needless to say, these algorithms can differ in several aspects.  An example of this is how certain constraints can produce different numbers for the same data.

Case Study: Secondary Structure Analysis of Lysozyme and α-Chymotrypsin

Figure 1 illustrates the degree to which the results of secondary structure analysis are affected by user choices. In this experiment, secondary structure analysis was carried out for CD spectra of lysozyme and α‑chymotrypsin, which are two fairly well-characterized proteins.

BeStSel obtained a fit that coincides quite well with the lysozyme’s experimental CD spectrum. However, the resultant fractions of principle secondary structure elements deviated largely from those known from the x-ray structure of the protein. In addition, the content of the α-helix diverges by 30% and the β-sheet content is overestimated by 15%.

BeStSel is a tool optimized for secondary structure determination of proteins dominated by β‑sheets. This suggests that it should perform better with α-chymotrypsin containing two characteristic β-barrels. In reality, the principle secondary structure elements’ relative fractions, as obtained by BeStSel, are closer to those from the x-ray structure. However, the fit clearly fails to capture a feature of the experimental CD spectrum at about 230 nm. This suggests the possible presence of secondary structure elements in the protein that cannot be represented by the reference set used by BeStSel. In fact, despite the distorted structure of its β-barrels, α-chymotrypsin contains two fairly uncommon elements: an extended ϵ-Helix [2] as well as a β‑bulge [3].

Conventional CD Analysis. Results of deconvolution of spectra for determination of secondary structure elements depend on the chosen model, reference set and algorithm.

Figure 1: Conventional CD Analysis. Results of deconvolution of spectra for determination of secondary structure elements depend on the chosen model, reference set and algorithm.

In essence, BeStSel fails to report the correct secondary structure of α-chymotrypsin, despite the fact it was developed with the purpose of overcoming the oversimplification of other tools for secondary structure analysis. By breaking down the principle secondary structure elements into sub-elements, BeStSel includes regular and distorted helices, left- and right-twisted antiparallel β-sheets, relaxed and parallel β-sheets, as well as turns and other elements. In fact, BeStSel reports only a fraction of 7.4% parallel β‑sheet (black box) for α-chymotrypsin, although it is clear from the x-ray structure that no parallel β‑sheets are present within the protein.

The lysozyme example demonstrates how a good fit arising from secondary structure analysis does not necessarily imply agreement between the estimated fractions of secondary structure elements and the actual structure. False or misleading results can manifest from choosing a model that considers underrepresented secondary structure elements in the protein.

From the α-chymotrypsin example shown above, it can be assume that even a model optimized for particular secondary structure elements has potential limitations when attempting to describe the experimental data correctly.

The same proteins were thus analyzed with CDPro [4], to illustrate the effect of the choice of  different reference sets and algorithms for secondary structure analysis.

CDPro revealed that the fractions of principle secondary structure elements obtained for lysozyme are roughly similar to those known from the x-ray structure. However, the choice of a different reference set or algorithm – even while maintaining other parameters  – leads to varying results. For instance, the α-helix content differs by 13% when using a different reference set and by 9% when using a different algorithm. Likewise, the β‑sheet content shows variations in the range of 11% and 8%, respectively.

Conversely, the differences that arise from varying the reference set or algorithm are less pronounced in the case of α-chymotrypsin. Nevertheless, the fractions obtained from the principle secondary structure elements are substantially different from those arising from the x-ray structure. In fact, despite the different combinations of reference sets and algorithms, CDPro-obtained results suggest that α-chymotrypsin has more α-helices than β‑strands.

It thus becomes clear from the examples above that conventional analysis by deconvolution of CD spectra can only provide a rough estimate about secondary structure contents. Moreover, there are additional limitations that arise from the use of various tools for analysis.

Most tools are unlike BeStSel because they report results only for the principle secondary structure elements. Moreover, reference sets are typically smaller - typically consisting of only a few dozen spectra - and they overrepresent certain secondary structure elements.

In the examples above, CDPro used reference sets comprising 37 and 43 spectra, respectively. K2D3 is a tool designed to overcome this limitation, since it uses a reference set consisting of calculated spectra based on known structures from the protein database [5]. However, when put to the test, this tool performed more poorly than BeStSel and CDPro. Thus, it is safe to conclude that CDPro provides a choice between different reference sets and algorithms, unlike most other tools.

In addition, most tools require a particular interval in data points (typically 1 nm) and wavelength range for analysis. For instance, CDNN - a tool that has not been developed since 1998 - could not be used to analyze the above examples, since CDNN requires data up to 260 nm. Moreover, CDSSTR (as implemented in CDPro) was also excluded from analysis, because, in the author’s experience, this algorithm lacks reproducibility and yields slightly different values when the same data is analyzed repeatedly.

Some tools, such as DichroWeb, are only accessible online and may even require registration. Naturally, this poses a problem for companies in the biopharmaceutical industry in particular, as they often have higher security and confidentiality protocols which prohibit the uploading of data.

In light of the above limitations, the fact is that results often depend on user choices, and users rarely report all of the parameters needed to reproduce their analysis (such as reference set and algorithm). This makes it difficult to compare results from deconvolution analysis for the determination of absolute secondary structure contents. Thus, reliance on a single tool is not recommended, as estimates for absolute secondary structure should be put into perspective.

Performing secondary structure analysis is usually motivated by data reduction and the likelihood of evaluating structural changes, based on a few numbers, that can easily be compared and reported. Herein lies the superiority of other approaches, such as the Weighted Spectral Difference (WSD), which outperform conventional analysis (Section 4).

The computation of the WSD as a measure for similarity is transparent and is based on a user-defined reference, without the need for complicated algorithms. Moreover, the WSD method provides values often relating linearly to spectral changes. Most importantly, the WSD method is independent of data spacing and places no limitations on the wavelength range. Thus, this technique can also be applied to near-UV CD spectra to obtain information about changes in tertiary structure.

To conclude, this approach enables statistical analysis in the determination of significant structural changes without making assumptions about the protein structure.

© Applied Photophysics Limited, 2018. All rights reserved. Chirascan™ is a trademark of Applied Photophysics Limited. All other trademarks are the property of their respective owners.


[1] J. Kardos and A. Micsonai, “BeStSel.” [Online]. Available: [Accessed: 27-Jun-2018].

[2] R. Srinivasan, R. Balasubramanian, and S. Rajan, “Extended helical conformation newly observed in protein folding,” Science (80-. )., vol. 194, no. 4266, pp. 720–722, Nov. 1976.

[3] P. Manavalan and W. C. Johnson, “Sensitivity of circular dichroism to protein tertiary structure class,” Nature, vol. 305, no. 5937, pp. 831–832, Oct. 1983.

[4] R. W. Woody, S. Y. Venyaminov, W. C. Johnson Jr., and S. W. Provencher, “CDPro.” [Online]. Available: [Accessed: 27-Jun-2018].

[5] H. M. Berman et al., “The Protein Data Bank,” Nucleic Acids Res., vol. 28, pp. 235–242, 2000.

About Applied Photophysics

Applied Photophysics is a leading provider of systems and accessories for the biophysical characterization of biomolecules. Headquartered in Leatherhead, Surrey, UK, the Company has been established for more than 40 years.

The SX-range of stopped-flow spectrometers, used to monitor changes in absorbance and fluorescence during fast biological reactions, is acknowledged globally as the gold standard for kinetic studies. In 2005, the Company introduced the first Chirascan™ system, using the phenomenon of circular dichroism (CD) to characterize changes in the higher order structure of proteins.

Since then, the company has continued to incorporate its in-depth knowledge and understanding of CD into a range of Chirascan products that are used in cutting-edge research and to support the development of innovator drugs and biosimilars in the biopharmaceutical industry. Compared to conventional CD instruments, the new generation of Chirascan systems ensures that every scientist gets the most from every CD analysis.

Sponsored Content Policy: publishes articles and related content that may be derived from sources where we have existing commercial relationships, provided such content adds value to the core editorial ethos of News-Medical.Net which is to educate and inform site visitors interested in medical research, science, medical devices and treatments.

Last updated: Jan 8, 2019 at 4:05 AM

Other White Papers by this Supplier