Infrared Spectroscopy of Proteus mirabilis Swarm ...

Mapping Bacterial Surface Population Physiology in Real-Time: Infrared Spectroscopy of Proteus mirabilis Swarm Colonies JULIE KEIRSSE, ELODIE LAHAYE, ANTHONY BOUTER, VIRGINIE DUPONT, CATHERINE BOUSSARD-PLE´DEL, BRUNO BUREAU, JEAN-LUC ADAM, ´ RIE MONBET, and OLIVIER SIRE* VALE Laboratoire des Polyme`res, Proprie´te´s aux Interfaces et Composites, Universite´ de Bretagne-Sud, Campus de Tohannic, BP573, 56017 Vannes Cedex, France (J.K., E.L., A.B., V.D., O.S.); Laboratoire des Verres et Ce´ramiques UMR-CNRS 6512, Institut de Chimie, Universite´ de Rennes 1,Campus de Beaulieu, 35042 Rennes Cedex, France (J.K., C.B.-P., B.B., J.-L.A.); and Laboratoire de Statistiques et ses Applications de Bretagne Sud, Universite´ de Bretagne-Sud, Campus de Tohannic, BP573, 56017 Vannes Cedex, France (V.M.)

We mapped the space–time distribution of stationary and swarmer cells within a growing Proteus mirabilis colony by infrared (IR) microspectroscopy. Colony mapping was performed at different positions between the inoculum and the periphery with a discrete microscopemounted IR sensor, while continuous monitoring at a fixed location over time used an optical fiber based IR–attenuated total reflection (ATR) sensor, or ‘‘optrode.’’ Phenotypes within a single P. mirabilis population relied on identification of functional determinants (producing unique spectral signals) that reflect differences in macromolecular composition associated with cell differentiation. Inner swarm colony domains are spectrally homogeneous, having patterns similar to those produced by the inoculum. Outer domains composed of active swarmer cells exhibit spectra distinguishable at multiple wavelengths dominated by polysaccharides. Our real-time observations agree with and extend earlier reports indicating that motile swarmer cells are restricted to a narrow (approximately 3 mm) annulus at the colony edge. This study thus validates the use of an IR optrode for real-time and noninvasive monitoring of biofilms and other bacterial surface populations. Index Headings: Swarming; Bacterial differentiation; Infrared spectroscopy; Optical IR-ATR sensor; Discriminant analysis.

INTRODUCTION The past ten years have witnessed the development of sensitive, rapid, and increasingly accurate physical techniques for identification of bacterial pathogens.1 These techniques include infrared (IR) and Raman spectroscopies,2,3 as well as different separation techniques (gas chromatography (GC), high-performance liquid chromatography (HPLC)). Since metabolic changes in bacterial populations precede morphological changes in any disease process, spectroscopy-based diagnosis can reveal the early stages of an infection. Recent technical developments in IR spectroscopy have dramatically increased its sensitivity. In particular, the use of microscopy and optical fibers simplifies sample preparation, and surface analysis is facilitated by the use of attenuated total reflection (ATR) modes. The principles of ATR–Fourier transform infrared (FT-IR) spectroscopy have been described earlier by Harrick,4 this technique becoming widely used for obtaining infrared spectra.5,6 Also, the use of remote optical fibers for infrared spectroscopy dates back to the 1990s. The first fibers7–10 were mainly polycrystalline and in AgCl/AgBr. They presented good mechanical properties but were light sensitive because they darken upon light exposure. Another approach was developed based on the use of chalcogenure glasses as Received 9 December 2005; accepted 29 March 2006. * Author to whom correspondence should be sent. E-mail: osire@ univ-ubs.fr.

Volume 60, Number 6, 2006

material for infrared optical fibers,11–19 more especially in the Naval Research Laboratory and in the Laboratoire Verres et Ceramiques. The present paper demonstrates the potential of IR spectroscopy for studying the spatial structure and molecular differentiation of bacterial biofilms in situ and in real time. Understanding the dynamics of biofilm development is a major challenge in health, industry, and biotechnology. Biofilms constitute the predominant lifestyle of bacteria that colonize such different surfaces as teeth, epithelia, water ducts, steel, or hulls.20 The switch between a planktonic, swimming phenotype and a sessile phenotype occurs frequently in the prokaryotic world and may constitute the rule rather than the exception.21 The swarming ability of Proteus mirabilis, a Gram-negative opportunistic pathogen of the human urinary tract, provides one of the most striking examples. This species is able to colonize an organic surface of several centimeters in less than a day through synchronous expansion waves, producing a characteristic ‘‘bull’s eye’’ swarm colony pattern.22–27 Studying P. mirabilis swarming dynamics requires the mapping, in space and time, of the distribution between swarming (SWAR) and vegetative (VEG) non-motile cells. Matsuyama et al.24 studied the internal population structure of a Proteus colony. The spectral analysis of this population internal structure was chosen as a validation test to demonstrate the potential of IR spectroscopy for real-time resolution of the phenotype distribution within an expanding bacterial colony. Earlier, Nichols et al.28 and Nivens et al.29,30 developed the principal design of an experiment by which a flow-through ATR cell is used to monitor in situ and in real time changes in biofilm formation. In order to monitor the development of a P. mirabilis colony, a previous study31 was performed to show that FT-IR spectroscopy was a potent tool to distinguish P. mirabilis phenotypes through specific spectral markers. IR spectroscopy coupled to microscopy (see Refs. 32–36 for pioneer studies) permits noninvasive molecular mapping of the surface population. IR spectra reflect the chemical groups, and hence the biomolecules, present in the sample. In liquid cell culture, large amounts of water, a strong chromophore in the IR domain, prevent the observation of weaker absorption bands. Fortunately, the water in biofilms is greatly reduced compared to batch cultures. Consequently, IR spectra from biofilms present well-structured patterns that eventually allow for sample identification. Membrane lipids, protein, and polysaccharide signals predominate in biofilm IR spectra. Hence, IR spectroscopy allows metabolic imaging of the whole Proteus colony, including the exopolysaccharide (EPS) con-

0003-7028/06/6006-000000$2.00/0 2006 Society for Applied Spectroscopy

APPLIED SPECTROSCOPY

FIG. 1. The microspectroscopic setup (A) consists of sampling a particular colony spot by establishing a close contact between the Ge micro-crystal tip and the cell surface population. Sampling localization is controlled by a motorized thermostated stage. The optical fiber setup (B) is also shown. A single fiber is used both as IR beam waveguide and IR-ATR sensor (optrode). At the sensing zone (about 9 mm) the optical fiber diameter is reduced to 100 lm to increase the sensitivity.

tinuum.31 The agar substrate contribution to the IR spectra is negligible for three reasons: (1) IR spectra are obtained without direct contact with agar; (2) even when data collection occurs in contact with the agar (fiber-optic evanescent wave spectroscopy (FEWS) setup), IR illumination only penetrates to a depth of about 2 cell diameters, far less than the actual thickness of the biofilm; and (3) no sharp structure is observed in pure agar IR spectra.37 In this paper we apply IR microspectroscopy to achieve the first real-time spectral mapping of P. mirabilis swarming. We extend our previous study of a homogeneous differentiating bacterial population by resolving events spatially and temporally in an actively growing P. mirabilis colony. The conclusions from spectral imaging were confirmed by morphological observations using in situ confocal microscopy. In addition, we demonstrate the use of an optical IR-ATR fiber sensor operating in the mid-infrared (MIR) domain utilizing FEWS.4,38 IR spectra can be collected by remote spectroscopy using an optical fiber made of chalcogenide glass. IR glass optical waveguides (optrodes) offer the potential to develop remote analysis for on-line biotechnological systems, such as food processing, as well as medical applications, such as endoscopy.

MATERIALS AND METHODS Strain, Media, and Culture Conditions. The P. mirabilis wild-type strain WT19 corresponds to the clinical isolate U6450 (1). WT19 was grown in LB medium at 37 8C. To perform mapping experiments, 3 lL of an overnight liquid culture (3.108 CFU/ mL) was spotted onto an LB agar (1.5%) plate, and the latter was incubated at 37 8C until the colony diameter reached

either 20 mm for time mapping or 45–60 mm for space mapping. In our experimental conditions, the periodic swarming resulted in terraces spaced by about 2 mm. Liquid culture experiments were similarly performed after an overnight incubation. Infrared Microspectroscopy. The FT-IR microspectroscopic experiments were performed using a Prote´ge´ 460 spectrometer coupled to a Continulm microscope (Nicolet) operated in the micro-ATR mode with a single reflection Ge micro-crystal exhibiting a refractive index n ¼ 4 and a narrow ([ ¼ 50 lm) active optical spot. The microscope was fitted with an MCT detector and a computer controlled stage. Petri plates were placed on the stage on a thermoplate base (WPI) maintained at 37 6 1 8C. For each acquisition, the microcrystal was put in surface contact with the colony for a few seconds, and then the stage was lowered and acquisition started after a one minute delay to allow most of the sample water to evaporate (Fig. 1A). Between each experiment, the Ge crystal was cleaned until a flat baseline (no residual bands) was observed. Each spectrum was acquired at a 4 cm1 resolution and represented the accumulation of 128 separate scans. Mapping. For space mapping, MIR spectra were collected along a radial axis from the colony periphery to the inoculation site in 2 mm steps. This increment roughly corresponds to the width of the terraces. Such series, referred to as space series, comprised about 20 sequential spectra, which were collected over about 120 min. Since the colonies mainly evolve at the edge, spectra were sequentially collected from the periphery towards the inoculum. Indeed, it was checked that the colony inner domains were not subject to change for long periods. Time mapping was performed on colonies whose diameter had reached at least 20–30 mm. Time series spectra were collected at a constant radius until the colony diameter was increased

FIG. 2. The figure displays the optrode setup for time mapping of Proteus colonies. The fiber-sensing zone (SZ) is initially (t1) placed slightly beyond the colony edge (FE); then, as the colony expands, the colony edge reaches the SZ point (t2), and finally monitors more inner domains (t3). I, inoculum.


FIG. 4. FT-IR difference spectrum between swarming and stationary cells. Gray bars feature the eleven frequency domains (see text) that have been semiempirically selected for phenotypic pattern recognition. See Table I for band assignments.

FIG. 3. Mean (top panel) and standard deviations (bottom panel) of normalized FT-IR spectra corresponding to the swimming (LIQC), stationary (STAC), and swarming (SWAR) P. mirabilis phenotypes. For each phenotype, about 40 spectra were collected from 9 to 11 distinct cultures as described in the text. Mean spectra are vertically offset for clarity.

two-fold. The radius was chosen to be slightly beyond the migration front at the start of the measurements. Hence, the first spectra mainly reflect the agar contribution, whereas subsequent spectra sequentially probe the colony edge and the inner terraces. Such series were acquired on a pool of 5 to 7 plates synchronously inoculated at the center (3 lL) and maintained at 37 8C to ensure swarming ability. Data from different plates were normalized according to the distance from the colony edge of the actual plate. This allows increasing the size of each data array for index calculation, though it may slightly increase the noise in the subsequent index calculations as a function of the distance from the edge. Laser Scanning Confocal Microscopy. Spectral mapping was correlated with bacterial visualization achieved by fluorescence labeling using the BacLight LIVE/DEAD bacterial viability staining kit (Molecular Probes). Hence, bacterial viability and cell lengths are simultaneously determined. A mixture (1:1) of stock solutions of stain (SYTO 9 and propidium iodide) was diluted 25-fold in water. Then 150 lL of the dye solution was directly spread on the colony surface, a cover glass was deposited on top, and agar was cut off along the cover glass sides and laid on a slide. Excitation was

performed using the argon laser line at 458 nm. Live (green) SYTO9-stained cells (475 nm long-pass filter for emission) and dead (red) propidium iodide-stained cells (570 nm long-pass filter for emission) were visualized with a Zeiss LSM510 confocal scanning laser device mounted on a Zeiss Axiovert200 M microscope (Carl Zeiss). The objective used was an oil immersion plan-apochromat lens (633, NA ¼ 1.4). Confocal images of green (SYTO 9) and red (propidium iodide) fluorescence were simultaneously collected for 2 s. Fiber Evanescent Wave Spectroscopy. An optical fiber sensor operating in the MIR domain was developed. It utilizes an optical fiber made of chalcogenide glass. The composition of the glass is Te2As3Se5, called TAS glass,38 in the chalcogenide family. This glass composition exhibits an excellent resistance to devitrification and a good chemical durability, especially against water and solvent corrosion,39 induces no protein denaturation and does not interfere with cell division and communication.40 The fiber spectral window encompasses the MIR spectral domain, with transparency from 4000 cm1 to 800 cm1. To improve sensitivity, the diameter of the fiber was locally reduced to create a tapered sensing zone, which was in contact with the sample to be analyzed.39 The experimental setup consisted of a Vector 22 FTIR spectrometer (Brucker Co.), coupled with a tapered fiber, and a HgCdTe detector (Fig. 1B). A single fiber was used both as a waveguide and sensing element. In this experimental configuration, the evanescent wave penetration depth allowed probing of only the first 2–3 lm into the sample. In the experiments presented here, the length of fiber in contact with the colony was about 9 6 1 mm. Spectral resolution was set to 4 cm1 and spectra resulted from the accumulation of 100 scans. The U-turn of the fiber-sensing zone lay in contact with the agar, and spectra were collected every 15 min before, during, and after the colony’s edge passed the fiber (Fig. 2). At the end of the monitoring period, the colony radius was roughly doubled. Data Treatment. Water Vapor Subtraction. Before further treatment, potential water vapor contribution to IR spectra was carefully subtracted by adjusting the subtraction factor for zeroing the 3853 cm1 water vapor band relative to a baseline ranging from 3859 to 3848 cm1. Spectra Normalization. Normalization is used to correct spectra for variations of the amount of biomass put into contact


TABLE I. Assignment of the main spectral differences between the SWAR and STAC phenotypes (see Fig. 4). Most assignments are from Refs. 48–50. Wavenumber (cm1)

Vibrational mode

Macromolecules

1700

m(C¼O)

1652 1625 1602 1514

m(C¼O) mas m(C¼C), m(C–N)

1485 1442 1407 1376 1355 1317 1242

das(CH3)3Nþ m(N¼N) ds(CH3)3Nþ ds(CH3)

Lipids (ester bond) and carboxylate Protein amide I Polysaccharides Proteins essentially Aromatic ring from amino acids Membrane lipids

1163 1116 1086 1023 991 968 ,900

m(C¼O) ms(PO2) ms(PO2)

Lipids Ring vibration

m(S¼O) mas(PO2)

mas(CH3)3Nþ

Phosphodiester from phospholipids and nucleic acids Polysaccharides Polysaccharides Polysaccharides Sugar ring modes Sugar ring modes Lipids ‘‘Fingerprint region’’

with the crystal surface. Accordingly, spectra were normalized on their absorbencies at 1465 cm1 (dCH2 bending mode) after zeroing the spectrum baseline between 2000 and 1800 cm1. This band is usually chosen as an internal standard since its ‘‘concentration’’ is roughly constant in most biomolecules. Discriminant Analysis. The primary purpose of discriminant analysis is to classify unknown samples into well-defined groups or categories based on a training set.41,42 This makes it possible to determine unambiguously the identity (or quality) of an unknown sample. In this study, the discriminant analysis was performed assuming that the reference groups had the same prior probabilities and that the discriminant variables (here absorbencies at different wavenumbers) were Gaussian. The distance of the sample spectrum to the spectra of the three reference classes (stationary cells, swarmers, and swimmers) was computed as described in the Appendix.

RESULTS Bacteria Infrared Microscopic Data. For this study, P. mirabilis colonies, or pelleted planktonic bacteria, were put into contact with a microscope-mounted germanium (Ge) micro-crystal, from whose surface the IR beam ([ ’ 50 lm) was reflected (see Fig. 1A). After drying, absorption spectra were collected in the ATR mode through a sample depth of 1–2 lm, depending on the wavelength. This depth penetration is analogous to the optical pathway in the conventional transmission mode. Only colony surface bacteria are analyzed, and the agar substrate does not influence the measurement. Phenotypic Spectral Discrimination. A previous study31 showed that swimming planktonic and four-hour-old swarmer cells were discriminated by using IR spectral markers characteristic of the two phenotypes. The present study aims to map a typical P. mirabilis colony periodically growing through alternate swarming and consolidation periods. To achieve such a task, it was necessary to apply chemometrics


FIG. 5. 2D plot of P. mirabilis phenotypes as resolved from spectral pattern recognition analysis. The plot features Mahalanobis distances between each reference spectrum (a point) and the barycenter of a particular cluster. The Mahalanobis distance is closely related to the euclidian distance but takes into account distinct variances. For clarity, a two 2D plot, instead of a 3D plot, is presented. The differentiation index presented in the text is calculated from the barycenters of the SWAR and STAC clusters shown in the right panel. ( ) LIQC; (¤) STAC; and (m) SWAR.

(see Appendix) using statistical, non-supervised methods to classify unknown spectra according to known standards. Reference spectra (Fig. 3) were collected for (1) planktonic swimmer cells (LIQC), (2) surface swarmer cells (SWAR), and (3) stationary surface cells (STAC). LIQC spectra were acquired on cells grown in liquid culture medium after conventional centrifugation (3000 3 g, 5 min). SWAR spectra were collected at the edge of P. mirabilis colonies ([ ’ 60 mm) undergoing massive migration on petri plates. Finally, STAC spectra were collected at the inoculum site from the same colonies that were used for the SWAR population. Approximately 40 reference spectra were collected for each phenotype from at least nine independent cultures. These spectra constituted the reference data bank for subsequent identification of ‘‘unknown’’ samples. Figure 3 displays the normalized mean spectra of LIQC, SWAR, and STAC cells along with their respective variance. The 1500–1000 cm1 frequency domain clearly exhibits pattern differences. Whereas mean spectra (top panel) allows the identification of spectral domains where inter-group variance is high, the corresponding frequencies being thus discriminative, the variances spectra (bottom panel) show the domains where intra-group variance is high. Particularly, it is clear that within a particular group, absorbencies in the ‘‘protein’’ (1600–1700 cm1) and the O–H stretching mode (2800–3600 cm1) domains vary too much to be useful for spectral discrimination between the distinct P. mirabilis phenotypes. Since the first aim of the present study was to map STAC and SWAR phenotypes, discriminant frequency domains were semi-empirically deduced from the SWAR

FIG. 6. Space distribution of the SWAR and STAC P. mirabilis phenotypes as revealed by fluorescence confocal microscopy and spectral pattern recognition analysis performed on growing colonies by IR microspectroscopy. The differentiation index I (bottom panel) reflects the spatial evolution from stationary (STAC) cells (I ! 0) towards swarming (SWAR) cells (I ! 1). The line is drawn to guide the eye. A scheme of the colony, featuring the inoculum (blue), the terraces (100 to 600 lm width), and the migration front (red), is presented as space markers to be related to cell length (vertical numbers) and differentiation index. Confocal microscopy imaging is also presented (top panel) to show P. mirabilis phenotypes. The biofilm has been dyed with the LIVE/DEAD fluorescent kit: living bacteria appear green, dead ones, red. (A, B, C, and D) frames (73 lm 3 73 lm) correspond to locations noted on the colony scheme. Cell lengths are the average and intrasample variance of 25 counts.

minus STAC difference spectrum. This difference spectrum (see Fig. 4 and Table I) shows the main discrepancies between the two phenotypes in the 1800–800 cm1 frequency range. In addition, the STAC and LIQC phenotypes are not identical. The main differences localize to frequency domains where polysaccharides dominate the IR spectrum. Differences in lipids composition also show up. Thus, a significant metabolic switch accompanies the transition between STAC and SWAR phenotypes. It should be emphasized that spectra are de-

termined by the most abundant molecular species in the sample. Thus, lipid absorption bands mostly originate from membrane phospholipids, and sugar bands mainly reflect EPS and membrane LPS. Since the discriminant analysis simultaneously evaluates the spectra at several wavelengths (here, sixty distinct values), the spectral patterns can be unambiguously discriminated despite significant variance at any one wavelength. We semiempirically selected eleven discriminating frequency domains


FIG. 7. Time distribution of the SWAR and STAC P. mirabilis phenotypes as revealed by pattern recognition analysis performed on actively growing colonies by IR microspectroscopy. The micro-crystal was used to sample, as a function of time, the differentiation process at a fixed location. The differentiation index I is plotted as a function of time and distance from the colony edge, this being initially nul. The line is drawn to guide the eye.

(see Fig. 4) on the basis of low intra-sample and large intersample variances. The frequency domains are: 1702–1712, 1440–1444, 1409–1413, 1397–1401, 1369–1373, 1314–1318, 1235–1239, 1116–1120, 1084–1088, 1021–1025, and 1161– 1165 cm1. The spectral dimensionality was thereby reduced from 3000 (3800–800 cm1) to 60 (sum of frequencies of the eleven domains). Figure 5 displays the cluster map in which each IR spectrum was reduced to a point in a complex space and positioned according to its Mahalanobis distance to the phenotype cluster barycenters (see Experimental Procedure and Appendix). For the purpose of clarity, the plot represents a twodimensional (2D) space, though the space is actually threedimensional (3D) since three phenotypes are considered. Note that all three spectral/phenotypic clusters are rigorously distinct and do not overlap. Space Mapping of Swarming and Stationary Cells. The acquisition of discrete SWAR and STAC reference spectra made it possible to investigate the spatial distribution of swarmer and stationary cells within a currently growing P. mirabilis colony. For each IR spectrum, a score index I was calculated to describe which phenotype (either SWAR or STAC) was dominant at that particular location. A low value (close to 0) reflects a pure STAC population, whereas a high value (close to 1) reflects a pure SWAR population. Figure 6 displays the evolution of I as a function of the distance from the inoculation site and cell size distribution snapshots for a typical experiment. Microscopic observations of cell length and motility showed that I values from 0–0.6 reflected the STAC phenotype, while a mixed STAC/SWAR population is observed in the I ¼ 0.6–0.8 range. At I . 0.8, the SWAR phenotype dominates. Short (,5 lm) stationary cells were prominent in the inner terraces, whereas cells as long as 40 lm dominate at the periphery. These two populations are separated by a mixed cell population in which cell lengths


FIG. 8. Mean (A) and standard deviation (B) of SWAR and STAC P. mirabilis IR spectra as observed by using the optical fiber sensor in the FEWS mode. The optical fiber was merely put into contact with the colony to acquire typical SWAR and STAC spectra. The corresponding reference spectra have been used for pattern recognition. The (C) panel shows the time evolution of the differentiation index as shown in Fig. 7. The fiber was initially positioned parallel to the migration front and successive IR spectra were collected as a function of time while the front reachs and overpasses the optrode. The line is drawn to guide the eye.

of about 18 6 5 lm (n ¼ 25) are observed within a particular colony. In this domain, very low motility (as deduced from video recordings in phase contrast microscopy), if any, was observed. The data in Fig. 6 thus confirm that the SWAR population is confined to a restricted area at the very edge of the colony. Time Mapping of Swarming and Stationary Cells. To monitor the swarming process in real time within a growing colony, IR spectra were collected periodically at a fixed location. Typically, the Ge micro-crystal was put into contact with the petri dish surface 20 mm from the inoculation site. Initially this position was outside the colony edge, and spectra were subsequently collected at the same radius (but at distinct locations displaced clockwise around the center) until the colony radius had roughly doubled to 20 mm. Thus, the spectra sequentially correspond to the colony edge, the active swarming domain, and later to dedifferentiation zones as the migration wave passed by. Figure 7 displays the evolution of the differentiation index I as a function of time and distance from the colony edge. It is clear that swarmer cells initially detected (I ’ 0.9) progressively undergo dedifferentiation as the cell population expands. Both ‘‘space’’ and ‘‘time’’ IR samplings and motility observations show that swarmer cells

are only present in high proportions at the edge and that cells inside the colony do not undergo re-differentiation. These results match what was previously observed by Matsuyama and co-workers.24 Design of an Optical Fiber Sensor for Detection and Monitoring of Biofilms. To meet the challenge of deriving both qualitative and quantitative information throughout the course of biofilm development, we designed a remote optrode based on the FEWS concept. The IR-ATR sensor is essentially an optical fiber made of chalcogenide TAS glass37 coupled to a conventional FT-IR spectroscope from Bruker (Fig. 1B). To validate the use of this IR-ATR sensor for biofilm studies, we carried out a pattern recognition method and time map of P. mirabilis swarm colony development under the same conditions as used for the IR microscope setup. For this, FEWS reference spectra were collected by placing the U-turn of the fiber (about 9 mm in contact with the cell surface population) either at the inoculum or at the periphery of growing colonies, with a diameter of 50–60 mm. Petri plates were thermostated at 37 8C during spectral acquisitions, and the fiber U-turn was placed in contact with the agar medium through a narrow slit in the cover plate to maintain humidity and temperature. Figures 8A and 8B display the mean SWAR and STAC FEWS spectra along with the SWAR-minus-STAC difference spectrum. This shows that FEWS spectra are of the same quality as those collected in the micro-ATR mode. By using these spectra as standards, we resolved the time distribution of both SWAR and STAC. For this purpose, the fiber-sensing zone was continuously put in contact with the agar slightly beyond the colony edge, and spectra were collected periodically during swarming (Fig. 2). Phenotype discrimination was achieved using the eleven frequency domains previously selected. The FEWS time map displayed in Fig. 8C clearly matches the ATR microspectroscopic results, showing that swarmer cells quickly dedifferentiate once they are no longer located at the migration front, the swarmer population being restricted to a 3–5 mm annulus around the colony periphery. This corresponds to the outermost (expanding) terrace plus the external, peripheral translucent zone. As opposed to ATR microscope experiments, in which the micro-crystal must be cleaned between measurements, the FEWS optical IR-ATR sensor continuously monitors the colony since it remains in contact with the colony throughout the expansion process.

DISCUSSION We have demonstrated how IR microspectroscopies can be used to resolve the spatio-temporal distribution of STAC and SWAR phenotypes in a growing P. mirabilis colony. We answered questions about the potential contribution of ‘‘inner’’ terraces to colony expansion25,43 by showing that the SWAR phenotype is only displayed at the colony edge. In contrast to the macroscopic pattern of successive terraces separated by structurally distinct zones, we observed the cell population within the colony to be spectrally homogeneous (Figs. 6 and 7). Hence, colony expansion mainly relies on a moving ‘‘annulus’’ at the leading edge where swarmer populations are located. Interestingly, cell motility, as observed from video recordings with phase contrast microscopy, is maximal close to the periphery, with individual cells exhibiting velocities up to 30 lm/s (to be compared to a whole-colony radial expansion rate of ’1 lm/s). This cell motility rapidly decreases a few millimeters from the periphery and completely vanishes in

inner domains. Hence, the spectroscopic SWAR phenotype and cell mobility are correlated in space. These results match the detailed study of Matsuyama and co-workers.24 In this study, the IR microscope was mostly used as an accurate pointer device. This perfectly matches the requirement for studying P. mirabilis dynamics. However, novel multichannel detectors based on photodiode arrays (PDA) allow for much faster 2Dmapping, this innovation being optimized for less stable biological samples.44 Interest in biofilm monitoring extends far beyond the study of P. mirabilis swarming. Biomedical and food safety are among the areas where real-time monitoring of latent or actively growing biofilms is critical. Accordingly, an optical IR-ATR sensor was developed to merge the potential of IR spectroscopy with the versatility of optical fibers. A dedicated ‘‘optrode’’ was used for time mapping of the P. mirabilis swarming process. Keeping the optrode continuously in contact with the substratum, we monitored in situ the colony expansion and the differentiation processes. Thus, the optrode is sensitive enough to not only detect the biofilm onset but also phenotypic changes within a particular population. Whereas IR microscopy is suitable for research, the optrode facilitates applications of IR spectroscopy on virtually any cell surface population. This fiber would be even more efficient if integrated beforehand into the structure susceptible to bacterial colonization, e.g., a catheter or a water pipe. Clearly, IR spectroscopy and optical fibers can contribute to metabolic imaging and early diagnostic approaches in the medical and industrial domains as well as in the laboratory. ACKNOWLEDGMENTS The authors are greatly indebted in Dr. James Shapiro for fruitful discussions and critical manuscript reading. This work was supported by the French De´le´gation Ge´ne´rale aux Armements and by the Re´gion Bretagne. 1. D. Naumann, C. P. Schultz, and D. Helm, in Infrared Spectroscopy of Biomolecules (Wiley-Liss, New York, 1996). 2. M. Jackson and H. H. Mantsch, in Infrared Spectroscopy of Biomolecules (Wiley-Liss, New York, 1996). 3. P. Ro¨sch, M. Harz, M. Schmitt, K.-D. Peschke, O. Ronneberger, H. Burkhardt, H.-W. Motzkus, M. Lankers, S. Hofer, H. Thiele, and J. Popp, Appl. Environ. Microbiol. 71, 1626 (2005). 4. N. J. Harrick, in Internal Reflection Spectroscopy (John Wiley and Sons, New York, 1967). 5. U. P. Fringeli and H. H. Gunthard, ‘‘Infrared Membrane Spectroscopy’’, in Membrane Spectroscopy, E. Grell, Ed. (Springer-Verlag, Berlin, 1981), pp. 270–332. 6. E. Goormaghtigh and J. M. Ruysschaeret, in Molecular Description of Biological Membranes by Computer Aided Conformational Analysis (CRC Press, Boca Raton, FL, 1990), vol. 1, p. 285. 7. R. Krska, E. Rosenberg, K. Taga, R. Kellner, A. Messica, and A. Katzir, Appl. Phys. Lett. 61, 1778 (1992). 8. R. Goebel, R. Krska, R. Kellner, J. Kastner, A. Lambrecht, M. Tacke, and A. Katzir, Appl. Spectrosc. 49, 1174 (1995). 9. R. Simhi, Y. Gotshal, D. Bunimovich, B. Sela, and A. Katzir, Appl. Opt. 35, 3421 (1996). 10. M. Jakusch, B. Mizaikoff, R. Kellner, and A. Katzir, Sens. Actuators, B 38, 83 (1997). 11. W. R. Moser, J. R. Berard, P. J. Melling, and R. J. Burger, Appl. Spectrosc. 46, 1105 (1992). 12. K. Taga, R. Kellner, U. Kainz, and U. B. Sleytr, Anal. Chem. 66, 35 (1994). 13. J. S. Sanghera, F. H. Kung, P. C. Pureza, V. Q. Nguyen, R. E. Miklos, and I. D. Aggarwal, Appl. Opt. 33, 6315 (1994). 14. J. S. Sanghera, F. H. Kung, L. E. Busse, P. C. Pureza, and I. D. Aggarwal, J. Am. Ceram. Soc. 78, 2198 (1995). 15. C. Blanchetiere, K. Le Foulgoc, H. L. Ma, X. H. Zhang, and J. Lucas, J. Non-Cryst. Solids 184, 200 (1995).


16. K. Foulgoc, L. Le Neindre, X. H. Zhang, and J. Lucas, Proc. SPIE-Int. Soc. Opt. Eng. 2836, 26 (1996). 17. S. Hocde, L. Le Neindre, K. Le Foulgoc, C. Boussard, P. Le Roux, X. H. Zhang, and J. Lucas, Proc. SPIE-Int. Soc. Opt. Eng. 3262, 144 (1998). 18. K. Li and J. Meichsner, J. Phys. D: Appl. Phys. 34, 1318 (2001). 19. J. S. Sanghera, L. Shaw, A. Brandon, and D. Ishwar, C. R. Chimie 5, 873 (2002). 20. M. Dworkin, in Bacteria as Multicellular Organisms (Oxford University Press, Oxford, 1997), p. 3. 21. J. A. Shapiro, in Bacteria as Multicellular Organisms (Oxford University Press, Oxford, 1997), p. 14. 22. C. Allison and C. Hughes, Mol. Microbiol. 5, 1975 (1991). 23. G. M. Fraser and C. Hughes, Curr. Opin. Microbiol. 2, 630 (1999). 24. T. Matsuyama, Y. Takagi, Y. Nakagawa, H. Itoh, J. Wakita, and M. Matsushita, J. Bacteriol. 182, 385 (2000). 25. O. Rauprich, M. Matsushita, C. J. Weijer, F. Siegert, S. E. Esipov, and J. A. Shapiro, J. Bacteriol. 178, 6525 (1996). 26. J. A Shapiro, Bioessays 17, 597 (1995). 27. J. A. Shapiro, Annu. Rev. Microbiol. 52, 81 (1998). 28. P. D. Nichols, J. M. Henson, J. B. Guckert, D. E. Nivens, and D. C. White, J. Microbiol. Methods 4, 79 (1985). 29. D. E. Nivens, J. Q. Chambers, T. R. Anderson, A. Tunlid, J. Smit, and D. C. White, J. Microbiol. Methods 17, 199 (1993). 30. D. E. Nivens, J. Schmitt, J. Sniatecki, T. R. Anderson, J. Q. Chambers, and D. C. White, Appl. Spectrosc. 47, 668 (1993). 31. M. Gue´, V. Dupont, A. Dufour, and O. Sire, Biochemistry 40, 11938 (2001). 32. C. R. Burch, Proc. Phys. Soc., London 59, 41 (1947). 33. R. Barer, A. R. H. Cole, and H. W. Thompson, Nature (London) 163, 198 (1949). 34. E. R. Blout, G. R. Bird, and D. S. Grey, J. Opt. Soc. Am. 40, 304 (1950). 35. E. R. Blout and G. R. Bird, J. Opt. Soc. Am. 41, 547 (1951). 36. E. R. Blout and M. J. Abbate, J. Opt. Soc. Am. 45, 1028 (1955). 37. D. Le Coq, K. Michel, J. Keirsse, C. Boussard-Ple´del, G. Fonteneau, B. Bureau, J. M. Le Que´re´, O. Sire, and J. Lucas, C. R. Chimie 5, 907 (2002). 38. S. Hocde´, C. Boussard-Ple´del, G. Fonteneau, and J. Lucas, Solid State Sci. 3, 279 (2001). 39. S. Hocde´, C. Boussard-Ple´del, G. Fonteneau, D. Lecoq, H. L. Ma, and J. Lucas, J. Non-Cryst. Solids 274, 17 (2000). 40. J. Keirsse, C. Boussard-Ple´del, O. Lore´al, O. Sire, B. Bureau, P. Leroyer, B. Turlin, and J. Lucas, Vib. Spectrosc. 32, 23 (2003). 41. M. F. Devaux, D. Bertrand, P. Robert, and M. Qannari, Appl. Spectrosc. 42, 1015 (1988). 42. I. T. Joliffe, Principal Component Analysis (Springer-Verlag, New York, 1986). 43. K. A. Bisset and W. I. Douglas, J. Med. Microbiol. 9, 229 (1976). 44. R. Bhargava and I. Levin, in Spectrochemical Analysis Using Infrared Multichannel Detectors (Sheffield Analytical Chemistry Series, Blackwell Publishing, 2005). 45. R. G. Brereton, in Chemometrics Data Analysis for the Laboratory and Chemical Plant (John Wiley and Sons, New York, 2002). 46. H. L. Mark, Anal. Chem. 59, 790 (1987). 47. R. O. Duda, P. E. Hart, and D. G. Stork, Pattern Classification (John Wiley and Sons, New York, 2000), 2nd ed.


48. K. Brandenburg and U. Seydel, in Infrared Spectroscopy of Biomolecules (Wiley-Liss, New York, 1996), p. 203. 49. R. N. A. H. Lewis and R. N. McElhaney, in Infrared Spectroscopy of Biomolecules (Wiley-Liss, New York, 1996), p. 159. 50. B. Stuart, in Biological Applications of Infrared Spectroscopy (John Wiley and Sons, New York, 1997).

APPENDIX Discriminant Analysis. An unknown sample is attached to the nearest group in the sense of the Mahalanobis distance.45 Briefly, Mahalanobis distance46 is a multidimensional distance D defined by the matrix equation D2 ¼ ðS SÞ 0 R1 ðS SÞ where S is a vector consisting of spectral energy at different selected wavenumbers (this vector describes the position in multidimensional space corresponding to the spectrum of a given sample), S is a vector describing the position of a reference point in space (e.g., the position of a known material), and R is the pooled covariance matrix describing distance measures in the multidimensional space of interest. Since some groups of data may be spread more than others, the Mahalanobis distance has the desirable property that it is based on the spread of the data in multidimensional space. The wavenumbers retained for discriminant analysis are those with significant differences between the classes. They are listed in the Phenotypic spectral discrimination section. A score (or normalized index) Ii is constructed to compare two classes (for instance, STAC and SWAR). According to Bayesian decision theory,47 which is a generalization of the discriminant analysis described above, this index may be defined by the a posteriori probability of membership in the STAC group as follows: Ii ¼

1 1 2 1 þ exp 2 ðdp di2 Þ

where dp (resp. di) is the Mahalanobis distance between the sample and the mean of the SWAR class (resp. STAC). This definition induces the minimal risk of misclassification. The index I decreases into [0, 1] when di or dp increases. It is equal to 1 if di is equal to 0, it is equal to 0 if di tends to infinity, and it is equal to ½ if di2 ¼ dp2 . The score Ii should be read as the probability that the membership belongs to the SWAR group.