Biosensors and Bioelectronics 22 (2007) 2643–2649

Detection and identification of bacteria using antibiotic susceptibility

and a multi-array electrochemical sensor with pattern recognition
Jason Karasinski a , Leslie White a , Yachao Zhang a , Eric Wang a , Silvana Andreescu a ,
Omowunmi A. Sadik a,∗ , Barry K. Lavine b , Mehul Vora b
a Department of Chemistry, State University of New York-Binghamton, P.O. Box 6000, Binghamton, NY 13902, USA
b Department of Chemistry, Oklahoma State University, Stillwater, OK 74078, USA

Received 14 June 2006; received in revised form 15 September 2006; accepted 31 October 2006
Available online 12 December 2006

Abstract
This work proposes the use of amperometric signals generated by a 96-well multi-array dissolved oxygen multi-electrode sensor (DOX)
coupled with principal component analysis for continuous monitoring, identification and differentiation of bacteria. Two types of differentiation
mechanisms were tested: (1) direct monitoring of respiratory activity via oxygen consumption and (2) quantification of the effect of three broad-
spectrum antibiotics on bacteria growth and respiration over time. Five species of bacteria were examined including: Escherichia coli, Escherichia
adecarboxylata, Comamonas acidovorans, Corynebacterium glutamicum and Staphylococcus epidermidis. The addition of small concentrations
of antibiotics to the growth medium alters the oxygen consumption of the cells and a unique fingerprint is created for a specific cell. This fingerprint
is shown to evolve over a specific concentration range that is dependant of instrumental constraints of the DOX system. The application of principal
component analysis (PCA) to classify the data was also examined. It was shown that bacteria could be classified simply by their oxygen consumption
rates over a varying concentration range. Discrimination between species can also be increased by the effects of the antibiotics on the oxygen
consumption of varying concentrations of cells. The proposed DOX–PCA system illustrates a generic template that can be tailored to meet specific
research goals by the selection of specific cell/antibiotic combinations and concentrations.
© 2006 Published by Elsevier B.V.

Keywords: Antibiotic susceptibility; Oxygen sensor; Multi-array sensor; Microbial sensor; Principal component analysis

1. Introduction
Over the last decade, almost every type of bacteria has
become less responsive to antibiotic treatment. The CDC, 2006
(www.cdc.gov) has labeled microbial antibiotic resistance as
one of the world’s most pressing public health problems and
surveillance programs have been implemented throughout (Fluit
et al., 2006; Bootsma et al., 2006; Whitlock et al., 2002). Despite
this concern, there is still no accepted system for classify-
ing bacteria based on responses to antibiotics. Conventional
methods for assessing antibiotic susceptibility include the disk
diffusion method (Prescott et al., 1999), the broth microdilu-
tion method (Nahed et al., 2006) and the antibiotic agar plate
method (Webster et al., 2004). These methods are reliable for
determining the minimal inhibitory concentration (MIC) for a
∗ Corresponding author. Tel.: +1 607 777 4132; fax: +1 607 777 4478.
E-mail address: osadik@binghamton.edu (O.A. Sadik).
particular drug but are time consuming (24–48 h) and prone to
human error because of the manual measurements that must be
performed. There are also several commercially available sys-
tems that provide antibiotic prepped frozen panels that can be
either analyzed in a special instrument or manually read (Rennie
et al., 2004; Murdoch et al., 2003; Jorgensen et al., 1998).
Rapid identification of bacteria and fungi has been reported
using capillary electrophoresis (Avaniss-Aghajani et al., 1994;
Turenne et al., 1999). However, these methods still rely on
the density of the cell culture and do not provide any infor-
mation regarding how long it takes for the drug to act on the
bacteria or how the bacteria responds to the drug at different
stages of the incubation period. Currently, the MIC as deter-
mined by these methods is the only way to qualify antibiotic
susceptibility.
The DOX-96 is a multi-channel electrochemical sensor
designed to measure the dissolved oxygen content in 96-well
plates (Andreescu et al., 2005, 2006). The system has been
successfully tested for determining the MIC’s for various

0956-5663/$ – see front matter © 2006 Published by Elsevier B.V.

doi:10.1016/j.bios.2006.10.037
2644 J. Karasinski et al. / Biosensors and Bioelectronics 22 (2007) 2643–2649
Fig. 1. Schematic of the DOX–PCA concept. (A–C) represent DOX responses for high medium and low cell concentrations, respectively.
antibiotics/cell combinations showing a 98.2% agreement with
the standard microdilution method (Kitahara et al., 2003). Not
only is this method high throughput, but also the determinations
are made based on the oxygen consumption (or lack thereof)
from the cells. For cells growing in medium, the DOX-96 will
measure a sharp, steady decrease in dissolved oxygen as the
cells grow. In the presence of the antibiotic at the MIC, the
oxygen consumption will cease. In this system, the MIC is
determined by manual examination of the shape of the current
versus time plots produced by the DOX software.
Recently, we have reported a method using the DOX-96 cou-
pled with pattern recognition techniques for the identification
and classification of 11 types of bacteria down to the species
level (Karasinski et al., 2005). In this work, small concentra-
tions of various broad-spectrum antibiotics were added to the
growth medium supporting microbial growth and respiration.
The antibiotic concentrations used were significantly lower than
the minimal inhibitory concentration for most cells. The result
was that cellular growth and/or respiration for the microbes
studied was only partially inhibited and thus the oxygen con-
sumption by the cells was decreased compared to cells growing
in antibiotic free medium. It was shown that the effects of the
antibiotics on oxygen consumption were unique for different
types of microbes. These effects were monitored continuously
in 96-well plates using the DOX system and the relevant data
was processed using principal component analysis (PCA). The
end goal of the work described above was to create a library of
oxygen consumption “fingerprints” that could be used to identify
unknown bacteria.
The application of this method as a typical sensor system is
limited because, like the conventional susceptibility methods,
it cannot be applied to mixtures of cells and the experiments
must be performed in controlled conditions (medium, temp,
etc.). However, the DOX antibiotic fingerprints provide qualita-
tive, reproducible information, specific enough to differentiate
between species, which may be utilized as an additional param-
eter for clinical applications such as monitoring or mapping
antibiotic resistance.
In the present work, we have expanded upon the original
DOX–PCA concept to include various concentrations of cells.
The proposed concept is that different concentrations of bacte-
ria growing in the presence of antibiotics will create different
responses at the dissolved oxygen electrodes. This can simplify
the identification of an unknown because its concentration would
not have to be adjusted before analysis. The method will also
avoid the instrumental constraints as illustrated in Fig. 1.
2. Materials and methods
Corynebacterium glutamicum (ATCC #13032), Staphylococ-
cus epidermidis (ATCC #12228), Escherichia adecarboxylata
(ATCC # 23216), Comamonas acidovorans (ATCC #51340)
and Escherichia coli (ATCC #25922) were purchased from
American Type Culture Collection (Manassas, VA). E. coli K12
 J. Karasinski et al. / Biosensors and Bioelectronics 22 (2007) 2643–2649
strain was donated from the Biological Sciences Department
at SUNY-Binghamton. The antibiotics chloramphenicol,
ampicillin and tetracycline and nutrient agar were purchased
from Sigma (St. Louis, MO). The growth medium used to for
cell cultivation and electrochemical experiments was Mueller
Hinton broth (Sigma). For experiments involving study of the
effect of antibiotics, stock 1 mg/ml solutions of each antibiotic
were prepared in ultra-pure water. For plate preparation,
tetracycline was diluted to 1 ␮g/ml in Mueller Hinton broth.
Ampicillin and chloramphenicol were diluted to 5 ␮g/ml.
2.1. Electrochemical instrumentation
The DOX-96 multi-array electrochemical sensor was pro-
vided by Daikin Industries, Ltd. The principle of operation and
a schematic representation of the DOX sensor was described
in detail previously (Kitahara et al., 2003). Basically, the DOX
functions as a classical amperometric oxygen sensor. Changes in
the oxygen level of selected wells were measured using a conven-
tional 96-well plate by placing 200 ␮l of growth medium with or
without bacteria in the presence or absence of antibiotics. Then,
96 gold-plated electrodes (Daikin Corp., Japan) were fitted onto
the top of the plate and placed in the measurement chamber of
the sensor where the oxygen reduction current was monitored
continuously at a fixed potential of −700 mV for more than
8 h. This potential was selected following the oxygen reduction
current (cyclic voltammetric profiles) obtained with a conven-
tional electrochemical system. This corresponds to a maximum
value of the oxygen reduction current, which should provide
maximum sensitivity in terms of recognition. Negative controls
wells loaded with 200 ␮l of broth only were tested for each
experiment. Each experiment was simultaneously repeated four
times. The electrodes were used as disposable devices and each
plate was used for a single measurement. For antibiotic exper-
iments, the wells were first loaded with 100 ␮l Mueller Hinton
broth containing either 5.0 ␮g/ml chloramphenicol, 5.0 ␮g/ml
ampicillin or 1.0 ␮g/ml tetracycline followed by the addition of
100 ␮l of serial diluted cell cultures. Controls with antibiotics
only (no cells) were prepared by adding 100 ␮l of broth only
to the antibiotic prepped wells. The plates were incubated at
37 ◦ C according to the manufacturer’s instructions. In addition,
the system was allowed to equilibrate for at least 1 h before each
experiment. We also placed the system inside a laminar flow
hood, with the air flow turned off to shield it from any drastic
changes in ambient temperature.
2.2. Bacteria preparation
All bacteria strains were cultivated under identical experi-
mental conditions, including growth medium, incubation time
and temperature. Before each experiment, bacteria cultures were
grown overnight at 37 ◦ C in Mueller Hinton broth. The fresh cul-
tures were diluted to an OD600 of approximately 0.2 and then
serial diluted 1:2 four times. Hundred microliters of each of
the dilutions was added to the antibiotic prepared plates. Each
experiment was repeated four times simultaneously at −700 mV
for 6 h. Bacteria were quantified with the standard plate count
2645
method (Prescott et al., 1999) by correlating the colony forming
units to the UV–vis absorbance measurement at 600 nm.
2.3. Data acquisition and processing
Each bacteria sample was represented as a data vector, x = (x1 ,
x2 , x3 , . . ., xj , . . ., x360 ) with the components of the vector denot-
ing the time of the electrode current measurement, e.g., x1 is the
electrode current measured at the end of 1 min and x360 is the
electrode current measured at the end of 6 h. For pattern recogni-
tion analysis, the data were autoscaled. Each variable had a mean
of zero and a standard deviation of unity. Autoscaling removes
any inadvertent weighing of the data that would otherwise arise
from differences in magnitude among the measurements (current
at specific time intervals). After autoscaling, all of the measure-
ments have equal weight and therefore an equal effect on the
analysis.
3. Results and discussion
3.1. Sensor concept, design and pattern recognition
Fig. 1A shows what a DOX response may look like when the
microbial cell concentration is very high. In this case, the oxy-
gen is consumed rapidly and before sufficient time is allowed for
the antibiotics to take effect on the cells. The result is that there
is no differentiation between cells growing in medium only and
cells growing in the presence of antibiotics. The dissolved oxy-
gen is rapidly consumed and the current from oxygen reduction
at the electrode quickly falls to zero. Fig. 1C shows the extreme
opposite case where the cell concentration is so low that no
oxygen consumption is registered by the DOX system until suf-
ficient time is allowed for the cells to grow. However, during
this time, the antibiotics may completely inhibit the growth of
the cells before they multiply to the required concentration to
be registered on the DOX. In this case, it may not be possible
to differentiate the signals produced in the antibiotic prepped
samples from samples containing medium only. Finally, Fig. 1B
illustrates the DOX “fingerprint” produced by an appropriate
concentration of cells. Different antibiotics will produce oxy-
gen consumption curves that fall between those of medium only
and those of cells only. In this work, five types of bacteria were
examined to determine the concentration range required to pro-
duce the DOX fingerprint as well as to determine if any unique
signals were produced within this concentration range.
A genetic algorithm (GA) for pattern recognition analysis was
used to analyze the DOX fingerprint data. The pattern recogni-
tion GA identifies specific time intervals in the DOX profiles
that optimize the classification of the bacterial species in a plot
of the two largest principal components of the data. Because
principal components maximize variance, the bulk of the infor-
mation encoded by the selected features is about differences in
the DOX profiles of the bacteria types represented in this data
set. The principal component plot used by the fitness function
of the pattern recognition GA acts as an embedded information
filter. Features (specific points on the DOX plots) are selected
on the basis of their principal component plots, with a good
2646 J. Karasinski et al. / Biosensors and Bioelectronics 22 (2007) 2643–2649
Table 1
Bacteria concentrations used in the DOX–PCA experiments
Species of bacteria Concentration (×106 cfu/ml)
C1
C. acidovorans (ATCC #51340) 112
C. glutamicum (ATCC #13032) 79.3
S. epidermidis (ATCC #12228) 81.7
E. adecarboxylata (ATCC #23216) 45
E. coli (ATCC #25922) 150
a Maximum cell concentration that can produce a DOX fingerprint.
principal component plot generated by features whose variance
or information is primarily about bacteria type. This limits the
search to these types of features, thereby significantly reducing
the size of the search space. In addition, the pattern recognition
GA focuses on those classes and or samples that are difficult to
classify by boosting their weights over successive generations.
Samples that consistently classify correctly are not as heavily
weighted in the analysis as samples that are difficult to classify.
Over the time, the algorithm learns its optimal parameters in
a manner similar to a neural network. The pattern recognition
GA integrates aspects of artificial intelligence and evolutionary
computations to yield a “smart” one-pass procedure for feature
selection and classification. The efficacy and efficiency of the
pattern recognition GA has been demonstrated in previous stud-
ies (Lavine et al., 2001, 2002, 2003a,b, 2004; Lavine and Vora,
2005).
3.2. Experimental design and differentiation principle
The main objective of this work was to expand upon the con-
cept of using computational analysis to classify viable microbes
based upon their oxygen consumption patterns in the presence of
antibiotics. When developing a system such as this, there are sev-
eral factors that must be considered: (1) nature of bacteria—the
bacteria being investigated must be an aerobic or facultative
anaerobic species. Obviously, the bacteria must consume oxy-
gen. (2) Antibiotic selection and concentration—the antibiotics
chosen for this study are broad-spectrum antibiotics that affect
protein synthesis (chloramphenicol and tetracycline) or cell wall
synthesis (ampicillin) on a wide variety of microbes. Lethal con-
centrations of selective antibiotics can theoretically be used to
inhibit certain types of cells from consuming oxygen and sim-
plify data analysis or reduce interference. Another important
factor is the concentration of the antibiotics. The concentrations
must be low enough so microbial growth and/or respiration is
not totally inhibited but high enough to induce some reduction
in microbial metabolism that will result in decreased oxygen
consumption. The concentrations used here, 2.5 ␮g/ml for ampi-
cillin and chloramphenicol and 0.5 ␮g/ml for tetracycline were
optimized as described previously (Karasinski et al., 2005). (3)
Bacteria concentration—this issue is addressed in Fig. 1. One
of the observations made was that it takes about 1–2 h for the
antibiotics to have an affect on the oxygen consumption. There-
fore, if the cell concentration is very high and/or if the cells
C2 C3 C4 C5
56a 28 14 7.5
39.65 19.83a 9.91 4.96
40.85 20.43a 10.21 5.10
22.5 11.25a 5.63 2.81
75 37.5a 18.8 9.4
multiply very rapidly, there will be little or no difference in the
signals created by the antibiotic-loaded wells because the cells
will consume all of the available oxygen before the antibiotics
have an effect. For the experiments performed here, the five
bacteria were serial diluted, 2-fold, from a starting OD600 of
about 0.1 (after addition of 100 ␮l into the plate wells contain-
ing 100 ␮l of medium). The actual concentrations of the cells,
in terms of colony forming units, as determined by the plate
counts are shown in Table 1. The objective of this study was
to determine the concentrations of cells that would produce the
unique oxygen consumption fingerprint as illustrated in Fig. 1B
and classify the data using computational methods.
3.3. DOX antibiotic susceptibility data
In this data set, we examined five different types of cells with
varying concentrations and the results confirmed that at high
concentrations, all oxygen is consumed before any antibiotic-
induced changes. As the concentration of the cells was diluted,
the antibiotic-induced changes became apparent. Fig. 2 shows an
example of the raw data output from the DOX system produced
from various concentrations of S. epidermidis. At concentration
1, the dissolved oxygen is depleted at the same rate for cells
growing alone in medium and for the cells growing in the pres-
ence of the three antibiotics. As the concentration is decreased
to C2 and C3, the effects of the three antibiotics can be differen-
tiated from the cells growing alone in medium at about 100 min.
Typically, this is the amount of time required for the antibiotics
to affect cellular oxygen consumption for most cells. At concen-
trations C4 and C5, it can be seen from the raw data 0.5 ␮g/ml
tetracycline and 2.5 ␮g/ml ampicillin slow the oxygen consump-
tion of S. epidermidis. In contrast, 2.5 ␮g/ml chloramphenicol
nearly totally inhibits cellular oxygen consumption as the slope
of the signal begins to resemble the slope of the controls con-
taining medium and antibiotics only. It can also be seen from
visualization of the data that the oxygen consumption curves
have a unique shape for each concentration of S. epidermidis.
This was also the case for the other four bacteria tested using this
concept. From the raw data it can be determined that the max-
imum concentration of cells that can be used to create a DOX
fingerprint is C3 or approximately 20.4 × 106 cfu/ml for S. epi-
dermidis. The maximum cellular concentration that produced a
DOX fingerprint for E. coli, E. adecarboxylata, C. glutamicum
and C. acidovorans are noted in Table 1.
 J. Karasinski et al. / Biosensors and Bioelectronics 22 (2007) 2643–2649
Fig. 2. Raw DOX data for S. epidermidis at varying concentrations (C1 = 81.7 × 106 cfu/ml, C2 = 40.9 × 106 cfu/ml, C3 = 20.43 × 106 cfu/ml, C4 = 10.2 × 106 cfu/ml,
C5 = 5.1 × 106 cfu/ml). Controls showed no interference from antibiotics.
In this study, none of the cells were diluted to a point where the
DOX output reached the case illustrated in Fig. 1C. Therefore,
the minimum cell concentration that could be used to produce a
DOX fingerprint was not determined. Although it is possible to
determine this concentration, it may require substantially more
time than the 10 h monitoring time used in these experiments
and this becomes a question of practicality. At optimal concen-
trations, differentiation in the DOX signals can be observed in
less than 5 h. Based on the responses from the lowest concen-
tration of cells used, the data supports the idea that as little as
106 cfu/ml can be used to produce a response within 5 h. This is
a small amount considering that the total volume in the well of
the plate is 200 ␮l.
3.4. Quantification of bacteria using the DOX system
Viable microbes can also be quantified using the DOX sys-
tem. Linear calibration curves can be constructed by selecting
a threshold current and measuring the amount of time it takes
for different concentrations of cells to pass through the thresh-
old. High concentrations of cells will consume oxygen faster
than low concentrations and thus will pass through the selected
threshold in less time. We have found in these experiments that
a threshold value of 200 nA produces the most linear calibration
curves. The curves are constructed by plotting the cell concen-
tration versus the time required for each concentration to reach
200 nA in the DOX system. These results show that it is not only
possible to classify bacteria using the DOX–PCA antibiotic sys-
tem, but the bacteria can be quantified as well. Detailed data is
included in the online supplementary information. Comparison
of the DOX system with conventional cell culture techniques
2647
(Andreescu et al., 2005) indicates that significant advantage of
the DOX system is its simplicity and the fact that it operates via a
conventional 96-well plate. Validation with Tryptan Blue, MTT
assay and Fluorescence techniques confirmed that the DOX sys-
tem is compatible with any commercially available equipment
employed in cell biology. The sensor could be useful for mea-
suring toxic activity in an extremely simple one-step procedure,
without additional preparation steps or addition of reagents.
3.5. Cyclic voltammetry
In the previously reported work, the DOX antibiotic suscep-
tibility concept was validated by UV–vis experiments as well as
performing the traditional tube dilution antibiotic susceptibility
method (Prescott et al., 1999). In this study, cyclic voltammetric
studies were performed on E. coli K12 not only to compare the
results, but also to compare the feasibility in terms of assay time
and user intervention between the DOX system and conventional
electrochemical methods. The cells, adjusted to an OD600 of 0.1,
were placed in medium containing the same concentration of the
tetracycline, chloramphenicol and ampicillin as used in the DOX
experiments. The results between the DOX and the CV correlate
well and a description of the experimental setup as well as more
detailed discussion can be found in the accompanying online
supplementary information.
The CV comparison illustrates the convenience of the multi-
array DOX format. Using a 96-well plate, the DOX can make
four replicate measurements on six different types of growing
alone or in the presence of the three antibiotics simultaneously.
Typically, the total monitoring time is around 5 h and the required
sample volume is 9.6 ml. In contrast, to duplicate these measure-
2648 J. Karasinski et al. / Biosensors and Bioelectronics 22 (2007) 2643–2649

ments using a single CV instrument, it would take 480 h; require

960 electrode cleanings and over 1 l of sample.

3.6. Classification methods using principal component

analysis

The first step was to apply principal component analysis

(Jackson, 2003) to the DOX data. Principal component anal-
ysis is the most widely used multivariate analysis technique in
science and engineering. It is a method for transforming the
original measurement variables into new, uncorrelated variables
called principal components. Each principal component is a lin-
ear combination of the original measurement variables. Using
this method is analogous to finding a new coordinate system bet- Fig. 3. Plot of the two largest principal components of the untreated bacteria for
ter at conveying information present in the data than axes defined concentrations C1–C4 using 12 time domain features identified by the pattern
by the original measurement variables. This new coordinate sys- recognition GA. 1, C. acidovorans; 2, C. glutamicum; 3, S. epidermidis; 4, E.
tem is linked to variation in the data. The basis vectors of this adecarboxylata; 5, E. coli. Separation of the bacteria samples by species is
evident.
new coordinate system are the principal components. Often, only
the two or three largest principal components are necessary to
explain all of the information present in a data set if the data
contains a large number of interrelated measurement variables.
Using principal component analysis, dimensionality reduction,
classification of samples, and identification of outliers in high
dimensional data is possible.
When the two largest principal components of the untreated
bacteria for concentrations C1–C4 (concentration C5 was omit-
ted from the PCA analysis because of the difficulty in modeling
the discriminatory information contained in the DOX curves for
both high and low concentrations of bacteria) are plotted using
all time domain features, there is a significant overlap between
the five species of bacteria. This overlap is not surprising in view
of the similarity of their DOX profiles when no antibiotics are
present in the growth medium. The plot is available with the
online supplementary material.
The pattern recognition GA was used to uncover features Fig. 4. Plot of the two largest principal components of C. acidovorans and C.
glutamicum in untreated media for concentrations C1–C5 using six time domain
characteristic of the DOX profile of each species of bacteria. The
features identified by the pattern recognition GA. 1, C. acidovorans and 2, C.
genetic algorithm identified features by sampling key feature glutamicum.
subsets, scoring their principal component plots, and tracking
those samples or classes that were difficult to classify. The boost-
ing routine used this information to steer the population to an
optimal solution. After 100 generations, the genetic algorithm
identified 12 time domain features whose PC plot showed clus-
tering on the basis of class (see Fig. 3). The principal component
map of these 12 time domain features suggests that information
is present within the DOX profiles characteristic of the bacteria
species.
The effect of antibiotics on the ability of the DOX curves
to discriminate among the five species of bacteria represented
in this data set was investigated. For the training set described
above, we did not observe any improvement in our ability
to classify the bacteria through PCA analysis of the DOX
profiles when antibiotics (tetracycline, chloramphenicol, and
ampicillin) were added to the growth medium. We attributed
this result to the differential effect that antibiotics have on the
Fig. 5. Plot of the two largest principal components of C. acidovorans and C.
respiration rate of the bacterial species studied. In an attempt to glutamicum in treated media (tetracycline) for concentrations C1–C5 using 10
assess the validity of this hypothesis, several binary classifica- time domain features identified by the pattern recognition GA. 1, C. acidovorans
tion problems involving species identification were undertaken. and 2, C. glutamicum.
 J. Karasinski et al. / Biosensors and Bioelectronics 22 (2007) 2643–2649
Our pattern recognition results showed that tetracycline was the
most discriminating of the antibiotics investigated. The time
domain features used to develop these principal component
plots were selected by the pattern recognition genetic algorithm.
An example of the results obtained for two species of bacteria,
C. acidovorans and C. glutamicum, is shown in Figs. 4 and 5.
In this case, tetracycline increased the information content of
the DOX curves as analyzed by the pattern recognition GA to
discriminate between two closely related species of bacteria. We
consider this to be a significant result since a large concentration
range of bacteria was spanned in these two binary classification
studies.
4. Conclusion
Multi-electrode array sensors coupled with principal com-
ponent analysis (DOX–PCA) was used to compare oxygen
consumption profiles of single concentrations of cells growing
in the presence of sub-lethal concentrations of broad-spectrum
antibiotics. The concentration range of the cells was expanded
to examine the instrumental constraints of the DOX and to
determine if the data produced would be more discriminat-
ing when applying PCA. Depending on the cell type, the
upper concentration limit to produce a DOX fingerprint falls
between 11 and 56 × 106 cfu/ml, while the lower limit must
be above 1 × 106 cfu/ml if one wishes to obtain results in
under 8 h.
This work demonstrates that different types of bacteria can
be identified by simple use of the oxygen consumption curves
generated by a range of cellular concentrations. Depending
on the cell type, the addition of antibiotics to the growth
medium can increase the discrimination. However, the sys-
tem seems to work best when comparing cells adjusted to a
single known concentration. The attractiveness of the method
is that the DOX–PCA system recognizes small changes in
growth and/or metabolism that conventional optical antibi-
otic susceptibility methods do not. It is shown that these
changes are unique enough to identify different types of bac-
teria. This system could provide a unique tool for monitoring
antibiotic resistance by comparing changes in the DOX–PCA
fingerprint. The method is also versatile, with the possible com-
binations of antibiotics, antibiotic concentrations and number
of sample replicates only being limited by the plate itself.
Antibiotic-induced DOX responses can be easily tailored to
meet specific requirements. Additionally, the DOX produces
linear calibration curves that can be used to quantify bacte-
ria. To increase the discriminatory power of the DOX profiles,
each bacterial species can be analyzed over the same concen-
tration range. An experimental design of this type would make
it possible to employ more advanced pattern recognition tech-
niques such as multi-way analysis (Smilde et al., 2004) which
would allow for even more efficacious exploitation of the DOX
data.
2649
Acknowledgements
We acknowledge the following agencies for funding: Envi-
ronmental Protection Agency through the STAR program
(RD-83090601), National Science Foundation CHE-0513470
and the NYS Center for Advanced Technology (IEEC).
Appendix A. Supplementary data
Supplementary data associated with this article can be found,
in the online version, at doi:10.1016/j.bios.2006.10.037.
References
Avaniss-Aghajani, E., Jones, K., Chapman, D., Brunk, C., 1994. Biotechniques
17 (1), 144–146, 148–149.
Andreescu, S., Sadik, O.A., McGee, D.W., 2005. Anal. Chem. 18 (3), 466–474.
Andreescu, S., Karasinski, J., Sadik, O.A., 2006. In: Grimes, G.A., Dickey, E.C.,
Pishko, M.V. (Eds.), Encyclopedia of Sensors, vol. 2. American Scientific
Publishers, Valencia, pp. 333–348.
Bootsma, M.C.J., Diekmann, O., Bonten, M.J.M., 2006. PNAS 103 (14),
5620–5625.
CDC, 2006. Center for Disease Control [online]. Available from:
http://www.cdc.gov/drugresistance/community/anitbiotic-resistance.htm
(accessed 15 April 2006).
Fluit, A.C., van der Bruggen, J.T., Aarestrup, F.M., Verhoef, J., Jansen, W.T.M.,
2006. Clin. Microbiol. Infect. 12, 410–417.
Jackson, E.J., 2003. A User’s Guide to Principal Components. John Wiley &
Sons, NY.
Jorgensen, J.H., McElmeel, M.L., Crawford, S.A., 1998. J. Clin. Microbiol. 36
(3), 788–791.
Karasinski, J., Andreescu, S., Sadik, O.A., Lavine, B., Vora, M.N., 2005. Anal.
Chem. 77 (24), 7941–7949.
Kitahara, T., Koyama, N., Matsuda, J., Hirakata, Y., Kamihira, S., Kohno, S.,
Nakashima, M., Sasaki, H., 2003. Biol. Pharm. Bull. 26 (9), 1229–1234.
Lavine, B.K., Davidson, C.E., Breneman, W., 2003a. J. Chem. Inf. Sci. 43,
1890–1905.
Lavine, B.K., Davidson, C.E., Moores, A.J., 2002. Vib. Spec. 28, 83–95.
Lavine, B.K., Davidson, C.E., Vander Meer, R.K., Lahav, S., Soroker, V., Hefetz,
A., 2003b. Chemolab 66, 51–62.
Lavine, B.K., Vora, M., 2005. J. Chromatogr. A 1096, 69–75.
Lavine, B.K., Davidson, C.E., Moores, A.J., Griffiths, P.R., 2001. Appl. Spec.
55, 960–966.
Lavine, B.K., Davidson, C.E., Rayens, W.T., 2004. Comb. Chem. H. Throughput
Screening 7, 115–131.
Murdoch, D.R., Mirrett, S., Harrrell, L.J., Donabedian, S.M., Zervos, M.J.,
Reller, L.B., 2003. J. Clin. Microbiol. 41 (6), 2703–2705.
Nahed, M.A., Papadopol, R., Asmar, B.I., Brown, W.J., 2006. Int. J. Antimicrob.
Agents 27 (5), 449–452.
Prescott, L.M., Harley, J.P., Klein, D.A., 1999. Microbiology, 4th ed.
WCB/Mcgraw-Hill, Boston.
Rennie, R., Brosnikoff, C., Turnbull, L., Shokoples, S., Lovgren, M., Galbraith,
J., Brown, C., Kipke, L., Hart, N. 2004. As presented at the 14th European
Congress of Clinical Microbiology and Infectious Disease.
Smilde, A., Bro, R., Geladi, P., 2004. Multi-way Analysis. John Wiley & Sons,
Chichester, West Sussex.
Turenne, C.Y., Sanche, S.E., Hoban, D.J., Karlowski, J.A., Kabani, A.M., 1999.
J. Clin. Microbiol. 37 (6), 1846–1851.
Webster, L.F., Thompson, B.C., Fulton, M.H., Chestnut, D.E., Van Dolah, R.F.,
Leight, A.K., Scott, G.I., 2004. J. Exp. Mar. Boil. Ecol. 298, 179–195.
Whitlock, J.E., Jones, D.T., Harwood, V.J., 2002. Water Res. 36, 4273–4282.