The animal feed industry in the United States is worth almost $300 billion annually in sales, and livestock animals consume over 17 million tons of feed each year in the state of Texas alone This industry affects people around the world who come into contact with various animal-sourced products. In order to meet the demand for these products, animals must grow quickly and remain healthy; this is commonly accomplished by adding antimicrobial drugs to the animals’ feed supply. The Food and Drug Administration (FDA) has carefully regulated the use of feed additives over the years and has allowed their frequent use due to their benefits to animal health and growth rate. However, these drugs could potentially cause adverse health effects such as antimicrobial resistance in animals and in humans who consume animal products. Due to the growing knowledge of these adverse effects, the FDA has become increasingly concerned with the widespread use of many antimicrobials and other drugs as additives. The FDA’s concerns have led to the limitation and, in some cases, the banning of these drugs in animal feed. Due to this increased concern and regulation, finding quick and reliable methods of detecting antimicrobial agents in feed has become extremely important.
IN ORDER TO MEET THE DEMAND FOR (AGRICULTURAL) PRODUCTS... ANIMALS MUST GROW QUICKLY AND REMAIN HEALTHY...
The currently used mainstream methods for antimicrobial detection in feed, such as High Performance Liquid Chromatography, are highly accurate, but they are also costly, inconvenient, and impractical for rapid testing on a large scale, such as that of the feed industry Therefore, analysis labs need a faster, more practical method of screening feed samples for contaminants. For regulatory purposes, the FDA’s limitations on antimicrobials must be met. This can only be done through the accurate classification and quantification of the concentration of antimicrobials in feed samples. If a feed sample is tested and found to have an antimicrobial concentration that is even slightly above the allowed limit, that batch of feed must be removed from the market as quickly as possible. This process is paramount to keep the public safe and healthy. To do this, my project explores the use of Raman spectroscopy, specifically Surface-Enhanced Raman Spectroscopy, as a testing method that offers new approaches to achieve the goal of protecting the public.
Raman spectroscopy is used to identify the molecular structures of compounds and the amount of those compounds present in feed samples. In my project, I used a Raman spectrometer to apply laser pulses to liquid samples that were enhanced with a nanoparticle solution. The laser pulses cause vibrations in sample molecules; these vibrations shift the light frequency by specific amounts depending on the shift in energy level of each chemical bond. This shift, called a Raman shift, is unique to each molecule and allows for the identification of specific target analytes or in my case, the different antimicrobials. A detector measures this shift in light frequency and outputs the data as unique spectral plots. This process is shown in a simplified form in Figure 1.
The ultimate objectives of this project are (1) to collect qualitative spectral data for four antimicrobials and (2) to create accurate classification and quantification models for antimicrobial detection through the use of Raman spectroscopy using the data from one antimicrobial. Ideally, these models can open doors to future utilization of Raman spectroscopy in the feed industry. By building on the results provided by the models, feed production facilities can ensure the safety of their products in a faster and more cost-effective way than current methods.
I chose four antimicrobials that are of concern to the FDA: monensin, decoquinate, lasalocid sodium, and chlortetracycline. Initially, the lab ordered and obtained these antimicrobials in a pure form. Lab technicians extracted the antimicrobials using different extraction solutions, and I used the resulting solutions as the calibration standards for testing. A solution of gold or silver nanoparticles was added to the sample antimicrobial solutions to greatly enhance the signal received by the detector in the Raman spectrometer. Samples were tested three times each for qualitative analysis and ten times each if they were to be analyzed quantitatively. During the testing process, the Raman spectrometer subjected samples to the laser in pulses and recorded the Raman shifts for each test. These Raman shifts and the corresponding intensities were recorded as peaks in spectra as shown in Figure 1.
For the qualitative analysis conducted on all four chosen antimicrobials, the spectral data were processed, and the graphs were visually analyzed for peaks unique to each antimicrobial. In addition, the unique peaks were analyzed to determine if the intensity increased as the concentration increased, as seen previously in Figure 2.
For the quantitative analysis conducted on the monensin data, the spectra were analyzed using statistical methods. These methods each used a different algorithm to simplify the complicated data into clear results that could be analyzed. For this, I first processed the raw spectra to obtain data that could be analyzed easily. Then the spectral data were converted into numerical data, and those data were processed with statistical analysis methods with the help of my mentor. The methods used for the classification models were basic linear discriminant analysis, k-nearest neighbor method, and partial least-squares discriminant analysis. The methods used for the quantification models were multiple linear regression, principle component regression, and partial least squares regression.
Once the above statistical methods were performed, the two models, classification and quantification, were created. The successes of each statistical method were compared to the other methods used. The classification models were made with six categories of samples: one for blank samples and one for each of the five different concentration levels in the calibration standards. The success of each model was measured by the percentage of correctness with which it classified the sample in its appropriate category. The quantification models were created to plot the known concentration of monensin in calibration standards versus the model’s predicted concentration of these same samples, and success was measured by the value of the correlation coefficient. Finally, the models were validated by testing other calibration standards of known concentration and determining how well the models classified these new samples or quantified the exact concentrations in the samples.
Upon analysis, the spectra revealed unique peaks for all four antimicrobials tested. For the most part, these peaks had increasing intensities according to the concentration of antimicrobial present. Each antimicrobial present was identified, and the relative concentration present in the samples was determined by visually analyzing the spectral plots. For more exact classification and quantification of the monensin content in samples, the statistical models for monensin were analyzed and compared. Each statistical method had varying degrees of success in creating a reliable model that could predict the concentration of monensin in the samples.
As stated previously, for the classification model, the success of each statistical method was rated by the percentage of correctness with which it classified test samples into preset categories based on concentration level. This correctness for each method is shown in Table 1.
As seen in the table, the k-nearest neighbor method and the partial least squares discriminant analysis both classified samples with 97.6% accuracy in the validation tests. This indicates an important result: the k-nearest neighbor method, the partial least squares method, or both would be best used to create an accurate and reliable classification model.
Success for quantification was measured by the correlation coefficient, or value, of each data set and how close its value was to one. A value very close to one indicated that the model’s predicted concentration matched the actual known concentrations of monensin in samples. The results for standard Raman testing are shown in Figure 3 and Table 2.
It may seem that the values listed in Table 2 are low values; however, it is acceptable to have numbers as low as 0.7 in the feed screening process due to variability of samples as well as the extremely low concentrations at which the tests are run. As seen from the table, the multiple linear regression method had the most success overall with the highest values. This indicates an important result: the multiple linear regression method provides the best method for giving an accurate quantification model for samples.
With this knowledge, the methods with the best results can be applied to other antimicrobial samples to create similar classification and quantification models. These models can then be used in simple and efficient testing of feed samples on a larger scale. Finally, it is important to note that both the classification and quantification models showed a promising ability to predict antimicrobial concentration down to the parts per billion level. This ability could be further improved upon with additional study and data collection. The added study would preferably include the testing of as many more samples as possible, as well as spreading out the scope of the project to include other antimicrobials or chemicals of concern to the FDA.
WITH THIS KNOWLEDGE, THE METHODS... CAN BE APPLIED TO OTHER ANTIMICROBIAL SAMPLES TO CREATE SIMILAR CLASSIFICATION AND QUANTIFICATION MODELS...
Ultimately, in this project, I showed the ability of Raman spectroscopy to accurately detect specific concentrations of four antimicrobials in feed samples. Not only was Raman spectroscopy shown to be as accurate as current methods, but it was found to be much more rapid and straightforward. In addition to demonstrating the benefits of Raman spectroscopy itself, I also was able to utilize statistical analysis methods, which created the classification and quantification models that can aid in detection of any banned or restricted antimicrobials. The models from this project can be easily applied to other antimicrobials in feed samples and can be scaled up to rapidly screen samples in an industrial setting. By utilizing Raman spectroscopy testing methods and prediction models such as the ones created in this project, the feed industry can save money and time on regulatory product testing. More importantly, feed producers can quickly detect and discard any contaminated and possibly harmful feed that may otherwise get through the production process undetected. Finally, this technique can reduce the risk of the adverse health effects from antimicrobial contamination for the people who consume or use animal products in the state of Texas and the U.S as well as around the world.
NOT ONLY WAS RAMAN SPECTROSCOPY SHOWN TO BE AS ACCURATE AS CURRENT METHODS, BUT IT WAS ALSO FOUND TO BE MUCH MORE RAPID AND STRAIGHTFORWARD.
I would like to thank Dr. Kyung-Min Lee, Dr. Timothy Herrman, and Prabha Vasudevan for their mentorship and instruction. I would also like to thank the Office of the Texas State Chemist for allowing me to use the laboratory space and equipment throughout the length of my project.
Danielle Yarbrough ‘19
Danielle is junior chemical engineering major with a minor in biochemistry from Winnsboro, Texas. Danielle performed the research for this article under the supervision of Dr. Timothy Herrman and plans to continue it by attending graduate school beginning in the fall 2019 semester where she will pursue a doctorate in biochemistry. Danielle’s goal is to begin a career research in the pharmaceutical or medical industry, specifically in the field of nanoparticles and their use in immunotherapy.
1. American Feed Industry Association. 2017. “Feed Industry Statistics.”http://www.afia.org/files/feed _industry_stats2018.pdf.
2. McEwen, Scott A., and Paula J. Fedorka-Cray. 2002. “Antimicrobial use and resistance in animals.” Clinical Infectious Diseases, no. 34 (June): S93-S106. Accessed October 23, 2017. https://doi.org/10.1086/340246
3. Food and Drug Administration. 2017. “CFR-Code of Federal Regulations Title 21.” Accessed October 25, 2017.https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfCFR /CFRSearch.cfm?CFRPart=556&showFR=1.
4. Boscher, A., Cdric Guignard, Terence Pellet, Lucien Hoffman, and Torsten Bohn. 2010. “Development of a multi-class method for the quantification of veterinary drug rasidues in feedingstuffs by liquid chromatography-tandem mass spectrometry.” Journal of Chromatography A, no. 1217 (August): 6394-6404. Accessed November 1, 2017. https://www.ncbi.nlm.nih.gov/pubmed/20810120.
5. Lee, K. M., T. J. Herrman, and U. Yun. 2014. “Application of Raman spectroscopy for qualitative and quantitative analysis of aflatoxins in ground maize samples.” Journal of Cereal Science, no. 59 (January): 70-78. Accessed September 26, 2017.https://www.sci encedirect.com/science/article/pii/S0733521013001653.
6. Lee, K. M., T. J. Herrman, Y. Bisrat, and S. C. Murray. 2014. “Feasibility of surface-enhanced Raman spectroscopy for rapid detection of aflatoxins in maize.” Journal of Agricultural and Food Chemistry, no. 62 (April): 4466–4474. Accessed September 26, 2017.https://pubs.acs.org/doi/abs/10.1021/jf5 00854u.