Toward Culture-Free Raman Spectroscopic Identification of

Article pubs.acs.org/ac

Toward Culture-Free Raman Spectroscopic Identification of Pathogens in Ascitic Fluid Sandra Kloß,†,⊥ Petra Rösch,†,⊥ Wolfgang Pfister,‡ Michael Kiehntopf,§ and Jürgen Popp*,†,⊥,∥ †

Institute of Physical Chemistry and Abbe Center of Photonics, University of Jena, Helmholtzweg 4, D-07743 Jena, Germany InfectoGnostics Forschungscampus Jena, Philosphenweg 7, D-07743 Jena, Germany ‡ Institute of Medical Microbiology, Jena University Hospital, Erlanger Allee 101, D-07747 Jena, Germany § Institute of Clinical Chemistry and Laboratory Diagnostics, Jena University Hospital, Erlanger Allee 101, D-07747 Jena, Germany ∥ Leibniz Institute of Photonic Technology e.V., Albert-Einstein-Straße 9, D-07745 Jena, Germany ⊥

S Supporting Information *

ABSTRACT: The identification of pathogens in ascitic fluid is standardly performed by ascitic fluid culture, but this standard procedure often needs several days. Additionally, more than half of the ascitic fluid cultures are negative in case of suspected spontaneous bacterial peritonitis (SBP). It is therefore important to identify and characterize the causing pathogens since not all of them are covered by the empirical antimicrobial therapy. The aim of this study is to show that pathogen identification in ascitic fluid is possible by means of Raman microspectroscopy and chemometrical evaluation with the advantage of strongly increased speed. Therefore, a Raman database containing more than 10000 single-cell Raman spectra of 34 bacterial strains out of 13 different species was built up. The performance of the used statistical model was validated with independent bacterial strains, which were grown in ascitic fluid.

P

combination with high-resolution melt analysis9 and microarrays.10 Raman spectroscopy represents also a conceivable opportunity to identify the pathogens present in ascitic fluid within several hours. Raman spectroscopy is based on the vibrational fingerprint of the chemical composition of a sample, which is, in contrast to other molecular sensitive methods, label- and destruction-free and needs only a minimal sample preparation. With Raman microspectroscopy, the combination of a Raman spectrometer with a microscope, a spatial resolution below one micrometer can be achieved, which allows the investigation of single bacterial cells. The Raman spectrum of a single bacterial cell represents a sum of the Raman spectra of all cell components. Since their composition varies from bacterial species to species, it was demonstrated that with Raman spectroscopy bacterial species can be distinguished with appropriate chemometrical methods based on comprehensive reference databases.11,12 Performing Raman measurements on single-cell level allows investigating single bacterial cells directly after the isolation from complex matrices. Hence, no timeconsuming cultivation steps are needed before investigating and

eritonitis can be divided in two groups: One is the secondary peritonitis, which is associated with a local or generalized peritoneal inflammation because of a disruption of the anatomical barrier.1 In contrast, spontaneous bacterial peritonitis (SBP), which represents a primary peritonitis, is a common complication in patients with decompensated liver cirrhosis and ascites, which is caused by bacterial translocation through the intestinal barrier without any physical disruption.2 Last time, a drift of the detected pathogens in SBP toward a higher abundance of Gram positive Enterococcus species was reported, which are not covered by the empirical first-line therapy with third generation cephalosporines.2−4 An early identification of pathogens, particularly in patients where the first-line therapy fails, is very important, since this failure is associated with an increased mortality.5 Identification of pathogens in ascitic fluid is typically done by ascitic fluid culture after a paracentesis. However, up to 65% of all ascitic fluid cultures in case of SBP are negative.6 Also, 20−25% of cultures are negative in cases of secondary peritonitis.1 In addition to problems caused by negative cultures, the identification of pathogens in positive ones is time-consuming and takes at least 1 day. Alternatives to the classical culture methods were reported. These consist of polymerase chain reaction (PCR)-based detection of pathogens either in ascitic fluid6−8 or in © 2014 American Chemical Society

Received: August 21, 2014 Accepted: December 13, 2014 Published: December 17, 2014 937

DOI: 10.1021/ac503373r Anal. Chem. 2015, 87, 937−943

Article

Analytical Chemistry

Figure 1. Median Raman spectra of the investigated Gram negative (A) and Gram positive (B) species (black), the area between the 25−75th percentile spectra (white), area which contains 95% of all spectral values (gray) of investigated species: (a) P. aeruginosa, (b) E. coli, (c) K. pneumoniae, (d) Enterob. aerogenes, (e) Enterob. cloacae, (f) Enteroc. faecalis, (g) Enteroc. faecium, (h) Staph. epidermidis, (i) Staph. saprophyticus, (j) Staph. aureus, (k) Strep. agalactiae, (l) Strep. bovis, and (m) Strep. pneumoniae.

identifying the species once a database has been set up.13,14 Moreover, identification of the species of antibiotically pretreated bacterial cells with an appropriate database of Raman spectra could be demonstrated.15,16 Account must, however, be taken of the fact that single-cell Raman spectra not only differ between different bacterial species but also between different culture conditions.17−19 Therefore, the culture conditions for the data of the reference database should be as similar as possible to the conditions, which are expected for the samples to be analyzed later. In this context it could be shown that it is in principle possible to identify the main containing pathogens in patient urine samples by means of Raman spectroscopy and application of an appropriate database.15 In the present study, the Raman-spectroscopic identification of different pathogens in ascitic fluid was demonstrated for the first time. To that end a reference database of bacteria cultured in ascitic fluid was implemented which allows a chemometric identification of bacteria based on a support vector machine algorithm. To construct a robust model, which could also be used for the identification of bacteria in patient samples of further studies, for most of the bacterial species different strains were included in the database.

the number of measured spectra is given in Table S-1 in the Supporting Information. All bacterial strains were provided by the Institute of Medical Microbiology, University Jena, and were originally purchased from the German Collection of Microorganisms and Cell Cultures (DSM), American Type Culture Collection (ATCC) or were isolates from patient samples. The bacteria were cultured on Columbia blood agar plates (CBA, Oxoid) for 24 h at 37 °C. One colony of these CBA cultures were used to inoculate 1 mL sterile filtered ascitic fluid (syringe filter 0.8/0.2 μm-pore size, Pall Corporation, New York, USA) and incubated for 24 h at 37 °C. The ascitic fluid culture was centrifuged for 5 min at 10000 × g (Eppendorf MiniSpin plus, Eppendorf AG, Hamburg, Germany) and the supernatant was discarded. The resulting pellet was resuspended in 1 mL autoclaved deionized water and centrifuged for 5 min at 10000 × g. This washing step was done twice. Afterward the pellet was suspended in 1 mL autoclaved deionized water and 10 μL of this suspension was spread on a nickel foil in small droplets and allowed to dry at room temperature. For each strain used for the database at least three independent batches were prepared. The independent ascitic fluid pools consisted of independent pooled residual ascitic fluid samples from clinical diagnostics. The independent test strains were directly cultured in sterile filtered residual ascitic fluid samples. Spectroscopic Instrumentation. All Raman spectroscopic measurements were performed with the Raman microscope BioParticleExplorer (MicrobioID 0.5, RapID, Berlin, Germany). A frequency-doubled solid-state Nd:YAG diode pumped laser (LCM-S-111, Laser-Export co. Ltd., Moscow, Russia) at 532 nm was used for excitation. The laser beam was focused with a 100× magnification objective (MPLFLN 100x, NA: 0.9, Olympus Corporation, Tokyo, Japan) on the sample with a maximal laser power of approximately 14 mW. To prevent burning effects, different gray filters were used to reduce the laser power on the sample to 4−11 mW. The backscattered light was focused on a single stage monochromator (HE 532, Horiba Jobin Yvon, Munich, Germany)

■

MATERIALS AND METHODS The study was approved by the ethics committee of the Jena University Hospital (Germany). Bacterial Culture. For the study 34 bacterial strains out of the 13 different species, namely, Escherichia coli (E. coli), Klebsiella pneumoniae (K. pneumoniae), Pseudomonas aeruginosa (P. aeruginosa), Enterobacter aerogenes (Enterob. aerogenes), Enterobacter cloacae (Enterob. cloacae), Enterococcus faecalis (Enteroc. faecalis), Enterococcus faecium (Enteroc. faecium), Staphylococcus epidermidis (Staph. epidermidis), Staphylococcus saprophyticus (Staph. saprophyticus), Staphylococcus aureus (Staph. aureus), Streptococcus agalactiae (Strep. agalactiae), Streptococcus bovis (Strep. bovis), and Streptococcus pneumoniae (Strep. pneumoniae) were used. An overview of all strains and 938

DOI: 10.1021/ac503373r Anal. Chem. 2015, 87, 937−943

Article

Analytical Chemistry

Figure 2. 3-Level SVM classification model of single bacterial cells cultured in two independent batches of ascitic fluid (the numbers in the circles represent the accuracies of the SVMs achieved by 10-fold cross validation).

function from the R package kernlab.28 This model was afterward used to predict the independent test data set. Median Raman spectra and percentile spectra were calculated using the preprocessed, vector-normalized single-cell spectra of all classification batches of one species.

which is equipped with an 920 lines/mm grating and collected with a thermoelectrically cooled CCD camera (DV401A-BV, Andor Technology, Belfast, Northern Ireland). The spectral resolution was about 10 cm−1. For each bacterial cell two consecutive Raman spectra were measured, which were afterward combined for spike removal. Integration times between 10 and 20 s for a single bacterial cell, within a wavenumber range of 3319−70 cm−1 were chosen. Data Preprocessing and Chemometrical Analysis. All preprocessing was done using the open source software Gnu R20 with in-house developed scripts.21 The first step consisted in removing the spectral background using a method based on the SNIP clipping algorithm.22 For this, we used a fourth order clipping filter. Since two consecutive spectra of one cell were always measured with the BioParticleExplorer, spikes could afterward be eliminated by a robust variant of the upper-bound spectrum algorithm.23 The single-cell Raman spectra were then wavenumber calibrated by using an acetaminophen Raman spectrum measured at the same day as reference.24,25 The spectra were truncated to wavenumbers between 3100 and 2650 cm−1 and 1750−450 cm−1, which were used for the chemometrical analysis. Finally, all spectra were vectornormalized. As has been shown previously, support vector machines (SVMs) are especially suited to classify and identify different bacterial species.13,26 SVMs belong to the group of maximum margin classifiers and find the optimal solution for given parameters efficiently. A classification model was built using the all-pairs approach for SVMs with a radial basis kernel.27 The parameter sigma was optimized using the sigest

■

RESULTS AND DISCUSSION For the built up of a reference database, the most prevalent Gram negative species in ascitic fluid, namely E. coli, K. pneumoniae, P. aeruginosa, Enterob. aerogenes, and Enterob. cloacae, were used.4,29 Because of the rising prevalence of Gram positive species in SBP also Enteroc. faecalis, Enertoc. faecium, Staph. epidermidis, Staph. saprophyticus, Staph. aureus, Strep. agalactiae, Strep. bovis, and Strep. pneumoniae were implemented.1,3 For most of the species different strains were used to enlarge the variability of the database (see Supporting Information Table S-1). The median Raman spectra of the 13 species are depicted in Figure 1 (black lines). By the bare eye a separation of the Raman spectra of the different species is not possible, since the differences are very small. The spectra are dominated by the CH3-/CH2-stretching vibrations in the wavenumber region between 3100 and 2800 cm−1. Characteristic bands for proteins, DNA/RNA, lipids and polysaccharides can be found in the fingerprint region between 1750 and 500 cm−1. A tentative band assignment can be found in the Supporting Information in Table S-2. Besides the interspecies variations also intraspecies variations (white area between the 25th and the 75th percentile spectra and gray area, which contains 95% of all spectral values) are visible in Figure 1. The highest 939

DOI: 10.1021/ac503373r Anal. Chem. 2015, 87, 937−943

Article

Analytical Chemistry

Table 1. Results of the 3-Level SVM Classification Model and the Validation of the Model with Bacteria Cultured in an Independent Ascitic Fluid Batch classification model first step top-level

Gram negative Gram positive

TPa/all

sens.a [%]

2491/2517 4195/4240

99.2 98.8

validation with independent batch accu.a [%]

TPa/all

sens.a [%]

1286/1341 2124/2134

97.6 98.8

98.9 second step genus-level Gram negative

P. aeruginosa Enterobacteriacea

528/559 1942/1958

94.5 99.2

Staphylococcus Streptococcus Enterococcus

1525/1532 1467/1478 1293/1230

99.5 99.3 97.0

98.1 280/328 996/1013

85.4 98.3

796/798 652/728 587/608

99.8 89.6 96.5

98.3 Gram positive

95.2

98.7 third step species-level Staphylococcus

Staph. epidermidis Staph. saprophyticus Staph. aureus

558/587 336/368 554/577

95.1 91.3 96.0

Streptococcus

Strep. agalactiae Strep. bovis Strep. pneumoniae

576/581 194/200 694/697

99.1 97.0 99.6

Enterococcus

Enteroc. faecalis Enteroc. faecium

633/638 563/592

99.2 95.1

95.4 196/282 163/205 230/311

69.5 79.5 74.0

222/323 64/100 303/305

68.7 64.0 99.3

247/300 294/308

82.3 95.5

94.5

73.8

99.1

97.2 a

accu.a [%]

80.9

89.0

TP: true positive spectra. sens.: sensitivity. accu.: accuracy.

variations within a single bacterial species were found for Staph. saprophyticus in the wavenumber region between 1180 and 1090 cm−1 which could be assigned to vibrations in proteins and for Staph. epidermidis in the region around the band at 781 cm−1 corresponding to vibrations in DNA and RNA. For an assignment of single-cell bacterial Raman spectra to a bacterial species, it is necessary to train a statistical model in a first step. Therefore, the spectra of two independent batches (different ascitic fluid pools for culture) of each bacterial strain (Supporting Information Table S-1) were used to train the model which is depicted in Figure 2. Six different SVMs were trained to build up a robust 3-level statistical model. The performance of the model was evaluated with a 10-fold cross validation. On the first level a SVM model was trained to distinguish between spectra of Gram negative and Gram positive bacteria. 6686 out of 6757 spectra (98.9%) could be assigned correctly with this model. In a second step, two SVMs were trained; one SVM model distinguishes between spectra from the Gram negative Enterobacteriacea and P. aeruginosa, whereby 95.2% of the spectra (2470 out of 2517) were assigned correctly. The other classifier on the second level was implemented to differentiate spectra from the Gram positive genera Staphylococcus, Streptococcus, and Enterococcus. With this model 4185 out of 4240 spectra were classified correctly (accuracy of 98.7%). To determine the species of the Gram positive bacteria three further SVMs were used on the third level (one for each genus). Hereby accuracies between 94.5% (Staphylococcus) and 99.1% (Streptococcus) were calculated via a 10-fold cross validation. A further discrimination of the Enterobacteriacea was not possible with this model. However, the differentiation between the nonfermenting genus Pseudomonas and fermenting Enterobacteriacea is important because

each of the two classes could be treated with different antibiotics. To validate the performance of the 3-step classification model, Raman spectra of independently cultured batches (in a different ascitic fluid pool) of each bacterial strain were queried against the classifiers. The results of this validation are summarized in Table 1. On the top-level 3410 out of 3475 spectra could be identified correctly (98.1%), which is well in the range of the classification model with 98.9%. An assignment of spectra to the classes Enterobacteriacea and P. aeruginosa was done with an accuracy of 95.2% in comparison to 98.3% in the classification model. For the genus differentiation of the Gram positive bacteria also an accuracy of 95.4% could be reached, whereby the sensitivities for the single genera ranged from 89.6% for Streptococcus to 99.8% for Staphylococcus. Although the best identification on genus-level was achieved for Staphylococcus, identification of the three representative species was achieved with a somewhat lower accuracy of 73.8%, with sensitivities ranging from 69.5% for Staph. epidermidis to 79.5% for Staph. saprophyticus. 80.9% of the spectra from the Streptococcus species were correctly identified on the specieslevel. The best results could be obtained for the Enterococcus species, where 89.0% of the spectra were assigned to the correct species. After the successful validation of the 3-level classification model with an independent batch of the bacterial strains implemented in the database, we decided to combine the data set of the classification model (2 batches per strain) and the independent validation data set (1 batch per strain) to train the 3-level SVM model once again. By combining the two data sets, also the spectral variations between the data sets, especially between the different pools of the ascitic fluids, are considered 940

DOI: 10.1021/ac503373r Anal. Chem. 2015, 87, 937−943

Article

Analytical Chemistry

Figure 3. Identification of unknown bacterial samples on genus-level (A) and on species-level (B).

samples, which shall be evaluated in further studies, are different to those which are used to construct the database. To test the performance of our model we prepared 13 independent test samples with bacterial strains not included in the model built before. Therefore, 12 different strains, ten of them patient isolates, were cultured in different residual patients’ ascitic fluid samples (from patients different to those used for the train data set) and isolated as described before, to simulate potential patients’ ascitic fluid samples. One strain was cultured in two different ascitic fluid samples to test the variability in the spectra induced by different ascitic fluid samples for culture. The numbering of these samples is summarized in Supporting Information Table S-1. For each sample approximately 100 Raman spectra of single bacterial cells were measured, preprocessed and afterward evaluated with the combined 3-level model. The bar graphs of the assignment on the genus-level and the species-level are depicted in Figure 3 A and B (the data are also summarized in Supporting Information Table S-3).

and could be taken into account for the evaluation of independent bacterial strains later. The combined model contains the information on more than 10000 individual Raman spectra of single bacterial cells and as a consequence information about possible variations between single cell Raman spectra of different bacterial species, between different strains of the same species, between different batches of the same bacterial strain and also individual variations within one batch of a single bacterial strain. The results of the 10-fold cross validation of the combined model are depicted in Supporting Information Figure S-1. The accuracies of the six different SVMs are very similar to those of the model with only two batches of each bacterial strain, although more variations are included. For further validation of the reference database, it is necessary that not only bacterial strains, which are already implemented, can be identified with the constructed model. Also the evaluation of unknown, not referenced, bacterial strains is very important since the bacterial strains in patient 941

DOI: 10.1021/ac503373r Anal. Chem. 2015, 87, 937−943

Article

Analytical Chemistry All spectra of sample 1 (E. coli) were assigned to the class Gram negative on the first level and to Enterobacteriacea on the second level. For sample 2 (K. pneumoniae) 98 out of 109 spectra (89.9%) were labeled as Gram negative on the first level. All of these 98 spectra were then correctly assigned to Enterobacteriacea on the second level. The 11 spectra, which were falsely labeled at the first level, were assigned to Streptococcus (5.8%) and Enterococcus (4.6%) on the genus level, and to Strep. agalactiae (5 spectra), Strep. pneumoniae (1 spectrum) and Enteroc. faecalis (5 spectra) on the species-level. Spectra of sample 3 (Enterob. cloacae) were predominantly labeled as Gram negative in the first step (105 out of 109 spectra) and as Enterobacteriacea in the second step (94.5%). All spectra of sample 4 were properly identified as being Gram negative and P. aeruginosa. To test the variability induced by different ascitic fluid samples as culture media, we applied two different ones to culture the Enteroc. faecalis strain Efca3 (5 and 6). For sample 5, 99.0% of the spectra were assigned correctly on the Gram level, 94.3% on the genus level and 92.4% on the species level in comparison to 100% of the spectra which were properly identified on all three levels for sample 6. Here the influence of the culture medium on the identification is visible, but the correct assignment of 92.4% of the spectra on the species level is still quite good. The tested Enteroc. faecium strain (7) was categorized as being Gram positive (100%) on the first, Enterococcus (89 out of 98 spectra) on the second and Enteroc. faecium (87.8%) on the third level. All spectra from the samples containing the different Staphylococcus species (8, 9, and 10) lead to the right assignment to Gram positive and Staphylococcus. On the species-level, 94.2% of the spectra from sample 10 (Staph. aureus) were allocated correctly. For the two coagulase negative Staphylococcus species (8 and 9) there were bigger overlaps between these two species: 24.8% of the spectra from sample 8 (Staph. epidermidis) were labeled as Staph. saprophyticus and 31.7% of the Staph. saprophyticus (9) spectra were labeled as Staph. epidermidis, but the differentiation from Staph. aureus is quite good for both of the samples. The unknown sample of Strep. agalactiae (11) was evaluated properly on the all three levels (100%, 100%, and 96.3%). To also test the capability to identify different strains of the same species, we analyzed two different Strep. pneumoniae strains (12 and 13). On the top-level and the genus-level, 96.1% of sample 12 and 98.0% of sample 13 were correctly labeled. Also on the species-level the assignment to Strep. pneumoniae was performed with quite similar accuracies of 73.8% (12) and 71.7% (13), respectively. Nearly all falsely identified spectra of these two samples were assigned to Strep. agalactiae. In summary, 98.4% of all spectra from the independent test samples were identified correctly on the top-level. On the second level (Figure 3A) 97.2% of the spectra were labeled properly. And even on the species-level (Figure 3B) 83.6% of the spectra from Gram positive pathogens could be assigned to the right species.

model, which eventually was validated with independent batches of each bacterial strain. We were able to show that the identification capability of the classifier was retained even when batches of bacterial strains were analyzed, which have not been referenced in the database before. In the case of Gram negative bacteria, a differentiation between P. aeruginosa and different species from Enterobacteriacea was possible with an accuracy of 96%. This is important, as different antibiotics can be taken into account for the treatment of these two groups of bacteria. 97.7% of the spectra from Gram positive bacteria were correctly assigned on the genus level and 83.6% even on the species level. Hereby, the discrimination of the Enterococcus species from the other Gram positive bacteria is of high importance because the Enterococcus species are not covered by the empirical first line antibiotic therapy with third-generation cephalosporins.2 From these results, it can be assumed that Raman spectroscopy can be applied successful to patients’ ascitic fluid, which should be further evaluated in future studies. After a Raman compatible isolation step from the ascitic fluid, which could be, for example, buoyant density centrifugation30 or capturing of bacteria with antibodies immobilized on aluminum substrates,31 bacteria could be analyzed directly via Raman spectroscopic investigation on a single-cell level without a timeconsuming culture step. Thus, direct identification of pathogens in approximately 3 h, including sample preprocessing, measurement and data analysis, will hopefully open new opportunities for timely initiation of early goal directed antimicrobial therapies.

■

ASSOCIATED CONTENT

S Supporting Information *

Implemented bacterial strains with number of spectra in the database and numbering and number of spectra of the independent test samples, tentative band assignment, combined 3-level SVM classification model (3 independent batches per strain, the numbers in the circles represent the accuracies of the SVMs achieved by 10-fold cross validation), and identification of the unknown bacterial samples on the 3 levels of the SVM model. This material is available free of charge via the Internet at http://pubs.acs.org.

■

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. Phone: +49-3641-948320, +49-3641-206300. Fax: +49-3641-948302. Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS Funding of the research project FastDiagnosis (13N11350) from the Federal Ministry of Education and Research, Germany (BMBF), is gratefully acknowledged. We also thank Bernd Kampe for help with the program Gnu R, Svea Sachse (Institute of Medical Microbiology) for help in biological questions, Monika Alexi (Institute of Medical Microbiology) for the collection of the bacterial isolates and Cora Richert (Institute of Clinical Chemistry and Laboratory Diagnostics) for the collection of the ascitic fluid samples. Dr. Stephan Stöckel and Dr. Thomas Mayerhöfer are gratefully acknowledged for the critical reading of the present manuscript.

■

CONCLUSIONS The present study demonstrates for the first time a single-cell Raman spectroscopic identification method to determine pathogens in ascitic fluid. As a first step, a reference database was established, consisting of more than 10000 individual Raman spectra of single bacterial cells from 34 strains out of 13 species, which were used to train a robust three-level SVM 942

DOI: 10.1021/ac503373r Anal. Chem. 2015, 87, 937−943

Article

Analytical Chemistry

■

REFERENCES

(1) Blot, S.; De Waele, J.; Vogelaers, D. Drugs 2012, 72, e17−e32. (2) Ginès, P.; Angeli, P.; Lenz, K.; Møller, S.; Moore, K.; Moreau, R. J. Hepatol. 2010, 53, 397−417. (3) Alexopoulou, A.; Papadopoulos, N.; Eliopoulos, D. G.; Alexaki, A.; Tsiriga, A.; Toutouza, M.; Pectasides, D. Liver Int. 2013, 33, 975− 981. (4) Reuken, P. A.; Pletz, M. W.; Baier, M.; Pfister, W.; Stallmach, A.; Bruns, T. Aliment. Pharmacol. Ther. 2012, 35, 1199−1208. (5) Umgelter, A.; Reindl, W.; Miedaner, M.; Schmid, R. M.; Huber, W. Infection 2009, 37, 2−8. (6) Bruns, T.; Sachse, S.; Straube, E.; Assefa, S.; Herrmann, A.; Hagel, S.; Lehmann, M.; Stallmach, A. Liver Int. 2009, 29, 1206−1214. (7) Rogers, G.; Russell, L.; Preston, P.; Marsh, P.; Collins, J.; Saunders, J.; Sutton, J.; Fine, D.; Bruce, K.; Wright, M. Eur. J. Clin. Microbiol. Infect. Dis. 2010, 29, 533−541. (8) Soriano, G.; Esparcia, Ó .; Montemayor, M.; Guarner-Argente, C.; Pericas, R.; Torras, X.; Calvo, N.; Román, E.; Navarro, F.; Guarner, C.; Coll, P. Aliment. Pharmacol. Ther. 2011, 33, 275−284. (9) Hardick, J.; Won, H.; Jeng, K.; Hsieh, Y.-H.; Gaydos, C. A.; Rothman, R. E.; Yang, S. J. Clin. Microbiol. 2012, 50, 2428−2432. (10) Sugihara, T.; Koda, M.; Maeda, Y.; Matono, T.; Nagahara, T.; Mandai, M.; Ueki, M.; Murawaki, Y. Int. Med. 2009, 48, 3−10. (11) Kusić, D.; Kampe, B.; Rösch, P.; Popp, J. Water Res. 2014, 48, 179−189. (12) Stöckel, S.; Meisel, S.; Elschner, M.; Melzer, F.; Rösch, P.; Popp, J. Anal. Bioanal. Chem. 2014, 1−8. (13) Meisel, S.; Stöckel, S.; Elschner, M.; Melzer, F.; Rösch, P.; Popp, J. Appl. Environ. Microbiol. 2012, 78, 5575−5583. (14) Meisel, S.; Stöckel, S.; Rösch, P.; Popp, J. Food Micorobiol. 2014, 38, 36−43. (15) Kloß, S.; Kampe, B.; Sachse, S.; Rösch, P.; Straube, E.; Pfister, W.; Kiehntopf, M.; Popp, J. Anal. Chem. 2013, 85, 9610−9616. (16) Münchberg, U.; Rösch, P.; Bauer, M.; Popp, J. Anal. Bioanal. Chem. 2014, 406, 3041−3050. (17) Harz, M.; Rösch, P.; Peschke, K. D.; Ronneberger, O.; Burkhardt, H.; Popp, J. Analyst 2005, 130, 1543−1550. (18) Walter, A.; Kuhri, S.; Reinicke, M.; Bocklitz, T.; Schumacher, W.; Rösch, P.; Merten, D.; Büchel, G.; Kothe, E.; Popp, J. J. Raman Spectrosc. 2012, 43, 1058−1064. (19) Huang, W.; Bailey, M.; Thompson, I.; Whiteley, A.; Spiers, A. Microbial Ecol. 2007, 53, 414−425. (20) R Development Core Team. Gnu R; R Foundation for Statistical Computing: Vienna, Austria, 2011. (21) Bocklitz, T.; Walter, A.; Hartmann, K.; Rösch, P.; Popp, J. Anal. Chim. Acta 2011, 704, 47−56. (22) Morhác, M.; Kliman, J.; Matoušek, V.; Veselský, M.; Turzo, I. Nucl. Instrum. Methods Phys. Res., Sect. A 1997, 401, 113−132. (23) Zhang, D.; Jallad, K. N.; Ben-Amotz, D. Appl. Spectrosc. 2001, 55, 1523−1531. (24) Dörfer, T.; Bocklitz, T.; Tarcea, N.; Schmitt, M.; Popp, J. Z. Phys. Chem. (Muenchen, Ger.) 2011, 225, 753−764. (25) Bocklitz, T.; Putsche, M.; Stüber, C.; Käs, J.; Niendorf, A.; Rösch, P.; Popp, J. J. Raman Spectrosc. 2009, 40, 1759−1765. (26) Stöckel, S.; Meisel, S.; Elschner, M.; Rösch, P.; Popp, J. Anal. Chem. 2012, 84, 9873−9880. (27) Chang, C.-C.; Lin, C.-J. ACM Trans. Intell. Syst. Technol. 2011, 2, 1−27. (28) Karatzoglou, A.; Smola, A.; Hornik, K.. R package kernlab; R Foundation for Statistical Computing: Vienna, Austria, 2013. (29) Ariza, X.; Lora-Tamayo, J.; Castellote, J.; Xiol, X.; Ariza, J. Scand. J. Gastroenterol. 2013, 48, 1213−1221. (30) Meisel, S.; Stöckel, S.; Elschner, M.; Rösch, P.; Popp, J. Analyst 2011, 136, 4997−5005. (31) Pahlow, S.; Kloß, S.; Blättel, V.; Kirsch, K.; Hübner, U.; Cialla, D.; Rösch, P.; Weber, K.; Popp, J. ChemPhysChem 2013, 14, 3600− 3605.

943

DOI: 10.1021/ac503373r Anal. Chem. 2015, 87, 937−943