Analytical Biochemistry 390 (2009) 197–202

Contents lists available at ScienceDirect

Analytical Biochemistry

journal homepage: www.elsevier .com/locate /yabio

Correcting for the inner filter effect in measurements of fluorescent proteinsin high-cell-density cultures

Chong Zhang, Min-Sheng Liu, Bing Han, Xin-Hui Xing *

Department of Chemical Engineering, Tsinghua University, Beijing 100084, People’s Republic of China

a r t i c l e i n f o

Article history:Received 18 March 2009Available online 24 April 2009

Keywords:GFPFluorescence intensityHigh cell densityInner filter effectMathematical model

0003-2697/$ - see front matter � 2009 Elsevier Inc. Adoi:10.1016/j.ab.2009.04.029

* Corresponding author. Fax: +86 10 6277 0304.E-mail address: xhxing@tsinghua.edu.cn (X.-H. Xin

1 Abbreviations used: FP, fluorescent protein; GFP, ginner filter effect; NADH, nicotinamide adenine dinucrecombinant green fluorescent protein; CFP, cyan fluoreb-D-1-thiogalactopyranoside; PBS, phosphate-bufferedcence units.

a b s t r a c t

Fluorescent proteins (FPs), such as green fluorescent protein (GFP) and its variants, are well-developedvisible markers for analyzing bioprocesses. Accurate measurement of fluorescence emitted from FPs inwhole cells is complicated by the inner filter effect (IFE), which is caused by intracellular light absorptionand scattering by cell particles. The IFE causes nonlinearity between fluorescence intensity and fluoro-phore concentrations in FP-harboring cells and can significantly influence the accuracy of FP-based anal-ysis, especially at high cell densities. A mathematical model based on detection of fluorescence intensityusing a fluorescence spectrophotometer was developed to provide a simple correction for the IFE in fluo-rescence intensity detection in high-density cultures. The parameters of this model were determined inthree different FP-harboring bacterial strains to give the ‘‘real fluorescence” intensity without the IFE.Using these parameters, accurate analysis of FP-labeled Escherichia coli at high cell density in pure cultureand in mixed cultures with fluorescent and nonfluorescent strains was easily and successfully achieved.

� 2009 Elsevier Inc. All rights reserved.

Rapid, convenient, and accurate quantification of cells and theirbiomolecules is a key concern in bioprocess control or analysis. No-vel molecular biology-based techniques, such as DNA probing andmarker gene expression, have been applied to meet increasingneeds for bioprocess control and optimization. Fluorescent pro-teins (FPs),1 such as green fluorescent protein (GFP) and its variants,have unique benefits such as real-time detection, low toxicity to thehost cells, lack of required cofactors, and the possibility of fusion totarget proteins [1]. These advantages make GFP a useful marker forquantifying cell concentration or the activity of a target enzyme ina bioprocess. Bentley and coworkers [2] developed a GFP sensor toquantitatively analyze the cell concentration of Escherichia coli cul-tures by measuring the fluorescence intensity in the first applicationof GFP as an analytical marker for monitoring and analyzing a bio-process. Using this method, real-time quantification of a process inbacterial cultures was achieved at high cell densities [3–5]. GFP fu-sion proteins were further developed to allow quantification ofGFP-fused enzymes. By detecting fluorescence intensity, the produc-tion of chloramphenicol acetyl transferase [6], organophosphorushydrolase [7], and heparinase [8] has been accurately quantified. In

ll rights reserved.

g).reen fluorescent protein; IFE,leotide, reduced form; GFPuv,scent protein; IPTG, isopropylsaline; RFU, relative fluores-

addition to measuring aerobic cultures, quantitative analysis ofanaerobic cultures by GFP was also achieved, taking advantage ofthe quick recovery of the fluorophore by reacting with oxygen afterGFP expression in the anaerobic cells [9,10].

Nonetheless, accurate measurement of the fluorescence emittedby FPs in whole cells is complicated by the inner filter effect (IFE)from intracellular light absorption and scattering by cell particles[11]. The IFE can affect the excitation or emission light and canbe caused by the cell wall or cell particles. As a result, the detectedfluorescence intensity for whole cells at high cell densities is notproportional to the GFP molecules inside the fluorescent cells. Inlow-cell-density cultures, the IFE is negligible [6–10]. However,in most practical applications, because high-cell-density culturesare often needed, the IFE becomes significant and must be ac-counted for to achieve accurate analysis.

In general, there are three ways to reduce the IFE. One is toempirically dilute the sample to a sufficiently low concentration.Dilution is complicated, however, and is not always possible, forexample, with online measurements. This method also leads toan increase in labor and an accumulation of errors. Alternatively,changing the spectrophotometric cuvette configurations [12] orshortening the light path length could diminish the IFE [13]. How-ever, these methods are not applicable for common fluorescencespectrophotometers; development of specially designed spectro-photometric cuvettes with appropriate configurations, or even mi-cro-bioreactors, would be required for this option. A better solutionis direct calculation with mathematical correction of the measuredfluorescence data. By establishing a mathematical model to link

198 Correcting for the inner filter effect / C. Zhang et al. / Anal. Biochem. 390 (2009) 197–202

the IFE to cell density, Srinivas and Mutharasan [14] developed anapproximate light intensity model for IFE correction for measure-ment of NADH (nicotinamide adenine dinucleotide, reduced form)fluorescence, and Konstantinov and Dhurjati [15] proposed a real-time algorithm for elimination of the IFE in monitoring biolumi-nescent bacterial cultures. However, IFE correction was linked tooptical density data, and the ‘‘real fluorescence” without the IFEcould not be acquired. Su and coworkers [16] developed an im-proved backscatter fluorescence probe model to eliminate theinfluence of the IFE on GFP concentration measured by online fluo-rescence detection of a culture. Without using any other opticaldata, direct reduction of the IFE on the fluorescence measurementwas achieved. The drawback of this approach was that additionaldevices besides a common fluorescence spectrophotometer wererequired. In this example, a state observer using an extended Kal-man filter was engaged.

These examples show that, for commercially available fluores-cence spectrophotometers, fluorescence of FPs without the IFEcannot be easily measured. Therefore, model correction of the IFEto give the real or theoretical fluorescence of FPs would be usefulfor rapid analysis of bioprocesses at high cell densities. In this study,we aimed to develop a mathematical model for simple IFE correctionto use in detecting fluorescence intensity of GFP and its variants witha commercial fluorescence spectrophotometer. The goal was accu-rate FP-based measurement in high-cell-density cultures. The modelwas established using basic parameters representing the absorbing/scattering effects of three different FPs—GFP, recombinant greenfluorescent protein (GFPuv), and cyan fluorescent protein (CFP)—and the penetration effects by the cell wall. Without linking to opti-cal density data, we successfully reduced the IFE and obtained thereal fluorescence of high-cell-density cultures and mixed culturesof fluorescent and nonfluorescent E. coli strains. This enabled theaccurate and rapid analysis of cultures by fluorescence intensitymeasurement within a wide range of cell densities.

Materials and methods

Bacterial strains and culture conditions

Three different FP-harboring strains, which were previouslydeveloped in our lab [17], were used: E. coli BL21(DE3) (pUC–GFP-mut1), E. coli BL21(DE3) (pUC–GFPuv), and E. coli BL21(DE3)(pET28a–CFP). These three strains express the fluorescent proteinsGFPmut1 (excitation 488 nm, emission 509 nm), GFPuv (excitation395 nm, emission 509 nm), and CFP (excitation 434 nm, emission477 nm), respectively.

Each strain was cultivated in Luria Broth (10 g/L tryptone, 5 g/Lyeast extract, and 10 g/L NaCl) until the OD600 reached 0.4 to 0.6.Isopropyl b-D-1-thiogalactopyranoside (IPTG) was then added toa final concentration of 0.5 mM to induce expression of the FPs.

Analytical methods

The OD600 for evaluating the cell concentration was measuredwith a spectrophotometer (Shimadzu UV-1206, Japan). For themeasurement of fluorescence intensity, a fluorescence spectropho-tometer was used (Hitachi F-2500, Japan).

For measuring the fluorescence intensity of whole cells, thecells were washed three times with phosphate-buffered saline(PBS [pH 7.4]: 10 mM Na2HPO4, 1 mM KH2PO4, 140 mM NaCl,and 3 mM KCl) and resuspended in the same volume of PBS toeliminate noise from the medium during fluorescence detection.

For measuring the fluorescence intensity of crude FP extracts,the influence of scattering by cell particles was eliminated by dis-rupting previously resuspended cells by sonication (300 s, 200 W)

and removing cell debris by centrifugation at 15,000 rpm and 4 �Cfor 10 min.

Cell number counting

To count the cell number, 200 ll of DAPI (40,6-diamidino-2-phenylindole, 0.5 M, Invitrogen, USA) was added to 1 ml of dilutedbacteria solution, and the mixture was incubated at 37 �C for 1 to2 h. Stained cells were carefully collected onto a filtration mem-brane (0.2 lm, Whatman 110656). The membrane surface waswashed repeatedly with a 20� volume of ultrapure water to re-move unincorporated fluorescent dye. The cell number wascounted on a fluorescence microscope (excitation 358 nm, emis-sion 461 nm, Nikon E600, Japan).

Model development

The model developed in this study was based on the followingassumptions. First, the excitation light was uniform and its inten-sity was steady. Second, the fluorescent cells were in a homoge-neous solution, and the amount of fluorescent protein harboredin each cell and physiological/physical properties of each cell wereidentical. Third, the shape of the cell approximated a sphere, not ashort rod. This study used a high number of cells per unit volume(N0 = 2.026 * 109/ml/OD600, where N0 is the cell number per OD ofunit volume) and the orientation of the long or short axis of the cellwas random, so that light scattering had no net orientation and aspherical cell shape was a reasonable approximation. Fourth, theexcitation light would be divided into two parts after scattering:the original excitation light and the scattered light (which has alonger wavelength than the original). The scattered light was nottaken into consideration in the fluorescence excitation.

In a system without scattering effects, the fluorescence inten-sity at any point, f(x, y), can be expressed as

f ðx; yÞ ¼ j � I0 � e � /; ð1Þ

where j is the instrumental constant, I0 is the exciting light inten-sity, e is the absorption efficiency of fluorescent protein, and u is thequantum efficiency of fluorescent protein. All four of these param-eters are constant.

In this work, FP-harboring strains at high density were used as amodel system to study scattering effects. As shown in Fig. 1,assuming each cell as a particle, the excitation light is scatteredthrough each layer of cells. Thus, the excitation light intensity atpoint x (I) can be expressed as

I ¼ I0 � ðkIÞx=d; ð2Þ

where kI is the scattering coefficient of exciting light and the dis-tance between cell particle layers (d) is given by

d ¼ 3ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1=ðc � N0Þ

p; ð3Þ

where c is the cell concentration (expressed as OD).Likewise, the light emitted by each fluorescent cell is scattered

by other cell particles. Thus, the emission fluorescence intensitycan be expressed as

f ðx; yÞ ¼ ðkFÞy=d � j � I � e � /; ð4Þ

where kF is the scattering coefficient of emission light.Considering the barrier effect of the cell wall of each cell particle,

the penetration coefficient of excitation and emission light should beadded as correction parameters. Thus, after taking the double inte-gral, the detected fluorescence intensity (F) can be expressed as

F ¼c � N0 � ðd2=lÞ � k � I0 � e � / �xI �xF � ½ððkIÞl=d � 1Þ=

lnðkIÞ� � ½ððkFÞl=d� 1Þ= lnðkFÞ�; ð5Þ

y

x

d

(x,y)I

F

l

l

F

I

Fluorescence Protein

Cell

CellMembrane

Fig. 1. Schematic diagram of the principle of fluorescence scattering by cellparticles.

450 500 550 600

0

2000

4000

6000

8000

10000

RFU

(-)

wavelength (nm)

in solution in high- density cells

Red shift

Fig. 2. Fluorescence spectrum of GFPmut1 in solution and in high-density culture(OD660 = 100).

Correcting for the inner filter effect / C. Zhang et al. / Anal. Biochem. 390 (2009) 197–202 199

where xI and xF are the penetration coefficients of exciting lightand emission light, respectively.

Ignoring the scattering effect by the cell particles and interfer-ence of cell walls, the fluorescence intensity inside the cells (F0)should be

F0 ¼ c � N0 � k � I0 � l � e � / ð6Þ

Combining Eqs. (5) and (6), the detected fluorescence intensity canbe expressed as follows:

F ¼F0 � ðd2=l2Þ �xI �xF � ½ððkIÞl=d � 1Þ= lnðkIÞ� � ½ððkFÞl=d� 1Þ=

lnðkFÞ�: ð7Þ

Our aim in developing the fluorescence scattering model was toeliminate the interference of cell walls and cell particles on fluores-cence quantification. Thus, the modified fluorescence intensity aftermodel correction by eliminating the scattering effect and interfer-ence of cell walls (Fm) should be equal to the fluorescence intensityinside the cells (F0), which can be expressed as follows:

Fm ¼ F0 ¼F=½ðd2=l2Þ �xI �xF � ðððkIÞl=d � 1Þ= lnðkIÞÞ

� ðððkFÞl=d � 1Þ= lnðkFÞÞ� ð8Þ

Results

IFE for GFP fluorescence detection in high-density E. coli cultures

Fig. 2 shows the fluorescence spectrum of GFPmut1 in a high-density culture (OD660 = 100) compared with the protein in solu-tion obtained by cell disruption with no scattering effect. In the

high-density culture, the peak excitation wavelength (488 nm) de-creased dramatically, whereas the peak emission wavelength wasclearly red-shifted from 509 to 520 nm. These shifts were due tothe IFE of high-density cells, resulting in energy losses in boththe excitation and emission.

Fig. 3 shows the influence of cell density on fluorescenceintensity detection of GFP-harboring cells. When the OD660 isin the range of 0.5 to 2, the linear relationship between the fluo-rescence intensity and cell concentration was quite strong,meaning that the IFE can be ignored in practical use. Otherwise(OD660 > 2), it should be taken into consideration (Fig. 3A). More-over, in an ideal situation, if the IFE could be ignored, the addi-tion of nonfluorescent E. coli BL21(DE3) cells to a constantamount of GFP-harboring E. coli BL21(DE3) (pUC–GFPmut1) cellswould have no effect on the detected fluorescence intensity(Fig. 3B). However, the actual detected fluorescence intensitywas much lower than the ideal value. As the IFE increased withhigher total cell concentrations, the detected fluorescence inten-sity of the whole cells ceased to be linearly proportional to theconcentration of the fluorescent protein.

Mathematical modeling of IFE for fluorescence detection of FPs in high-density cultures

Three different FP-harboring strains were used to determine theparameters in Eq. (8): E. coli BL21(DE3) (pUC–GFPmut1), E. coliBL21(DE3) (pUC–GFPuv), and E. coli BL21(DE3) (pET28a–CFP),which express the fluorescent proteins GFPmut1, GFPuv, and CFP,respectively. The host cells used in this study were the same foreach fluorescent protein, thereby ensuring identical cell wall andparticle effects on excitation and emission. Fig. 4 shows the fluo-rescence spectra of the different FP-harboring strains. Each proteinhad its own characteristic excitation and emission wavelengths.

The data of F0, F, and N0 in Eq. (8) could be measured directly.The fluorescence intensity of disrupted cells was used to representthe value of the fluorescence intensity inside the cells (F0) becausethe fluorescent proteins had been released from the cells and thecell debris had been removed, thereby eliminating the scatteringeffect of cell walls and particles. Meanwhile, the actual fluores-cence intensity of the cells (F) could be detected directly fromthe whole cells. Using the method described in Materials andMethods, N0 was first determined, giving a value of 2.026 * 109/ml/OD600 (n = 8). Based on F0, F, and N0, the parameters of kI, kF,xI, and xF in Eq. (8) could be determined.

0 2 4 6 8 10 12 14 16 18 200

500

1000

1500

2000

2500

3000

3500

4000

RFU

(-)

c/OD600

2 4 6 8 10 12

140

160

180

200

220

240

RFU

(-)

OD600

detected fluorescence intensity theoretical fluorescence intensity

B

A

Fig. 3. Influence of cell density on fluorescence intensity detection of GFP-harboring cells. (A) Increasing the concentration of GFP-harboring E. coli BL21(DE3)(pUC–GFPmut1) cells. (B) Adding nonfluorescent E. coli BL21(DE3) cells to aconstant amount of GFP-harboring E. coli BL21(DE3) (pUC–GFPmut1) cells.

200 Correcting for the inner filter effect / C. Zhang et al. / Anal. Biochem. 390 (2009) 197–202

To determine the parameters kI and kF, mixtures of differentconcentrations of crude fluorescent protein extracts and nonfluo-rescent bacterial strains were used as a model system. Here F wasthe detected fluorescence intensity of the mixtures and F0 was the

350 400 450 500 550 6000

2000

4000

RFU

(-)

wavelength (nm)

GFPmut1 GFPuv CFP

Fig. 4. Fluorescence spectra of three different fluorescent proteins: GFPmut1,GFPuv, and CFP.

fluorescence intensity of the crude fluorescent protein extracts.Because the fluorescent proteins were in solution, the influence ofthe cell wall on fluorescence detection could be ignored, so that

xI ¼ xF ¼ 1: ð9Þ

Based on Mie scattering theory,

kI ¼ j=k3I ; kF ¼ j=k3

F ; ð10Þ

where j is the scattering constant of scattering media.Considering Eqs. (8)–(10), kI and kF could be determined. The re-

sults are shown in Table 1.Likewise, to determine the parameters of xI and xF, different

concentrations of FP-harboring cells were used as the model sys-tem. Here F was the detected fluorescence intensity of the cellsand F0 was the fluorescence intensity of crude FP extracts obtainedby cell disruption.

Based on Beer’s law,

xI ¼ eð�s=kIÞ; xF ¼ eð�s=kF Þ; ð11Þ

where s is the absorption constant.Using Eqs. (8) and (11), and the values of kI and kF in Table 1, xI

and xF were determined. The results are shown in Table 2. kF (xF)values of GFPmut1 and GFPuv were nearly the same because oftheir identical emission wavelengths.

According to the above-determined parameters, the modifiedfluorescence intensity (Fm) could be deduced from that of the de-tected value (F) using Eq. (8) to eliminate the scattering effectand interference of the cells. Calculated fluorescence intensity(Fm) should be identical to the ideal fluorescence intensity (F0) ofFP at the given cell concentration (Fig. 5). In the respective opticaldensity ranges of 0 to 20 for GFP and GFPuv and 0 to 5 for CFP, andin the fluorescence intensity range of 10,000 to 45,000 relativefluorescence units (RFU), the Fm values of the three FP-harboringstrains all were identical to F0. Also, as shown in Table 3, the linearrelationship between Fm and cell concentration (c) was quitestrong, with a linear correlation coefficient of more than 0.999.However, the linear relationship between actual fluorescenceintensity (F) and c is not good, with a linear correlation coefficientof less than 0.977. Moreover, the slopes of Fm–c and F0–c are nearlythe same and give a maximum error of 0.95%. Because we focusonly on the measurement of cell concentration by using the de-duced Fm, the error between the slopes of Fm–c and F0–c is accept-able for the accurate quantification in the whole range. In mostcases, Fm values of the three FP-harboring strains are identical toF0, but for GFPuv Fm at higher OD660 (e.g., OD660 > 18) is lower thanF0 with a relative error of 8% (Fig. 5). This would result in an accu-mulation of errors in quantifying the concentration of fluorescentstrains at high density. Thus, we evaluate such errors in the follow-ing model experiments.

Table 1Scattering coefficients for GFPmut1, GFPuv, and CFP in E. coli BL21 (n = 50).

GFPmut1 GFPuv CFP

Wavelength (nm) 488 509 395 509 434 477Scattering coefficient 0.9460 0.9524 0.8870 0.9472 0.9070 0.9299

Table 2Absorption coefficients of GFPmut1, GFPuv, and CFP in E. coli BL21 (n = 50).

GFPmut1 GFPuv CFP

Wavelength (nm) 488 509 395 509 434 477Absorption coefficient 0.6641 0.6754 0.6222 0.6919 0.6295 0.6564

0 5 10 15 200

10000

20000

30000

40000

RFU

(-)

OD600

OD600

OD600

F0

Fm

F

A

0 5 10 15 200

10000

20000

30000

40000

RFU

(-)

RFU

(-)

F0

Fm

F

500

10000

20000

30000

40000

50000

F0

Fm

F

B

C

Fig. 5. Comparison of fluorescence intensities calculated by the mathematicalmodel (Fm) with the measured results at different cell concentrations, where F isfluorescence intensity of whole cells and F0 is fluorescence intensity of the cellextract after disrupting the cells followed by removal of the cell debris: (A) GFP; (B)GFPuv; (C) CFP.

Table 3Application of the mathematical model to a high-cell-density culture (n = 56).

GFPmut1 GFPuv CFP

Slope of F0–c 1997 2091 5909Slope of Fm–c 2001 2086 5966Error between slope of Fm–c and F0–c 0.21% 0.29% 0.95%Linear correlation coefficient of Fm–c 0.9996 0.9992 0.9994Linear correlation coefficient of F–c 0.9751 0.9585 0.9765

0 2 4 6 8 10 12 14 16 180123456789

101112131415161718

Cal

cula

ted

OD

600

Cal

cula

ted

OD

600

Cal

cula

ted

OD

600

Real OD600

Real OD600

Real OD600

Average error = -4.99%

0 2 4 6 8 100

1

2

3

4

5

6

7

8

9

10

Average error = -6.25%

0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.000.00

0.25

0.50

0.75

1.00

1.25

1.50

1.75

2.00

Average error = -8.51%

A

B

C

Fig. 6. Comparison of the calculated and real cell concentrations in a bacterialmixture with different proportions of fluorescent and nonfluorescent strains: 9:1(A), 1:1 (B), and 1:9 (C).

Correcting for the inner filter effect / C. Zhang et al. / Anal. Biochem. 390 (2009) 197–202 201

Quantification of specific fluorescent microorganisms in a mixture offluorescent and nonfluorescent strains at high density

One of the most promising applications of FP as a fluorescentreporter is to quantify specific fluorescent microorganisms orproteins in mixed culture systems. The presence of other nonfluo-

202 Correcting for the inner filter effect / C. Zhang et al. / Anal. Biochem. 390 (2009) 197–202

rescent microorganisms will also greatly influence the accurateanalysis of FP due to the IFE.

The fluorescent strain E. coli BL21(DE3) (pUC–GFPmut1) and thenonfluorescent strain E. coli BL21(DE3) were mixed at a high celldensity (maximum OD600 = 20) as a model system to test the appli-cation of the above mathematical model on analysis of a mixedculture. The parameters of both strains were identical because theyused the same host cells. Thus, Eq. (8) was also suitable for this spe-cial case. By detecting the fluorescence intensity of the mixture, theconcentration of fluorescent cells could be easily deduced from thelinear relationship between Fm and c. Figs. 6 shows the calculated cellconcentration with different proportions of fluorescent and nonfluo-rescent strains (9:1, 1:1, and 1:9 for Figs. 6A, B, and C, respectively).The calculated values (triangles) were identical to the actual concen-tration of added fluorescent strains (line), indicating that the devel-oped mathematical model could also be used in a mixed culture ofhigh cell density. Incidentally, at different proportions of fluorescentand nonfluorescent strains—9:1, 1:1, and 1:9—the errors for thededuced Fm at high concentrations were �4.99, �6.25, and �8.51%,respectively. Although the model calculation would underestimatethe cell concentrations at high concentrations, the systematicalerrors as calculated above could be acceptable for the measurementsof fluorescent proteins in high-cell-density cultures.

Discussion

Using fluorescence intensity measurements of FPs from a fluo-rescence spectrophotometer, a mathematical model for simpleelimination of the IFE from fluorescence detection of GFP, GFPuv,and CFP in high-density cultures was developed.

The model was based on basic parameters representing theabsorbing/scattering effects of the three FPs and penetration ef-fects of the cell wall. The parameters kI and kF represented the lossratios of excitation and emission light intensity after scattering bythe cells, and xI and xF represented the loss ratios of excitationand emission light intensity after absorption by the cell walls. Allparameters reflected the basic cell characteristics responsible forthe IFE of FP fluorescence. To the best of our knowledge, this isthe first report to give IFE parameters for different FPs.

Unlike previous mathematical models that were based on thesample optical density [13,14], this calculation gives the real directfluorescence intensity of FPs inside the cells by eliminating the IFEof cell particles and cell walls. Moreover, the previous empiricalmodel that used absorbance at the excitation and emission wave-lengths to correct for the IFE in homogeneous samples [18] cannotbe used in the high-cell-density cultures examined in our study be-cause the IFE comes from mostly the cell walls and cell particles.

With the method developed here, bioprocesses in high cell den-sities could be accurately monitored. Particularly relevant to bio-augmentation using mixed cultures, the concentration of thetarget strain could be accurately quantified with fluorescent mark-ers. The host cells used in the mixed cultures studied here were ofthe same strain, but this model could be extended for applicationto analysis of mixed cultures with different host cell types if the

parameters pertaining to the IFE of the bacterial cells could be esti-mated beforehand.

Acknowledgments

This study was supported in part by projects of the NationalNatural Science Foundation of China (20836004) and the 863 Planof Ministry of Science and Technology of China (2006AA02Z203).

References

[1] L. Poppenborg, K. Friehs, E. Flaschel, The green fluorescent protein is a versatilereporter for bioprocess monitoring, J. Biotechnol. 58 (1997) 79–88.

[2] L. Randers-Eichhorn, C.R. Albano, J. Sipior, W.E. Bentley, G. Rao, On-line greenfluorescent protein sensor with LED excitation, Biotechnol. Bioeng. 55 (1997)921–926.

[3] H.J. Chae, M.P. DeLisa, H.J. Cha, J.C. Li, W.A. Weigand, J.J. Valdes, G. Rao, W.E.Bentley, On-line optimization and control of recombinant protein expressionin high cell density Escherichia coli cultures using GFP-fusion monitoring,Abstr. Pap. Am. Chem. Soc. 219 (2000) U213.

[4] H.J. Chae, M.P. DeLisa, H.J. Cha, W.A. Weigand, G. Rao, W.E. Bentley, Frameworkfor online optimization of recombinant protein expression in high-cell-densityEscherichia coli cultures using GFP-fusion monitoring, Biotechnol. Bioeng. 69(2000) 275–285.

[5] M.P. DeLisa, J.C. Li, G. Rao, W.A. Weigand, W.E. Bentley, Monitoring GFP–operon fusion protein expression during high cell density cultivation ofEscherichia coli using an on-line optical sensor, Biotechnol. Bioeng. 65 (1999)54–64.

[6] C.R. Albano, L. Randers-Eichhorn, W.E. Bentley, G. Rao, Green fluorescentprotein as a real time quantitative reporter of heterologous protein production,Biotechnol. Prog. 14 (1998) 351–354.

[7] C.F. Wu, H.J. Cha, G. Rao, J.J. Valdes, W.E. Bentley, A green fluorescent proteinfusion strategy for monitoring the expression, cellular location, and separationof biologically active organophosphorus hydrolase, Appl. Microbiol.Biotechnol. 54 (2000) 78–83.

[8] Y. Chen, X.H. Xing, F.C. Ye, Y. Kuang, Soluble expression and rapidquantification of GFP–hepA fusion protein in recombinant Escherichia coli,Chin. J. Chem. Eng. 15 (2007) 122–126.

[9] C. Zhang, X.H. Xing, Quantification of a specific bacterial strain in an anaerobicmixed culture for biohydrogen production by the aerobic fluorescencerecovery (AFR) technique, Biochem. Eng. J. 39 (2008) 581–585.

[10] C. Zhang, X.H. Xing, K. Lou, Rapid detection of a GFP-marked Enterobacteraerogenes under anaerobic conditions by aerobic fluorescence recovery, FEMSMicrobiol. Lett. 249 (2005) 211–218.

[11] S.A. French, P.R. Territo, R.S. Balaban, Correction for inner filter effects in turbidsamples: Fluorescence assays of mitochondrial NADH, Am. J. Physiol. CellPhysiol. 44 (1998) C900–C909.

[12] S.M. Kao, A.N. Asanov, P.B. Oldham, A comparison of fluorescence inner-filtereffects for different cell configurations, Instrum. Sci. Technol. 26 (1998) 375–387.

[13] A. Zanzotto, P. Boccazzi, N. Gorret, T.K. Van Dyk, A.J. Sinskey, K.F. Jensen, In situmeasurement of bioluminescence and fluorescence in an integratedmicrobioreactor, Biotechnol. Bioeng. 93 (2006) 40–47.

[14] S.P. Srinivas, R. Mutharasan, Inner filter effects and their interferences in theinterpretation of culture fluorescence, Biotechnol. Bioeng. 30 (1987) 769–774.

[15] K.B. Konstantinov, P. Dhurjati, Real-time compensation of the inner filter effectin high-density bioluminescent cultures, Biotechnol. Bioeng. 42 (1993) 1190–1198.

[16] W.W. Su, B. Liu, W.B. Lu, N.S. Xu, G.C. Du, J.L. Tan, Observer-based onlinecompensation of inner filter effect in monitoring fluorescence of GFP-expressing plant cell cultures, Biotechnol. Bioeng. 91 (2005) 213–226.

[17] M.S. Liu, Green Fluorescence Protein as a Visible Marker for the Quantificationof Bioprocess, Ph.D. Thesis, Tsinghua University, Beijing, China, 2008.

[18] J.R. Lakowicz, Principles of Fluorescence Spectroscopy, third ed., Springer, NewYork, 2006.