Location: Quality and Safety Assessment Research Unit
2019 Annual Report
Objectives
The goal of the research is to develop and validate early, rapid, sensitive and/or high-throughput methods and techniques for detecting biological and physical hazards in poultry (food) products with optical sensing methods and instruments. Thus, the nature of the research is to combine chemistry and engineering disciplines (optical, agricultural, and food) with microbiological techniques to solve food safety detection problems in poultry (food). Specific objectives are:
Objective 1: Develop high-speed imaging methods for rapid detection of pathogens in live poultry flocks, and foodborne hazards, including foreign materials, in processed poultry products.
Sub-objective 1A: Develop Salmonella surveillance system for early detection of diseased birds.
Sub-objective 1B: Develop high-speed hyperspectral imaging methods and system for foreign material detection.
Objective 2: Develop rapid methods and protocols for early detection, identification, and quantification of pathogens in poultry products (foods) using imaging spectroscopy.
Sub-objective 2A: Develop hyperspectral microscope imaging (HMI) methods and system for early detection and identification of pathogen at the cellular level.
Sub-objective 2B: Develop fluorescence in-situ hybridization (FISH) imaging methods to identify pathogenic bacteria at the cellular level.
Sub-objective 2C: Develop nanobiosensor for pathogen detection with surface enhanced Raman spectroscopy (SERS) at the cellular level.
Sub-objective 2D: Develop methods for intervention carryover for Salmonella detection.
Sub-objective 2E: Develop methods for plate detection with optimized agar media at the colony level.
Objective 3: Develop methods to detect biofilms in poultry processing facilities with optical technologies. In developing the methods assess if any biomarkers can be identified to enhance or improve the detection sensitivity or specificity.
Approach
Ensuring poultry meat is safe to eat is of utmost importance to producers and consumers alike and rapid and early detection of foodborne pathogenic bacteria and foreign material in poultry products is needed. This research, which is divided into three objectives, primarily investigates optical sensors for rapid or improved detection of pathogenic bacteria with imaging and spectroscopic methods. Obj. 1A: an early-warning imaging surveillance system will be developed to detect bile in poultry droppings from laying hens in their cages. These higher levels of bile have been linked to birds with very high levels of Salmonella. Spectra will be collected to optimize key wavelengths and then a color-imaging system will be optimized for wireless real-time monitoring. Obj. 1B: Building on the success of a high speed hyperspectral imaging system developed within the unit, research will be expanded to detect foreign materials in various processed poultry products. Spectral libraries of normal meat features (muscle, fat, skin) and foreign material (rubber, metal, plastics, bone) will be used to develop algorithms suitable for high-speed use and then tested in real time. Obj. 2A: Hyperspectral microscope imaging (HMI) will be used to classify and quantify pathogens commonly found in poultry and other meats. The focus will be on identifying single bacteria cells from chicken rinsate by combining cell morphology and spectral profiles into an automated method for counting and classifying pathogenic bacteria. Additionally, markers will be used to enhance detection or means to separate and concentrate the bacteria will be implemented (immunomagnetic beads). Obj. 2B: Multiplex fluorescence in-situ hybridization (m-FISH) will be combined with HMI to further enhance detection with new protocols that will combine multiplexed probes and enhanced HMI detection resulting in broader, more robust methods of identification. Obj. 2C: Surface enhanced Raman spectroscopy (SERS), utilizing aptamers or antibodies and nano-enhanced surfaces, will be studied for Salmonella detection in broiler meat. Both labeled and label-free SERS will be evaluated. Obj. 2D: At FSIS’s request, research to neutralize sanitizers, frequently used to reduce pathogens while processing poultry meat, will be conducted to prevent interference of those sanitizers on bacterial analysis. Four potential neutralizing agents (quaternary ammonium, peroxyacetic acid, acidified sodium chlorite, acid solution, and dibromodimethylhydantoin sanitizers) will be screened for efficacy. Obj. 2E: Hyperspectral imaging (HI) systems will be used to classify pathogenic serovars growing in agar plates and collaborations will explore additional agar additives (both chromogenic and non-chromogenic) that will help differentiate serovars of E. coli O157:H7 and other shiga-toxin producing E. coli (STEC). Obj. 3: First in the lab, and then in processing plants, HI systems will be used and paired with spray-on markers to enhance the detection of biofilms on equipment surfaces. This research is potentially collaborative with ARS Beltsville and will help to discriminate biofilms from other organic material.
Progress Report
Evaluate factors causing low recovery of Campylobacter during regulatory sampling following acidified sodium chlorite (ASC) treatment of broiler carcasses and subsequent carry-over into neutralizing Buffered Peptone Water (nBPW) rinses. Since some poultry processing plants use ASC treatments on broiler carcasses to reduce pathogens in their final products, ARS has developed a nBPW rinse to reduce the effect of these treatments on samples sent to the Food Safety Inspection Service (FSIS) for testing. Solution alkalinity was shown to positively correlate with the presence of un-reduced chlorite anion in the recovery broth, implying that low recovery of Campylobacter from rinses may be due to residual chlorite, an oxidizing agent. To demonstrate the susceptibility of Campylobacter to residual chlorite, three strains were inoculated into pH= 7.5 nBPW with or without addition of sodium chlorite and stored for 24 hours at 4oC prior to culturing. Microbial counts from the solutions indicated that residual chlorite can decrease recovery by up to 4 log CFU/ml relative to controls. Acceptable recovery of Campylobacter from nBPW rinses containing residual chlorite may require development of a suitable neutralizing agent in the nBPW recovery medium.
Reduction of Campylobacter on poultry thighs using sequential treatments of antimicrobials. Campylobacter is a major concern for poultry processors as USDA performance standards have become stricter. This study evaluated the use of a low pH processing aid and peracetic acid (PAA) applied either individually or in consecutive dip treatments to reduce Campylobacter in thighs. Thighs were inoculated with a C. coli marker strain and each dipped into bags containing 1 L of treatment 1 for 6 s. Thighs were allowed 5 s to drip, placed onto foil for 60 s, and dipped into treatment 2 for 6 s. After 5 s drip time, each was placed in a bag with 150 mL buffered peptone water and hand shaken for 60 s; controls involved the same procedure with no treatment. Rinsates were serially diluted, plated onto Campy Cefex agar with 200 ppm gentamicin and incubated microaerobically for 48 h at 42°C. Procedures were replicated 5 times. Significant reductions (P<0.05) compared to untreated using consecutive dips of low pH and PAA were 98.2% and 99.3%, respectively. Treatments of low pH followed by peracetic acid reduced Campylobacter 99.2% from untreated thighs. Peracetic acid followed by low pH showed significant reductions compared to all other treatments (99.9% from untreated). This data suggests that treatment with an oxidizing agent (PAA) following by an acidic treatment maximizes Campylobacter reduction. Treating with this sequence may allow processors to meet the strict performance standards on Campylobacter in broiler parts.
Machine learning with convolutional neural networks enhanced classification performance of bacteria with hyperspectral microscope imagery. Illnesses caused by foodborne pathogens have become a global issue in the food industry. A hyperspectral microscope imaging (HMI) method, combined with a Machine Learning technique called convolutional neural networks (CNN), demonstrated the potential to classify foodborne bacterial species at the cellular level. HMI can acquire unique spatially-resolved spectral features of different bacterial cells. For HMI data analysis, ARS researchers at Athens, Georgia applied two different CNN models, including U-net and one-dimensional CNN (1D-CNN), to classify foodborne pathogenic bacteria. The U-net algorithm was used for automatic segmentation of bacterial cells, which performed better for generating an accurate mask with a faster processing time than conventional segmentation methods using Otsu and Watershed algorithms. For classification with bacterial species including Campylobacter, E. coli, Listeria, Salmonella and Staphylococcus, the 1D-CNN algorithm achieved a higher accuracy of 93.9% than conventional classification algorithms such as k-Nearest Neighbor (79.6%) and Support Vector Machine (82.8%). Thus, these two novel machine learning methods for classification improved performance of HMI for classification of foodborne bacteria.
Foodborne pathogenic bacteria identified and classified by visible/NIR hyperspectral microscopic imaging with deep learning algorithm. A hyperspectral microscope imaging (HMI) system, operating at wavelengths between 450-800 nm, allows the collection of spatial and spectral images from foodborne bacterial samples within 45 sec. Using these high-resolution contiguous spectral images, ARS researchers in Athens, Georgia utilized a novel deep learning method to identify and classify non-O157 Shiga toxin-producing Escherichia coli (STEC) including the “Big Six” (O26, O45, O103, O111, O121, and O145) at the serogroup level. The deep learning algorithm used stacked auto-encoder and soft-max regression (SAE-SR), to classify bacterial serogroups at the cellular level. The SAE-SR deep-learning algorithm performed with 97% classification accuracy and was much better than other classical algorithms such as linear discriminant analysis (LDA) with only 87.8% accuracy and support vector machine (SVM) with 94.5% accuracy. Thus, the newly developed deep learning algorithms classified the serogroups better than conventional chemometric models.
High-throughput Shiga-toxin detection with immune-sensing technology and surface plasmon resonance imaging. Shiga toxin-producing Escherichia coli (STEC) are responsible for gastrointestinal illness and even death. Current detection methods, such as real-time PCR and enzyme immunoassay (EIA), lack the rapid detection with high-volume samples that are needed for regulatory agencies. ARS researchers in Athens, Georgia, have demonstrated an optical method with surface plasmon resonance imaging (SPRi) that has the potential for rapid, label-free, and simultaneous screening of multiple pathogenic bacteria. Recently, they extended the evaluation of SPRi for detection of Shiga-toxins (Stx1, Stx2) produced by E. coli. By collaborating with ARS researchers in Albany, California, they selected the optimal Shiga-toxin antibodies among eight different antibodies (Stx1pAb, 1-2mAb, 1d-3mAb, 1e-4mAb, Stx2pAb, 2-1mAb, 2-2mAb, and 2-10mAb) for use with a SPRi gold (50 nm) coated sensor chip through a mercaptoundecanoic acid monolayer and carbodiimide crosslinking, and subsequently blocked with 1% skim-milk proteins. Shiga-toxins were detected by SPR-sensorgram analysis of the difference images between target molecules and references. Among the eight antibodies tested, Stx1pAb and Stx1d-3mAb for toxoid 1A and Stx2-1mAb and Stx2-2mAb for toxoid 2A performed well with a 50-ng/mL limit-of-detection and detection time of 20 min. The results suggest that the SPRi method with selected antibodies has the potential for rapid and high-throughput Shiga-toxin detection.
Deep learning techniques for detection of microbial colonies on agar plates. Research was conducted to evaluate the feasibility of deep learning-based hyperspectral image analysis to classify 15 different strains of the “Big Six” non- O157:H7 Shiga toxin producing Escherichia coli (STEC) plus O157:H7 E. coli colonies on each of two agar media types including modified Rainbow and modified MacConkey agar. Initial results indicated that the deep learning technique, known as 3D convolutional neural network, resulted in a detection accuracy of 88-92%, compared to earlier chemometric-based detection methods with 49-60% accuracy.
Development of database for phenotyping of biological materials with artificial intelligence (AI) and big data analytics. An international collaborative research project is being conducted to develop a generic data models and architecture for a database system that can be optimized for phenotyping plants and pathogenic microorganisms via Artificial Intelligence and big data analytics. Preliminary research indicated that a database-type, known as NoSQL, was necessary for managing unstructured data such as images, metadata and research notes coming from various sources including laboratory instruments, field work, and sensors.
Accomplishments
1. Reduction of Campylobacter on poultry thighs using sequential treatments of antimicrobials. Campylobacter is a major concern for poultry processors, as USDA performance standards have become stricter. ARS researhers in Athens, Georgia, evaluated the use of a low-pH processing aid and peracetic acid (PAA) applied in either individual- or consecutive-dip treatments to reduce Campylobacter in thighs. Thighs were inoculated with a marker strain of C. coli and then dipped into bags containing either the low pH processing aid or the PAA. Combinations of dual low-pH dips, dual PAA dips, low pH then PAA, and PAA then low-pH were all evaluated against a control of dual buffer dips. Peracetic acid followed by low pH dips showed significant reductions compared to all other treatments (99.9% from untreated). This data suggests that treatment with an oxidizing agent (PAA) following by an acidic treatment (low pH) maximizes Campylobacter reduction. Treating with this sequence may allow processors to meet the strict performance standards on Campylobacter in broiler parts.
Review Publications
Eady, M.B., Gayatri, S., Park, B. 2018. Detection of Salmonella from chicken rinsate with hyperspectral microscope imaging compared against RT-PCR. Talanta. 195:313-319. https://doi.org/10.1016/j.talanta.2018.11.071.
Park, B., Eady, M.B., Oakley, B., Yoon, S.C., Lawrence, K.C., Gamble, G.R. 2019. Hyperspectral microscope imaging methods for multiplex detection of Campylobacter. Journal of Spectral Imaging. 8:a6. https://doi.org/10.1255/jsi.2019.a6.
Kang, R., Park, B., Chen, K. 2019. Identifying non-O157 Shiga toxin-producing Escherichia coli (STEC) using deep learning methods with hyperspectral microscope images. Spectrochimica Acta. https://doi.org/10.1016/j.saa.2019.117386.
Landrum, M., Cox Jr, N.A., Cosby, D.E., Berrang, M.E., Gamble, G.R., Da Costa, M.J., Pesti, G.M. 2018. Low pH processing aid to lower the presence of naturally occurring campylobacter on whole broiler carcasses. Advanced Food and Nutritional Sciences. 3:7-13.
Gamble, G.R., Berrang, M.E., Cosby, D.E., Cox, N.A., Hinton, A. 2019. Neutral pH Sodium Chlorite decreases recovery of Campylobacter in neutralizing buffered peptone water from simulated broiler carcass rinses. Journal of Food Safety. https://doi.org/10.1111/jfs.12656.
Landrum, M.A., Cox Jr, N.A., Wilson, J.L., Berrang, M.E., Gamble, G.R., Harrison, M.A., Fairchild, B.D., Kim, W.O., Hinton Jr, A. 2019. Reduction of campylobacter on poultry thighs using sequential treatments of antimicrobials. Advanced Food and Nutritional Sciences. 4:1-7. https://doi.org/10.21065.
