Accepted Manuscript Title: Identification of water-conditioned Pseudomonas aeruginosa by Raman microspectroscopy on a single cell level Author: Anja Silge Wilm Schumacher Petra R¨osch Paulo A. Da Costa Filho C´edric G´erard J¨urgen Popp PII: DOI: Reference:
S0723-2020(14)00081-2 http://dx.doi.org/doi:10.1016/j.syapm.2014.05.007 SYAPM 25627
To appear in: Received date: Revised date: Accepted date:
29-10-2013 30-4-2014 2-5-2014
Please cite this article as: A. Silge, W. Schumacher, P. R¨osch, P.A.D.C. Filho, C. G´erard, J. Popp, Identification of water-conditioned Pseudomonas aeruginosa by Raman microspectroscopy on a single cell level, Systematic and Applied Microbiology (2014), http://dx.doi.org/10.1016/j.syapm.2014.05.007 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Identification of water-conditioned Pseudomonas aeruginosa by Raman microspectroscopy on a single cell level
ip t
Anja Silgea,c, Wilm Schumachera,c, Petra Röscha,c, Paulo A. Da Costa Filhod, Cédric Gérarde, Jürgen Poppa,b,c a
M
an
us
cr
Institute of Physical Chemistry and Abbe Center of Photonics, Friedrich-SchillerUniversität Jena, Helmholtzweg 4, D-07743 Jena, Germany b Leibniz Institute of Photonic Technology, Albert-Einstein-Strasse 9, D-07702 Jena, Germany c InfectoGnostics Research Campus Jena, Center for Applied Research, Philosophenweg 7, D-07743 Jena, Germany d Analytical Sciences Department, Nestlé Research Center, Vers-chez-les-blanc, 1000 Lausanne 26, Switzerland e Food Safety & Quality Department, Nestlé Research Center, Vers-chez-les-blanc, 1000 Lausanne 26, Switzerland
Abstract
The identification of Pseudomonas aeruginosa from samples of bottled natural mineral
d
water by the analysis of subcultures is time consuming and other species of the
te
authentic Pseudomonas group can be a problem. Therefore, this study aimed to investigate the influence of different aquatic environmental conditions (pH, mineral
Ac ce p
content) and growth phases on the cultivation-free differentiation between waterconditioned Pseudomonas spp. by applying Raman microspectroscopy. The final dataset was comprised of over 7,500 single-cell Raman spectra, including the species Pseudomonas aeruginosa, P. fluorescens and P. putida, in order to prove the feasibility of the introduced approach. The collection of spectra was standardized by automated measurements of viable stained bacterial cells. The discrimination was influenced by the growth phase at the beginning of the water adaptation period and by the type of mineral water. Different combinations of the parameters were tested and they resulted in accuracies of up to 85% for the identification of P. aeruginosa from independent samples by applying chemometric analysis. Keywords: Pseudomonas aeruginosa, natural mineral water, cultivation-free identification, Raman microspectroscopy chemometrics 1
Page 1 of 25
1. Introduction Natural mineral waters are not free of microbes. The autochthonous aquifer microflora of ground water sources largely consists of attached bacteria belonging to the alpha-, beta-
ip t
and gamma-subclasses of the Proteobacteria [22]. The bacterial community is a highly preserved natural component of natural water and has no impact on public health [22].
cr
Changes in the composition of the bacterial population indicate a vulnerability of the hydrogeological protection barrier of the source. In addition, these changes reveal an
us
inadequate control of the bottling process [22]. Pseudomonas aeruginosa does not belong to the normal microflora of natural mineral waters and is an important indicator of water quality vulnerability. It is a ubiquitous environmental bacterial species, able to pass
an
through deficient groundwater source protection mechanisms at very low densities. P. aeruginosa has a high adaptability to environmental changes and proliferates under very countries as an indicator organism.
M
low nutrient concentrations [6]. Therefore, P. aeruginosa is monitored in various P. aeruginosa can cause a range of infections but rarely causes serious illness in
d
healthy individuals with no predisposing factors, and it is therefore categorized as an
te
opportunistic human pathogen. The main route of infection is by skin contact, but P. aeruginosa can also orally infect immunosuppressed or cancer patients [8, 26].
Ac ce p
However, there is no evidence that normal use of drinking water is a source of infection in the general population [12, 38].
The identification of P. aeruginosa isolated from natural mineral water is not easy since other species of the authentic Pseudomonas rRNA group I are normal members of the microbial flora of such waters [2, 19]. The international organization for standardization specifies a method for the detection and identification of P. aeruginosa from samples of bottled water (ISO 16266) [9]. The procedure aims to differentiate between the autochthonous Pseudomonas spp., such as P. fluorescens and P. putida, and the indicator species P. aeruginosa by membrane filtration and subsequent incubation on a selective medium. The confirmation of P. aeruginosa presence or absence depends on the type of colonies present, and the presence of blue/green colonies may lead to the identification of P. aeruginosa in 24 hours. However, other colonies need to be subcultured and tested, which takes a total of 8 days to obtain a result [9]. This 2
Page 2 of 25
excessive amount of time and effort causes high costs. In addition, certain drawbacks hamper the reliability of the current standard method. For example, the appearance of the blue/green pigment pyocyanin varies depending on the nutrient status of the cell [39], and water samples declared P. aeruginosa positive in agreement with ISO 16266 estimations have a negative impact on the monitoring process.
ip t
have shown false positive rates of up to 71% [5,16]. Consequently, such false
cr
The frequently described phenomenon of the viable but non-culturable (VBNC) state of bacteria stressed by aquatic environments poses an additional challenge for the
us
identification of waterborne germs. Cells in the VBNC state are considered to be metabolically active but temporarily devoid of the ability to form colonies on agar media [22]. Consequently, these cells stay undetected on bacteriological culture media, and a
an
lot of research has been carried out to improve the detection and identification of waterborne P. aeruginosa [28, 32].
M
The concept of vibrational microspectroscopy in combination with fluorescence microscopy offers a new prospect for the assessment of drinking water quality. Infrared and Raman spectroscopy have been successfully applied in the field of environmental
d
and clinical microbiology [1, 3, 15, 23, 29, 31]. Raman microspectroscopy is especially
te
suited for the rapid identification and characterization of microorganisms on a single cell level [24, 30, 33, 37]. The identification strategy is quite simple: a Raman microscope
Ac ce p
focuses a laser beam on the sample and collects the scattered light via a microscope objective. With this technique it is possible to achieve a spatial resolution of approximately 1 μm, which correlates with the size range of a bacterial cell. The resulting Raman spectrum is a superposition of the biochemical contents and a spectral fingerprint of the measured cell. This spectral pattern can be used for taxonomic identification. However, the differences between the Raman spectra of bacterial cells grown under comparable conditions are not visible by unaided observation. Therefore, chemometric techniques have been established in order to discriminate between the Raman spectra of bacteria belonging to different genera, species or even strains [7, 15, 23, 33, 35, 37]. In combination with fluorescence staining methods, Raman measurements can be limited to cells of particular interest [21]. Since only viable cells have hazard potential,
3
Page 3 of 25
fluorescence probes (e.g. carboxyfluorescein diacetate (CFDA)) can be used for viability staining, and successful Raman measurements can then be performed on these cells. The aim of the present study was to investigate water-conditioned P. aeruginosa, P. fluorescens and P. putida with regard to the classification and identification of the
ip t
species P. aeruginosa by means of Raman microspectroscopy on a single cell level. An experimental study of selected organisms was performed, which included reference
cr
organisms of a culture collection and validly described environmental isolates from drinking quality source water samples. The isolates were chosen in order to generate a
us
matrix of interest from real samples. The selection of strains was made based on the framework of the current ISO standard, and the fact that P. putida and P. fluorescens identification requires additional confirmatory tests. The extent of biological and different
an
aquatic environmental parameters on the identification of the targeted bacteria was investigated. The analysed parameters were strain diversity, growth physiology and
M
mineral content, in combination with the pH value of representative water types. A defined experimental framework was chosen in order to image the spectral variations of pure cultures of the Pseudomonas influenced by different parameter combinations.
d
Furthermore, the study focused on the automation of sample analysis, so that viable
te
stained cells could be detected by fluorescence microscopy and then measured automatically. A self-acting outlier detection algorithm was applied for operator-
Ac ce p
independent data selection that was accessed in subsequent data analyses. The results, the potential and the prospects for spectrometric identification of environmentalconditioned P. aeruginosa are discussed. 2. Experimental setup
2.1. Raman spectroscopy
A Raman setup was coupled to a fluorescence device and motorized measuring stage for automatic measurements (Bio Particle Explorer; rap.ID Particle Systems GmbH, Berlin, Germany, Märzhäuser Wetzlar GmbH & Co, KG). In fluorescence mode, a LED provided a fixed excitation wavelength of 470 nm, and the visual inspection of the sample allowed only the detection of green fluorescent particles due to a single bandpass filter (Semrock, FF01-530/43-25). In automated mode, the positions of the detected bacteria were measured successively. 4
Page 4 of 25
The CCD camera (F-145B/C, Stingray) was coupled to a microscope (MPFLN-BD, Olympus) for sample observation. For Raman measurements, a solid-state frequency doubled Nd:YAG module (LCM-S-111-NNP25; Laser Export Co. Ltd.) with an excitation wavelength of 532 nm was used. An Olympus MPFLN-BD 100x (NA = 0.9) objective
ip t
focused the Raman excitation light onto the sample with a lateral spot size of ≤ 1 μm. A laser power of approximately 10 mW was applied to the sample. After removal of
cr
Rayleigh scattering, the 180° back-scattered Raman light was diffracted with a singlestage monochromator (HE 532, Horiba Jobin Yvon) with a 920-line/mm grating and was
us
collected with a thermoelectrically cooled CCD camera (DV401A-BV; Andor technology) with a spectral resolution of approximately 10 cm−1. Spectra were collected from 276 to 3,204 cm−1. For single cell measurements, two subsequent spectra of 10 s exposures
an
each were recorded, and the two spectra were compared for spike removal (see data
M
pre-processing).
2.2. Preparation of bacteria-spiked bottled mineral water samples The types of water were categorized as high, medium and low mineral water based on
d
their content of Ca2+, Mg2+ and Na+. The grouping of the waters into high, medium and
te
low pH waters was based on the content of HCO−3 and the measured pH (see supplementary section A1 for details). The following Pseudomonas strains were
Ac ce p
selected from the Collection de l’Institut Pasteur (CIP): P. aeruginosa (CIP 82118 (Pae01)) and P. fluorescens (CIP 6913T (Pfl01)). The Nestlé Research Centre provided environmental isolates from drinking quality source water samples that had been isolated and confirmed based on the application of NF EN ISO 16266 (August 2008) and further validated using ID 32 GN. Each individual isolate of P. aeruginosa (Pae) and P. putida (Ppu) had a different biochemical profile: Pae03 API ID 32 GN
99.6%
profile 20573063072
Pae04 API ID 32 GN
99.9%
profile 20573067072
Ppu01 API ID 32 GN
99.5%
profile 40472077073
Ppu02 API ID 32 GN
99.5%
profile 42072067073
5
Page 5 of 25
For pre-cultivation, all bacteria were incubated for approximately 16 h in 20 mL of nutrient agar medium (NA) (0.5% peptone, 0.3% meat extract). These and all following incubations were carried out at 25 °C with shaking at 100 rpm. Bacteria from two different growth phases were prepared and submitted to a water adaptation period of 72
ip t
h. For the stationary phase, bacteria were used directly from the pre-culture. For the exponential growth phase, 5 mL of the pre-culture were transferred into 15 mL of fresh
cr
NA medium and incubated for 90 min. Bacterial cultures synchronized to the respective growth phase were washed twice by centrifugation at 3,000 rpm for 15 min at 4 °C with
us
resuspension of the pellet in the respective type of sterile mineral water in order to obtain the spiking solution. For creating the model dataset, 18 mL of each type of mineral water were sterilized by membrane filtration (0.2 μm membrane) and spiked with
an
bacteria to a final concentration of approximately 105 cfu mL-1. The water adaptation period (72 h) started from this inoculation time point. For evaluation, the analysis volume
M
was increased to 500 mL and the bacterial count was adjusted to approximately 10 cfu
2.3. Sample preparation
d
mL-1.
te
After the water adaption period, bacterial samples were prepared as follows. Fluorescence staining with carboxyfluorescein diacetate (CFDA) was carried out
Ac ce p
according to the proposal of Hoefel et al. and is described in detail in supplementary material section A2 [17]. Bacteria were isolated by membrane filtration (0.8 μm pore size) on a polycarbonate filter coated with nickel (rap.ID Particle Systems GmbH, Berlin, Germany). For single cell Raman analysis, the bacteria were transferred from the filter onto a nickel foil (rap.ID Particle Systems GmbH, Berlin, Germany). 2.4. Data pre-processing and chemometric analysis All data processing, including outlier detection and chemometric analysis, was performed using the R program [34]. For pre-processing, background correction, normalization, spike removal and reduction of the spectral regions of interest were carried out. For more details see supplementary material section A3. An important point of the present investigation was outlier detection for data selection, since non-bacterial spectra can still be recorded, for example, by occasional non-specific fluorescence 6
Page 6 of 25
staining or auto-fluorescent particles. A different spectral fingerprint characterizes such spectra, and these so called outliers have to be removed from the spectral dataset. For this purpose, the sign method, introduced by Filzmoser et al. and implemented as the function sign2 in the R package mvoutlier, was used [11] (see supplementary material
ip t
section A4 for details).
For data evaluation, a principal component analysis (PCA; implemented in the R-
cr
package prcomp) was carried out in order to reduce the dimensionality of the data while maintaining spectra variability. A supervised classification algorithm, the support vector
us
machine (SVM) as implemented in R package e1071, was applied in the classification mode with a radial basis kernel [4]. The first 25 principal components (PCs) of the dataset were introduced to the SVM. The identification performance of the classifier was
an
validated by a leave-one-block-out-cross-validation (LOBO-CV) [36]. The data were divided into subsets of independent batch cultures in order to provide both a training
M
dataset and an identification dataset. In doing so, the dataset was first split along the “batch/age variance axis” shown in Fig. 1. The SVM model was trained with 5 batches. The group labels for the data of the left out batch were predicted. To provide an
d
overview of the results, the number of spectra assigned to either group A or B were
te
summed for the three exponential and three stationary batches, respectively. The stationary phase batches were based on measurements of 4,458 single cells, and the
Ac ce p
exponential phase batches included 3,130 spectra of single cells. For a second LOBOCV, the dataset was split along the “environmental variance axis” shown in Fig. 1 in bacteria from the given water types. These were called water-blocks in the subsequent part of the study. The SVM model was trained with three of the water-blocks, and the group assignment for the data of the remaining water-block was predicted. The spectral characteristics contributed by the influence of the left out water type were therefore not present in the model. For validation of the setup, the collected spectra of two blind samples were rotated into the PCA-space of the reference dataset and the first 25 PCs were introduced to the trained SVM model. 3. Results and discussion 3.1. Collection of the dataset
7
Page 7 of 25
The organization of the data is shown in Fig. 1 with the axes indicating the considered parameters of the study. Each button visualizes a sample characterized by a combination of the parameters. All buttons from one layer display one biologically independent batch. Layers two to six are indicated by only one button in order to be able
ip t
to visualize the overall picture. The genetic variances of the Pseudomonas isolates were represented by three strains of the target species P. aeruginosa. For the statistical
cr
analysis, single cell spectra of these three strains were pooled and designated as group ”A”, the target-bacteria. P. fluorescens and P. putida were chosen as background-
us
bacteria with single cell spectra of these strains subsequently pooled and designated as group ”B”. The characteristics of the mineral waters (different mineral contents and pH values) determined the respective aquatic environment of each sample. Different types
an
of mineral waters, therefore, constituted an environmental variance in the study. The batch and age variances were divided into six batches. Batches one to three
M
represented three biologically independent replicates of the exponential growth phase pre-cultures. Batches four to six represented three replicates of the stationary phase pre-cultures. In total, 144 measurement series were carried out. The sample preparation
d
for the Raman analysis in the study included viability staining (ca. 30 min), as well as the
te
filtration and application of the cells to the measurement substrate (ca. 30 min). The detection and measurement time of approximately 50-100 representative single cells per
Ac ce p
sample took approximately 45 min. The subsequent data pre-processing and the chemometric analysis required another 30 min. Thus, the analysis time per sample amounted to between 2 and 3 h (depending on the number of measured cells). 3.2. Classification and identification results The classification algorithm support vector machine (SVM) was applied for the classification and identification of single cell Raman spectra. Each spectrum of the final dataset (a total of 7,588 spectra) was hereby considered as a single data point, characterized by its spectral features (Fig. 2, the averaged group-spectra). The grey zone around each spectrum reflected statistical spread and the complexity of the dataset due to intra-group variations. It was particularly pronounced in the CH-stretching region with approximately 2,930 cm−1, which represented a mixture of contributions from all cell components [13]. The diversity in the fingerprint region was also discernible. Peak 8
Page 8 of 25
positions of the most prominent bands in this region are denoted in the spectra (1,660 cm−1: amide I (protein), 1,572 cm−1: guanine/adenine ring stretching, 1,448 cm−1: CH2 deformation, 1,240 cm−1: C-C and C-N assigned to adenine/thymine and amide III, 1,003 cm−1: ring breathing mode of phenylalanine, 780 cm−1: O-P-O stretch mode phosphate
ip t
backbone DNA, 720 cm−1: adenine) [15, 18, 23].
The SVM separated the data points by a boundary that was defined by the largest
cr
margin between groups A and B. The decision boundary was displayed by the SVM dvalue of zero. All points with a SVM d-value >0 were assigned to group A, and 0 are assigned to group A and
S0723-2020(14)00081-2 http://dx.doi.org/doi:10.1016/j.syapm.2014.05.007 SYAPM 25627
To appear in: Received date: Revised date: Accepted date:
29-10-2013 30-4-2014 2-5-2014
Please cite this article as: A. Silge, W. Schumacher, P. R¨osch, P.A.D.C. Filho, C. G´erard, J. Popp, Identification of water-conditioned Pseudomonas aeruginosa by Raman microspectroscopy on a single cell level, Systematic and Applied Microbiology (2014), http://dx.doi.org/10.1016/j.syapm.2014.05.007 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Identification of water-conditioned Pseudomonas aeruginosa by Raman microspectroscopy on a single cell level
ip t
Anja Silgea,c, Wilm Schumachera,c, Petra Röscha,c, Paulo A. Da Costa Filhod, Cédric Gérarde, Jürgen Poppa,b,c a
M
an
us
cr
Institute of Physical Chemistry and Abbe Center of Photonics, Friedrich-SchillerUniversität Jena, Helmholtzweg 4, D-07743 Jena, Germany b Leibniz Institute of Photonic Technology, Albert-Einstein-Strasse 9, D-07702 Jena, Germany c InfectoGnostics Research Campus Jena, Center for Applied Research, Philosophenweg 7, D-07743 Jena, Germany d Analytical Sciences Department, Nestlé Research Center, Vers-chez-les-blanc, 1000 Lausanne 26, Switzerland e Food Safety & Quality Department, Nestlé Research Center, Vers-chez-les-blanc, 1000 Lausanne 26, Switzerland
Abstract
The identification of Pseudomonas aeruginosa from samples of bottled natural mineral
d
water by the analysis of subcultures is time consuming and other species of the
te
authentic Pseudomonas group can be a problem. Therefore, this study aimed to investigate the influence of different aquatic environmental conditions (pH, mineral
Ac ce p
content) and growth phases on the cultivation-free differentiation between waterconditioned Pseudomonas spp. by applying Raman microspectroscopy. The final dataset was comprised of over 7,500 single-cell Raman spectra, including the species Pseudomonas aeruginosa, P. fluorescens and P. putida, in order to prove the feasibility of the introduced approach. The collection of spectra was standardized by automated measurements of viable stained bacterial cells. The discrimination was influenced by the growth phase at the beginning of the water adaptation period and by the type of mineral water. Different combinations of the parameters were tested and they resulted in accuracies of up to 85% for the identification of P. aeruginosa from independent samples by applying chemometric analysis. Keywords: Pseudomonas aeruginosa, natural mineral water, cultivation-free identification, Raman microspectroscopy chemometrics 1
Page 1 of 25
1. Introduction Natural mineral waters are not free of microbes. The autochthonous aquifer microflora of ground water sources largely consists of attached bacteria belonging to the alpha-, beta-
ip t
and gamma-subclasses of the Proteobacteria [22]. The bacterial community is a highly preserved natural component of natural water and has no impact on public health [22].
cr
Changes in the composition of the bacterial population indicate a vulnerability of the hydrogeological protection barrier of the source. In addition, these changes reveal an
us
inadequate control of the bottling process [22]. Pseudomonas aeruginosa does not belong to the normal microflora of natural mineral waters and is an important indicator of water quality vulnerability. It is a ubiquitous environmental bacterial species, able to pass
an
through deficient groundwater source protection mechanisms at very low densities. P. aeruginosa has a high adaptability to environmental changes and proliferates under very countries as an indicator organism.
M
low nutrient concentrations [6]. Therefore, P. aeruginosa is monitored in various P. aeruginosa can cause a range of infections but rarely causes serious illness in
d
healthy individuals with no predisposing factors, and it is therefore categorized as an
te
opportunistic human pathogen. The main route of infection is by skin contact, but P. aeruginosa can also orally infect immunosuppressed or cancer patients [8, 26].
Ac ce p
However, there is no evidence that normal use of drinking water is a source of infection in the general population [12, 38].
The identification of P. aeruginosa isolated from natural mineral water is not easy since other species of the authentic Pseudomonas rRNA group I are normal members of the microbial flora of such waters [2, 19]. The international organization for standardization specifies a method for the detection and identification of P. aeruginosa from samples of bottled water (ISO 16266) [9]. The procedure aims to differentiate between the autochthonous Pseudomonas spp., such as P. fluorescens and P. putida, and the indicator species P. aeruginosa by membrane filtration and subsequent incubation on a selective medium. The confirmation of P. aeruginosa presence or absence depends on the type of colonies present, and the presence of blue/green colonies may lead to the identification of P. aeruginosa in 24 hours. However, other colonies need to be subcultured and tested, which takes a total of 8 days to obtain a result [9]. This 2
Page 2 of 25
excessive amount of time and effort causes high costs. In addition, certain drawbacks hamper the reliability of the current standard method. For example, the appearance of the blue/green pigment pyocyanin varies depending on the nutrient status of the cell [39], and water samples declared P. aeruginosa positive in agreement with ISO 16266 estimations have a negative impact on the monitoring process.
ip t
have shown false positive rates of up to 71% [5,16]. Consequently, such false
cr
The frequently described phenomenon of the viable but non-culturable (VBNC) state of bacteria stressed by aquatic environments poses an additional challenge for the
us
identification of waterborne germs. Cells in the VBNC state are considered to be metabolically active but temporarily devoid of the ability to form colonies on agar media [22]. Consequently, these cells stay undetected on bacteriological culture media, and a
an
lot of research has been carried out to improve the detection and identification of waterborne P. aeruginosa [28, 32].
M
The concept of vibrational microspectroscopy in combination with fluorescence microscopy offers a new prospect for the assessment of drinking water quality. Infrared and Raman spectroscopy have been successfully applied in the field of environmental
d
and clinical microbiology [1, 3, 15, 23, 29, 31]. Raman microspectroscopy is especially
te
suited for the rapid identification and characterization of microorganisms on a single cell level [24, 30, 33, 37]. The identification strategy is quite simple: a Raman microscope
Ac ce p
focuses a laser beam on the sample and collects the scattered light via a microscope objective. With this technique it is possible to achieve a spatial resolution of approximately 1 μm, which correlates with the size range of a bacterial cell. The resulting Raman spectrum is a superposition of the biochemical contents and a spectral fingerprint of the measured cell. This spectral pattern can be used for taxonomic identification. However, the differences between the Raman spectra of bacterial cells grown under comparable conditions are not visible by unaided observation. Therefore, chemometric techniques have been established in order to discriminate between the Raman spectra of bacteria belonging to different genera, species or even strains [7, 15, 23, 33, 35, 37]. In combination with fluorescence staining methods, Raman measurements can be limited to cells of particular interest [21]. Since only viable cells have hazard potential,
3
Page 3 of 25
fluorescence probes (e.g. carboxyfluorescein diacetate (CFDA)) can be used for viability staining, and successful Raman measurements can then be performed on these cells. The aim of the present study was to investigate water-conditioned P. aeruginosa, P. fluorescens and P. putida with regard to the classification and identification of the
ip t
species P. aeruginosa by means of Raman microspectroscopy on a single cell level. An experimental study of selected organisms was performed, which included reference
cr
organisms of a culture collection and validly described environmental isolates from drinking quality source water samples. The isolates were chosen in order to generate a
us
matrix of interest from real samples. The selection of strains was made based on the framework of the current ISO standard, and the fact that P. putida and P. fluorescens identification requires additional confirmatory tests. The extent of biological and different
an
aquatic environmental parameters on the identification of the targeted bacteria was investigated. The analysed parameters were strain diversity, growth physiology and
M
mineral content, in combination with the pH value of representative water types. A defined experimental framework was chosen in order to image the spectral variations of pure cultures of the Pseudomonas influenced by different parameter combinations.
d
Furthermore, the study focused on the automation of sample analysis, so that viable
te
stained cells could be detected by fluorescence microscopy and then measured automatically. A self-acting outlier detection algorithm was applied for operator-
Ac ce p
independent data selection that was accessed in subsequent data analyses. The results, the potential and the prospects for spectrometric identification of environmentalconditioned P. aeruginosa are discussed. 2. Experimental setup
2.1. Raman spectroscopy
A Raman setup was coupled to a fluorescence device and motorized measuring stage for automatic measurements (Bio Particle Explorer; rap.ID Particle Systems GmbH, Berlin, Germany, Märzhäuser Wetzlar GmbH & Co, KG). In fluorescence mode, a LED provided a fixed excitation wavelength of 470 nm, and the visual inspection of the sample allowed only the detection of green fluorescent particles due to a single bandpass filter (Semrock, FF01-530/43-25). In automated mode, the positions of the detected bacteria were measured successively. 4
Page 4 of 25
The CCD camera (F-145B/C, Stingray) was coupled to a microscope (MPFLN-BD, Olympus) for sample observation. For Raman measurements, a solid-state frequency doubled Nd:YAG module (LCM-S-111-NNP25; Laser Export Co. Ltd.) with an excitation wavelength of 532 nm was used. An Olympus MPFLN-BD 100x (NA = 0.9) objective
ip t
focused the Raman excitation light onto the sample with a lateral spot size of ≤ 1 μm. A laser power of approximately 10 mW was applied to the sample. After removal of
cr
Rayleigh scattering, the 180° back-scattered Raman light was diffracted with a singlestage monochromator (HE 532, Horiba Jobin Yvon) with a 920-line/mm grating and was
us
collected with a thermoelectrically cooled CCD camera (DV401A-BV; Andor technology) with a spectral resolution of approximately 10 cm−1. Spectra were collected from 276 to 3,204 cm−1. For single cell measurements, two subsequent spectra of 10 s exposures
an
each were recorded, and the two spectra were compared for spike removal (see data
M
pre-processing).
2.2. Preparation of bacteria-spiked bottled mineral water samples The types of water were categorized as high, medium and low mineral water based on
d
their content of Ca2+, Mg2+ and Na+. The grouping of the waters into high, medium and
te
low pH waters was based on the content of HCO−3 and the measured pH (see supplementary section A1 for details). The following Pseudomonas strains were
Ac ce p
selected from the Collection de l’Institut Pasteur (CIP): P. aeruginosa (CIP 82118 (Pae01)) and P. fluorescens (CIP 6913T (Pfl01)). The Nestlé Research Centre provided environmental isolates from drinking quality source water samples that had been isolated and confirmed based on the application of NF EN ISO 16266 (August 2008) and further validated using ID 32 GN. Each individual isolate of P. aeruginosa (Pae) and P. putida (Ppu) had a different biochemical profile: Pae03 API ID 32 GN
99.6%
profile 20573063072
Pae04 API ID 32 GN
99.9%
profile 20573067072
Ppu01 API ID 32 GN
99.5%
profile 40472077073
Ppu02 API ID 32 GN
99.5%
profile 42072067073
5
Page 5 of 25
For pre-cultivation, all bacteria were incubated for approximately 16 h in 20 mL of nutrient agar medium (NA) (0.5% peptone, 0.3% meat extract). These and all following incubations were carried out at 25 °C with shaking at 100 rpm. Bacteria from two different growth phases were prepared and submitted to a water adaptation period of 72
ip t
h. For the stationary phase, bacteria were used directly from the pre-culture. For the exponential growth phase, 5 mL of the pre-culture were transferred into 15 mL of fresh
cr
NA medium and incubated for 90 min. Bacterial cultures synchronized to the respective growth phase were washed twice by centrifugation at 3,000 rpm for 15 min at 4 °C with
us
resuspension of the pellet in the respective type of sterile mineral water in order to obtain the spiking solution. For creating the model dataset, 18 mL of each type of mineral water were sterilized by membrane filtration (0.2 μm membrane) and spiked with
an
bacteria to a final concentration of approximately 105 cfu mL-1. The water adaptation period (72 h) started from this inoculation time point. For evaluation, the analysis volume
M
was increased to 500 mL and the bacterial count was adjusted to approximately 10 cfu
2.3. Sample preparation
d
mL-1.
te
After the water adaption period, bacterial samples were prepared as follows. Fluorescence staining with carboxyfluorescein diacetate (CFDA) was carried out
Ac ce p
according to the proposal of Hoefel et al. and is described in detail in supplementary material section A2 [17]. Bacteria were isolated by membrane filtration (0.8 μm pore size) on a polycarbonate filter coated with nickel (rap.ID Particle Systems GmbH, Berlin, Germany). For single cell Raman analysis, the bacteria were transferred from the filter onto a nickel foil (rap.ID Particle Systems GmbH, Berlin, Germany). 2.4. Data pre-processing and chemometric analysis All data processing, including outlier detection and chemometric analysis, was performed using the R program [34]. For pre-processing, background correction, normalization, spike removal and reduction of the spectral regions of interest were carried out. For more details see supplementary material section A3. An important point of the present investigation was outlier detection for data selection, since non-bacterial spectra can still be recorded, for example, by occasional non-specific fluorescence 6
Page 6 of 25
staining or auto-fluorescent particles. A different spectral fingerprint characterizes such spectra, and these so called outliers have to be removed from the spectral dataset. For this purpose, the sign method, introduced by Filzmoser et al. and implemented as the function sign2 in the R package mvoutlier, was used [11] (see supplementary material
ip t
section A4 for details).
For data evaluation, a principal component analysis (PCA; implemented in the R-
cr
package prcomp) was carried out in order to reduce the dimensionality of the data while maintaining spectra variability. A supervised classification algorithm, the support vector
us
machine (SVM) as implemented in R package e1071, was applied in the classification mode with a radial basis kernel [4]. The first 25 principal components (PCs) of the dataset were introduced to the SVM. The identification performance of the classifier was
an
validated by a leave-one-block-out-cross-validation (LOBO-CV) [36]. The data were divided into subsets of independent batch cultures in order to provide both a training
M
dataset and an identification dataset. In doing so, the dataset was first split along the “batch/age variance axis” shown in Fig. 1. The SVM model was trained with 5 batches. The group labels for the data of the left out batch were predicted. To provide an
d
overview of the results, the number of spectra assigned to either group A or B were
te
summed for the three exponential and three stationary batches, respectively. The stationary phase batches were based on measurements of 4,458 single cells, and the
Ac ce p
exponential phase batches included 3,130 spectra of single cells. For a second LOBOCV, the dataset was split along the “environmental variance axis” shown in Fig. 1 in bacteria from the given water types. These were called water-blocks in the subsequent part of the study. The SVM model was trained with three of the water-blocks, and the group assignment for the data of the remaining water-block was predicted. The spectral characteristics contributed by the influence of the left out water type were therefore not present in the model. For validation of the setup, the collected spectra of two blind samples were rotated into the PCA-space of the reference dataset and the first 25 PCs were introduced to the trained SVM model. 3. Results and discussion 3.1. Collection of the dataset
7
Page 7 of 25
The organization of the data is shown in Fig. 1 with the axes indicating the considered parameters of the study. Each button visualizes a sample characterized by a combination of the parameters. All buttons from one layer display one biologically independent batch. Layers two to six are indicated by only one button in order to be able
ip t
to visualize the overall picture. The genetic variances of the Pseudomonas isolates were represented by three strains of the target species P. aeruginosa. For the statistical
cr
analysis, single cell spectra of these three strains were pooled and designated as group ”A”, the target-bacteria. P. fluorescens and P. putida were chosen as background-
us
bacteria with single cell spectra of these strains subsequently pooled and designated as group ”B”. The characteristics of the mineral waters (different mineral contents and pH values) determined the respective aquatic environment of each sample. Different types
an
of mineral waters, therefore, constituted an environmental variance in the study. The batch and age variances were divided into six batches. Batches one to three
M
represented three biologically independent replicates of the exponential growth phase pre-cultures. Batches four to six represented three replicates of the stationary phase pre-cultures. In total, 144 measurement series were carried out. The sample preparation
d
for the Raman analysis in the study included viability staining (ca. 30 min), as well as the
te
filtration and application of the cells to the measurement substrate (ca. 30 min). The detection and measurement time of approximately 50-100 representative single cells per
Ac ce p
sample took approximately 45 min. The subsequent data pre-processing and the chemometric analysis required another 30 min. Thus, the analysis time per sample amounted to between 2 and 3 h (depending on the number of measured cells). 3.2. Classification and identification results The classification algorithm support vector machine (SVM) was applied for the classification and identification of single cell Raman spectra. Each spectrum of the final dataset (a total of 7,588 spectra) was hereby considered as a single data point, characterized by its spectral features (Fig. 2, the averaged group-spectra). The grey zone around each spectrum reflected statistical spread and the complexity of the dataset due to intra-group variations. It was particularly pronounced in the CH-stretching region with approximately 2,930 cm−1, which represented a mixture of contributions from all cell components [13]. The diversity in the fingerprint region was also discernible. Peak 8
Page 8 of 25
positions of the most prominent bands in this region are denoted in the spectra (1,660 cm−1: amide I (protein), 1,572 cm−1: guanine/adenine ring stretching, 1,448 cm−1: CH2 deformation, 1,240 cm−1: C-C and C-N assigned to adenine/thymine and amide III, 1,003 cm−1: ring breathing mode of phenylalanine, 780 cm−1: O-P-O stretch mode phosphate
ip t
backbone DNA, 720 cm−1: adenine) [15, 18, 23].
The SVM separated the data points by a boundary that was defined by the largest
cr
margin between groups A and B. The decision boundary was displayed by the SVM dvalue of zero. All points with a SVM d-value >0 were assigned to group A, and 0 are assigned to group A and
