D

Ma

b

a

ARR2AA

KNIBFL

1

srdcteav2ciep(cc

PT

0d

Biosensors and Bioelectronics 33 (2012) 172– 178

Contents lists available at SciVerse ScienceDirect

Biosensors and Bioelectronics

jou rn al h om epa ge: www.elsev ier .com/ locate /b ios

esign and field application of a UV-LED based optical fiber biofilm sensor

atthias Fischera, Martin Wahla, Gernot Friedrichsb,∗

GEOMAR | Helmholtz Centre for Ocean Research Kiel, Kiel, GermanyInstitute of Physical Chemistry, Christian-Albrechts-University Kiel, Kiel, Germany

r t i c l e i n f o

rticle history:eceived 11 November 2011eceived in revised form2 December 2011ccepted 25 December 2011vailable online 3 January 2012

eywords:atural biofilm

ntrinsic tryptophan fluorescenceiofilm formation dynamics

a b s t r a c t

Detecting changes in the formation dynamics of biofilms stemming from bacteria and unicellular microor-ganisms in their natural environment is of prime interest for biological, ecological as well as anti-foulingtechnology research. We developed a robust optical fiber-based biofilm sensor ready to be applied innatural aquatic environments for on-line, in situ and non-destructive monitoring of large-area biofilms.The device is based on the detection of the natural fluorescence of microorganisms constituting thebiofilm. Basically, the intrinsic fluorescence of the amino acid tryptophan is excited at a wavelength of� = 280 nm and detected at � = 350 nm utilising a numerically optimized sensor head equipped with aUV-LED light source and optical fiber bundles for efficient fluorescence light collection. Calibration wascarried out with tryptophan solutions and two characteristic marine bacteria strains revealing linear sig-nal response, satisfactory background suppression, wide dynamic range, and an experimental detection

3 2

iber optical sensorED light source

limit of 4 × 10 cells/cm . Successful field experiments in the Baltic Sea accomplished over a period oftwenty-one days provided for the first time continuous observation of biofilm formation dynamics ina natural habitat. Starting from the first adhering bacteria, the measurement yielded the characteristicthree phases of biofilm formation up to a fully developed biofilm. The sensor system holds potential forapplications in aquatic sciences including deep sea research and, after further miniaturisation, in the

l field

industrial and biomedica

. Introduction

In natural and artificial aqueous environments such as lakes,treams, oceans, or technical water systems, surfaces becomeapidly covered by microorganisms. Following a biochemical con-itioning phase, the biofilm formation starts when pioneer bacteriaells adhere to a surface. The ability of bacterial cells to attacho a surface is controlled by environmental factors such as nutri-nt level, pH, and temperature (Costerton et al., 1995; Daveynd O’Toole, 2000). In natural ecosystems they enable or pre-ent further colonisation by micro- and macroorganisms (Harder,009; Wahl, 1989). Biofilms can mature to highly complex celllusters formed by bacteria and multicellular eukaryotic organ-sms and reach a thickness of up to 300 �m after 25 days (Battint al., 2003). Also in industrial applications such as bioreactors oripelines, biofilms have advantageous as well as detrimental effects

Flemming et al., 1998). In a medical context, biofilms are a cru-ial factor in provoking chronic diseases and bacterial infections inatheterised applications or clinical water systems (Stickler, 2008).

∗ Corresponding author at: Christian-Albrechts-Universität Kiel, Institut fürhysikalische Chemie, Olshausenstr. 40, 24098 Kiel, Germany.el.: +49 431 880 2749; fax: +49 431 880 1704.

E-mail address: friedrichs@phc.uni-kiel.de (G. Friedrichs).

956-5663/$ – see front matter © 2012 Elsevier B.V. All rights reserved.oi:10.1016/j.bios.2011.12.048

.© 2012 Elsevier B.V. All rights reserved.

Ideally, biofilm sensors can operate continuously, in situ and with-out mutual interference with the microbial community. For onlinebiofilm monitoring applications, several microscopic, spectro-scopic, fiber-optical, electrochemical, and piezoelectric techniqueshave been developed that assess diverse biofilm characteristics(Denkhaus et al., 2006; Flemming, 2003; Janknecht and Melo, 2003;Nivens et al., 1995; Wolf et al., 2002). Though many of these tech-niques provide excellent sensitivity and selectivity, owing to theirtechnical limitation or complexity, the majority of methods are onlyapplicable in the laboratory rather than in field studies or otherharsh environments. Continuous monitoring of biofilm formationdynamics in the field has to meet challenges which differ fromthose for highly sophisticated laboratory instrumentation. In gen-eral, biofilm field sensors call for a robust and reliable measuringtechnique, e.g., being immune to vibrations or fluctuating tempera-ture and humidity conditions. More specifically, the sensor conceptshould allow for non-destructive detection in order to facilitateon-line measurements, selective detection of the biofilm to dis-tinguish between organic and inorganic material on the surface,background suppression of signal attributable to organic mate-rial in the overlaying bulk water, and sufficient penetration depth

accounting for the biofilms three-dimensional structure. Moreover,the capability to probe large substrate surfaces allows compensat-ing for the inhomogeneous settlement characteristics of biofilmswith highly patchy cell clusters of several hundred micrometer

d Bioelectronics 33 (2012) 172– 178 173

ddtFtr

neNtsiaDaon2acawtlltedflTsatl

obgiUbbtits

2

2

sdIltfooisbUF

Fig. 1. Spectroscopic layout of the biofilm sensor. The combination of a narrow

M. Fischer et al. / Biosensors an

iameter (Dalton et al., 1996; Stoodley et al., 2001). Often, a wideynamic range is required to quantify the entire range from ini-ially adsorbed bacteria cells up to complex biofilm communities.inally, autonomous operation with sufficiently long operationalime, rapid and easy signal acquisition are likewise desirable crite-ia.

The stated requirements are best met by optical detection tech-iques such as time-resolved photoacoustic spectroscopy (Schmidt al., 2002) and optical fiber based fluorimetry (Haisch andiessner, 2007). In particular, the high sensitivity, fast response

ime, and the capability of monitoring large areas in situ withoutample contact make fluorescence methods attractive. Monitor-ng of biomolecules using intracellular fluorophores, for instancemino and nucleic acids, has long been known (Arrage et al., 1995;etermann et al., 1998; Sohn et al., 2009; Tartakovsky, 1996). Suit-ble sensors have been proven useful for quantitative detectionf biomass and cellular activity (Angell et al., 1993) and yield sig-als that correlate with cell number and bacteria growth (Ji et al.,004). The most commonly used intrinsic fluorophores in bacteriabsorb light at wavelengths in the mid- or near-UV range. Typi-ally, protein detection is based on fluorescence excitation of theromatic amino acids phenylalanine, tyrosine, and tryptophan atavelengths about 280 nm. Due to a high absorption cross sec-

ion and quantum yield, often tryptophan fluorescence at 350 nmargely dominates the overall signal. Gas discharge and fluorescentamps as well as lasers have been widely applied as adequate excita-ion light sources (Angell et al., 1993; Beyenal et al., 2004; Bonnint al., 2007; Estes et al., 2003). More recently, light-emitting UViodes (UV-LEDs) have been utilised for studying intrinsic proteinuorescence as well (Barbieri et al., 2005; McGuinness et al., 2005).heir key advantages for field applications are the low power con-umption, the narrow bandwidth spectral emission characteristics,nd low cost. Moreover, UV-LEDs can be switched on and off instan-aneously within several milliseconds yielding highly reproducibleight intensities.

Along those lines, the objective of this work was the devel-pment of a novel on-line, non-invasive in situ optical sensorased on intrinsic protein fluorescence for monitoring biofilmrowth dynamics in the field. The sensor concept relies on (i) back-lluminating a biofilm establishing on a transparent substrate by aV-LED and (ii) fluorescence collection by means of a large num-er of optical fibers. Detailed simulations of the optical setup haveeen performed in order to optimize the spatial arrangement inerms of illumination and collection efficiency. Prior to field testsn the Baltic Sea, the sensor prototype has been calibrated by dilu-ion series of tryptophan and artificial biofilms of marine bacterialtrains.

. Experimental

.1. Sensor layout

A prototype instrument has been assembled comprising theensor unit, the electronics, and a waterproof housing deployableown to 50 m water depth (see photograph in the Supplemental

nformation). The sensor unit (5 cm in diameter and 25 cm inength) consisted of a photomultiplier tube (PMT) for light detec-ion and a patent-pending sensor head including the substrateor biofilm establishment, the light source and the collecting fiberptics. A UV transparent culture dish or a quartz plate with anptional disposable foil for optimized bacteria growth attached tot (Greiner Bio-One, Germany, Lumox film 25) was used as settling

ubstrate. The intrinsic fluorescence of the biofilm was excited byack-illumination through the substrate using a continuous waveV-LED (Sensor Electronic Technology, USA, UVTOP275-TO39-W). The 120◦ angle of aperture resulted in a uniform and large

bandwidth UV-LED with interference filters facilitates selective and sensitive detec-tion of biofilm emission mainly stemming from the intrinsic fluorescence of theamino acid tryptophan.

excitation area of approximately 0.5 cm2.The normalized spectralemission characteristics of the UV-LED, the transmission curves ofthe employed optical filters as well as the absorption and fluores-cence spectrum of tryptophan are illustrated in Fig. 1. The outputwavelength of the 280 nm UV-LED with a narrow spectral band-width of 14 nm full width at half maximum (FWHM) was chosen tocoincide with the absorption maximum of tryptophan. In order tosuppress interfering residual red-shifted emission, which is inher-ent to state-of-the-art UV-LED technology, a 280 nm band passfilter (Semrock, USA, BrightLine HC 280 ± 15 nm) was placed infront of the LED. Typically, the biofilm was exposed to an opti-cal output power of about 600 �W. Possible DNA damage of themicrobial community (Elasri and Miller, 1999) was excluded byswitching off the UV-LED between the measurements resultingin short exposure times of several tens of milliseconds. Repro-ducible UV excitation intensities have been assured by using anLED driver with a temperature compensated high-precision con-stant current source with a precision of 0.02%/V. Due to ageingprocesses of the UV-LED, the excitation intensity slowly decreasedover time. According to the data sheet, the typical lifetime of theUV-LED was about one hundred hours operation time. For high-precision or long-term studies, the resulting bias in fluorescenceintensity would have to be corrected by monitoring the UV out-put or by regularly measuring calibration standards. However, inthis study, due to the short duty cycle of the LED during the mea-surements, the degradation of the light source was negligible anda corresponding correction was not necessary. The emitted flu-orescence light of the biofilm was collected and guided by 540fused silica optical fibers (custom-made, Biolitec, Germany) to thedetector. Each individual fiber was a 15 cm long multi-mode, step-index fiber with 185 �m core diameter, a numerical aperture ofNA = 0.26, and a spectral response within the a limited wavelengthrange of 350 nm < � < 2.5 �m. 30 fibers each were placed in eigh-teen bundles that were arranged hemispherically and in two ringsaround the LED. At the end of the combined fiber bundles, thecollected fluorescence light was spectrally separated and discrim-inated from the incident and the reflected or scattered excitationlight by a combination of two interference filters in front of thedetector. These filters were centered at the maximum of the tryp-tophan fluorescence peak at 350 nm (AHF, Germany, H 350 ± 25 nm

and Asahi Spectra, USA, 350 ± 5 nm FWHM). Finally, a PMT operat-ing in single-photon counting mode (Hamamatsu Photonics, Japan,H9319) was used for sensitive light detection. The PMT was setin complete darkness before running a measurement to account

1 d Bioe

ftcot1cl

2

se1pctlHotwf(tba

B

a

cetp

3

pp

Fpsi

74 M. Fischer et al. / Biosensors an

or dark counts. Overall, the combined spectral characteristics ofhe optical components sufficiently suppressed background signalontributions, hence facilitating selective and sensitive detectionf biofilm emission. The timing of the measurements, the elec-ronics, the readout of the detector, and the data recording on a

GB SD memory card was accomplished by a programmable micro-ontroller making the sensor package ready for use as a field dataogger.

.2. Fluorescence collection efficiency

A crucial issue of a fiber based sensor is the optimization of thepatial arrangement of the entire optical system in terms of lightmission and collection performance (Bünting and Karlitschek,998; DiBerardino, 2002). Its overall performance is governed byarameters of the optical fibers as well as geometrical issues andan be specified as so-called collection efficiency (CE). Accordingo Bünting et al. (1999), the CE defines the amount of fluorescenceight detectable by a sensor as function of its setup parameters.owever, in our case the challenge is to detect a fluorescence signalf an extended planar surface without significantly deterioratinghe signal by potential background fluorescence from the bulkater. Consequently, next to the signal-to-noise ratio stemming

rom the total detected light intensity attributable to the biofilmS/N ∝

√Ibiofilm), a measure for Biofilm-CE (BCE) has to account for

he biofilm specificity, i.e., the ratio of the signal resulting from theiofilm (Ibiofilm) and the background signal from the bulk (Ibulk), justs well:

CE ∝√

Ibiofilm × Ibiofilm

Ibulk

Moreover, a uniform sensing of the biofilm area is important tollow for adequate averaging over its heterogeneous texture.

Instead of following experimental testing of various sensor headonfigurations, a promising sensor geometry was modelled andvaluated by using home-made FORTRAN software based on ray-racing methodology. Details of the simulations and optimizationrocedure will be discussed below.

. Material and methods

For sensor calibration experiments, dilution series of l-trypto-han (Carl Roth, Germany) dissolved in artificial seawater wererepared in the concentration range 5 × 10−9 M ≤ c ≤ 1 × 10−4 M.

ig. 2. Left: Schematic longitudinal cross-sectional view of the cylindrical sensor head indattern. Model parameters: a distance between biofilm and LED; b thickness of the optiubstrate; e side limit; f half radiation angle of LED; g numerical aperture of the fibers; hntensity profile (lower) through the fluorescence detection efficiency field at biofilm leve

lectronics 33 (2012) 172– 178

Additionally, model biofilms of two different strains of commonmarine bacteria were investigated, namely the marine gram-negative Pseudoalteromonas carrageenovora (DSM 6820) and thegram-positive Bacillus subtilis (DSM 4181). All microorganismswere obtained from the Leibniz Institute DSMZ–German Collec-tion of Microorganisms and Cell Cultures. Bacterial strains werecultured in marine nutrient (5 g peptone, 5 g yeast extract in 1 Lseawater) at 27 ◦C overnight. In the stationary phase, cells werecentrifuged at 2600 G for 5 min, re-suspended at different concen-tration levels in artificial seawater, and thereafter transferred to asterile UV transparent Lumox tissue culture dish (Greiner Bio-One,Germany). After a settling period of 2 h the substrate was rinsedwith sterile seawater and the fluorescence intensity of adheringbacteria was measured. For each data point the fluorescence signalon the PMT was integrated ten times over a period of 10 ms. All mea-surements have been performed in a dark box to suppress ambientlight contributions. To quantify the corresponding accumulated celldensity of the artificial biofilm, the bacteria were stained by thefluorescent dye DAPI (4′,6-diamidino-2-phenylindole dihydrochlo-ride, Sigma–Aldrich, Germany) for 10 min and then dip-rinsed indistilled water. Thereafter the removable substrate of the dishcontaining the attached bacteria was used for quantification bymicroscopic imaging. Twenty random images of the substrate werecaptured by epifluorescence microscopy (Carl Zeiss MicroImaging,Germany, Axio Scope.A1) and the cell numbers were counted by aprogram implemented in the software ImageJ (National InstituteHealth, USA, Abramoff et al., 2004). The software also determinedthe coverage area of bacteria by estimating the proportion of pixelsoriginating from labelled bacteria relative to the background.

4. Results and discussion

4.1. Simulation and optimization

The schematic cross-sectional view of the implemented cylin-drical sensor head design is illustrated in Fig. 2. Highlightedare the direct irradiation of the biofilm from underneath (back-illumination) and the configuration of the light collecting fiberbundles, which were arranged in two staggered circles aroundthe UV-LED. The geometry of the conically bored mount that

holds the light source, the fibers, and the substrate was essentiallydetermined by the angle of radiation of the UV-LED. The fiber con-figuration allows for a dense packing of the fiber bundles, a goodoverlap of the fiber acceptance cones with the volume excited by

icating simulation parameters and a color-coded fluorescence detection efficiencycal window; c, d tilt angles between respective fiber bundles and the plane of the

thickness of the fiber bundles. Right: Corresponding meridionial cut (upper) andl.

d Bioelectronics 33 (2012) 172– 178 175

tb

stttwtsscaorfldsaiabcopTfiatictdfl

itotacbMtadtopnafififattttc

wiorn

M. Fischer et al. / Biosensors an

he UV-LED, and a short distance between the fiber tips and theiofilm.

As the aim was to maximize the BCE value for a certain sen-or layout, the optimal combination of the setup parameters hado be identified. Model parameters outlined in Fig. 2 were the dis-ance a between biofilm and the light source, the thickness b ofhe optical window (determining the pressure resistance for deepater applications), the inclination angles c and d (between respec-

ive fiber bundles and the plane defined by the substrate), and theide limit e (constraining the width of the detection volume). Theimulation program calculated BCE according to a three-step pro-edure. First, a 3D excitation light field was numerically set up with

grid size down to 0.2 mm taking into account the radiation anglef the light source, its angular emission characteristics, and theefraction at the substrate window. Most important, the photonux per element of area decreases quadratically with the normalistance above the light source. Due to overall low absorption ofeawater at � = 280 nm (Wozniak and Dera, 2007), additional lightbsorption has not been taken into account. Secondly, a field spec-fying the collection efficiency of a spherically emitted photon atny point above the substrate has been calculated. The ray-tracingased algorithm accounted for the distance between the fluores-ence source and the fiber tip, Fresnel reflection and refractionccurring at the two sides of the substrate window, potential clip-ing of the light beam, and the numerical aperture of the fibers.hirdly, multiplication of the excitation and collection efficiencyelds resulted in a field specifying the overall detection efficiency of

(tryptophan) fluorophor. This detection efficiency field was usedo extract the desired Ibiofilm and Ibulk intensities by level-by-levelntegration over all grid points. In this way, the biofilm versus bulkontrast (biofilm contrast) could be easily obtained by comparinghe detection efficiency of the first level above the substrate to theetection efficiencies of the remaining levels representing the bulkuorescence.

A longitudinal cut through a typical detection efficiency fields shown in the left scheme of Fig. 2 as color-coded overlay. Notehat the region below the substrate does not contribute to the flu-rescence signal; it is only shown to illustrate the complexity ofhe simulated field. As expected for the inclined fiber arrangementssumed for Fig. 2, the collected fluorescence mainly stems from theenter region directly above the light source. Highest signal contri-utions are obtained from the biofilm at the surface of the substrate.oreover, as can be seen from the meridional cut illustrated in

he right scheme of Fig. 2, a uniform spatial detection efficiency ischieved at the level of the biofilm. The effective detection area,efined as the area of strongest fluorescence adding up to 95% ofhe total fluorescence, corresponds to 0.55 cm2. Of course, detailsf the detection efficiency field strongly rely on the geometricalarameters chosen for the sensor head. On the one hand, an alter-ative arrangement with fibers in parallel and close to the opticalxis would have yielded better overlap of the aperture cones of thebers with the excitation light field of the UV-LED. However, theber tips would have to be placed at a rather long distance away

rom the substrate to avoid shading effects of the excitation lightnd, more importantly, the biofilm contrast would be inferior. Onhe other hand, inclination angles close to 0◦ would have improvedhe biofilm contrast at the cost of fluorescence intensity losses dueo clipping effects, total reflection, and again long light paths. Fur-her images of detection light fields resulting from selected fiberonfigurations can be found in the Supplemental Information.

The numerical output of an optimization run for our prototypeith the two-ring arrangement is shown in Fig. 3. Here, normal-

zed BCE values are plotted as a function of the inclination anglesf the fiber bundles corresponding to the inner and outer ring,espectively. For this specific sensor head geometry, optimum incli-ations are about 30–45◦ for the inner ring and 25–40◦ for the outer

Fig. 3. Optimization run. Normalized BCE values for different combinations of fiberbundle inclination angles. The black cross marks the optimal fiber configuration.

ring. For the reasons outlined above and indicated by the low BCEvalues in Fig. 3, exceedingly low or high inclination angles are dis-advantageous. A 45◦/30◦ (inner/outer fiber ring) configuration withBCE = 0.9 was chosen for our prototype as a compromise in termsof optimization, possible bending radius of the fibers, and technicalpracticability in terms of water pressure resistance.

4.2. Laboratory performance

4.2.1. Dynamic rangeThe dynamic range of the sensor was determined by measuring

the fluorescence intensity of dilution series of tryptophan solutions.Linear correlations between fluorescence intensity and tryptophanconcentration with correlation coefficients of R2 ≈ 0.98 have beenfound. A calibration plot is given in the Supplementary Information.The dynamic range of the sensor covers at least four orders ofmagnitude enabling the quantitative detection of tryptophan fromthe nanomolar to the millimolar range. Typical concentrations ofdissolved free amino acids in fresh and seawater are in the nanomo-lar range (Mopper and Lindroth, 1982; Reynolds, 2003). Owingto the high sensitivity and with a fiber geometry optimized forbulk water detection, the presented probe might therefore proveuseful in studies of microbial activity as well since it overcomesthe temporal and spatial limitations of the usually utilised highpressure liquid chromatography method (Mopper and Lindroth,1982).

4.2.2. Biofilm specificityThe height dependence of the collected fluorescence has

been determined in order to obtain an evidence for the biofilmspecificity of the sensor head. For this purpose, small volumes ofa 10−7 M tryptophan solution were pipetted cumulatively into acuvette placed on the sensor substrate. As illustrated in Fig. 4, withincreasing liquid level the contribution of added solution to themeasured total fluorescence signal strongly decreases. The solidcurve corresponds to the predicted height dependence using thenumerical simulation program. Keeping in mind the simple exper-imental procedure, overall good agreement was found. Remainingdiscrepancies may arise from possible artefacts such as adsorption

or reflection effects in the experiments occurring especially at lowliquid levels. 50% of the maximum fluorescence signal level arisesfrom the first 3.4 mm above the substrate. For ideal backgroundfluorescence discrimination, which however is not feasible with

176 M. Fischer et al. / Biosensors and Bioelectronics 33 (2012) 172– 178

Fig. 4. Fluorescence signal of a 10−7 M tryptophan solution at different liquid levels.Cat

ttTubiwdlsc

4

flicibeDatf

bracsiusa

atmbflT12fit

Fig. 5. Double-logarithmic plots of the fluorescence intensity as function of (a) bac-teria cell density and (b) coverage area for artificial biofilms of bacteria B. subtilis

urves represent the predicted height dependence for different fiber inclinationngles (inclination angle of inner/outer fiber ring). BCE values have been referencedo the optimum value set to BCE = 1.

he applied sensor head concept excluding any imaging optics,he collected fluorescence should be restricted to < 1 mm height.he dashed curves in Fig. 4 correspond to simulation resultssing different sets of inclination angles of the fibers. Although aetter biofilm specificity would have been obtained with lower

nclination angles (10◦/10◦ for inner/outer fiber ring, 50% signalithin 1.8 mm), due to the overall lower intensity, the BCE valuerops to 0.2. Similarly, despite higher attainable signal levels, with

arge inclination angles (80◦/80◦, 50% signal within 7 mm) biofilmpecificity and the BCE value deteriorates due to increasing signalontributions from the bulk water.

.2.3. Signal linearityThe sensor performance has also been tested by determining the

uorescence signal of two marine bacteria strains. Calibration plotsllustrating the measured fluorescence intensity versus the bacteriaell count and bacteria coverage as well as examples of the underly-ng epifluorescence microscopy images are shown in Fig. 5. For theacteria under investigation it was possible to directly relate thepifluorescence images with a coverage area since their labelledNA is uniformly distributed across the cell. Note, however, thatn entirely monolayered coverage area of 100% may not poten-ially exist in natural biofilms owing to their complexity, patchyormation and three-dimensional structure (Costerton et al., 1995).

In all cases, the double logarithmic plots reveal good linearityetween fluorescence signal and cell number or surface coverage,espectively. Moreover, the data of the two different gram-negativend -positive strains show no significant difference in their fluores-ence characteristics. Certainly, variable environmental conditionsuch as the cell’s nutrient level are expected to influence thentrinsic fluorescence signal. Nevertheless, the experimental resultnderlines the common perception that tryptophan fluorescenceerves as a good and quite universal indicator for bacterial numbernd biomass.

The lower detection limit of the sensor was estimated to bebout 4000 bacteria cells/cm2 from Fig. 5a by taking into accounthe statistical 2� error and the reproducibility of the measure-

ents. At such low bacteria numbers, the surface coverage is wellelow 0.01% and only single bacteria are visible on a typical epi-uorescence microscopy image as shown in the insert of Fig. 5b.his detection limit compares very well with a typical number of

× 105 found for other fiber optical devices (Janknecht and Melo,003) and reveals that the field sensor is capable to monitor therst attachments of cells prior to evolving to a biofilm. At the sameime, owing to the wide dynamic range of the sensor, it is possible

(solid circles) and P. carrageenovora (open circles). The insets in (b) show imagesof bacteria stained with DAPI at two levels of establishment. Error bars representmeans (n = 15) ± standard error.

to detect the adherence of first pioneer cells up to a fully coveredmonolayer of bacteria.

4.3. Field performance

In the first field deployment, the dynamics of a biofilm estab-lishment has been continuously monitored by exposing the sensorin the Baltic Sea (Kiel Bight). In a water depth of 1 m, the sen-sor unit was installed in a dark flow cell with steady exchange ofthe seawater. In biofilm logging mode, the fluorescence intensitywas measured hourly (five times 10 ms measurement time each)over a period of twenty-one days. Physical water parameters suchas temperature (T), salinity (S), dissolved oxygen concentrationand pH-value were measured every 10 min over the experimen-tal time from mid April to May 2011. The average values wereT = 12.7 ◦C, S = 26.3 mS, 93% oxygen saturation and pH 8.1. Theresulting biofilm dynamics curve is shown in Fig. 6 (open cir-cles). Reference sample substrates were placed inside the flow cellunder the same flow conditions. Each day one of these subsam-ples has been analyzed by means of epifluorescence microscopy.Determined biofilm coverages are superimposed on the data of thebiofilm logger (triangles). A striking feature of the experimentalsensor data is that the initial signal level, which is attributableto bulk water contributions, turned out be low. Obviously, theoften problematic background fluorescence is sufficiently sup-pressed. Overall, the fluorescence intensity increases as the biofilmmatures. The trends of the continuously logged biofilm sensor data

and single point microscopic measurements of the surface cover-age are in close agreement showing that even for measurementsof natural biofilms in the field the sensor exhibits linear signalresponse.

M. Fischer et al. / Biosensors and Bioe

Fig. 6. Normalized fluorescence signal of long-term monitoring of biofilm growthin the Baltic Sea (open circles). Analysis of covered substrate area using DAPIs(p

acinbdtieaAbpprtqbshiccitfisto

scae

5

thtfb

tained subsamples by optical microscopy (triangles). Error bars represent meansn = 20) ± standard error of the coverage area. The numbers indicate the three mainhases of biofilm formation dynamics.

The data clearly reveal the expected three-phase biofilm growthnd dynamics as defined by Bryers and Characklis (1982). In ourase, following an approximately four days long conditioning andnduction phase, denoted by number (I) in the figure, the expo-ential accumulation stage commences (II). During this phase, theiofilm grows continuously. Twelve days later, the detachment,ispersal and sloughing phase (III) begins; a phase that is charac-erized by a high variability of the biofilm nature. Although theres good correlation between fluorescence signal and bacteria cov-rage area, it is problematic to convert fluorescence intensity to

total cell number in general (Lewandowski and Beyenal, 2003).part from the fact that the corresponding calibration was onlyased on bacteria, particular care must be taken not to over inter-ret the signal response of the complex matured biofilm duringhase (III). However, as long as biomass settling on a surface orelative consideration of biofilm formation dynamics is concerned,he interpretation of the data should be straightforward. Conse-uently, the up and down turns seen in the sensor signal of a matureiofilm can be attributed to a real effect and are not due to sen-or instabilities. Further assessment and interpretation of the dataas to await more thorough field campaigns. For example, dur-

ng the imaging of subsamples we discovered that the microbialomposition of the biofilm varied from bacterial to heterogeneousommunities and even grazing organisms at later stages. Thereforet might prove necessary to occasionally draw subsamples for quan-ification and identification of complex natural biofilms during theeld study, except in manageable environments. Nevertheless, forhort settling times with a natural biofilm dominated by bacteria,he number of cells ranged from 104 to 107 bacteria cells/cm2 afterne and eleven days, respectively.

The obtained field data are a proof of principle of the sen-or’s capability to guarantee reliable operation under harsh fieldonditions. In the meantime, the sensor unit was tested at temper-tures ranging from 1 to 35 ◦C without experiencing any deleteriousffects.

. Conclusion

An optical fiber biofilm sensor based on detecting intrinsic tryp-ophan fluorescence of bacteria has been developed. The sensor

ead principle allows for large-area illumination and detection andhus ensures the required averaging over the patchy structure ofresh natural biofilms. The spatial arrangement of the fibers haseen optimized in terms of signal intensity and biofilm specificity.

lectronics 33 (2012) 172– 178 177

Linear response and wide dynamic range has been demonstratedby laboratory measurements with tryptophan solutions and arti-ficial biofilms of two different bacteria strains in the laboratory.With a detection minimum of 4000 cells/cm2, the sensor is capableof monitoring biofilms from the first attachment cells up to fullydeveloped complex film. Furthermore, the sensor was applied ina field study over a period of twenty-one days in the Baltic Sea todemonstrate its capability to measure biofilm development. Back-ground fluorescence was found to be sufficiently low and the threestages of biofilm development could be clearly resolved. The flexi-bility, robustness and sensitivity of the sensor offer a high potentialfor applications in aquatic biology, biotechnology, and medicine.Extended field studies in marine environments are currently under-way. Further improvement and upgrading of the sensor is feasible.For example, multichannel detection based on LED arrays operat-ing at different excitation wavelengths and/or spectrally resolvedfluorescence measurements by integrating sensitive microspec-trometers could be easily implemented. For example, such anupgrade would allow for simultaneous measurement of biomassand cell activity by means of tryptophan and adenosine triphos-phate (ATP) fluorescence.

Acknowledgements

We thank Tim Lachnit (Kiel University) and Annett Klemm (Uni-versity of St Andrews) for helpful discussions and Friedrich Temps(Kiel University) for providing easy access to spectroscopic equip-ment, especially during the important set-up phase of the project.Financial support by the German Science Foundation (DFG–EC 80)in the framework of the cluster of excellence The Future Ocean isgratefully acknowledged.

The sensor head concept is patent pending: M. Fischer,G. Friedrichs. M. Wahl, Großflächiger Biofilmsensor/Large-areabiofilm sensor, DE 102011101934.4.

Appendix A. Supplementary data

Supplementary data associated with this article can be found, inthe online version, at doi:10.1016/j.bios.2011.12.048.

References

Abramoff, M.D., Magelhaes, P.J., Ram, S.J., 2004. Biophotonics Int. 11, 36–42.Angell, P., Arrage, A., Mittelman, M., White, D., 1993. J. Microbiol. Methods 18,

317–327.Arrage, A.A., Vasishtha, N., Sundberg, D., Bausch, G., Vincent, H.L., White, D.C., 1995.

J. Ind. Microbiol. 15, 277–282.Barbieri, B., Terpetschnig, E., Jameson, D.M., 2005. Anal. Biochem. 344, 298–300.Battin, T.J., Kaplan, L.A.J., Hansen, C.M.E., 2003. Nature 426, 439–442.Beyenal, H., Yakymyshyn, C., Hyungnak, J., Davis, C.C., Lewandowski, Z., 2004. J.

Microbiol. Methods 58, 367–374.Bonnin, C., Matoga, M., Garnier, N., Debroche, C., de Vandière, B., Chaminade, P.,

2007. J. Chromatogr. A 1156, 94–100.Bryers, J.D., Characklis, W.G., 1982. Biotechnol. Bioeng. 24, 2451–2476.Bünting, U., Karlitschek, P., 1998. Spectrochim. Acta, Part A 54, 1369–1374.Bünting, U., Lewitzka, F., Karlitschek, P., 1999. Appl. Spectrosc. 53, 49–56.Costerton, J.W., Lewandowski, Z., Caldwell, D.E., Korber, D.R., Lappin-Scott, H.M.,

1995. Annu. Rev. Microbiol. 49, 711–745.Dalton, H.M., Goodman, A.E., Marshall, K.C., 1996. J. Ind. Microbiol. Biotechnol. 17,

228–234.Davey, M.E., O’Toole, G.A., 2000. Mol. Biol. Rev. 64, 847–867.Denkhaus, E., Meisen, S., Telgheder, U., Wingender, J., 2006. Microchim. Acta 158,

1–27.Determann, S., Lobbes, J.M., Reuter, R., Rullkötter, J., 1998. Mar. Chem. 62, 137–156.DiBerardino, D., 2002. Opt. Commun. 210, 233–243.Elasri, M.O., Miller, R.V., 1999. Appl. Environ. Microbiol. 65, 2025–2031.Estes, C., Duncan, A., Wade, B., Lloyd, C., Ellis, W., Powers, L., 2003. Biosens. Bioelec-

tron. 18, 511–519.Flemming, H.C., 2003. Water Sci. Technol. 47, 1–8.Flemming, H.C., Tamachkiarowa, A., Klahre, J., Schmitt, J., 1998. Water Sci. Technol.

38, 291–298.Haisch, C., Niessner, R., 2007. Water Res. 41, 2467–2472.

1 d Bioe

H

JJLM

MNRS

Tartakovsky, B., 1996. Water Res. 30, 2941–2948.Wahl, M., 1989. Mar. Ecol.: Prog. Ser. 58, 175–189.

78 M. Fischer et al. / Biosensors an

arder, T., 2009. In: Flemming, H.C., Murthy, P.S., Venkatesan, R., Cooksey, K. (Eds.),Marine and Industrial Biofouling. Springer, Berlin Heidelberg, pp. 219–231.

anknecht, P., Melo, L.F., 2003. Rev. Environ. Sci. Biotechnol. 2, 269–283.i, J., Schanzle, J.A., Tabacco, M.B., 2004. Anal. Chem. 76, 1411–1418.ewandowski, Z., Beyenal, H., 2003. Water Sci. Technol. 47, 9–18.cGuinness, C.D., Sagoo, K., McLoskey, D., Birch, D.J.S., 2005. Appl. Phys. Lett. 86,

261911-261911.

opper, K., Lindroth, P., 1982. Limnol. Oceanogr. 27, 336–347.ivens, D.E., Palmer, R.J., White, D.C., 1995. J. Ind. Microbiol. 15, 263–276.eynolds, D.M., 2003. Water Res. 37, 3055–3060.chmid, T., Panne, U., Haisch, C., Hausner, M., Niessner, R., 2002. Environ. Sci. Technol.

36, 4135–4141.

lectronics 33 (2012) 172– 178

Sohn, M., Himmelsbach, D.S., Barton, F.E., Fedorka-Cray, P.J., 2009. Appl. Spectrosc.63, 1251–1255.

Stickler, D.J., 2008. Nat. Clin. Pract. Urol. 5, 598–608.Stoodley, P., Wilson, S., Hall-Stoodley, L., Boyle, J.D., Lappin-Scott, H.M., Costerton,

J.W., 2001. Appl. Environ. Microbiol. 67, 5608-5608.

Wolf, G., Crespo, J.G., Reis, M.A.M., 2002. Rev. Environ. Sci. Biotechnol., 227–2512002.Wozniak, B., Dera, J., 2007. Light Absorption in Sea Water, Atmospheric and Oceano-

graphic Sciences Library, vol. 33. Springer, New York, USA.