J. Comb. Chem. XXXX, xxx, 000–000
A
Automated Image-Based Method for Laboratory Screening of Coating
Libraries for Adhesion of Algae and Bacterial Biofilms
Eraldo Ribeiro,† Shane J. Stafslien,*,‡ Franck Cassé,§ James A. Callow,§
Maureen E. Callow,§ Robert J. Pieper,‡ Justin W. Daniels,‡ James A. Bahr,‡ and
Dean C. Webster‡
Computer Sciences, Florida Institute of Technology, Melbourne, Florida 32901, Center for Nanoscale
Science and Engineering, North Dakota State UniVersity, Fargo, North Dakota 58102, and School of
Biosciences, The UniVersity of Birmingham, Birmingham B15 2TT, U.K.
ReceiVed March 20, 2008
Assessment and down-selection of non-biocidal coatings that prevent the adhesion of fouling organisms in
the marine environment requires a hierarchy of laboratory methods to reduce the number of experimental
coatings for field testing. Automated image-based methods are described that facilitate rapid, quantitative
biological screening of coatings generated through combinatorial polymer chemistry. Algorithms are described
that measure the coverage of bacterial and algal biofilms on coatings prepared in 24-well plates and on
array panels, respectively. The data are used to calculate adhesion strength of organisms on experimental
coatings. The results complement a number of physical and mechanical methods developed to screen large
numbers of samples.
Introduction
The majority of antifouling coatings in current use are
based on biocides, such as copper,1 and organic compounds
that readily degrade once released from the paint.2 Increasing
environmental concerns and escalating costs of registration
and regulatory compliance have resulted in the need to find
alternative environmentally benign technologies to control
biofouling.3,4 Non-biocidal coatings, the so-called foulingrelease coatings exemplified by silicone elastomers, are not
inherently antifouling but, organisms attach only weakly and
therefore are released by hydrodynamic forces like those
experienced by a fast moving boat.5 Screening the antifouling
potential of biocidal antifouling paints is relatively straightforward. New formulations are typically painted onto large
panels that are hung from rafts, and the presence of different
types of fouling organisms is scored over time. Non-biocidal
formulations necessitate more complex methods of assessment, and laboratory evaluations are frequently employed
to aid down-selection of formulations for field testing.
Because the settlement of fouling organisms is rarely
inhibited on non-toxic coatings, a measure of the relative
adhesion strength of the attached organisms is required.
Moreover, because different fouling organisms appear to
adhere to different degrees,6,7 a number of test organisms
need to be employed.
North Dakota State University has developed a highthroughput capability to produce novel environmentally
benign coatings for the marine environment.8–10 Because the
* To whom correspondence should be addressed. Phone: 701-231-5826.
Fax: 701-231-7916. E-mail: Shane.Stafslien@ndsu.edu.
†
Florida Institute of Technology.
‡
North Dakota State University.
§
The University of Birmingham.
10.1021/cc800047s CCC: $40.75
number of formulations generated from one concept could
be as many as several hundred, down-selection on the basis
of laboratory assays is essential. Typically, the first downselection is based on various surface and mechanical tests,8,10
followed by a second down-selection based on a number of
biological assays.11–15 Coatings can be deposited in a range
of formats for subsequent evaluation including 24-well
plates11–15 and array panels.16
This paper reports methods to quantify biofilm coverage
on coating surfaces using new algorithms that allow quantification of a large number of samples. The biofilm
quantification methods presented in this paper are based on
Gaussian mixture-model color segmentation techniques. The
technique was successfully applied to two main types of
biofilm quantification. The first was the detection of retracted
bacterial biofilm on coatings cast in 24-well plates. The
second was the quantification of algal biomass on 12-patch
array panels. Quantification results obtained in this study are
promising and demonstrate the effectiveness of the automated
biofilm quantification methodology.
Two groups of organisms, namely, bacteria and algae were
employed in two formats, namely, 24-well plates and array
panels, respectively. Bacterial biofilm coverage on coatings
deposited in wells of 24-well plates was quantified after
drying and staining. Coverage of patches of coatings
deposited in groups of 12 on array plates was determined
after a lawn of young plants of the green macroalga UlVa
cultured on the array panels was sprayed at specific impact
pressures by a water jet. The set of coatings used to
demonstrate the methods was a group of 24 siloxane-acrylicpolyurethane coatings.17
The marine bacterium Cellulophaga lytica (formerly
Cytophaga lytica) was used to evaluate microbial biofilm
XXXX American Chemical Society
Published on Web 06/20/2008
B
Journal of Combinatorial Chemistry, XXXX Vol. xxx, No. xx
Ribeiro et al.
Figure 1. Experimental design of combinatorial acrylic polyol
library.
adhesion to the siloxane-polyurethane coating surfaces.13
Measurement of the strength of attachment of young plants
(sporelings) of the green macroalga UlVa to surfaces provides
an indication of the fouling-release properties of higher
fouling organisms.18–21
Experimental Section
Coating Library Synthesis. The preparation of the
siloxane-acrylic-polyurethane coating library has been reported in detail elsewhere.17 Briefly, a library of 24 acrylic
polyols was synthesized by solution free radical polymerization. The monomers consisted of hydroxyl ethyl acrylate,
which was varied from 5 to 20 wt %, and butyl acrylate and
butyl methacrylate, which were varied systematically from
a ratio of 0:100 to 100:0. The experimental design is shown
in Figure 1. The acrylic polyols were then incorporated into
a siloxane-acrylic-polyurethane coating formulation consisting of 10% of a 3-aminopropyl-terminated PDMS (10,000
g/mol)22 and an isocyanate cross-linker (Tolonate HDT from
Rhodia). The ratio of isocyanate to amine + hydroxyl
functional groups was 1.1:1.0. The formulation also consisted
of 2,4-pentanedione as a pot life extender, dibutyl tin
diacetate as cure catalyst, and solvents toluene, methyl amyl
ketone, and ethyl ethoxy propionate. Control silicone elastomers DC 3140 and T2 Silastic were received from
Ellsworth Adhesives. DC 3140 and T2 Silastic were solvent
reduced with MIBK and called silicone A and silicone B,
respectively.
Coating Application and Curing. For bacterial analysis,
the siloxane-urethane coatings were solvent cast into modified 24-well polystyrene plates as described previously.11
Each unique coating formulation was deposited into one
column of a 24-well plate, providing four coating replicates
for analysis. The first column of each array plate contained
the standard siloxane coating A.
A Symyx Coating Application system was used to apply
formulations to 4 × 8 in. aluminum panels for algal analysis;8
0.1 mL of each coating formulation was deposited on the
panels using disposable pipettes to prevent cross contamination, and then a robotic doctor blade spread the coating at a
wet film thickness of 200 µm. The automated doctor blade
was washed with toluene under sonication three times after
each application and dried with an air knife. After completion
of deposition of all the coatings, the panels were removed
from the coating application station, placed into an enclosed
drying cabinet with ventilation, and left overnight to cure at
Figure 2. Digital imaging station used to capture high resolution
photographs of 24-well coating array plates.
ambient temperature. Coatings were cured at 100 °C for one
hour the following day.
Bacterial Bioassay. The measure of bacterial biofilm
adhesion to the siloxane-urethane coating surfaces was
determined by using a high-throughput bacterial biofilm
retention and retraction assay.13 An overnight culture of the
marine bacterium Cellulophaga lytica (formerly Cytophaga
lytica) was resuspended in sterile artificial seawater supplemented with nutrients (∼107 cells ml-1). Rows 2-4 were
inoculated with 1 mL of the C. lytica suspension, while row
1 received nutrient medium only. The coating array plates
were incubated statically for 18 h at 28 °C in a temperature
controlled water bath. After 18 h of incubation, the array
plates were rinsed three times with deionized water and dried
for ∼1 h at ambient conditions. The array plates were then
stained with a biomass indicator dye, crystal violet (CV),
for 15 min, and rinsed three times with deionized water to
remove excess CV.
High-resolution images of the CV-stained coating array
plates were acquired with a Nikon D200 digital single lens
reflex camera set at 10 megapixels per image. The camera
was permanently mounted to a modified 1 m high copy stand
and controlled by Nikon Camera Control Pro software to
provide a stable hands-off configuration (Figure 2). A
Tamron 28-300 mm macrozoom lens was used to capture
the entire array plate in a single image and at a distance
sufficient enough to minimize obstruction from the well
walls. Array plates were loaded into the recess on the copy
stand deck and images were collected through the software.
After digital images of each array plate were taken, CV
was extracted from the biofilm retained on the coating
surfaces by addition of 0.5 mL of 33% glacial acetic acid.11
0.15 mL of the resulting eluates were transferred to a 96well plate and measured for absorbance at 600 nm with a
multiwell plate reader. The absorbance value reported for
each siloxane-urethane coating was the mean value of three
replicate wells. Error bars represent one standard deviation
of the mean.
Screening of Coating Libraries for Adhesion
Journal of Combinatorial Chemistry, XXXX Vol. xxx, No. xx C
Algal Bioassay. Zoospores were obtained from fertile UlVa
linza plants as described in Callow et al.,23 and the spore
concentration was adjusted to 1 × 106 spores ml-1. Three
replicates of each of the three array panels were incubated
in trays (three plates in each tray) of 30.5 × 21 cm with 300
mL of zoospore suspension for 3 h in darkness at ∼20 °C.
During this time, a proportion of the total number of spores
settled, that is, adhered to the surface panels. Unattached
(swimming) spores were removed by washing the plates
gently in artificial seawater.
The panels were transferred to glass tanks (19 × 20 × 60
cm) containing 10 L of 10% strength enriched seawater
medium.24 Three plates were incubated in each of three tanks,
and the position of the plates was changed daily. Each tank
was fitted with a pump that recirculated the medium inside
the tank at 200 L h-1. Illumination was provided by two
daylight fluorescent tubes placed above the tank (Boyu PL18 W) that provided an irradiance of 32 µmol m-2 s-1. After
5 days, the spores had germinated, and the sporelings (young
plants) formed a green mat over the surface of the plates
(Figure 3).
Each set of three plates was exposed to an automated water
jet at a different uniform surface pressure. The water jet25
was programmed to raster six times across the middle of
each of the three rows of samples. The surface impact
pressures used were 93, 151, and 171 kPa. Photographs were
taken of each plate before and after jet washing using a digital
camera (Sony DSC-S75) set at 1.8 Mega Pixels.
Image Segmentation for Biofilm Retraction Quantification. The biofilm quantification algorithm described in this
section uses a color segmentation method based on a
Gaussian mixture model.26 In this model, the color of biofilm
pixels is represented by probability distributions. The work
in this paper describes two applications of the procedure.
The first of these is the detection of retracted bacterial biofilm
on coatings cast in 24-well plates. The second is the
quantification of algal biomass on 12-patch array panels.
The goal of the algorithm is to classify all image pixels
according to their estimated color models. The result of this
classification procedure is an image segmented into two
distinct regions: biofilm regions and non-biofilm regions.
Once these regions are at hand, quantifications such as
percent biofilm surface coverage estimation can be obtained
by means of a straightforward frequency count of the labeled
pixels.
Modeling Biofilm Color Distributions. The key assumption of the method is pixel color is a realization of a
multimodal distribution of color segments. This multimodal
distribution can be represented by a Gaussian mixture
model26 of K components given by
k
p(x|Θ) )
∑ Ripi(x|θi)
(1)
i)1
where x is a color vector,Ri represents the mixing weights
k
such that Σi)1
Ri ) 1, Θ represents the collection of parameters (R1, ..., RK, θi, ..., θK) with θ ) (µ, Σ), and pi is a
multivariate Gaussian density function as given by
F(xi| j) )
1
2π|
∑j|
1⁄2
{
1
exp - (xi - µj)T
2
∑ j-1 (xi - µj)} (2)
where xi ) [H,S]T is the color of the i-th pixel in the image,
µj represents the two-dimensional mean vector of the j color
segment, and Σj its 2 × 2 covariance matrix. |Σj| is the
determinant of Σj. In this paper, images are represented using
the HSV (hue, saturation, and value) intensity decoupling
color model.27 The color of each image pixel is represented
by a vector formed by the hue and saturation components in
the HSV model (i.e., chromaticity components). Neglecting
the relative luminance component V helps reduce the
algorithm’s sensitivity to small illumination variations.
Biofilm segmentation is accomplished by estimation of the
parameters of the densities describing the image’s color
populations (eq 1). The final pixel labeling procedure consists
of determining the color segment model with the maximum
likelihood for each image pixel. Each component of the
Gaussian mixture represents a color segment in the image.
If x is assumed to be sampled independently from p(x), then
the probability of the whole image is given by
K
L (Θ) )
N
k
p(x|Θ) ) ∏ ∑ R ip(x| j)
∏
i)1
i)1 j)1
(3)
In general, there are two ways to obtain the parameters of
the above density functions. If prior information about the
labeling of pixels belonging to the biofilm and non-biofilm
classes is available, the density estimation is called superVised. In this case, it is assumed that the user is able to
provide a selection of regions containing only biofilm pixels,
as well as a region containing only non-biofilm pixels. Once
these regions are at hand, the estimates of the mean vector
and the covariance matrix in eq 2 for each color population
are given by
N
µ̂j )
∑
1
x
N i)1 i
and
N
∑
1
Σˆ j )
(x - µ̂j)(xi - µ̂j)T
N i)1 i
(4)
where N is the total number of pixels in the selected image
region. On the other hand, if no labeling information is
available, the estimation is said to be unsuperVised. Unsupervised classification allows for reduced interaction and fully
automated segmentation methods. In this case, the input to the
segmentation algorithm is simply the image to be segmented.
The algorithm must estimate pixel membership and density
parameters simultaneously by maximizing eq 3 with respect to
the model parameters. This maximization problem is solved
using the expectation-maximization (EM) algorithm.28 In the
segmentation method proposed in this paper, a K-means
algorithm29 is used to obtain the initial values for the unknown
mean vectors µ1, ..., µk. The initial covariance matrices Σ1, ...,
Σk are chosen to be the identity matrices.
Detecting Retracted Biofilm Regions. Digital images of
the coating array plates were loaded into the graphical user
interface (GUI) of the percent coverage software program. A
mask was then generated and applied to the loaded image
(Figure 4). The percent biofilm coverage for each well of the
Journal of Combinatorial Chemistry, XXXX Vol. xxx, No. xx
Ribeiro et al.
Figure 3. Biomass of UlVa sporelings on array panels after 5 days
growth and after being sprayed with a hose at three different impact
pressures. The top row shows the configuration and codes of the
36 paint patches (24 novel formulations) on the three array panels.
Two patches of two standard siloxanes (standard A and B) are
represented on the left-hand side of every panel. The first row of
images shows a set of three panels before spraying. The second,
third, and fourth row of images shows sets of panels after hosing
at 93, 151, and 171 kPa of impact pressure, respectively. The panels
sprayed at 93 kPa were those shown before they were sprayed (top
row of images). The black crosses indicate coating failure hence
these patches were excluded from the analysis.
plate was determined and exported to Microsoft Excel by
executing the software. The percent coverage value for each
coating was reported as the mean value of three replicate wells.
Error bars represent one standard deviation of the mean.
The concentration of the crystal violet staining color on
the coating surfaces is the main discriminative visual
information used for the detection of bacterial biofilm.
Traditionally, bacterial biofilm detection is achieved by
visually distinguishing test well regions in which the staining
color appears distinctively darker than regions in corresponding control wells. Color variations in the control wells results
from the crystal violet dye binding or interacting with the
coating material itself. The actual color distribution on the
coating surfaces is found to be multimodal, having at least
three main dominant components. The first component is the
bright color of the clean background. This color is usually
the most frequent in the control wells. It is also present in
regions from which the bacterial biofilm has retracted on
the coating surface. The second color segment is the retracted
bacterial biofilm represented by the crystal violet dye used
in the experiments. Nevertheless, some of the violet dye
might also appear in the control wells. In this case, it appears
in significantly low frequencies. The third color segment
consists of noisy pixel colors originated from illumination
artifacts such as reflections and specularities.
The biofilm segmentation method estimates the parameters
of this three-mode color pattern distribution in the multiwell
plates using the EM algorithm28 as described previously. This
procedure is performed separately for each well in the plate.
The majority of pixels in the control wells should belong to
the background color class. This information is used to
determine the label of the background color segment. The
other two labels are easily determined by using the strong
blue component of the violet dye as a disambiguating clue.
At the end of this step, the labeled model densities and prior
distributions represent the colors for the biofilm, background,
and noise regions, respectively.
However, the sporadic presence of violet color pigmentation
creates the need for a further comparison of the distributions
of the labeled biofilm components. Here, the focus in on the
violet color density only. This additional step is performed to
determine a threshold indicating the main difference between
violet color distributions in the control well and the one in the
corresponding test wells. This procedure is performed separately
for each column of the multiwell plate (i.e., a set of corresponding control and test wells).
To obtain a simple threshold value, a combined chromaticity map of the biofilm pixels is created. This map is given
by y(i) ) h(i) × s(i) and allows for a one-dimensional
distribution of the violet color with values ranging from
brighter to darker violet. Consequently, the value of the
segmentation threshold for of the settled biofilm with respect
to a given control well can be set to the maximum value of
the combined chromaticity map, that is, T ) maxiyc(i), where
yc(i) is the combined chromaticity map of the biofilm color
component in the corresponding test well. The retracted
biofilm pixel map is then given by
D
Screening of Coating Libraries for Adhesion
br(i) )
{
1 if yt(i) g T
0 otherwise
Journal of Combinatorial Chemistry, XXXX Vol. xxx, No. xx E
(5)
Here, yt(i) is the combined chromaticity map of the biofilm
color component in the control well. Finally, the calculation
of percent cover is given by pt ) (100/N)∑ibr(i), where N is
the total number of pixels in the region of interest in the test
well image. Algorithm 1 summarizes the main steps of the
process (Figure 5A).
Quantifying Biomass of UlWa. Quantification of biomass
of the green alga UlVa is another application of the color
segmentation methods described in this paper. The biofilm’s
dominant color is green instead of violet. In addition, the
segmentation problem is mostly a binary one. The goal of
the quantification is to determine the area percentage of the
high-pressure water jet removed algae on the test panels.
The supervised segmentation approach is used in this case.
The user provides two selected image subregions as input
to the program. The two image subregions contain pixels
from algae and non-algae regions, respectively. The program
estimates the density parameters of each pixel color population using the equations in 4. The resulting labeling of algal
pixels is accomplished by a maximum likelihood classification. The percent of remaining algal biomass is calculated
as described previously in this paper.
Another important part of the method to quantify algal
biomass is the perspective distortion removal step. This step
allows for the alignment of UlVa panels that present
perspective distortion because of the angle at which the
images are acquired (Figure 6). The rectification step helps
ensure the repeatability of the experiments. The details of
Figure 4. Coating array plate image loaded into the graphical user interface (top). Mask applied to the plate image (bottom). White regions
of the mask show the precise area used to calculate percent surface coverage for each well of the coating array plate.
F
Ribeiro et al.
Journal of Combinatorial Chemistry, XXXX Vol. xxx, No. xx
Figure 5. Software algorithm used to quantify (A) bacterial biofilm and (B) algal percent surface coverage on multiwell coating plates and
coating array panels, respectively.
the planar rectification process30 is summarized as follows.
The perspective distortion observed for the UlVa panel
images corresponds to a plane-to-plane transformation and
can be represented by a 3 × 3 planar projective transformation, H, also called planar homography. Under planar
homography, image points are mapped as
x ) Hx′
(6)
where x and x′ are represented in homogeneous coordinates.
Pixel coordinates in the distorted images are represented by
x ) (x, y, 1)T. Pixel coordinates in the undistorted (i.e.,
rectified) image are given by x ) (x1, x2, x3)T. Here, we focus
on the case of planar projective transformation. The geometric transformation has 8 degrees of freedom. The image
points x and x′ correspond to the same point X in the world.
To solve for the transformation H, we need at least four pairs
of corresponding points in the images. In our current
implementation, the user manually selects these points using
the mouse. Once the four pairs of points are determined, the
algorithm estimates the transformation H and rectifies the
panel images to a fronto-parallel view using the homography
mapping in eq 5. Algorithm 2 summarizes the main steps of
the process (Figure 5B).
Results
Twenty-four-Well Plate Evaluations. Figure 7 shows the
images of the coating array plates after crystal violet staining.
A large variation in the degree of surface coverage of retained
C. lytica biofilm on the siloxane-polyurethane coatings was
observed. To facilitate direct comparison, a representative image
of each coating is also shown in Figure 7. The data in Figure
8A and B illustrates two methods of analysis of the biofilm on
the test coatings. Figure 8A shows total biomass quantified
following extraction of crystal violet. In general, the amount
of retained biomass was similar throughout the siloxanepolyurethane coating library. The percent coverage measurements for each experimental coating using the automated
software program are shown in Figure 8B. Several coatings
exhibited a high degree of biofilm retraction (i.e., low surface
coverage) indicating reduced C. lytica biofilm adhesion to those
surfaces.13 Examination of the coating array plate images shows
that a significant amount of crystal violet dye bound to three
of the experimental siloxane-polyurethane compositions (B6,
C6, and D6) (Figure 7). In the case of coating D6, the binding
of crystal violet dye did not significantly interfere with the
percent coverage analysis as the biofilm retraction was moderate
on this surface. However, a high degree of biofilm retraction
was observed on coatings B6 and C6. The dense areas of
retracted biofilm on these coating surfaces were adequately
detected by the chromaticity threshold determination method
described above (Figure 9).
Twelve-Patch Array Panel Evaluations. There was a
complete coverage of algal sporelings over all nine panels
after 5 days growth (only one set is shown in Figure 3).
Each set of panels was exposed to a different surface
impact pressure by hosing with a calibrated water jet. The
sets of panels after they were sprayed at 93, 151, and 171
kPa, respectively, are shown in Figure 3. It can be seen
that different formulations release the algal biomass
growing on the surface more effectively than others.
Figure 3 also shows that biomass removal increases with
increasing impact pressure.
The percent coverage of biomass (green color) on every patch
was quantified using the software described above. Figure 10A
shows percent removal at the three impact pressures for five
formulations, chosen to represent the range of fouling release
performance from good to bad. From such plots, the critical
impact pressure for removal of 50% coverage of biomass can
be determined. Figure 10B shows the experimental formulations
in terms of critical pressure. Six experimental coatings coatings
(A2, B5, C3, C4, D1, D3) and standards A and B are excluded
from the plot because 50% removal was not achieved at the
highest impact pressure used.
Discussion
Large numbers of experimental coatings can be generated
very quickly using a combinatorial, high-throughput ap-
Screening of Coating Libraries for Adhesion
Journal of Combinatorial Chemistry, XXXX Vol. xxx, No. xx G
Figure 6. Illustration of the image analysis program that calculates percentage coverage of UlVa sporelings on the panels. The image shows
panel 2 after it was sprayed at 151 kPa of impact pressure. The zone of interest has been selected on patch B4, followed by the foreground
and background. The percent cover calculation highlights the green biomass inside the zone of interest and therefore provides the percentage
cover inside the selected area.
Figure 7. C. lytica biofilm surface coverage on the siloxaneurethane coatings after staining with crystal violet. The first column
of each array plate contained standard siloxane A (SA). Columns
2-6 contained experimental siloxane-urethane coatings. The top
row of each array plate was inoculated with nutrient medium only.
Rows 2-4 were inoculated with a nutrient medium suspension of
C. lytica, providing three replicate samples for analysis. The bottom
right image shows a representative well for each of the 24 unique
siloxane-urethane coating compositions.
proach.31 This necessitates the development of automated
toolsets and assays to quickly screen coating libraries for
various properties and identify promising compositions that
may warrant further investigation. In the case of antifouling
marine coatings development, one of the primary performance screens is typically a biological laboratory assay.
These assays use one or more array formats (i.e., 24-well
plates, 12-patch panels) that allow for rapid and efficient
analysis of experimental coatings in parallel.
The high-throughput biological screening methods developed
to date use a variety of spectrophotometric and imaging methods
to quantify the biomass of marine organisms biomass on the
surface of coating arrays. Although effective, the imaging
methods employed are tedious, time-consuming, and require
manual manipulation and analysis of each coating image by
the investigator. This includes the determination of percent
surface coverage of the marine organism for measurement of
adhesion strength on fouling-release coatings.13,16
We have reported here on the development of an automated imaging software tool to quantify bacterial and algal
percent coverage on coating arrays. It is important to
emphasize that biofilm coverage of the coating surface and
not total biomass is being quantified by the algorithms
described here. The difference is shown clearly for the
bacterial data, where extraction of dye revealed only small
differences between the samples (Figure 8A) compared to
quantification of the area of surface covered by biofilm
(Figure 8B). A measure of surface coverage is the more
relevant parameter when down-selecting coatings on the basis
of their fouling release potential.
The accuracy of analysis is greatly improved by the imaging
software as an algorithm is consistently applied to each coating
image, rather than a subjective calculation made by the
investigator. In addition, the throughput of coating analysis is
H
Journal of Combinatorial Chemistry, XXXX Vol. xxx, No. xx
Ribeiro et al.
Figure 8. Assessment of C. lytica biofilm retention and retraction on the siloxane-urethane coatings using the HTBRRA. (A) C. lytica biofilm
retention as determined by absorbance measurements (600 nm) of crystal violet extractions in 33% glacial acetic acid. (B) C. lytica biofilm
retraction as determined by percent surface coverage measurements obtained from digital images of crystal violet stained coating array plates.
Each data point was reported as the mean value of three replicate coating samples. Error bars represent one standard deviation of the mean.
significantly enhanced because each coating of an array plate
is analyzed simultaneously by the software program and
exported to a common database. As many as 200-300 patches
on array panels or 24-well plates can be analyzed in one
experiment (180-360 experimental patches/wells). The limiting
factor in the number of coating arrays that can be analyzed in
one experiment is the amount of time needed for plate
processing (i.e., rinsing, staining, and water jetting, etc.), rather
than the time needed to process the images. The only requirement is that good quality digital images of the coating array
plates are obtained for accurate and efficient analysis. In certain
instances, reflected light may be captured on the coating surface
during the imaging process and compromise the ability of the
imaging software to adequately quantify percent coverage. This
phenomenon can potentially be avoided by the use of an
appropriate lighting tent.
The ultimate goal of the automated imaging software
tool is to aid in the rapid and accurate identification of
promising coatings compositions from large numbers of
candidate materials. Because there are many species of
marine organisms responsible for fouling, it is important
that the primary biological laboratory screening assays
include more than one particular organism to appropriately
down-select coatings.
Figure 9. (A) HSV-based color distribution models for a siloxaneurethane control and test well. Color distribution models are used
to accurately discriminate between crystal violet stained biofilm
(test well) and crystal violet stained coating surface (control well).
(B) Images of control wells, test wells, and detected regions of
retracted biofilm for siloxane-urethane compositions B6 and C6.
Screening of Coating Libraries for Adhesion
Figure 10. Fouling release performance as assessed by the ease of
removal of UlVa sporelings. (A) Five patches (A3, A6, B1, C6,
and D5) were selected out of 24 experimental coatings to illustrate
the range of release performance at three different impact pressures.
(B) Plot showing the impact pressure required to achieve 50%
removal (critical pressure) from each coating surface. Coatings A2,
B5, C3, C4, D1, D3 and standards A and B are not included because
less than 50% biomass was removed at the highest pressure.
As shown in Figure 10B, analysis of the coatings with
UlVa indicates that two coatings, A6 and B6, had superior
fouling release performance (i.e., lowest critical removal
pressure). These two coatings also exhibited a low bacterial
biofilm surface coverage (i.e., high degree of biofilm
retraction). However, several other coatings that showed a
high degree of biofilm retraction (B1, B5, C1, C2, and C6)
performed quite poorly in terms of the release of UlVa
biofilm. If the amount of coating material, labor, or cost of
testing, etc., is not a limiting factor, all coatings that
performed well in both assays could be scaled up for ocean
testing. In reality, the data from these two bioassays would
be considered alongside adhesion data for pseudobarnacles,32
diatoms (i.e., slime forming unicellular algae),15 and barnacle
adhesion.33 Thus, the 200-300 coating formulations evaluated with bacteria and algae would typically be reduced to
approximately 10 for full ocean testing.
Acknowledgment. Financial support from the Office of
Naval Research through ONR Grants N00014-05-1-0822 and
N00014-06-1-0952 is gratefully acknowledged.
References and Notes
(1) Finnie, A. A. Biofouling 2006, 22, 279–291.
(2) Turley, P. A.; Fenn, R. J.; Ritter, J. C.; Callow, M. E.
Biofouling 2005, 21, 31–40.
(3) Swain, G. E. Paint Coat Eur. 1999, 18–25.
(4) Genzer, J.; Efimenko, K. Biofouling 2006, 22, 339–360.
Journal of Combinatorial Chemistry, XXXX Vol. xxx, No. xx I
(5) Kavanagh, C. J.; Quinn, R. D.; Swain, G. W. J. Adhes. 2005,
81, 843–868.
(6) Holland, R.; Dugdale, T. M.; Wetherbee, R.; Brennan, A. B.;
Finlay, J. A.; Callow, J. A.; Callow, M. E. Biofouling 2004,
20, 323–329.
(7) Holm, E. R.; Kavanagh, C. J.; Meyer, A. E.; Wiebe, D.;
Nedved, B. T.; Wendt, D.; Smith, C. M.; Hadfield, M. G.;
Swain, G.; Darkangelo, C.; Truby, K.; Stein, J.; Montemarano,
J. Biofouling 2006, 22, 233–243.
(8) Webster, D.; Bennett, J.; Keubler, S.; Kossuth, M.; Jonasdottir,
S. JCT Coat. Technol. 2004, 1, 35–39.
(9) Webster, D. C.; Chisholm, B. J.; Stafslien, S. J. Biofouling
2007, 23, 179–192.
(10) Chisholm, B. J.; Webster, D. C.; Bennett, J. C.; Berry, M.;
Christianson, D.; Kim, J.; Mayo, B.; Gubbins, N. ReV. Sci.
Instrum. 2007, 78.
(11) Stafslien, S. J.; Bahr, J. A.; Feser, J. M.; Weisz, J. C.; Chisholm,
B. J.; Ready, T. E.; Boudjouk, P. J. Comb. Chem. 2006, 8, 156–
162.
(12) Stafslien, S.; Daniels, J.; Chisholm, B.; Christainson. D.
Biofouling 2007, 23, 37–44.
(13) Stafslien, S.; Daniels, J.; Mayo, B.; Christianson, D.; Chisholm,
B.; Ekin, A.; Webster, D.; Swain, G. Biofouling 2007, 23, 45–
54.
(14) Stafslien, S. J.; Bahr, J. A.; Daniels, J. W.; Vander Wal, L.;
Nevins, J.; Smith, J.; Schiele, K.; Chisholm, B. ReV. Sci.
Instrum. 2007, 78.
(15) Cassé, F.; Stafslien, S. J.; Bahr, J. A.; Daniels, J.; Finlay, J. A.;
Callow, J. A.; Callow, M. E. Biofouling 2007, 23, 121–130.
(16) Cassé, F.; Ribeiro, E.; Ekin, A.; Webster, D. C.; Callow, J. A.;
Callow, M. E. Biofouling 2007, 23, 267–276.
(17) Pieper, R. J.; Ekin, A.; Webster, D. C.; Casse, F.; Callow, J. A.;
Callow, M. E. J. Coat. Technol. Res. 2007, 4 (4), 453–461.
(18) Chaudhury, M. K.; Finlay, J. A.; Chung, J. Y.; Callow, M. E.;
Callow, J. A. Biofouling 2005, 21, 41–48.
(19) Tang, Y.; Finlay, J. A.; Kowalke, G. L.; Meyer, A. E.; Bright,
F. V.; Callow, M. E.; Callow, J. A.; Wendt, D. E.; Detty,
M. R. Biofouling 2005, 21, 59–71.
(20) Krishnan, S.; Ayothi, R.; Hexemer, A.; Finlay, J. A.; Sohn, K. E.;
Perry, R.; Ober, C. K.; Kramer, E. J.; Callow, M. E.; Callow,
J. A.; Fischer, D. A. Langmuir 2006, 22, 5075–5086.
(21) Statz, A.; Finlay, J.; Dalsin, J.; Callow, M.; Callow, J. A.;
Merrersmith, P. B. Biofouling 2006, 22, 391–7399.
(22) Ekin, A.; Webster, D. C. J. Polym. Sci., Part A: Polym. Chem.
2006, 44, 4880–4894.
(23) Callow, M. E.; Callow, J. A.; Pickett-Heaps, J. D.; Wetherbee,
R. J. Phycol. 1997, 33, 938–947.
(24) Starr, R. C.; Zeikus, J. A. J. Phycol. 1987, 23, S1–S27.
(25) Finlay, J. A.; Callow, M. E.; Schultz, M. P.; Swain, G. W.;
Callow, J. A. Biofouling 2002b, 18, 251–256.
(26) Figueiredo, M. A. T.; Jain, A. K. Unsupervised learning of
finite mixture models. IEEE Transactions on Pattern Analysis
and Machine Intelligence I.; IEEE: New York, 2002; Vol.
24, pp 381-396.
(27) Gonzalez, R. C.; Woods, R. E. Digital Image Processing;
Addison-Wesley Longman Publishing Co., Inc.: Boston, MA,
2001
(28) Dempster, A.; Laird, N. M.; Rubin, D. B. J. R. Stat. Soc.,
Ser. B 1977, 39, 1–38.
(29) Duda R., H. P.E.; S. D.G. Pattern Classification; John Wiley
and Sons, Inc.: New York, 2001.
(30) Hartley, R. I.; Zisserman, A. Multiple View Geometry in
Computer Vision; Cambridge University Press: Cambridge,
U. K., 2000; ISBN 0521623049.
(31) Webster, D. C. JCT Coat. Technol. 2005, 2, 24–29.
(32) Kim, J.; Chisholm, B. J.; Bahr, J. Biofouling 2007, 23, 113–
120.
(33) Rittschof, D.; Orihuela, B.; Stafslien, S.; Daniels, J.; Chisholm,
B.; Christianson, D.; Holm, E. Biofouling 2008, 24, 1–9.
CC800047S