International Journal of Food Microbiology 204 (2015) 66–74

Contents lists available at ScienceDirect

International Journal of Food Microbiology journal homepage: www.elsevier.com/locate/ijfoodmicro

Short communication

Rapid detection of Listeria monocytogenes in milk using confocal micro-Raman spectroscopy and chemometric analysis Junping Wang a,b, Xinfang Xie a, Jinsong Feng c, Jessica C. Chen c, Xin-jun Du a, Jiangzhao Luo a, Xiaonan Lu c,⁎, Shuo Wang a,⁎⁎ a b c

Key Laboratory of Food Nutrition and Safety, Ministry of Education, Tianjin Key Laboratory of Food Nutrition and Safety, Tianjin University of Science and Technology, Tianjin 300457, China Synergetic Innovation Center of Food Safety and Nutrition, Harbin 150030, China Food, Nutrition, and Health Program, Faculty of Land and Food Systems, University of British Columbia, Vancouver, British Columbia V6T 1Z4, Canada

a r t i c l e

i n f o

Article history: Received 15 December 2014 Received in revised form 16 March 2015 Accepted 22 March 2015 Available online 1 April 2015 Keywords: Raman spectroscopy Listeria monocytogenes Chemometric analysis Milk

a b s t r a c t Listeria monocytogenes is a facultatively anaerobic, Gram-positive, rod-shape foodborne bacterium causing invasive infection, listeriosis, in susceptible populations. Rapid and high-throughput detection of this pathogen in dairy products is critical as milk and other dairy products have been implicated as food vehicles in several outbreaks. Here we evaluated confocal micro-Raman spectroscopy (785 nm laser) coupled with chemometric analysis to distinguish six closely related Listeria species, including L. monocytogenes, in both liquid media and milk. Raman spectra of different Listeria species and other bacteria (i.e., Staphylococcus aureus, Salmonella enterica and Escherichia coli) were collected to create two independent databases for detection in media and milk, respectively. Unsupervised chemometric models including principal component analysis and hierarchical cluster analysis were applied to differentiate L. monocytogenes from Listeria and other bacteria. To further evaluate the performance and reliability of unsupervised chemometric analyses, supervised chemometrics were performed, including two discriminant analyses (DA) and soft independent modeling of class analogies (SIMCA). By analyzing Raman spectra via two DA-based chemometric models, average identification accuracies of 97.78% and 98.33% for L. monocytogenes in media, and 95.28% and 96.11% in milk were obtained, respectively. SIMCA analysis also resulted in satisfied average classification accuracies (over 93% in both media and milk). This Raman spectroscopic-based detection of L. monocytogenes in media and milk can be finished within a few hours and requires no extensive sample preparation. © 2015 Elsevier B.V. All rights reserved.

1. Introduction Listeria are Gram-positive, facultatively anaerobic, and non-spore forming microbes. The genus Listeria consists of six closely related species: Listeria monocytogenes, Listeria grayi, Listeria innocua, Listeria seeligeri, Listeria ivanovii, and Listeria welshimeri (Davis and Mauer, 2011). In addition, several proposed novel species have been identified in recent years (den Bakker et al., 2014; Lang Halter et al., 2013; Leclercq et al., 2010; Weller et al., in press). Of these, only L. monocytogenes is a serious threat to humans causing a life threatening disease known as listeriosis in susceptible populations (Drevets and Bronze, 2008). A report from US FoodNet concluded that listeriosis was related to 30% of foodborne deaths from 1996 to 2005 and resulted in a high case of fatality rate of 16.9% ⁎ Correspondence to: X. Lu, Food, Nutrition, and Health Program, Faculty of Land and Food Systems, The University of British Columbia, Vancouver, British Columbia V6T 1Z4, Canada. Tel.: +1 604 822 2551; fax: +1 604 8225143. ⁎⁎ Correspondence to: S. Wang, Key Laboratory of Food Nutrition and Safety, Ministry of Education of China, Tianjin University of Science and Technology, Tianjin, 300457, China. Tel.: +86 22 60601430; fax: +86 22 60601332. E-mail addresses: [email protected] (X. Lu), [email protected] (S. Wang).

http://dx.doi.org/10.1016/j.ijfoodmicro.2015.03.021 0168-1605/© 2015 Elsevier B.V. All rights reserved.

(Barton Behravesh et al., 2011). In addition, this pathogenic bacterium represents a major concern to food safety because it can grow in refrigerated ready-to-eat food products (Gandhi and Chikindas, 2007). Globally, fluid milk and dairy products have been implicated in many large-scale listeriosis outbreaks (Büla et al., 1995; Fleming et al., 1985; Fretz et al., 2010; Koch et al., 2010; Linnan et al. 1988; Macdonald et al., 2005Makino et al., 2005). Further, risk assessment data from the United States ranks several categories of dairy products as high-tomoderate risk in terms of predicted listeriosis cases on a per serving basis (Food and Drug Administration et al., 2003). Given this, research on the development of applicable tools for the rapid detection of L. monocytogenes in fluid milk is warranted. To date, numerous rapid detection methods have been developed for Listeria and L. monocytogenes, including polymerase chain reaction (PCR) and RT-PCR based methods (Bickley et al., 1996; Nogva et al., 2000), loop mediated isothermal amplification assays (Cho et al., 2014), phage based assays (Loessner et al., 1997), biosensors (Ohk and Bhunia, 2013), immunoassays (Magliulo et al., 2007), matrix-assisted laser desorption ionization time-of-flight (MALDI-TOF) mass spectroscopy (Barbuddhe et al., 2008), Fourier transform infrared (FT-IR) spectroscopy (Holt et al.,

J. Wang et al. / International Journal of Food Microbiology 204 (2015) 66–74

1995; Rebuffo et al., 2006), and Raman spectroscopy (Oust et al., 2006). Among these methods, vibrational spectroscopy (i.e., FT-IR and Raman spectroscopies) is a non-destructive tool because there is no requirement to add chemical dyes or labels for bacterial identification. Spectroscopic techniques have been extensively employed to detect and discriminate different microorganisms (Bosch et al., 2008; Kuhm et al., 2009; Lu et al., 2012; Maquelin et al., 2002; Preisner et al., 2010) and further validated to reliable typing methods for bacterial epidemiology (Kirschner et al., 2001; Willemse-Erix et al., 2009) in a real-time manner. Compared to FT-IR spectroscopy (measuring IR absorption of a sample), Raman spectroscopy collects the inelastically scattered light of molecules by laser excitation (mainly 532 nm, 633 nm, and 785 nm). Raman spectroscopy is suitable for the analysis of biological samples in an aqueous environment, which is a major challenge for IR absorption spectroscopy. Further, Raman spectral bands are narrow and thus more information can be recorded within spectral fingerprinting regions (defined as 400–1800 cm−1) than with IR spectroscopy. Macromolecules such as nucleic acids, proteins, lipids, and carbohydrates are Raman active molecules. Thus, Raman spectroscopy can elucidate chemical and molecular composition, structure, and interactions within bacterial cells (Lu et al., 2011). Advanced chemometric analysis is required to interpret vibrational spectra because differences in the raw spectral features among biological specimens are always minor (Bocklitz et al., 2011; Goodacre, 2003). Chemometric analysis can further amplify the minor spectral variations and establish segregation and/or quantitative analysis models. There are three major pattern recognition methods: unsupervised principal component analysis (PCA), hierarchical cluster analysis (HCA), and supervised discriminant analysis (DA) and these provide either cluster plots or dendrogram structures for segregation and discrimination. Recently, soft independent modeling of class analog (SIMCA) has also been extensively employed for bacterial identification and speciation (Lu et al., 2011). In this study, we aimed to use segregation models to establish a “three-level” classification approach to identify L. monocytogenes. For “first-level” approach, we differentiated Raman spectra between Gram-positive and Gram-negative bacteria. For “second-level” approach, we constructed chemometric models to discriminate different Listeria species. Both segregation models were developed using bacterial cells cultivated in media. For “third-level” approach, we aimed to detect and differentiate L. monocytogenes directly from milk, a representative food sample. Raman spectral patterns of bacterial cells grown in different nutrient media (e.g., media versus food) may result in differences in the chemical composition of bacterial cell membranes and cell walls because bacterial growth may be affected by variation in the composition of nutrient media (Hutsebaut et al., 2004). Thus, the spectra used to construct chemometric models between bacteria grown in liquid media and milk are different, and separate models were constructed for the detection and assessment of Listeria and L. monocytogenes in milk. In the present study, we confirmed that confocal micro-Raman spectroscopy coupled with chemometric analysis is a fast and reliable method to identify and discriminate Listeria spp. in milk, including human pathogen L. monocytogenes.

2. Material and methods 2.1. Bacterial strains The following bacterial strains were evaluated in this study: six closely related Listeria species, namely L. grayi CICC21670, L. monocytogenes CICC21663, L. innocua CICC10417, L. ivanovii CICC21663, L. seeligeri CICC 21671, and L. welshimeri CICC21672. Staphylococcus aureus ATCC49521, one Salmonella enterica clinical isolate, and one Escherichia coli clinical isolate were also used as representatives of Gram-positive and Gramnegative bacteria for model construction and testing.

67

2.2. Bacterial sample preparation Bacteria were grown aerobically at 37 °C for 14 to 16 h; Listeria was grown in Brain Heart Infusion (BHI) broth, while other bacteria (i.e., S. aureus, S. enterica and E. coli) were cultivated in Luria-Bertani (LB) broth. Bacterial cultures were then harvested and 1 ml of culture was centrifuged at 12,000 ×g for 5 min at 22 °C. For Raman spectral collection of bacteria grown in media, the supernatant was discarded while the pellet was washed three times using phosphate buffer saline (pH = 7.4) and resuspended in 1 ml deionized water. For Raman spectral collection of bacteria grown in milk, the overnight culture was 10-fold diluted and 1 ml aliquot was inoculated into UHT-treated milk [fat content 3.7% (w/v)] yielding a final concentration of 108 CFU/ml, as determined by conventional plating assay. The inoculated milk samples were incubated at 22 °C for 24 h, allowing for bacterial adaptation to the new matrix (i.e., milk). Then, a 1 ml aliquot was centrifuged at 7000 ×g for 5 min at 22 °C. The top layer was discarded, followed by addition of 1 ml phosphate buffer saline containing 0.5% (w/v) Tween-20. This step was repeated four times and the resultant bacterial pellet was resuspended in deionized water. Five microliters of each resultant bacterial sample was transferred to a glass microarray slide coated with a thin film of gold (Thermo Scientific Inc., Waltham, MA) and dried at 22 °C. This gold-coated microarray slide has low fluorescence, providing a great background for Raman spectral collection with a high signal-to-noise ratio (Lu et al., 2012). 2.3. Raman spectroscopic instrumentation A Renishaw inVia Raman system (Renishaw, Gloucestershire, UK) equipped with a Leica microscope (Leica Biosystems, Wetzlar, Germany) was used. This system is equipped with a 785-nm nearinfrared diode laser. Raman scattered light was collected and dispersed by a diffraction grating, and finally the Raman scattering signal was collected by a CCD array detector (576- by 384-pixel). Gold-coated microarray chips covered with bacterial samples (i.e., Listeria, S. aureus, S. enterica and E. coli) from either media or milk were mounted onto the microscope stage under the exposure of incident laser (300 mW). Raman spectra were collected by using a 50× objective over a 10-s exposure time with wavenumbers of 400–1800 cm−1. 2.4. Databases Two databases were created to analyze all spectra for each bacterial sample, including Listeria. Ten Raman spectra were collected for each sample in one experiment (N = 3). All spectra for bacteria grown in media were collected and introduced into a database named “media” (240 spectra), while all spectra for bacteria grown in milk were centralized into a second database named “milk” (240 spectra). In sum, three batches of each bacterial sample (30 spectra per sample) were used to construct the classification model, while an independent fourth batch of bacterial sample (10 spectra per sample) was used for model testing. 2.5. Data preprocessing and chemometrics 2.5.1. Data preprocessing Data preprocessing using OPUS (Bruker, Germany) was conducted before the construction of chemometric models. We first conducted a baseline subtraction to remove background fluorescence from Listeria cells coated on gold-coated microarray slides. In addition, this baseline subtraction can also maximally reduce Gaussian noise and CCD background noise. A polynomial background fit was applied to determine and correct the fluorescence background, as described in a previous study (Lieber and Mahadevan-Jansen, 2003). Spectral smoothing was then performed using a 9-point Savitzky–Golay algorithm. Other spectral preprocessing treatment was followed, including normalization and first derivative transformation.

68

J. Wang et al. / International Journal of Food Microbiology 204 (2015) 66–74

2.5.2. Classification chemometric models Chemometrics reduce the dimensions of data sets into fewer independent parameters for easy interpretation (Goodacre, 2003). Thus, chemometric models were developed using Matlab (The MathWorks, Inc., Natick, MA, USA). Unsupervised principal component analysis (PCA) and hierarchical cluster analysis (HCA) investigated the chemical variations in bacterial samples (i.e., Listeria species) by constructing a cluster-based or tree-based segregation model (Huang et al., 2004). To evaluate the performance (e.g., recognition rate) and reliability of PCA and HCA models, supervised chemometric models were subsequently developed. Here we used Mahalanobis distance-based distance discriminant analysis (DDA) and naïve Bayesian discriminant analysis, as two representative discriminant analysis (DA), to classify different Listeria species. We also applied soft independent modeling of class analogies (SIMCA) for a further testing because SIMCA may identify samples as belonging to multiple classes and not produce a classification of samples into non-overlapping classes (Lu et al., 2011). The wavenumbers between 400 and 1800 cm−1 were selected for the construction of all chemometric models. 3. Results 3.1. Differentiation of bacteria in media using Raman spectra-based PCA model Unsupervised chemometric analysis was applied to analyze Raman spectra derived from bacterial samples. Fig. 1 shows the PCA cluster model for the segregation of six different Listeria species (i.e., L. innocua, L. monocytogenes, L. grayi, L. seeligeri, L. ivanovii, and L. welshimeri), one S. aureus strain as well as two representative Gram-negative bacterial strains (S. enterica and E. coli). Gram-positive bacteria (Listeria spp. and S. aureus) and Gram-negative bacteria (S. enterica and E. coli) were clearly segregated by PC 1 and PC 2 scores. In addition, Listeria spp. was well separated from other bacteria. 3.2. Raman spectral features of Listeria bacteria There are several factors that can affect Raman spectral reproducibility, including bacterial cultivation time, growth temperature, media use, and wavenumber selection (Lu et al., 2011; Lu et al., 2012). We collected 40 Raman spectra from four independent experiments for each bacterial isolate grown in media. Average Raman spectra and standard deviation

are shown in Fig. 2. The low standard deviation of Raman spectra demonstrates the high reproducibility of the spectral dataset. The average Raman spectra of six different Listeria species grown in media were compared (Fig. 2) and band assignment is summarized in Table S1. The Raman spectral fingerprinting region used (i.e., 400–1800 cm−1) illustrates the chemical compositions of bacterial cell membrane and cell wall. Several prominent bands were observed from all Listeria species, including the ones at 724, 780, 1003, 1335, 1453, and 1667 cm− 1. The band at 724 cm−1 is mainly attributed to the deformational vibration of adenine (Lu et al., 2011), while the band at 1003 cm− 1 is derived from phenylalanine (the symmetric ring breathing mode) (Maquelin et al., 2002). The band at 780 cm−1 is assigned to cytosine or uracil (nucleic acids) (Oust et al., 2006), and the band at 1335 cm−1 is associated with CH3CH2 vibrational modes (Lu et al., 2011). The bands at 1453 cm−1 and 1667 cm−1 are assigned to C–H deformation and amide I, respectively (Maquelin et al., 2002). In general, different Listeria species showed similar Raman spectral patterns, indicating that a further investigation by chemometric analysis was necessary.

3.3. Chemometric analysis of L. monocytogenes in media and milk 3.3.1. Identification of L. monocytogenes grown in media To investigate the applicability of using Raman spectroscopy to identify and discriminate Listeria on the species level, unsupervised PCA and supervised DA were performed. Using PCA, six closely related Listeria species were successfully separated from each other (Fig. 3). L. monocytogenes was clearly segregated from the other five Listeria species, indicating variations in cell chemical compositions. To validate PCA modeling, supervised DDA and naïve Bayesian classifier were applied for further analysis of Raman spectral features of six different Listeria species. Calibration datasets were developed by combining Raman spectra obtained from three independent experiments, with a total of 30 spectra for each bacterial isolate. To validate the developed supervised chemometric models (i.e., DDA and naïve Bayesian classifier), spectra collected from another experiment were used for testing, and predicted data projection was conducted. Table 1 summarizes DDA and naïve Bayesian classifier results in terms of percentage of correct classification, respectively. In brief, DDA and naïve Bayesian classifier had 97.78% and 98.33% average classification accuracy, respectively, indicating a good reliability of using Raman spectroscopy for bacterial speciation.

Fig. 1. PCA scores plot of the nine species of selected bacteria (n = 15). PCA models were constructed using spectra in the wavenumber of 400–1800 cm−1 using Matlab. In this model, Sent: Salmonella enterica, Saur: Staphylococcus aureus, Ecol: Escherichia coli, LM: Listeria monocytogenes, LG: Listeria grayi, LIN: Listeria innocua, LSE: Listeria seeligeri, LIV: Listeria ivanovii, and LWE: Listeria welshimeri.

J. Wang et al. / International Journal of Food Microbiology 204 (2015) 66–74

69

Fig. 2. Mean and standard deviation of Raman spectra of Listeria spp. grown in media: Listeria innocua (A), Listeria monocytogenes (B), Listeria ivanovii (C), Listeria grayi (D), Listeria seeligeri (E), and Listeria welshimeri (F) (n = 40).

Fig. 3. PCA scores plot of six Listeria species in media (n = 20). PCA models were constructed using spectra in the wavenumber of 400–1800 cm−1 using Matlab. In this model, LM: Listeria monocytogenes, LG: Listeria grayi, LIN: Listeria innocua, LSE: Listeria seeligeri, LIV: Listeria ivanovii, and LWE: Listeria welshimeri.

70

J. Wang et al. / International Journal of Food Microbiology 204 (2015) 66–74

Table 1 Classification accuracy (CA) for Raman spectra of six Listeria species determined by distance discriminant analysis (DDA) (A) and naïve Bayesian classifier (B): database “media”. Calibration datasets were developed by combining spectral data from three independent experiments (A, B and C), resulting in a total of 30 spectra for each bacterial strain. The spectral data obtained from another experiment D was used as prediction dataset. LM: Listeria monocytogenes, LG: Listeria grayi, LIN: Listeria innocua, LSE: Listeria seeligeri, LIV: Listeria ivanovii, and LWE: Listeria welshimeri. (A) Classification result of DDA Predicted

LIN LM LIV LGR LSE LWE

Calibration dataset (n = 180)

Prediction dataset (n = 60)

LIN

LM

LIV

LGR

LSE

LWE

CA (%)

LIN

LM

LIV

LGR

LSE

LWE

CA (%)

28 0 0 0 0 0

0 30 0 0 0 0

0 0 30 0 0 0

0 0 0 30 0 0

1 0 0 0 30 0

1 0 0 0 0 30

93.3 100 100 100 100 100

8 0 0 0 0 0

0 10 0 0 0 0

0 0 10 0 0 0

1 0 0 10 0 0

1 0 0 0 10 0

0 0 0 0 0 10

80 100 100 100 100 100

Overall classification accuracy Average classification accuracy (B)

98.89 97.78

96.67

Classification result of Naïve Bayes classifier Predicted

LIN LM LIV LGR LSE LWE

Calibration dataset (n = 180)

Prediction dataset (n = 60)

LIN

LM

LIV

LGR

LSE

LWE

CA (%)

LIN

LM

LIV

LGR

LSE

LWE

CA (%)

30 0 0 0 0 0

0 30 0 0 0 0

0 0 30 0 0 0

0 0 0 30 0 0

0 0 0 0 30 0

0 0 0 0 0 30

100 100 100 100 100 100

8 0 0 0 0 0

0 10 0 0 0 0

0 0 10 0 0 0

0 0 0 10 0 0

0 0 0 0 10 0

2 0 0 0 0 10

80 100 100 100 100 100

Overall classification accuracy Average classification accuracy

100 98.33

Further validation was performed using SIMCA and the results of percentage of correct classification were calculated and summarized in Table 2. SIMCA is a non-linear multivariate classification method based on PCA model and the classifier in the SIMCA model identifies spectra as members of multiple groups (De Maesschalck et al., 1999). An average classification accuracy of 93.06% was achieved by using SIMCA for the determination of six different Listeria species. Therefore, all the supervised chemometric models (i.e., DDA, naïve Bayesian classifier and SIMCA) demonstrated a high recognition rate. 3.3.2. Identification of L. monocytogenes grown in milk As the results of identifying and discriminating L. monocytogenes grown in media were promising, we applied this approach for further determination of L. monocytogenes in real food samples. Milk samples were spiked with Listeria species and analyzed after 24 h storage. The resultant biomass was used to collect Raman spectra. A total of 240 Raman

96.67

spectra were collected and divided into calibration (180 spectra) and prediction (60 spectra) datasets. Unsupervised PCA model shows clear segregation of 6 different Listeria species recovered from milk samples (Fig. 4a). This result is in agreement with the HCA model shown in Fig. 4b. The dissimilar properties among six different Listeria species can be easily discerned. This HCA method is based upon the principle of analysis of variance with the Ward linkage method and Euclidean distance measurement, resulting in individual cluster of parallel bacterial samples. Supervised chemometric analysis was further conducted and classification accuracy results are summarized in Table 3. Most spectra (175/180) in the calibration dataset of DDA model were correctly classified, resulting in an accuracy of 97.22%. Only 4 out of 60 spectra in the prediction dataset were wrongly classified (classification accuracy of 93.33%). Incorrectly classified L. monocytogenes were assigned to L. ivanovii and this may be due to the similarity in the chemical compositions of bacterial

Table 2 Classification accuracy (CA) for Raman spectra of six Listeria species in media determined by SIMCA. Calibration datasets were developed by combining spectral data from three independent experiments (A, B and C), resulting in a total of 30 spectral data for each bacteria strain. The spectral data obtained from another experiment D was used as prediction dataset. LM: Listeria monocytogenes, LG: Listeria grayi, LIN: Listeria innocua, LSE: Listeria seeligeri, LIV: Listeria ivanovii, and LWE: Listeria welshimeri. Classification result of SIMCA Predicted

LIN LM LIV LGR LSE LWE

Calibration dataset (n = 180)

Prediction dataset (n = 60)

LIN

LM

LIV

LGR

LSE

LWE

CA (%)

LIN

LM

LIV

LGR

LSE

LWE

CA (%)

28 0 1 0 0 0

0 25 0 0 0 0

0 0 29 1 0 0

0 0 0 29 0 0

0 0 0 0 23 0

2 5 0 0 7 30

93.33 83.33 96.67 96.67 76.67 100

9 0 0 0 0 0

0 10 0 0 0 0

0 0 10 0 0 0

0 0 0 10 0 0

0 0 0 0 8 0

1 0 0 0 2 10

90 100 100 100 80 100

Overall classification accuracy Average classification accuracy

91.11 93.06

95.00

J. Wang et al. / International Journal of Food Microbiology 204 (2015) 66–74

71

Fig. 4. (a) PCA scores plot of six Listeria species in milk (n = 10). (b) The dendrogram obtained using HCA (Ward linkage method) based on the Raman spectra of six Listeria species in milk (n = 10). Both PCA and HCA models were constructed using spectra in the wavenumber of 400–1800 cm−1 using Matlab. In this model, LM: Listeria monocytogenes, LG: Listeria grayi, LIN: Listeria innocua, LSE: Listeria seeligeri, LIV: Listeria ivanovii, and LWE: Listeria welshimeri.

cell membrane and cell wall (Day and Basavanna, 2015; Allerberger, 2003). Taken together, the average classification accuracy of DDA model was 95.28% (Table 3A). For naïve Bayesian classifier, most Raman spectral data (178/180) in the calibration dataset were classified correctly (classification accuracy of 98.89%). The average classification accuracy for the prediction dataset was 93.33% (56/60). Again, the incorrectly classified L. monocytogenes were assigned to L. ivanovii. Taken together, the average classification accuracy of naïve Bayesian classifier was 96.11% (Table 3B), which was slightly higher than DDA. We also applied SIMCA for classification of the samples and the results of percentage of correct classification were summarized in Table 4. An average classification accuracy of 96.67% was achieved by using SIMCA for the determination of six different Listeria species in milk. Incorrectly classified L. monocytogenes (4/10) were assigned to L. ivanovii as well. In sum, both unsupervised and supervised chemometric analysis confirmed the feasibility to apply confocal micro-Raman spectroscopy to identify and differentiate L. monocytogenes in milk.

4. Discussion L. monocytogenes is one of the major leading foodborne pathogens that can survive and grow in a variety of food matrices, such as fresh produce, seafood, meat and dairy products. Dairy products can serve as a major vehicle for its dissemination to consumers. Outbreaks of listeriosis and gastroenteritis due to L. monocytogenes contamination in milk have been reported (Dalton et al., 1997; Fleming et al., 1985). Discrimination of the pathogen L. monocytogenes from other Listeria species is important. Though, detection of other Listeria species can serve as indicator organisms for L. monocytogenes contamination in food processing environments (Tortorello, 2003). Routine culture-based laboratory testing of L. monocytogenes in foods is time consuming, laborious and sometimes biochemical tests employed are unreliable (Jadhav et al., 2012). Molecular and immunological testing techniques have been developed, but most of these techniques can only detect genus Listeria or only target L. monocytogenes. Moreover, these techniques may provide a false negative testing result

72

J. Wang et al. / International Journal of Food Microbiology 204 (2015) 66–74

Table 3 Classification accuracy (CA) for Raman spectra of six Listeria species determined by distance discriminant analysis (DDA) (A) and naïve Bayesian classifier (B): database “milk”. Calibration datasets were developed by combining spectral data from three independent experiments (A, B and C), resulting in a total of 30 spectral data for each bacterial strain. The spectral data obtained from another experiment D was used as prediction dataset. LM: Listeria monocytogenes, LG: Listeria grayi, LIN: Listeria innocua, LSE: Listeria seeligeri, LIV: Listeria ivanovii, and LWE: Listeria welshimeri. (A) Classification result of DDA Predicted

LIN LM LIV LGR LSE LWE

Calibration dataset (n = 180)

Prediction dataset (n = 60)

LIN

LM

LIV

LGR

LSE

LWE

CA (%)

LIN

LM

LIV

LGR

LSE

LWE

CA (%)

29 0 0 0 0 0

0 28 1 0 0 0

1 2 29 1 0 0

0 0 0 29 0 0

0 0 0 0 30 0

0 0 0 0 0 30

96.66 93.33 96.66 96.66 100 100

8 0 0 0 0 0

0 9 0 0 0 0

1 1 9 0 0 0

1 0 1 10 0 0

0 0 0 0 10 0

0 0 0 0 0 10

80 90 90 100 100 100

Overall classification accuracy Average classification accuracy (B)

97.22 95.28

93.33

Classification result of Naïve Bayes classifier Predicted

LIN LM LIV LGR LSE LWE

Calibration dataset (n = 180)

Prediction dataset (n = 60)

LIN

LM

LIV

LGR

LSE

LWE

CA (%)

LIN

LM

LIV

LGR

LSE

LWE

CA (%)

30 0 0 0 0 0

0 29 1 0 0 0

0 1 29 0 0 0

0 0 0 30 0 0

0 0 0 0 30 0

0 0 0 0 0 30

100 96.66 96.66 100 100 100

9 0 0 0 0 0

0 9 0 0 0 0

0 1 10 0 0 0

1 0 0 10 0 0

0 0 0 0 10 2

0 0 0 0 0 8

90 90 100 100 100 80

Overall classification accuracy Average classification accuracy

98.89 96.11

due to the interference from food matrices (Gasanov et al., 2005). A Listeria microarray has the potential for laboratory routine analysis, but it is expensive and requires highly trained personnel (Volokhov et al., 2002). As a rapid and non-destructive tool, Raman spectroscopy has been widely applied for the identification and speciation of different foodborne pathogens, such as Campylobacter (Lu et al., 2012) and Listeria (Oust et al., 2006). A major limitation to these studies was that the target bacteria were required to be purified before Raman spectroscopic detection because food components could contribute to spectral features of bacteria, leading to false analysis results (Lu et al., 2011). With simple sample preparation and confocal techniques integrated into Raman spectroscopic platform, Meisel and coworkers validated that confocal microRaman spectroscopy coupled with multivariate analysis could directly identify and speciate Brucella spp. in milk on the basis of single cell level (Meisel et al., 2012). In our preliminary study, a further adjustment of the pinhole size in the Raman spectrometer can further improve the detection sensitivity of bacterial cells (data not shown).

93.33

In the current study, two different databases were created for the identification of L. monocytogenes in media and milk, respectively. Bacteria may be at different physiological states on solid media and this could lead to heterogeneity (e.g., different ages and sizes) of the microcolonies, resulting in low reproducibility of Raman spectra (Choo-Smith et al., 2001). Therefore, we used liquid media rather than solid media for the cultivation of bacteria before Raman spectroscopic characterization. For experiments in milk, samples were spiked with 108 CFU/ml of L. monocytogenes in this study. Thus, no further isolation step was necessary because the bacterial biomass was concentrated. After simple sample preparation, bacterial samples were resuspended into distilled water and allowed for cell measurement using confocal micro-Raman spectroscopy. This high starting concentration was selected to generate high intensity of Raman spectra for the construction of “milk” database and advanced chemometric analyses. This is presumably orders of magnitude higher than the detection limit of this Raman spectroscopic method. Future study is aimed to reduce the initial inoculation level and cultivation time for method optimization.

Table 4 Classification accuracy (CA) for Raman spectra of six Listeria species in milk determined by SIMCA. Calibration datasets were developed by combining spectral data from three independent experiments (A, B and C), resulting in a total of 30 spectral data for each bacteria strain. The spectral data obtained from another experiment D was used as prediction dataset. LM: Listeria monocytogenes, LG: Listeria grayi, LIN: Listeria innocua, LSE: Listeria seeligeri, LIV: Listeria ivanovii, and LWE: Listeria welshimeri. Classification result of SIMCA Predicted

LIN LM LIV LGR LSE LWE

Calibration dataset (n = 180)

Prediction dataset (n = 60)

LIN

LM

LIV

LGR

LSE

LWE

CA (%)

LIN

LM

LIV

LGR

LSE

LWE

CA (%)

30 0 0 0 0 0

0 30 0 0 0 0

0 0 30 0 0 0

0 0 0 30 0 0

0 0 0 0 30 0

0 0 0 0 0 30

100 100 100 100 100 100

10 0 0 0 0 0

0 10 4 0 0 0

0 0 6 0 0 0

0 0 0 10 0 0

0 0 0 0 10 0

0 0 0 0 0 10

100 100 60 100 100 100

Overall classification accuracy Average classification accuracy

100 96.67

93.33

J. Wang et al. / International Journal of Food Microbiology 204 (2015) 66–74

Unsupervised PCA and HCA achieved a clear segregation of all the six Listeria species grown in both media and milk (Figs. 3 and 4). Supervised DDA and naïve Bayesian classifier models were then applied to analyze the spectral data from two independent databases (i.e., media and milk) and validated the segregation results from unsupervised chemometric analyses. A high correct identification rate of L. monocytogenes was obtained in both media (N93%) (Tables 1 and 2) and milk models (N 95%) (Tables 3 and 4) by DA-based classification methods and SIMCA. In addition, both the calibration model and the prediction model had a similarly high correct identification rate, indicating the robustness of the constructed models for identification of L. monocytogenes. In conclusion, confocal micro-Raman spectroscopy and chemometric analyses could identify L. monocytogenes in both media and milk in a rapid and reliable manner. Sample preparation is very simple and the analysis could be finished within a few hours depending upon the size of the constructed databases. Supplementary data to this article can be found online at http://dx. doi.org/10.1016/j.ijfoodmicro.2015.03.021. Acknowledgments This work was supported by funds awarded to J.W. by the Ministry of Science and Technology of the People's Republic of China (Project No. 2012AA101604), funds awarded to X.L. by the Natural Sciences and Engineering Research Council of Canada (NSERC RGPIN-2014-05487), and funds awarded to S.W. by the National Science Foundation for Distinguished Young Scholars of China (Project No. 31225021) and the Program for Changjiang Scholars and Innovative Research Team in University (Project No. IRT1166). References Allerberger, F., 2003. Listeria: growth, phenotypic differentiation and molecular microbiology. FEMS Immunol. Med. Microbiol. 35, 183–189. Barbuddhe, S.B., Maier, T., Schwarz, G., Kostrzewa, M., Hof, H., Domann, E., Chakraborty, T., Hain, T., 2008. Rapid identification and typing of Listeria species by matrix-assisted laser desorption ionization-time of flight mass spectrometry. Appl. Environ. Microbiol. 74 (17), 5402–5407. Barton Behravesh, C., Jones, T.F., Vugia, D.J., Long, C., Marcus, R., Smith, K., Thomas, S., Zansky, S., Fullerton, K.E., Henao, O.L., Scallan, E., FoodNet Working Group, 2011. Deaths associated with bacterial pathogens transmitted commonly through food: foodborne diseases active surveillance network (FoodNet), 1996–2005. J. Infect. Dis. 204 (2), 263–267. Bickley, J., Short, J.K., McDowell, D.G., Parkes, H.C., 1996. Polymerase chain reaction (PCR) detection of Listeria monocytogenes in diluted milk and reversal of PCR inhibition caused by calcium ions. Lett. Appl. Microbiol. 22 (2), 153–158. Bocklitz, T., Walter, A., Hartmann, K., Rösch, P., Popp, J., 2011. How to pre-process Raman spectra for reliable and stable models? Anal. Chim. Acta 704 (1), 47–56. Bosch, A., Miñán, A., Vescina, C., Degrossi, J., Gatti, B., Montanaro, P., Messina, M., Franco, M., Vay, C., Schmitt, J., Naumann, D., Yantorno, O., 2008. Fourier transform infrared spectroscopy for rapid identification of nonfermenting gram-negative bacteria isolated from sputum samples from cystic fibrosis patients. J. Clin. Microbiol. 46 (8), 2535–2546. Büla, C.J., Bille, J., Glauser, M.P., 1995. An epidemic of food-borne listeriosis in western Switzerland: description of 57 cases involving adults. Clin. Infect. Dis. 20 (1), 66–72. Cho, A.-R., Dong, H.-J., Seo, K.-H., Cho, S., 2014. Development of a loop-mediated isothermal amplification assay for detecting Listeria monocytogenes prfA in milk. Food Sci. Biotechnol. 23 (2), 467–474. Choo-Smith, L.-P., Maquelin, K., Van Vreeswijk, T., Bruining, H., Puppels, G., Ngo Thi, N.A., Kirschner, C., Naumann, D., Ami, D., Villa, A.M., Orsini, F., Doglia, S.M., Lamfarraj, H., Sockalingum, G.D., Manfait, M., Allouch, P., Endtz, H.P., 2001. Investigating microbial (micro) colony heterogeneity by vibrational spectroscopy. Appl. Environ. Microbiol. 67 (4), 1461–1469. Dalton, C.B., Austin, C.C., Sobel, J., Hayes, P.S., Bibb, W.F., Graves, L.M., Swaminathan, B., Proctor, M.E., Griffin, P.M., 1997. An outbreak of gastroenteritis and fever due to Listeria monocytogenes in milk. N. Engl. J. Med. 336 (2), 100–105. Davis, R., Mauer, L.J., 2011. Subtyping of Listeria monocytogenes at the haplotype level by Fourier transform infrared (FT-IR) spectroscopy and multivariate statistical analysis. Int. J. Food Microbiol. 150 (2–3), 140–149. Day, J.B., Basavanna, U., 2015. Magnetic bead based immune-detection of Listeria monocytogenes and Listeria ivanovii from infant formula and leafy green vegetables using the Bio-Plex suspension array system. Food Microbiol. 46, 564–572. De Maesschalck, R., Candolfi, A., Massart, D., Heuerding, S., 1999. Decision criteria for soft independent modelling of class analogy applied to near infrared data. Chemometr. Intell. Lab. Syst. 47 (1), 65–77. den Bakker, H.C., Warchocki, Wright, E.M., Allred, A.F., Ahlstrom, C., Manuel, C.S., Stasiewicz, M.J., Burrell, A., Roof, S., Strawn, L., Fortes, E.D., Nightingale, K.K.,

73

Kephart, D., Wiedmann, M., 2014. Five new species of Listeria (L. floridensis sp. nov., L. aquatica sp. nov., L. cornellensis sp. nov., L. riparia sp. nov., and L. grandensis sp. nov.) from agricultural and natural environments in the United States. Int. J. Syst. Evol. Microbiol. 64, 1882–1889. Drevets, D.A., Bronze, M.S., 2008. Listeria monocytogenes: epidemiology, human disease, and mechanisms of brain invasion. FEMS Immunol. Med. Microbiol. 53 (2), 151–165. Fleming, D.W., Cochi, S.L., MacDonald, K.L., Brondum, J., Hayes, P.S., Plikaytis, B.D., Holmes, M.B., Audurier, A., Broome, C.V., Reingold, A.L., 1985. Pasteurized milk as a vehicle of infection in an outbreak of listeriosis. N. Engl. J. Med. 312 (7), 404–407. Food and Drug Administration, United States Department of Agriculture, Centers for Disease Control and Prevention, 2003. Quantitative assessment of the relative risk to public health from foodborne Listeria monocytogenes among selected categories of ready-to-eat foods. U.S. Food and Drug Administration, Washington, DC (http://www.foodsafety.gov/~dms/lmr2-toc.html). Fretz, R., Sagel, U., Ruppitsch, W., Pietzka, A.T., Stöger, A., Huhulescu, S., Heuberger, S., Pichler, J., Much, P., Pfaff, G., Stark, K., Prager, R., Flieger, A., Feenstra, O., Allerberger, F., 2010. Listeriosis outbreak caused by acid curd cheese ‘Quargel’, Austria and Germany 2009. Eurosurveillance. 15(5) (19477). Gandhi, M., Chikindas, M.L., 2007. Listeria: a foodborne pathogen that knows how to survive. Int. J. Food Microbiol. 113 (1), 1–15. Gasanov, U., Hughes, D., Hansbro, P.M., 2005. Methods for the isolation and identification of Listeria spp. and Listeria monocytogenes: a review. FEMS Microbiol. Rev. 29 (5), 851–875. Goodacre, R., 2003. Explanatory analysis of spectroscopic data using machine learning of simple, interpretable rules. Vib. Spectrosc. 32 (1), 33–45. Holt, C., Hirst, D., Sutherland, A., MacDonald, F., 1995. Discrimination of species in the genus Listeria by Fourier transform infrared spectroscopy and canonical variate analysis. Appl. Environ. Microbiol. 61 (1), 377–378. Huang, W.E., Griffiths, R.I., Thompson, I.P., Bailey, M.J., Whiteley, A.S., 2004. Raman microscopic analysis of single microbial cells. Anal. Chem. 76 (15), 4452–4458. Hutsebaut, D., Maquelin, K., De Vos, P., Vandenabeele, P., Moens, L., Puppels, G., 2004. Effect of culture conditions on the achievable taxonomic resolution of Raman spectroscopy disclosed by three Bacillus species. Anal. Chem. 76 (21), 6274–6281. Jadhav, S., Bhave, M., Palombo, E.A., 2012. Methods used for the detection and subtyping of Listeria monocytogenes. J. Microbiol. Meth. 88 (3), 327–341. Kirschner, C., Maquelin, K., Pina, P., Thi, N.N., Choo-Smith, L.-P., Sockalingum, G., Sandt, C., Ami, D., Orsini, F., Doglia, S.M., Allouch, P., Mainfait, M., Puppels, G.J., Naumann, D., 2001. Classification and identification of enterococci: a comparative phenotypic, genotypic, and vibrational spectroscopic study. J. Clin. Microbiol. 39 (5), 1763–1770. Koch, J., Dworak, R., Prager, R., Becker, B., Brockmann, S., Wicke, A., Wichmann-Schauer, H., Hof, H., Weber, D., Stark, K., 2010. Large listeriosis outbreak linked to cheese made from pasteurized milk, Germany, 2006–2007. Foodborne Pathog. Dis. 7 (12), 1581–1584. Kuhm, A.E., Suter, D., Felleisen, R., Rau, J., 2009. Identification of Yersinia enterocolitica at the species and subspecies levels by Fourier transform infrared spectroscopy. Appl. Environ. Microbiol. 75 (18), 5809–5813. Lang Halter, E., Neuhaus, K., Scherer, S., 2013. Listeria weihenstephanensis sp. nov., isolated from the water plant Lemna trisulca taken from a freshwater pond. Int. J. Syst. Evol. Microbiol. 63, 641–647. Leclercq, A., Clermont, D., Bizet, C., Grimont, P.A.D., Le Flèche-Matéos, A., Roche, S.M., Buchrieser, C., Cadet-Daniel, V., Le Monnier, A., Lecuit, M., Allerberger, F., 2010. Listeria rocourtiae sp. nov. Int. J. Syst. Evol. Microbiol. 60, 2210–2214. Lieber, C.A., Mahadevan-Jansen, A., 2003. Automated method for subtraction of fluorescence from biological Raman spectra. Appl. Spectrosc. 57, 1363–1367. Linnan, M.J., Mascola, L., Lou, X.D., Goulet, V., May, S., Salminen, C., Hird, D.W., Yonekura, M.L., Hayes, P., Weaver, R., Audurier, A., Plikaytis, B.D., Fannin, S.L., Kieks, A., Broome, M.D., 1988. Epidemic listeriosis associated with Mexican-style cheese. N. Engl. J. Med. 219 (13), 823–828. Loessner, M.J., Rudolf, M., Scherer, S., 1997. Evaluation of luciferase reporter bacteriophage A511:luxAB for detection of Listeria monocytogenes in contaminated foods. Appl. Environ. Microbiol. 63 (8), 2961–2965. Lu, X., Al-Qadiri, H.M., Lin, M., Rasco, B.A., 2011. Application of mid-infrared and Raman spectroscopy to the study of bacteria. Food Bioprocess Technol. 4 (6), 919–935. Lu, X., Huang, Q., Miller, W.G., Aston, D.E., Xu, J., Xue, F., Zhang, H., Rasco, B.A., Wang, S., Konkel, M.E., 2012. Comprehensive detection and discrimination of Campylobacter species by use of confocal micro-Raman spectroscopy and multilocus sequence typing. J. Clin. Microbiol. 50 (9), 2932–2946. MacDonald, P.D., Whitwam, R.E., Boggs, J.D., MacCormack, J.N., Anderson, K.L., Reardon, J.W., Saah, J.R., Graves, L.M., Hunter, S.B., Sobel, J., 2005. Outbreak of listeriosis among Mexican immigrants as a result of consumption of illicitly produced Mexican-style cheese. Clin. Infect. Dis. 40 (5), 677–682. Magliulo, M., Simoni, P., Guardigli, M., Michelini, E., Luciani, M., Lelli, R., Roda, A., 2007. A rapid multiplexed chemiluminescent immunoassay for the detection of Escherichia coli O157: H7, Yersinia enterocolitica, Salmonella typhimurium, and Listeria monocytogenes pathogen bacteria. J. Agric. Food Chem. 55 (13), 4933–4939. Makino, S.-I., Kawamoto, K., Takeshi, K., Okada, Y., Yamaski, M., Yamamoto, S., Igimi, S., 2005. An outbreak of food-borne listeriosis due to cheese in Japan, during 2001. Int. J. Food Microbiol. 104 (2), 189–196. Maquelin, Kees, et al., 2002. Identification of medically relevant microorganisms by vibrational spectroscopy. J. Microbiol. Methods 51 (3), 255–271. Meisel, S., Stöckel, S., Elschner, M., Melzer, F., Rösch, P., Popp, J., 2012. Raman spectroscopy as a potential tool for detection of Brucella spp. in milk. Appl. Environ. Microbiol. 78 (16), 5575–5583. Nogva, H.K., Rudi, K., Naterstad, K., Holck, A., Lillehaug, D., 2000. Application of 5′-nuclease PCR for quantitative detection of Listeria monocytogenes in pure cultures, water, skim milk, and unpasteurized whole milk. Appl. Environ. Microbiol. 66 (10), 4266–4277.

74

J. Wang et al. / International Journal of Food Microbiology 204 (2015) 66–74

Ohk, S.-H., Bhunia, A.K., 2013. Multiplex fiber optic biosensor for detection of Listeria monocytogenes, Escherichia coli O157:H7 and Salmonella enterica from ready-to-eat meat samples. Food Microbiol. 33 (2), 166–171. Oust, A., Møretrø, T., Naterstad, K., Sockalingum, G.D., Adt, I., Manfait, M., Kohler, A., 2006. Fourier transform infrared and Raman spectroscopy for characterization of Listeria monocytogenes strains. Appl. Environ. Microbiol. 72 (1), 228–232. Preisner, O., Guiomar, R., Machado, J., Menezes, J.C., Lopes, J.A., 2010. Application of Fourier transform infrared spectroscopy and chemometrics for differentiation of Salmonella enterica serovar Enteritidis phage types. Appl. Environ. Microbiol. 76 (11), 3538–3544. Rebuffo, C.A., Schmitt, J., Wenning, M., von Stetten, F., Scherer, S., 2006. Reliable and rapid identification of Listeria monocytogenes and Listeria species by artificial neural network-based Fourier transform infrared spectroscopy. Appl. Environ. Microbiol. 72 (2), 994–1000.

Tortorello, M.L., 2003. Indicated organisms for safety and quality-uses and methods for detection: minireview. J. AOAC Int. 86 (6), 1208–1217. Volokhov, D., Rasooly, A., Chumakov, K., Chizhikov, V., 2002. Identification of Listeria species by microarray-based assay. J. Clin. Microbiol. 40 (12), 4720–4728. Weller, D., Andrus, A., Wiedmann, M., den Bakker, H., 2014. Listeria booriae sp. nov., and Listeria newyorkensis sp. nov., from food processing environments in the United States. Int. J. Syst. Evol. Microbiol. http://dx.doi.org/10.1099/ijs.0.0708390 (in press). Willemse-Erix, D.F., Scholtes-Timmerman, M.J., Jachtenberg, J.-W., van Leeuwen, W.B., Horst-Kreft, D., Schut, T.C.B., Deurenberg, R.H., Puppels, G.J., van Belkum, A., Vos, M.C., Maquelin, K., 2009. Optical fingerprinting in bacterial epidemiology: Raman spectroscopy as a real-time typing method. J. Clin. Microbiol. 47 (3), 652–659.