University of Nebraska - Lincoln DigitalCommons@University of Nebraska - Lincoln

Faculty Papers and Publications in Animal Science Animal Science Department

2016

Determination of yolk contamination in liquid egg white using Raman spectroscopy

K. Cluff

G. Konda Naganathan

D. Jonnalagada

I. Mortensen

R. Wehling

See next page for additional authors

Follow this and additional works at: https://digitalcommons.unl.edu/animalscifacpub

Part of the Genetics and Genomics Commons, and the Meat Science Commons

This Article is brought to you for free and open access by the Animal Science Department at DigitalCommons@University of Nebraska - Lincoln. It has been accepted for inclusion in Faculty Papers and Publications in Animal Science by an authorized administrator of DigitalCommons@University of Nebraska - Lincoln. Authors K. Cluff, G. Konda Naganathan, D. Jonnalagada, I. Mortensen, R. Wehling, and Jeyamkondan Subbiah Determination of yolk contamination in liquid egg white using Raman spectroscopy

K. Cluff,∗ G. Konda Naganathan,† D. Jonnalagada,† I. Mortensen,† R. Wehling,‡ and J. Subbiah†,‡,1

∗Bioengineering, Wichita State University, Wichita, KS 67260-0035; †Biological Systems Engineering, University of Nebraska, Lincoln, NE 68583-0726; and ‡Food Science and Technology, University of Nebraska, Lincoln, NE 68588-0115

ABSTRACT Purified egg white is an important in- dard normal variate transformation. Samples were ran- gredient in a number of baked and confectionary foods domly divided into calibration (n = 77) and validation because of its foaming properties. However, yolk con- (n = 38) data sets. A partial least squares regression tamination in amounts as low as 0.01% can impede (PLSR) model was developed to predict yolk contami- the foaming ability of egg white. In this study, we nation levels, based on the Raman spectral fingerprint. used Raman spectroscopy to evaluate the hypothesis Raman spectral peaks, in the spectral region of 1,080 that yolk contamination in egg white could be detected and 1,666 cm−1, had the largest influence on detecting based on its molecular optical properties. Yolk contam- yolk contamination in egg white. The PLSR model was inated egg white samples (n = 115) with contamina- able to correctly predict yolk contamination levels with tion levels ranging from 0% to 0.25% (on weight ba- an R2 = 0.90 in the validation data set. These results sis) were prepared. The samples were excited with a demonstrate the capability of Raman spectroscopy for 785 nm laser and Raman spectra from 250 to 3,200 cm−1 detection of yolk contamination at very low levels in egg were recorded. The Raman spectra were baseline cor- white and present a strong case for development of an rected using an optimized piecewise cubic interpolation on-line system to be deployed in egg processing plants. on each spectrum and then normalized with a stan- Key words: foaming properties, functional properties, yolk contamination, Raman spectroscopy, partial least squares regression 2016 Poultry Science 95:1702–1708 http://dx.doi.org/10.3382/ps/pew095

INTRODUCTION and Wells, 1955;Smith,1959). Therefore, quality con- trol methods that are both rapid and accurate at low In the United States, 75 billion eggs are produced concentration levels are needed to ensure appropriate per year, which account for 10% of the world produc- egg white purity. tion (American Egg Board, 2015). Approximately 60% Earlier methods of measuring yolk contamination fo- of these eggs are used by consumers, 9% by food ser- cused on physico-chemical or gravimetric measurements vice sectors such as restaurants, and the rest are used (Murphy and Uhing, 1959; Cunningham and Cotterill, for manufacturing egg products (American Egg Board, 1964). More recent methods have used enzymatic deter- 2015). Purified egg white is an important egg product mination of phospholipids, or other yolk lipid compo- that is used extensively in the bakery and confectionary nents, or colorimetric determination of yolk cholesterol industries because of its foaming properties. Producing (Nielsen, 2000). The current industry standard relies pure egg white, without contamination of egg yolk, has on the physical dispersal of lipids in an aqueous solvent become a challenge because of the mechanization and (Bergquist and Wells, 1955); however, this method is speed of egg handling and breaking in egg processing fa- time consuming and requires an experienced technician cilities (Wang and Wang, 2009). Yolk contamination is to obtain reliable results. a critical issue because it affects the egg white’s ability Egg yolk is comprised of approximately 50% wa- to form stable and voluminous foams. Yolk contamina- ter, 16% protein, and 32% fat, whereas, egg white tion levels as low as 0.01% in egg white can significantly consists of nearly 90% water and only 10% protein affect foam volume (St. John and Flor, 1930; Bergquist (Li-Chan and Nakai, 1989). Due to the abundance of lipids in egg yolk, a recent study (Mortensen, et al., 2007) used near-infrared (NIR) spectroscopy to de- C 2016 Poultry Science Association Inc. tect yolk contamination in egg white by measuring Received August 1, 2015. Accepted February 4, 2016. the molecular vibrations of lipids. Mortensen et al. 1Corresponding author: [email protected] (2007) collected transmission spectra from 25 samples

1702 DETECTION OF YOLK CONTAMINATION IN EGG WHITE 1703 of egg white with yolk contamination levels ranging Poultry farm. After hand breaking the eggs, the egg from 0% to 0.25%. Partial least squares regression mod- white and egg yolk were separated using an egg sep- els were developed to predict yolk contamination with arator and pooled in separate beakers. Careful atten- an R2 value of 0.69 (Mortensen et al., 2007). While tion was given to ensure that no yolk was accidentally these results from the NIR technique are highly en- mixed in with the egg white. The egg white and yolk couraging, NIR is limited by the presence of water were then homogenized separately using a household molecules. The presence of water produces a broad ab- blender (Model: Hamilton Beach, Southern Pines, NC). sorption in NIR spectra, and thus can mask portions The blending speed was regulated using a voltage con- of the molecular vibrational spectral peaks of other troller such that there was minimal foaming of samples molecules. during blending operation to avoid denaturation of the Raman spectroscopy is a complementary technique egg white and yolk proteins. This process provided ho- to NIR spectroscopy, but is governed by different selec- mogenized pure egg white and yolk samples. tion rules. Raman spectroscopy responds to molecular Approximately 100 g of pure egg white was mea- vibrations that may not be observable by NIR absorp- sured out using an analytical balance with a precision tion. Raman and infrared (IR) spectroscopy both rely of ± 0.0001 g. To obtain different levels of yolk contam- on vibrational energies of molecules for detection, but ination, varying amounts of pure egg yolk were added each have different selection rules (Sasic and Ozaki, to the pure egg white using a disposable pipette. Each 2010). For example, vibrations that produce Raman egg white–yolk mixture was blended using a household spectral peaks frequently will not appear in IR spectra blender at low speeds with the help of a voltage reg- and vice versa. These selection rules are useful because ulator to obtain homogenized yolk contaminated egg they make the two spectroscopic methods complemen- white samples. The amount of egg yolk contaminant tary to each other. The selection rules are related to was recorded then divided by the amount of egg white the symmetry of the molecule. Molecules with a sym- and multiplied by 100% to obtain the percent contami- metric configuration are said to be Raman active be- nation. A total of 115 liquid egg white samples with yolk cause they will produce a peak in the Raman spectra contamination levels ranging from 0% to 0.25% were (Sasic and Ozaki, 2010). For example, a symmetric car- prepared. There were 31, 29, 19, 18, and 18 samples bon double-bond (C = C) will produce a Raman spec- with yolk contamination percentages of 0.00 to ≤0.05, tral peak around 1,657 cm−1 (Movasaghi, et al., 2007). 0.06 to ≤0.10, 0.11 to ≤0.15, 0.16 to ≤0.20, and 0.21 Conversely, molecules that have an asymmetrical con- to ≤0.25, respectively. figuration are said to be IR active, because they will produce peaks in IR spectra (Sasic and Ozaki, 2010). For example, a carbonyl (C = O), which has an asym- metrical configuration, will produce an IR spectral peak − at around 1,700 cm 1 (Sasic and Ozaki, 2010). Further- Instrumentation and Raman Spectra more, in contrast to NIR, Raman spectroscopy is not Collection affected by the presence of water molecules, which is critical for detecting yolk contamination in egg white. The instrumentation consisted of a Raman spec- In addition, the C = C stretching modes of unsatu- trometer (Model: EZRaman-M, Enwave Optronics Inc, rated fatty acids and CH2 scissoring modes of satu- Irvine, CA), with a fiber-optic probe, 785 nm excitation rated fatty acids, found in egg yolk, produce Raman laser, and sample holder. The fiber-optic probe had a peaks near 1,657 and 1,443 cm−1 respectively (Ozaki, 200 μm diameter collection fiber. The Raman spectrom- et al., 1992; Movasaghi, et al., 2007). Yolk contains eter had a spectral range of 250 cm−1 to 3,200 cm−1 and large amounts of lipids, while egg white contains no a spectral resolution of 6 cm−1. The Raman spectrome- lipids. Thus, Raman spectroscopy has a high potential ter was interfaced and controlled by a laptop computer for detecting yolk contamination in egg white and may using EZRaman software (Enwave Optronics). yield improved results compared to NIR. In this study, The Raman spectrometer was set up with a 90 second we evaluate the hypothesis that yolk contamination in integration time, three-point smoothing, and a three egg whites can be detected based on its optical proper- spectrum average. Approximately 1 mL of the yolk con- ties. The objective of this study was to optically probe taminated liquid egg white samples were transferred yolk contaminated egg white and develop a model to to standard Raman spectral cuvettes (10 mm2), and detect contamination levels based on Raman spectral Raman spectra were recorded. The Raman spectrom- fingerprints. eter collected 2,090 data points between wavenumbers 250 to 3,200 cm−1, approximately every 1.4 cm−1. Prior MATERIALS AND METHODS to recording the spectrum of a yolk contaminated egg white sample, a dark scan was taken by switching off the Yolk Contaminated Egg White Samples laser exciter. The spectral data collection took approx- imately 1 to 2 minutes per sample. The temperature A wide variety of chicken eggs were purchased from of the eggs and component pools during analysis was local supermarkets and the University of Nebraska 20◦C. 1704 CLUFF ET AL.

Figure 1. Baseline correction of raw Raman spectrum performed by fitting an iteratively optimized piecewise cubic polynomial.

Data Pre-processing yolk contamination levels. The PLSR algorithm com- bines techniques used in principal component analysis Data pre-processing of the Raman spectra included and multiple linear regression and attempts to quantify baseline correction and normalization. The Raman the strength of the relationship between the response spectra were baseline corrected using an optimized variable and a set of predictor variables (Geladi and piecewise cubic interpolation, shown in Figure 1.The Kowalski, 1986). PLSR searches for principal compo- baseline correction is fitted by an iteratively modified nents (called factors) that are orthogonal to each other multi-polynomial fitting algorithm that removes intrin- and tries to relate them to the response variable. The sic auto-fluorescence background signals and improves variation present in the predictor variables can be sum- signal-to-noise ratios (Lieber and Mahadevan-Jansen, marized into a few PLS factors. Decomposition of the 2003; Afseth, et al., 2006; Zhao, et al., 2007; Beier and PLS factors, using the loading vectors, can identify Berger, 2009). Following baseline correction, the spec- which variables had the most weight in the predictive tra were normalized using the standard normal vari- PLSR model. A 10-fold cross-validation, which is a com- ate (SNV) technique. SNV normalization is a common mon statistical analysis technique for estimating model mathematical transformation for spectral data and is performance, was performed on the calibration data set designed to remove multiplicative error and preserve to evaluate the performance of the predictive model. the linear relationship between the spectral signal and Cross-validation is a standard multivariate statistical sample concentration (Barnes, et al., 1989; Rinnan, technique often used on small data sets to assess model et al., 2009). The baseline correction, normalization, stability and determine how well it will perform on fu- and the subsequent statistical analyses were performed ture data sets (Hastie, et al., 2009). The cross-validation with custom in-house algorithms developed using Mat- technique rotates the membership of the samples (dur- lab R14 (R2011b, The Mathworks Inc, MA) software. ing calibration) to ensures that the results are not mem- bership dependent (i.e. calibration group and validation group) and to ensure that the model is not overfitting Data Analysis the data. Finally, after the model was calibrated, it was tested using the samples in the validation set. The data set was divided into calibration and val- idation data sets for model development and testing. Validation samples were selected by ranking the sam- RESULTS AND DISCUSSION ples based on the yolk contamination levels and then every third sample was placed in the validation data Raman Spectra set. Overall, there were 115 samples. The calibration data set contained 77 samples and the validation set The yolk contaminated egg white samples, prepared consisted of 38 samples. A partial least squares re- from pure egg white and yolk, represented commonly gression (PLSR) model was developed using the en- encountered contamination levels in commercial egg tire spectral region (250 cm−1 to 3,200 cm−1)topredict white products. Figure 2 presents Raman spectra of DETECTION OF YOLK CONTAMINATION IN EGG WHITE 1705

Figure 2. Raman spectra of pure and yolk-contaminated liquid egg white. Notable changes in spectral peaks reflect biochemical differences between yolk contaminated egg white and pure egg white and concentration differences at various levels of contamination.

Figure 3. A) The first two partial least squares (PLS) factors accounted for 82.4% of the variance in the calibration data set (n = 77). B) The mean squared error of prediction (MSEP) 10-fold cross-validation of the model. pure egg white at 0.1% and 0.2% yolk contaminated the calibrated PLSR model was able to predict egg yolk liquid egg white. Marked differences were noted, within contamination levels with a mean squared error of pre- the fingerprint region (700 to 1,800 cm−1)intheRa- diction cross-validation (MSEP-CV) of 0.0018 (Figure man peak intensities, as the yolk contamination lev- 3B) and an R2 of 0.86 (Figure 4). This indicates that the els increased. Raman peaks around 1,443 cm−1 and model is accurate, parsimonious, and stable because it −1 1,666 cm were attributed to CH2 vibrational stretch- only needed two PLS factors and achieved a high level of ing (known as scissoring), and amide I bands, respec- predictive accuracy, regardless of training group mem- tively (Ozaki, et al., 1992). Differences in the 1,080 cm−1 bership, while applying a 10-fold cross-validation anal- were ascribed to the imidazole moiety of tryptophan ysis of the model performance. Furthermore, the model (Harada and Takeuchi, 1986), and peak differences in was tested on a validation data set (n = 38), where the the 1,744 cm−1 were ascribed to carbonyl vibrations of PLSR model was able to predict egg yolk contamination fatty acid esters. levels with an R2 of 0.90, with a mean squared error of prediction (MSEP) of 0.0007, shown in Figure 5. The PLS regression coefficients (Figure 6)givean Prediction Model indication of which wavenumbers had the most im- Most of the variance (82.4%) in the egg white Ra- portant impact on predicting egg yolk contamina- man spectra, due to yolk contamination was accounted tion. From Figure 6, we can see that wavenumbers −1 for by the first two PLS factors (Figure 3A). Based on 1,666 and 1,080 cm had the largest beta coefficients the first two PLS factors and 10-fold cross-validation, and thus indicate that these are key Raman spectral 1706 CLUFF ET AL.

Figure 4. Calibration results (n = 77) using the first two partial least squares (PLS) factors and 10-fold cross-validation. The partial least squares regression (PLSR) model was able to predict egg yolk contamination levels with an R2 of 0.86 and a mean squared error of prediction (MSEP) 10-fold CV of 0.0018.

Figure 5. Validation results (n = 38) from the predictive partial least squares regression (PLSR) model. The PLSR model was able to predict yolk contamination levels with an R2 of 0.90 and a mean squared error of prediction (MSEP) of 0.0007. DETECTION OF YOLK CONTAMINATION IN EGG WHITE 1707

Figure 6. The partial least squares (PLS) regression coefficients used to calculate the response value (yolk contamination levels) from the Raman spectra.

Figure 7. A score plot (from the validation data n = 38) of the first and second partial least squares (PLS) factor scores, display how the data group together based on the yolk contamination levels in liquid egg white. 1708 CLUFF ET AL. peaks which capture biochemical information that re- REFERENCES flect yolk contamination levels. A score plot (Figure 7), of the first and second PLS factors, reveals that yolk Afseth, N. K., V. H. Segtnan, and J. P. Wold. 2006. Raman spectra of biological samples: A study of preprocessing methods. Appl. contamination levels, from five subcategories, rang- Spectrosc. 60:1358–1367. ing from 0.00−0.05, 0.06−0.10, 0.11−0.15, 0.16−0.20, American Egg Board 2015. History of Egg Production: Today. 0.21−0.25, cluster together as contamination levels in- http://www.incredibleegg.org/egg-facts/basic-egg-facts/history- of-egg-production/today. Accessed July 4th 2015. crease from left to the right on the first PLS factor. Barnes, R. J., M. S. Dhanoa, and S. J. Lister. 1989. Standard Normal This natural clustering of the data could be used in an Variate Transformation and De-trending of Near-Infrared Diffuse on-line quality control setting by identifying an accept- Reflectance Spectra. Appl. Spectrosc. 43:772–777. able contamination level as a cutoff value. For example, Beier, B. D., and A. J. Berger. 2009. Method for automated back- ground subtraction from Raman spectra containing known con- if a cutoff value of 0.1% yolk contamination was set as taminants. Analyst. 134:1198–1202. a quality control measure, the model would be able to Bergquist, D. H., and F. Wells. 1955. The monomolecular surface perfectly separate and classify contaminated egg white film method for determining small quantities of yolk or fat in egg from pure egg white based on the natural clustering and albumen in Institute of Food Technologists, Columbus, OH. Cunningham, F. E., and O. J. Cotterill. 1964. Effect of Centrifug- grouping of the PLS factor scores. ing Yolk-Contaminated Liquid Egg White on Functional Perfor- mance. Poult. Sci. 43:283–291. Geladi, P., and B. R. Kowalski. 1986. Partial least-squares regression: Conclusions and Significance a tutorial. Anal. Chim. Acta. 185:1–17. Harada, I., and H. Takeuchi. 1986. Advances in Spectroscopy of Bi- Purified egg white is an important ingredient in a ological Systems. John Wiley & Sons, Chichester, U.K. number of baked and confectionary foods because of Hastie, T., R. Tibshirani, and J. Friedman. 2009. The Elements of its foaming properties. Voluminous and stable foam- Statistical Learning: Data mining, Inference, and Prediction (2nd Ed). Springer, New York. ing in egg white is impeded by yolk contamination in Li-Chan, E., and S. Nakai. 1989. Biochemical basis for the properties amounts as low as 0.01%. The findings in this study are of egg white. Critical reviews in poultry biology. 2:21–58. significant because we were able to accurately and non- Lieber, C. A., and A. Mahadevan-Jansen. 2003. Automated method destructively predict yolk contamination levels with an for subtraction of fluorescence from biological Raman spectra. 2 Appl. Spectrosc. 57:1363–1367. R of 0.90 in a validation data set, based on optical Mortensen, I., J. Subbiah, R. L. Wehling, V. L. Schlegel, and G. properties of yolk contaminated egg white. Notable dif- W. Froning. 2007. Near-infrared Transmission Spectroscopy to ferences in spectral peak heights were found in the pure Detect Yolk Contamination in Egg White in American Society of Agricultural and Biological Engineers, Minneapolis, MN. egg white compared to the yolk contaminated spec- Movasaghi, Z., S. Rehman, and I. U. Rehman. 2007. Raman Spec- tra. Raman spectral peaks in the region of 1,080 and troscopy of Biological Tissues. Appl. Spectrosc. Rev. 42:493–541. − 1,666 cm 1 had the largest influence on the yolk con- Murphy, J., and E. Uhing. 1959. Method of processing egg albumen. tamination predictive model. These peaks reflect bio- Swift and Co., assignee. Pat. No. 2,892,720. Nielsen, H. 2000. Application of Chemical Methods to the Determi- chemical differences between yolk and pure egg white nation of Egg Yolk Contamination in Commercial Productions of including concentration differences at various stages of Egg White Compared to Enzymatic Determination. LWT - Food contamination. Based on these results, Raman spec- Science and Technology. 33:151–154. troscopy could be used to reliably determine yolk con- Ozaki, Y., R. Cho, K. Ikegaya, S. Muraishi, and K. Kawauchi. 1992. Potential of Near-Infrared Fourier Transform Raman Spec- tamination levels in liquid egg white. troscopy in Food Analysis. Appl. Spectrosc. 46:1503–1507. Finally, this method of yolk contamination assess- Rinnan, A.,˚ F. v. d. Berg, and S. B. Engelsen. 2009. Review of the ment is both rapid and cost effective. The speed at most common pre-processing techniques for near-infrared spectra. which each sample could be assessed was on the or- TrAC, Trends Anal. Chem. 28:1201–1222. Sasic, S., and Y. Ozaki. 2010. Raman, Infrared, and Near-infrared der of 1 to 2 minutes and most low-cost bench top Chemical Imaging. John Wiley & Sons, Inc, Hoboken, NJ. Raman spectrometers ($6 K to $12 K) would be suf- Smith, C. F. 1959. Shell Egg Deterioration: Diffusion of Yolk Lipids ficient for this analysis. These findings may aid in into Albumen as the Natural Cause of Failures in Performance. the development of an online quality control measure Poult. Sci. 38:181–192. St. John, J. L., and I. H. Flor. 1930. A Study of Whipping and for bakery and confectionary industries. We conclude Coagulation of Eggs of Varying Quality. Poult. Sci. 10:71–82. from these findings that optical Raman spectroscopy Wang, G., and T. Wang. 2009. Effects of yolk contamination, shear- provides information-rich spectra about the yolk con- ing, and heating on foaming properties of fresh egg white. J. Food Sci. 74:C147–156. taminated egg white which could be used as a foam- Zhao, J., H. Lui, D. I. McLean, and H. Zeng. 2007. Automated aut- ing quality control measure in an on-line industrial ofluorescence background subtraction algorithm for biomedical setting. Raman spectroscopy. Appl. Spectrosc. 61:1225–1232.