METHODS FOR MICROSCOPIC
CHARACTERIZATION OF ORAL BIOFILMS:
ANALYSIS OF COLONIZATION, MICROSTRUCTURE,
AND MOLECULAR TRANSPORT PHENOMENA
T
S. SINGLETON
R. TRELOAR
P. WARREN
G.K. WATSON
R. HODGSON
C. ALLISON
Unilever Dental Research
Port Sunlight Laboratory
Quarry Road East
Bebington
Wirral, United Kingdom L63 3JW
Adv Dent Res 11(1):133-149, April, 1997
Abstract—Assessment of the role of biofilm microstructure
in biofilm-specific activities requires non-destructive
measurement techniques for parameterization of structural
characteristics in parallel with relevant biochemical and
physiological data. This paper briefly reviews some current
methods for biofilm structural analysis, with emphasis on
new developments in optical imaging and mathematical
modeling methods. Fluorescence imaging studies of bacterial
colonization events occurring on exposed model tooth
surfaces indicated that bacterial adhesion to sessile organisms
was of central importance to the early colonization process
and that this occurred in a non-random manner. Structural
studies of mature biofilms by confocal microscopy
demonstrated the spatial distribution of individual species
using fluorescent antibodies. Biofilms grown under different
physiological conditions exhibited differences in structure,
and methods were developed for parameterizing the spatial
orientations of the bacteria. Diffusive processes within
biofilm microstructures were studied using a random walk
model in both 2-D and 3-D. Modeling of convective flow
within biofilm microstructures was achieved by application of
lattice Boltzmann methodology.
Key words: microscopy, fluorescence, confocal, antibody
labeling, image analysis, biofilm, diffusion, flow, structural
analysis, connectivity, visualization.
Presented at the 14th International Conference on Oral
Biology, "Biofilms on Oral Surfaces: Implications for Health
and Disease", held March 18-20, 1996, in Monterey,
California, organized by the International Association for
Dental Research and supported by Unilever Dental Research
Downloaded from adr.sagepub.com at MCMASTER UNIV LIBRARY on February 21, 2016 For personal use only. No other uses without permission.
he development of oral biofilms is characterized by
the progression from a sparse sub-monolayer film
of bacteria attached to the tooth surface pellicle to
complex three-dimensional structures with inherent
ecological, physiological, and biochemical heterogeneity
(Gibbons and van Houte, 1980; Newman, 1980; Cobb and
Killroy, 1990). Biofilm communities exhibit specialized
biofilm-specific activities which are an elaboration of the
combined effects of species diversity, surface attachment,
and physico-chemical gradients (Costerton et aL, 1987, 1994;
van Loosdrecht et aL, 1990). Importantly, bacterial growth in
oral biofilms is strongly influenced by physico-chemical
environmental pressures—for example, pH, oxygen
availability, and redox have been implicated as key variables
that may affect the colonization of cariogenic and periodontal
pathogens (van Beelen et aL, 1986; Marsh, 1991, 1994). In
this context, the physical microstructure of the biofilm may
influence the formation of pH and Eh gradients through its
impact on diffusive and convective fluid flow characteristics.
These transport phenomena influence the access of substrates
for bacterial metabolism and, in turn, the removal of
metabolic end-products from the biofilm (Dibdin, 1981).
Additionally, the ability of antimicrobial agents to act against
biofilm organisms requires penetration to target sites (Anwar
et aL, 1992). Hence, considerable indirect evidence exists
indicating that biofilm microstructure may represent a key
determinant of biofilm growth, physiology, species diversity,
and susceptibility to antimicrobial strategies.
Methods for parameterization of biofilm microstructure
are required for study of the relative importance of
physiological, physico-chemical, and structural factors.
Classification and parameterization of structure in two or
three dimensions have been attempted in several fields
(Sahimi, 1994). Traditional parameterization approaches
have relied heavily upon stereological methods (Gundersen
et aL, 1988a,b), wherein distinguishing parameters such as
connectivity (Zhao and MacDonald, 1993) and pore and neck
sizes (MacDonald et aL, 1986) have been used. More
recently, attempts to use fractal dimensions as structural
classifiers have been presented (Pancorbo et aL, 1994).
However, natural systems rarely conform to regular fractal
behavior (Russ, 1993). A more direct approach for 3-D
structures has been the use of the Voronoi tesselation, which
has been used to dissect space-filling mosaic-like structures
that commonly arise from growth processes. This method has
found successful application in both the biological and
geographical sciences (Hahn and Lorz, 1994; Eils et aL,
133
134
1995). Alternative tools for the study of growth phenomena
include various second-order stereological measurements
(Mattfeldt et al, 1993), including the correlation and radial
distribution function. However, at present, there are no
useful, conceptually straightforward descriptions of a system
that links to unique spatial arrangements. This remains a
frustration in biofilm research and has resulted in reports of
biofilm appearance and morphology being essentially
qualitative descriptions (Newman, 1980; Caldwell et al,
1993; Korber et al, 1993; Wolfaardt et al., 1994).
Colonization and succession of bacterial species in dental
plaque and the role of adherence have been studied in detail
and modeled in vitro (Socransky et al., 1977; Marsh, 1994;
Marsh et al., 1995). Flow chamber studies have followed
adhesion of physiologically undefined bacterial suspensions
under laminar flow conditions to various surfaces by means
of radial distribution functions (Bos et al., 1994). This work
has been carried out on single- and two-species suspensions
for the assessment of adhesion and co-adhesion of organisms
to surfaces (Cowan and Busscher, 1993). In addition,
bacterial co-aggregations have been extensively characterized
in genetic and physiological studies by Kolenbrander and
others (Mergenhagen et al., 1987; Kolenbrander and London,
1993). These studies have indicated that specific inter- and
intra-species co-aggregative interactions may play a role in
colonization of the tooth surface and plaque biofilm
development. However, the precise role of these adhesion
molecules in the formation of realistic complex biofilm
structures has not been established. Furthermore, while
mathematical methods for describing spatial relationships
have been used in simple in vitro adherence systems, these
have not been extended to more complex multispecies
systems (Bos et al, 1994).
The quantitative characterization of the arrangement of the
microstructural features of a system and methods for
describing the arrangement of these features, including
parameters such as position, size, shape, and orientation, have
been presented in many fields and to different degrees of
sophistication (Ripley 1977, 1979; Baudhuin et al, 1979).
Using species-specific probes and appropriate imaging and
statistical techniques, one should be able to identify and
quantitate the key interrelationships between bacteria during
colonization of the tooth surface. There is a burgeoning
literature on the use of fluorescent methods for the direct
visualization of bacteria in micro-environmental systems
(Kepner and Pratt, 1994). These fluorescent methodologies
provide sensitive detection in complex environments (Bloem
et al, 1995). Furthermore, individual fluorescent stains have
been shown to be sensitive to changes in environmental
factors such as pH, Eh, and ion concentration within biofilm
systems (Costerton et al, 1994). Methods for the specific
microscopic identification of oral micro-organisms have been
developed with use of fluorescence detection, with targeting
either via antibody technology or nucleic acid probes
(Zambon et al, 1985; Grnur, 1988; Savitt et al, 1988;
Moncla et al, 1991; Amann et al, 1992). While both
methods have shown applications for bacterial detection on
glass slides, the targeted disclosure of bacteria within 3-D
Downloaded from adr.sagepub.com at MCMASTER UNIV LIBRARY on February 21, 2016 For personal use only. No other uses without permission.
SINGLETON ET AL. ADV DENT RES APRIL 1997
oral biofilm systems has not been widely reported (Singleton
etal, 1995).
Electron microscopy has been used extensively for the
examination of both surface structures and sectioned samples
of biofilms (Lie, 1978; Newman, 1980; Cobb and Killroy,
1990; Newman and Barber, 1995). Considerable information
on biofilm structure has been gained from these techniques,
particularly when applied in conjunction with specific
antisera for the detection of individual bacterial species
(Barber et al, 1990). Recently, the introduction of Confocal
Laser Scanning Microscopy (CLSM) has provided a step
change in the capability for non-invasive imaging of biofilms
(Caldwell et al, 1992a; Wolfaardt et al, 1994; Singleton et
al, 1995). This technique allows the 3-D extent of the
biofilm matrix at single-bacterium resolution to be visualized
by the gathering of serial optical section image data. By the
application of appropriate image processing methods,
realistic estimates of actual microstructures may be obtained.
These images may then be further processed to generate a 3-
D image which provides excellent visual information on
actual structures. CLSM therefore offers a non-invasive tool
for assessment of physiological or biochemical events deep
within biofilm matrices.
A particularly important need, deriving from biofilm
structural studies, is the understanding of the influence of the
microstructure on molecular transport through biofilms.
Transport phenomena include both convective flow and
diffusive processes, and these have been proposed as
important factors influencing the influx and efflux of
carbohydrate substrates and hydrogen ions, respectively, in
relation to the etiology of dental caries (Dibdin, 1981).
Studies of molecular transport through biofilms have used
measurements in homogenized or sedimented mono-species
bacterial systems (Dibdin, 1981; Tatevossian, 1985) and
more recently the use of fluorescent dextrans in conjunction
with confocal microscopy (Lawrence et al, 1994). The
homogenous, pelleted bacterial models have lost the original
biofilm structural characteristics; however, confocal
techniques have the advantage of preserving the biofilm
structure and are promising to be especially powerful when
carried out in a quantitative mode. For example, recently,
fluorescence recovery after photobleaching (FRAP) analysis
was carried out with fluorescent dextran in an oral biofilm
model (Birmingham et al, 1995).
The three-dimensional simulation of flow-through porous
media has been studied previously in a number of fields such
as geology and the oil industry (Jernot et al, 1992). An
effective way of simulating the diffusive process through
realistic and hence complex microstructures is based on a
variant of the continuum random walk approach first
described by Torquato and Kim (1989). Computer simulation
of the complete molecular transport properties (diffusion and
convective flow) over time is, in principle, achievable by
means of Computational Fluid Dynamics (CFD) codes.
However, since characterization of the complete molecular
dynamic interactions requires simulation over fundamental
time steps (fs), these codes are inappropriate when bulk
transport properties over longer (ps-ns) time scales are
VOLll(l) MICROSCOPIC CHARACTERIZATION OF BIOFILMS
considered. For time scales characteristic of observable
diffusive events, almost all the correlations of the
intramolecular dynamic processes will completely decay
(Deutch et ai, 1987). Furthermore, such codes cannot be
easily manipulated to deal with complex geometries (Deutch
etaU 1987).
Data presented in this paper show how optical methods for
imaging biofilm microstructure may be linked with image
analysis programs and mathematical models to parameterize
three key structural aspects of mixed-species biofilms grown
in vitro:
(1) Mapping early colonization events and spatial relation-
ships between different bacterial species over time.
(2) Imaging of specific biofilm populations in situ in three
dimensions and parameterization of biofilm microstruc-
tures grown under different physiological conditions.
(3) Modeling molecular diffusion and convective flow in
reconstructed three-dimensional biofilm microstructures.
MATERIALS AND METHODS
To be able to meet satisfactorily the extensive measurement
objectives set, we had to combine several methodologies.
Physical components of the experimental systems will be
discussed, together with the individual experimental
procedures, followed by a description of the analytical
methods used.
Microbiological methods
Continuous culture. A consortium of bacterial species
representative of dental plaque populations in health and
disease—Streptococcus sanguis SB 179, Streptococcus
mutans R9, Lactobacillus rhamnosus AC413, Actinomyces
naeslundii WVU627, Neisseria subflava A1078,
Fusobacterium nucleatum ATCC 10953, Veillonella parvula
ATCC 17745, Prevotella nigrescens T588, and
Porphyromonas gingivalis W50—was grown in continuous
culture (D = 0.1 h"1) in a complex medium (BM) containing
hog gastric mucin under anaerobic conditions at 37°C and
/., 1985).
Flow cell biofilm cultures. Glass flow cells (20-mL
working volume) containing glass holders carrying
hydroxyapatite (HA) discs were fed by means of a peristaltic
pump (Watson Marlow 505 DU) with inoculum from the
primary continuous culture and with fresh medium (BM,
mucin-containing) to produce a dilution rate of 0.75 h"1 at a
ratio of 1 part inoculum (complex community) to 9 parts
fresh medium. Flow cell contents were recirculated at 25
mL/min. Anaerobic conditions were maintained in the flow
cell with an atmosphere of 5% CO2 in nitrogen. The pH of
flow cell contents was continuously measured on an Orion
720A pH meter equipped with a Russell probe (serial number
CTW 711). In experiments to investigate the effect of sucrose
on biofilms, solutions (either sucrose or water) were pulsed
into flow cells directly by means of a peristaltic pump
(Watson Marlow). Conditions for the primary continuous
culture were unchanged.
Downloaded from adr.sagepub.com at MCMASTER UNIV LIBRARY on February 21, 2016 For personal use only. No other uses without permission.
135
Measurement methods
Bacterial staining procedures. The HA discs were
removed from flow cells and incubated at 37°C for 20 min in
0.5 mL IF buffer (2.5% Marvel in borate-buffered saline).
Specific anti-Veillonella parvula monoclonal antibody was
then added (20 pL), and after further incubation for 20 min, a
mixture of biotin-labeled antibodies (anti-mouse IgM, u
chain-specific diluted 1/150, and anti-mouse IgG, whole
molecule diluted 1/200) was added, and incubation continued
for 20 min. Finally, avidin-FITC solution (Sigma, diluted
1/500) was added, followed by a further 20-minute
incubation. The discs were then removed, placed on a
microscope slide, and counterstained with ethidium bromide
(1-2 pL in titer steps of a 10-pM solution). A similar
procedure was used for S. mutans labeling and
counterstaining of mature biofilms. For general structural
imaging where no specific labeling was required, either
ethidium bromide or BacLight™ (Molecular Probes,
Eugene, OR) was used at 10 uM.
Colonization mapping experiments. Mapping experiments
of bacterial colonization onto hydroxyapatite surfaces were
performed at time intervals of one-, three-, and five-hour
immersion in the flow cell apparatus. Following fluorescent
staining of the bacteria on the disk surface, samples were
mounted onto an Olympus BHS-BH2 fluorescence
microscope fitted with a Marshauzer (X,Y,Z) translation
stage and interfaced to a Kontron IBAS image-processing
system. Macro programs controlled the (X,Y) stage
movement and image-sampling procedure of the
fluorescently stained bacteria. Typically, two equivalent
image areas were taken of the same field of view but with
different fluorescent bandpass filters such that antibody-
labeled (green, fluorescein) bacteria and total bacterial
biomass (stained red; ethidium bromide) could be
sequentially detected. The success of these methods was
found to be crucially dependent on the quality of the antibody
labeling that could be achieved, i.e., low non-specific
labeling and high-brightness antibodies were required. Since
sampling of large areas was involved, control of optimum
focus and fluorescence filter interchange was prompted by
macro program and performed manually. Images were
typically gathered by means of a Leitz NPL Fluotar 50x/1.0
NA water immersion objective. With conventional
microscope optics, effective sampling of the HA disk biofilm
was possible only in 2-D and up to the depth of field of the
objective in use, i.e., less than 5 um total thickness. Digitized
image viewing fields (512 x 512 pixels at 8-bit intensity
resolution) were obtained by means of a low-light-level
Falcon SIT camera (Agar Scientific, Stanstead, UK) attached
to the trinocular head of the microscope and were typically
85 x 85 microns in (X,Y) dimension. Absolute reproducible
system (X,Y) translation of the scanning stage was measured
to be ± < 0.5 um. Hence, consecutive images representing a
square raster pattern could easily be produced within the
program environment. This procedure allowed for surface
area mapping in the range of 1 mm2 to 2 mm2 with high
precision and resolution of the co-ordinates of individual
136
Composite Reduced Image
of Raster Image Set
850 Microns
• •
1 \ <H
4
- 1
850 Microns
Jk. r 85 Microns
n
Fig. 1—(a) Schematic representation of 'early' colonization
processes occurring on HA surfaces immersed in the
continuous culture biofilm system. (b,c) Parameterization of
the colonization process is achieved by measurement of the
Euclidean distances of nearest, next nearest, etc., neighbors
of the bacteria of interest with respect to (either) the same
(or another) specifically labeled target species. These
Euclidean measurements are then compared with the prior
Poisson distribution for randomly distributed point
processes on an equivalent surface area. The method may be
used to map the development of colony size factors and
various component species over time (the Fig. schematically
describes the mapping o/V. parvula against total biomass as
described in the text).
bacterial cells.
Subsequent image processing of the consecutive-image-
pair mozaic allowed for segmentation (recognition) of
specific bacteria (green objects) and biomass material (red
objects). For each recognized object, information about size,
shape, and position was obtained. Hence, we were able to
make measurements of surface area coverage, spatial
distributions, and spatial coincidences in the patterns of the
green-labeled and/or the red-stained organisms, shown
schematically in Fig. 1.
Structural imaging. Three-dimensional fluorescence
imaging of more mature biofilms (up to 50-pm thickness) on
the HA surfaces (typical time point, 72 hr of immersion in
flow cell) were performed by means of a Noran Instruments
Odyssey Confocal Laser Scanning Microscope, capable of
video rate performance (25 sampled frames per second, each
frame composed of 512 x 512 pixels and digitized at 8-bit
intensity resolution). Images were gathered primarily by
means of a Leitz NPL Fluotar 100x/1.2 NA water immersion
objective. The square viewing fields measured approximately
46 um x 46 um. Optical sections were
Downloaded from takenat MCMASTER
adr.sagepub.com at 0.5-pm
UNIV LIBRARY
SINGLETON ET A L . ADV DENTRES APRIL 1997
Individual Image of
Raster Image Set
85 Mirrnna
Examples of Bacteria
associated with the
general biomass
Examples of species specific
nearest neighbours
ft
Non-specifically stained general biomass material
Antibody labelled organism,
(mechanical distance) through, wherever possible, the full
depth of the biofilm. The biofilm depth varied considerably
across the HA surface, and sampling of biofilm volume
segments was undertaken following microscopic visual
inspection of the biofilm such that from 6 to 10
'representative' segments (rather than mozaic maps) were
recorded. This fluorescence imaging technique allowed for
high-contrast imaging of individual cells; however, the
ultimate image quality was found to be governed by the
bacterial density of the biofilm and its depth, factors which
influence the extent of scattering of the probe beam
(Wolfaardt et al., 1994; Birmingham et al., 1995). Typical
signal-to-noise ratio conditions in these experiments were
such that the axial (z axis) optical resolution was ca. 0.8 um.
Image segmentation. Image data obtained from both the
colonization mapping and the 3-D microstructural
investigations were manipulated by means of the Kontron
IBAS image processing system. Object segmentation
(separating bacteria from background and converting into
binary images) from the grey-scale fluorescence image
stepson February
normally incorporated
21, 2016 For personal a design-based
use only. No other uses without permission. filtering step followed
VOLll(l) MICROSCOPIC CHARACTERIZATION OF BIOFILMS
by a thresholding step and a combination of binary opening
and closing procedures to produce smoother surface
representations of the bacteria. The optimal filter design for
separating objects from background was found to be the
Laplacian. Both the magnitude of the filter and its component
elements were varied depending on the relative optical
magnification (apparent bacterial size) in the images. To
separate biomass from background in colonization mapping
experiments, a simple threshold condition was applied, since,
for these complex objects, resolution of individual bacteria
was not required.
Simulation, modeling, and graphic visualization
methods. For the simulation/modeling and graphic
visualization tasks associated with this work, a Silicon
Graphics 2XZ workstation was used (160 Mb RAM, 200
MHz CPU, 2-Gb hard disk). Fast volume rendering was
performed by means of Voxel View 2.2 and Voxel Math 2.1
(Vital Images Inc., Fairfield, IA, USA). Flow and surface
visualization was performed by means of AVS 5.0 (Advanced
Visual Systems Inc., Waltham, MA, USA). The diffusion
model based on the Continuum Random Walk method was
written by one of the authors (RT) within the PV-WAVE
environment (Visual Numerics Inc., Houston, TX, USA). The
lattice Boltzmann method was written by one of the authors
(PW) using the C language. Three-dimensional image proces-
sing functions were performed by means of the CImages3D
package (Foster Findley Associates, Newcastle,UK).
Analytical methods
Colonization mapping. In this work, a simple procedure
which generates a description of the arrangements of objects
on planar features/surfaces and based on the analysis of
nearest-neighbor distances was adopted. This approach
classifies objects according to either random, clustered, or
regular arrangements and also allows for the separation of
clusters from random backgrounds and a description of the
degree of clustering or regularity (Schwarz and Exner, 1981).
The model assumes that the objects of interest on the planar
surface can be represented as points (e.g., using their
centroids). This assumption will be acceptable, provided that
the surface dimensions are large compared with the object
dimensions. In addition, edge effects caused by sampling will
be reduced if sufficiently large sample areas are taken.
A random point process can be well-described by a
Poisson distribution (Schwarz and Exner, 1981). That is, the
frequency, f(r^, of distances rj of the first nearest neighbor
takes the form:
f(rx) = 27Cpr1 exp(-7Cpr12)
and for distances r2 of the second nearest neighbors takes the
form
/(r 2 ) = 27C2p2r23 • exp(-7lpr22)
where p is the point density (number of points N per unit area
A). Experimentally, the estimator E(p) = N/A = NA.
Downloaded from adr.sagepub.com at MCMASTER UNIV LIBRARY on February 21, 2016 For personal use only. No other uses without permission.
137
Hence, if one is to test whether an arrangement of points
deviates from random behavior, then the simplest criteria are
to compare the observed mean and the standard deviation of
the observed distances with the values from the ideal random
arrangement. The expected mean, E(r2), and variance, E(Sj2),
of the Poisson distribution follow from knowledge of
E(rx) = 2Vp
p(~NA), and for the nearest neighbors, these values are found
to be:
1
471 P
Now the ratio of the observed (rj) and expected E(rj) mean,
Q = TJ/EO'J), and the observed (Sj2) and expected variance
E(Sj2), R = s J 2 /E(s 1 2 ), serve to indicate the nature of the
spatial distribution of the point objects.
For a random experimental point set, expect Q ~ 1, R ~ 1.
For a regular experimental point set, expect Q > 1, R « 1.
For a clustered experimental point set, expect Q < 1, R < 1.
For sets of clusters with a superimposed background,
expect Q < 1, R > 1.
For further segmentation of the degree of clustering and the
arrangement of any clusters within the surface, an appropriate
first step is to eliminate the random background points. This
can be achieved if one assumes that the overall point
distribution is a superposition of at least two independent
Poisson processes. In this case, the interest will be in a
process for pc covering cluster areas A c , and another pB
covering background areas. Hence, the density functions of
the overall process are then simply the weighted sum of the
two individual functions such that:
J{r)B
N N
where N is the total number of points. Then, Nc/N = C and
NB/N = 1 - C. So, for the first nearest neighbors:
f(rx) = 2nCpcrl • exp(-7Cpcr12) + 2n(l - C)pBr{ • exp(-7tpsr12)
and for the second nearest neighbors:
/(r2) = 27T2Cp2cr23 • exp(-7cpcr22) + 2TC2(1 - C)pB2r23 • exp(-7iP/?r22)
The unknown parameters, p c , p B , and C, follow from the
evaluation of the first three moments of either the first or
second nearest-neighbor distribution functions.
Assignment of objects to either clusters or background
then depends simply on the definition of the clustering
distance r*, where if the neighbor is at a distance r < r*, it is
assumed that the point belongs to a cluster, whereas,
alternatively, if r > r*, it is assumed to belong to background.
I3S SINGLETON ET AL. ADV DEM-RES APRIL 1997

in turn models the funda-

mental Brownian Diffusion
process at a microscopic level.
This technique is applicable
to either two- or three-dimen-
sional structures and proceeds
in 2-D by initiating all
trajectories at an identical spot
(in 3-D, randomly selected
points in a 2-D plane are
chosen) and then evolving
until the region boundary is
reached, whereupon each
trajectory is absorbed. The
time associated with each
trajectory (the passage time) is
calculated by assuming that
Total Colony Area (Square microns) the diffusion coefficient in the
free pore space is the same as
the known free solution value
Fig. 2—Area number density of bacterial biomass during colonization onto HA surfaces DQ. Deff is then estimated for
immersed in a continuous culture biofilm system for 1,3, or 5 hrs. the system by analysis of the
statistical properties of the set
of passage times. In each case,
The combination of values of p c , pB, and C allows for the geometry ensures that each trajectory has a fixed mean
quantitative descriptions of the clustering in terms of areas square displacement, and thus a properly defined diffusion
covered by the clusters A c = N c /p c and the number of points coefficient can be calculated.
belonging to the clusters N c = CN. The passage times for each trajectory are "binned" into a
histogram—the diffusive passage time distribution p(t).
Transport modeling. The three-dimensional charac- Typically, several thousand trajectories are required to yield
terization of transport through porous media is of interest in accurate descriptions of p(t), which is the probability that a
many fields (Jernot et al., 1992). Typically, several different randomly chosen trajectory will have a passage time in the
modes of transport will be important in any given system. We range t to t+dt. Consequently, the mean passage time (T ) is
focus on two common modes, diffusion and convective flow, given by:
and on how the microstructure alone interacts independently
with these processes. In particular, no account is taken of
extra-celluar material which may influence transport but
cannot be sampled by the imaging techniques used here.
=h J
Since the simulation occurs over fixed mean square
Modeling of diffusion. In this scenario, the diffusion displacements L2, the familiar equation for the Deff (in this
process can be represented macroscopically by the diffusion case, in 2-D) is obtained as:
equation, which, in vector form, is: D - ^

at Generally, these results are quoted as the ratio of effective

where Deff is the Effective Diffusion Coefficient for diffusion
through the microstructure, and c(r,t) is the concentration of
the diffusive species at time t and position r.
This equation can be solved analytically in only a limited
number of cases within easily defined geometries. For
diffusion through complex microstructures, a simulation
approach must be used which can numerically generate Deff.
The approach taken here is based on a variant of the
continuum random walk method (Torquato and Kim, 1989),
Which generates a set of non-interacting trajectories through
the pore space. Each trajectory is constructed to represent an
individual random walk through the microstructure,
Downloaded from adr.sagepub.com at MCMASTER UNIV LIBRARY on February 21, 2016 For personal use only. No other uses without permission.
and free diffusion coefficients, that is, as Deff/D0.
Modeling of 3-D convective flow. At the macroscopic
level, the convective flow component of the transport process
can be described by means of the Navier-Stokes equations for
the velocity vector v(r,t):
—- = F - ~ \p + %\zy
Dt p
where: F represents the external body forces (such as gravity)
acting on each fluid element; p is the (thermodynamic)
and this pressure, a sealer quality; p is the fluid density; and d is the
VOLll(l) MICROSCOPIC CHARACTERIZATION OF BIOFILMS
TABLE
EFFECT OF TIME OF IMMERSION IN THE CONTINUOUS CULTURE BIOFILM SYSTEM
ON COLONIZATION PARAMETERS DETERMINED BY MEANS OF STANDARD IMAGE ANALYSIS METHODS
AND THE NEAREST-NEIGHBOR APPROACH FROM OPTICAL MEASUREMENTS
OF V. parvula-SPECIFIC FLUORESCENCE (green, G) AND TOTAL BIOMASS FLUORESCENCE (red, R)
Time of Relative Surface Coverage Fraction of
Immersion (%) V. parvula (G)
in Culture Coincident with
(h) V. parvula Total Biomass General Biomass (R)
1 0.003 0.070 0.32
3 0.008 0.078 0.45
5 0.046 2.89 0.56
fluid kinematic viscosity.
Numerical solution of these equations is a much more
complex task than in the previous case. However, because of
the need to understand flow in a wide range of applications
(e.g., aerodynamics), a vast array of computational
techniques for numerically solving these equations has been
developed under the banner of Computational Fluid
Dynamics (CFD). Unfortunately, these are usually optimized
for large-scale flows past simple geometries which can be
efficiently described by a relatively short list of polygons
covering the surfaces. Accurate covering of the surfaces of
the complex structures that we are investigating would
require a huge number of polygons with consequent
degradation in the performance time of the simulation.
Instead, in a spirit similar to that of the diffusive modeling,
we have adopted the lattice Boltzmann method, which is
founded upon a microscopic description of the flow process
(Frisch et al, 1986; d'Humieres et al., 1986; Qian et al, 1992).
In essence, the system (in this case the biofilm) is
described by a lattice which is thought to consist of a large
number of identical structureless lattice particles which 'live'
on the nodes of a (2-D or 3-D) regular lattice space and that
move along the regular lattice links. Hence, the state of the
lattice at any time may be described by indicating which links
of which nodes are occupied by particles. The description of
the physical process (e.g., flow) is embedded within the set of
rules which determine what happens as particles enter a node
and undergo a collision. These rules are fixed such that the
proper physics of the desired flow model is eventually
obtained. By changing the collision rules, different aspects of
the physical model can be accounted for (e.g., boundary and
initial conditions and interactions with solid bodies defined
on the lattice). Since the method is lattice-based, it is ideal for
the incorporation of microstructures obtained from 3-D
imaging experiments, since they are also based on a lattice of
points.
The technique has been fully validated (Behrend et al.,
1994), confirming that the Navier-Stokes equations are
properly satisfied and hence fluid dynamic behavior
accurately modeled. Indeed, it can be shown that it is
Downloaded from adr.sagepub.com at MCMASTER UNIV LIBRARY on February 21, 2016 For personal use only. No other uses without permission.
139
Population Mean "Q" Statistic "R" Statistic
Ratio Nearest-neighbor
G/R (%) Distance between
V. parvula Cells
4.7 20.66 0.85 1.66
1.0 8.99 0.57 1.60
2.9 4.16 0.62 2.00
formally equivalent to the CFD methods described above—
with the advantage that it is more efficient and simpler to
implement for complex geometries. Further, the control of
parameters such as diffusion and temperature is easier and the
incorporation of complex structures essential to this study is
straightforward, in marked contrast to the intermediate steps
that would be required if the microstructure were to be
prepared for a CFD simulation.
RESULTS
Colonization mapping
Experiments were carried out for assessment of the relative
surface coverage and distribution of Veillonella parvula
(fluorescently labeled green by FITC-conjugated V. parvula-
specific Mabs) and of total organisms of the remaining
biomass (labeled red by ethidium bromide) over an area of
1.04 mm2 of HA disk in the flow cell biofilm system (shown
schematically in Fig. 1).
The results suggested a non-linear rate of surface coverage
with time (Table). The surface coverage of V. parvula
organisms corresponded to 108 detected organisms in 1.04
mm2 (0.003% coverage) at 1 hr. After 5 hrs of immersion in
the continuous culture system, the total surface area coverage
was still low (Table). The distribution of colony areas
showed that the dominant colony size at all time points was
small (primarily 0-3 um 2 range) and therefore represented
relatively few attached organisms per cluster (Fig. 2). The
data were expressed in a form which shows the physical
distribution of biomass in terms of colony area (Fig. 3).
Clearly, at 1 hr the majority of the biomass existed only in the
smallest of colony sizes (0-3 um2); however, after a further 4
hrs of colonization, the majority of the biomass was
associated with larger colony sizes (up to 500 um2 [Fig. 3]).
Furthermore, it was evident that the V. parvula organisms had
become increasingly associated with the general biomass
over time, even though the total biomass surface coverage
remained low (Fig. 4, Table). The proportions of V. parvula
organisms relative to the total biomass fell, between 1 and 3
hrs of colonization, to 1.0 and subsequently increased at 5 hrs
140 SINGLETON ET AL. ADV DENT RES APRIL 1997

3-D microstructural
imaging and
species-specific localization
Experiments were undertaken
to obtain representative con-
focal images of biofilms and
to extract realistic bacterial
microstructures. These data
were subsequently analyzed
for 3-D structural or distri-
butional statistics and for use
as realistic base structures for
the modeling of molecular
transport properties. Fig. 5
shows a composite volume
segment image of the bottom
20-um segment of a 40-um
Total Colony Area (Square microns) biofilm grown for 72 hrs in
the continuous culture system.
Various bacterial structural
Fig. 3—Normalized colony area distribution ofbiomass during colonization of HA surfaces morphologies were evident
immersed in a continuous culture biofilm system. within the ethidium-bromide-
labeled general biomass
material, and a small
to 2.9 (Table). microcolony of S. mutans cells was observed near the HA
Results from examination of the first nearest-neighbor surface when an S. mutans-specific FITC-conjugated
distribution of V. parvula organisms indicated a strong time- monoclonal antibody probe was used.
dependence. At 1 hr, V. parvula demonstrated a wide
distribution of nearest neigh-bors, with a mean separation Effect of carbohydrate availability
distance of 20.7 um. How-ever, after 5 hrs, a mean separation on biofilm microstructure
distance of 4.2 um was observed (Table). The statistical Experiments in which parallel identical flow cells were red
analysis of the propensity of the V. parvula organisms to from the same continuous culture system at a constant
distribute randomly (Q and R statistics) showed that, at all dilution rate (0.75 h"1) were carried out to investigate the
time periods studied, the V. parvula cells tended to 'cluster' effect of sucrose availability on biofilm microstructure (Figs.
together more than that predicted by a random distribution 6A, 6B). The mucin-limited biofilm was characterized by
(Table). relatively sparse biofilm
growth, with open areas
and 'string-like' structures
(Fig. 6A). In contrast, the
1 hr sample D 3hr sample El 5hr sample sucrose-supplemented
biofilm showed a signi-
ficantly higher bacterial
density and more extensive
depth (up to 50 pm) of
growth and a greater
diversity of dominant bac-
terial morphologies (Fig.
6B). Mucin-limited and
sucrose-supplemented
biofilm microstructures
were measured for dis-
tribution of particle
volumes and principal
lengths (Fig. 7). Results
suggested that the struc-
Nearest Neighbour Distance (microns) tures were composed of
Fig. 4—Nearest-neighbor distribution ofV. parvula during colonization of HA surfaces
from adr.sagepub.com at MCMASTER UNIV LIBRARY on February 21, 2016 For personal use only. No other uses without permission.multiple, morphologically
immersed in a continuousDownloaded
culture biofilm system.
similar subunits, with
VOL.U(l) MICROSCOPIC CHARACTERIZATION OF BIOFILMS 141

characteristic particle dimensions. A dominant bacterial

morphology (cocci) was particularly evident in the mucin-
limited biofilms, whereas data from sucrose-supplemented
biofilms indicated the presence of multiple cell morphologies
(Fig. 7).

Transport modeling
Diffusion. Analysis of extracted microstructural segments by
the continuum-random-walk method yielded two insights into
the influence of biofilm microstructure and molecular 20um
diffusion. First, the effective diffusion coefficient in the pore (114)

space, with respect to 'free particle' diffusion calculated in

both 2-D and 3-D simulations, showed that the effective
diffusion coefficient was a factor of ca. 2 slower than the free
particle diffusion. For example, a 2-D simulation carried out
with a transmission electron microscopy image of a bacterial
biofilm showed Defl/D0 « 0.49 (3000 walks, DQ = W6 m V )
(Fig. 8).
Three-dimensional data sets obtained from CSLM images
of an in vitro oral biofilm were used in a simulation of the 3-
D effective diffusion coefficient. These data showed an
effective slowing of diffusion by the biofilm to Derf/D() =
0.70, hence confirming the 2-D simulated data and a low
level of hindrance (Fig. 9). An essential feature of the
calculation of the effective diffusion coefficient in this model
is the generation of the Euclidean Distance Map (EDM)
applied to the pore space of the microstructure (the EDM is
the Euclidean distance of every pore volume element to the
nearest surface). Hence, analysis of this map readily reveals
information about the pore size distribution within the pore
space of the microstructure. An example of this analysis is
shown in Fig. 10, whereby the connectivity of the void space
is found to be a strong function of the defined pore neck size
in these in vitro biofilm structures.
50um
(512)
S.mutans cells labelled b) Mab I 111
General hiofilm mass labelled with Kthidium Bromide
Fig. 5—Confocal imaging with volume visualization of an
oral biofilm grown for 120 hrs in the nine-species continuous
culture biofilm system showing specific labeling with anti-S.
mutans Mab-FITC and ethidium bromide staining of non-
specifically-labeled cells.
Convective flow. We analyzed the microstructure by
considering convective transport through the void space. In
Fig. 11 A, flow streamlines generated by the lattice Boltzman
simulation are shown penetrating the structure. It is evident
from the broken flow lines (Fig. 1 IB) that the proper
description of bulk liquid transport through such complex
structures will require an additional diffusive term. Since the
properties of the fluid propagating through the microstructure

A: Min in Limited Biofilm B: Sucrose Supplemented Biofilm

Fig. 6—Comparison of biofilm microstructures grown in a continuous culture biofilm system under either (A) mucin-limited or
(B) sucrose-supplemented conditions. Images represent typical volume-rendered segments from confocal analysis of at least
two duplicate biofilms stained with ethidium bromide.

Downloaded from adr.sagepub.com at MCMASTER UNIV LIBRARY on February 21, 2016 For personal use only. No other uses without permission.
 SINGLETON ET AL. ADV DEW RES APRIL 1997

in bacterial populations associated with increasing biofilm

age and different pathological states (Marsh, 1994).
A Additional insight has been gained by descriptions of the
Mucin Limited binding forces and kinetics related to adhesion and to the
Sucrose Supplemented
ultrastructural detail relating to specific binding interactions
(Bos et al., 1994). Recently, methods for the in situ
examination of biofilm phenomena, including the monitoring
£ 0.6 of direct bacterial-surface interactions, have became
o available. Such methods allow for a much greater and more
O
fundamental understanding of biofilm systems. Hence, it is
now conceivable that biomathematical modeling could be
applied to the understanding of biofilm growth processes, and
to the conditional simulation of underlying 3-D structure.
Decision analysis methods may then be used to predict the
impact of environmental and physiological factors on biofilm
parameters.

0 500 1000 1500 2000 Colonization

Volume of Particle (voxel units)
Mucin Limited
Sucrose Supplemented
£ 0.6
D
O
0 20 40 60
Principal Length of Particle (voxel units)
Fig. 7—Comparison of the distribution of (A) particle volume
and (B) principal length within extracted microstructures of
mucin-limited and sucrose-supplemented biofilms.
are known, it is possible to calculate the net normal and
tangential forces applied to the microstructure such that an
investigation of the effects of these forces on bacterial
adherence can be made (data not shown).
DISCUSSION
Current understanding of complex bacterial interactions
within the oral cavity stems from classic microbiological
Downloaded from adr.sagepub.com at MCMASTER UNIV LIBRARY on February
methods and sampling procedures. Examples include
observations of sequential species succession and the changes
The time-dependent development of bacterial biomass on HA
surfaces in vitro may arise as a result of at least three
mechanisms: attachment to exposed (free) surface; growth of
bacteria already attached to the surface; and adhesion to
biomass already on the surface. (In the latter mechanism, we
B make no initial distinction between species-specific and non-
specific adhesive interactions.) So that the relative
importance of each process could be assessed under
controlled conditions of nutrient availability within a defined
consortium of interacting organisms, experiments were
carried out for the development of an analysis for the
parameterization of the colonization process. This analysis
was tested in the first instance by image mapping of V.
parvula organisms adhering to greater than 1 mm 2 HA
surface, following immersion into a continuous culture flow
cell system over a five-hour time course. It would be
expected that if attachment to 'exposed' surfaces was the
dominant mechanism of biomass development, then this
should be a random process uninfluenced by the presence of
A \
other bacteria. Collisions of planktonic bacteria with sessile
bacteria will occur but will be of low frequency, given the
80 100
low surface coverage shortly after immersion. In addition, it
would be expected that the surface coverage would increase
essentially linearly with time. In contrast, the data showed
that the rate of change of surface coverage was far from
linear and indeed that at no time did the distribution of V.
parvula approach a random distribution (Table). Similarly,
nearest-neighbor analysis of the total biomass showed non-
random behavior (Q = 0.6, R = 2.0). V. parvula is known to
co-aggregate strongly with Fusobacterium nucleatum but
weakly or not at all with the other members of the consortium
(D. Bradshaw, personal communication). When Poiseuilles'
equation was applied (Hiemenz, 1977), the shear rate in the
flow cell system was estimated to be low (< 3 s"1) and may be
considered unlikely to influence the overall colonization
pattern. A surface-adhesion-dominated process would be
expected to produce a biomass distribution with
predominantly small colony fragments. Hence, the data
showing
largeusebiomass
21, 2016 For personal aggregates
only. No other uses without permission. indicated a role for inter-
and intra-species adherence mechanisms in the colonization
VOLll(l) MICROSCOPIC CHARACTERIZATION OF BIOFILMS 143

B:

Results.
Simulated Random Walk rv
C: within the Intercellular Space.
Passage Time Distribution.

( seconds)

Stability of Effective Diffusion coefficient.

O.0

Fig. 8—Composite Fig. representing the route to evaluation of the 2-D effective diffusion coefficient from an experimental
microstructure. (A) Generation of raw data; (B) segmentation of image to obtain the surface of the extracted microstructure;
(C) calculation of the diffusive trajectories (for example, diffusive path shown in green, absorbing boundary shown in red);
and (D) evaluation of results.

Downloaded from adr.sagepub.com at MCMASTER UNIV LIBRARY on February 21, 2016 For personal use only. No other uses without permission.
 144 SINGLETON ET AL. ADVDEKT RES APRIL 1997

extent of local clustering on random

Reconstructed biomatrix background to be assessed
(Schwarz and Exner, 1981). It
should then prove possible to map
the cluster features onto the
physical co-ordinates of actual
biomass segments and hence
examine local bacterial segments
more closely.

Simulation Region
Sample Trajectories of
diffusing species.
Fig. 9—Three-dimensional diffusion simulation in an extracted biofilm microstructure.
of HA surfaces in vitro.
If surface attachment followed by bacterial growth was the
dominant mechanism, then a random spatial distribution with
a nonlinear time dependence to surface coverage would be
expected. However, the observed distribution was not
random. Furthermore, the rate of change of surface coverage
cannot be explained by the expected growth rates of sessile-
or planktonic-phase bacteria in these in vitro systems (Rogers
et al., 1986; Allison et al., unpublished data), indicating that
growth of adhered organisms was not likely to dominate the
formation of the observed colonization pattern. Non-random
distributions of colonizing bacteria were observed, strongly
suggesting that bacteria already attached to the surface exert
significant influence on further adherence. This influence
could arise from combinations of the additional collision
cross-section presented by bacterial surface normal to the
flow, from charge-related interactions, or from the influence
of extracellular material projected into the flowing media. It
has previously been shown that many organisms exhibit a
two-step adhesion process to surfaces (van Loosdrecht et al.,
1989) and that, in the first step, bacterial migration over the
surface remains possible. In our experiments, there was an
increased propensity for bacteria to exist in larger biomass
segments with time (Fig. 3, Table). However, the nearest-
neighbor analysis used in this study did not address the
relative significance of colony growth of very closely
associated organisms and adherence mechanisms. Further
Downloaded from adr.sagepub.com at MCMASTER UNIV LIBRARY on February 21, 2016 For personal use only. No other uses without permission.
development of the nearest-neighbor method will enable the
Structure
Currently, there are no general,
clearly defined methods which are
sensitive and unambiguous enough
for the useful classification of
complex geometries such as biofilm
microstructures. In order to study
biofilm structures and their
response to the environment, most
reports have relied on semantic
descriptions which are therefore at
best qualitative and subjective.
Some of the most detailed non-
destructive structural discussions to
date have used confocal imaging
optics to provide the necessary
image clarity (Caldwell et al.,
1992b, 1993). Work described in
this paper further demonstrated the
utility of confocal microscopy for
the mapping of biofilm microstructures. With antibody
methods, it was shown that direct species-specific
visualization within biofilm segments was achievable (Fig. 5).
Comparative structural studies of complex oral biofilms
grown under different nutrient conditions showed that
nutrient-limited biofilms contained significantly lower
bacterial density than nutrient-supplemented systems (Figs.
6A, 6B; as seen previously in other biofilm systems
[Caldwell et al., 1992b]). Spatial distributions and dominant
morphologies were less obvious and structural distributions
apparently more chaotic in sucrose-supplemented biofilms,
whereas under mucin limitation, the predominant morpho-
logical type appeared to be coccoid (ca. 1-um dimension).
These data from parameterization of the 3-D structure agreed
closely with viable counts of bacteria colonizing parallel
biofilms which showed that, while S. sanguis was the
dominant organism under mucin limitation, sucrose-
supplemented biofilms contained high numbers of S. mutans
and Lactobacillus rhamnosus (data not shown). By appro-
priate 3-D image analysis methods, it was possible for the
angular, physical, and spatial distributions of all objects to be
measured within the volume segment. Initial results from 3-D
parametric analysis of biofilm structures indicated that
several structural classifiers were required for structural
parameterization of complex biofilms and, in particular, for
characterization of the orientation of multicellular assemblies.
In the development of this approach, similar morphologies or
sets of object parameters may be grouped and replotted such
VOL.U(l) MICROSCOPIC CHARACTERIZATION OF BIOFILMS 145

that local associations may be

analyzed. In the future, multi-
parametric descriptions of biofilm
A:
types may be developed, which
may include building relations to
patterns of convective flow (see
below). These methods may give
further quantitative evaluation of
biofilm structures and assist in the
development of biomathematical
models.

Transport
Many characteristics of biofilms
are likely to be transport-
dependent, including the avail-
ability of nutrients, cell-cell
signaling, environmental factors
such as pH, pO 2 , and redox
gradients. Two forms of transport
would be expected to be impor-
tant in the complex structural
geometries found in biofilms:
first, the overall permeability of
the porous structure, where
convective flow descriptions are
appropriate and where diffusive B:
processes are essentially irrele-
vant; and second, in 'dead end'
geometries, where diffusional
processes must dominate.
Modeling of purely diffusive
events within the 2-D and 3-D
representations of biofilm
matrices showed that the effective
hindrance was increased by a
factor of only 2-4 (Figs. 8 and 9).
These results are in agreement
with related studies (Dibdin,
1990; Birmingham et ai, 1995)
but not with those of Lawrence et
al. (1994), where it appears that,
in the latter study, interpretation
of the data did not include all
possible parameters and hence
was not sensitive to diffusive
events only. Direct comparisons
of results obtained in this study
with experimental measurements Fig. 10—Analysis of pore space connected by (A) paths at least 0.6 }im in radius, and (B)
which necessarily contain paths at least 0.5 ytm in radius for the extracted volume segment shown in Fig. 9.
information pertinent to the
ultrastructural effects of extra-
cellular material are consistent and hence confirm relatively method of microstructural analysis, especially in terms of
little hindrance (Birmingham et al., 1995). connectivities and 'pore neck' characterization of different
A prerequisite intermediate step in the optimized biofilm areas (Fig. 10). Such an analysis is considered to be
continuum random walk model used in this work is to particularly useful in the identification and comparison of
calculate the complete Downloaded
Euclidean distance map of the differing micro-environments such as, for example, those that
from adr.sagepub.com at MCMASTER UNIV LIBRARY on February 21, 2016 For personal use only. No other uses without permission.
microstructure void space. This map is a particularly useful might be associated with anaerobic (dense) and aerobic (more
146 SINGLETON ET AL. ADV DENT RES APRIL 1997

mation of all fluid elements

A: over a bacterial surface, it is
possible to describe com-
pletely the total surface forces
experienced by an individual
bacterium (results not shown).
It is therefore theoretically
possible to determine the
forces transmitted to indi-
vidual biofilm components
under different conditions of
flow and shear and to relate
these to the kinds of binding
interactions experienced by
different bacteria in the
matrix.
Transport through real
biomatrices, however, is a
3 um more complex process than
B: we have considered here. In
reality, diffusion and con-
vection do not operate in
isolation, and each process is a
more complex physical
phenomenon than we have
illustrated here. In particular,
the active gradients known
to exist within biofilms
(Cronenberg and Van den
Heuvel, 1991; Lewandowski
et al., 1991) cannot be
explained by the relatively
modest diffusional retardation
produced by biofilm structures
alone—there is a need to in-
clude bio-chemical interactions
within the computational
models. One possibility is the
Fig. 11—Propagation analysis of flow through an extracted 3-D biofllm microstructure. extension of the lattice
Flow lines are shown skirting around bacteria and in some cases terminating at particularly Boltzmann technique to
dense areas. Flow direction is from the front to the back of the Fig. include more sophisticated
transport processes. For
example, recent developments
(Ladd, 1994) allow for the
open) structures. incorporation of further collision rules that will enable
A representative simulation of purely convective flow diffusion and convection-diffusion to be modeled within this
through a particular layer of the biofilm segment is shown in methodology.
Fig. 11. The paths taken by individual 'streamlines' can Nevertheless, we believe that the modeling presented here
clearly be seen to be 'sculpting' around bacteria and provides additional insight into the three-dimensional nature
bacterial aggregates. Importantly, it is clear that in certain of the transport process and its interaction with micro-
areas the proper flow propagation has been interrupted. This structure. Finally, note that the focus does not have to be on
phenomenon is related to both bacterial density and actual transport—the simulations when analyzed in detail can be used
physical blocking, indicating that such analysis may prove as comparative measures of different microstructural types.
useful for the identification and investigation of specific
"low-flow" areas (such as anaerobic sites) within biofilms. CONCLUSIONS
Furthermore, the Navier-Stokes equations include the com-
plete description of both the normal and tangential forces This paper describes a combination of experimental,
acting on each fluid element. Hence, by appropriate sum- analytical, and simulation methods applied to the
Downloaded from adr.sagepub.com at MCMASTER UNIV LIBRARY on February 21, 2016 For personal use only. No other uses without permission.
VOLll(l) MICROSCOPIC CHARACTERIZATION OF BIOFILMS
understanding of the formation and structural characterization
of model oral biofilms. The combination of optical
microscopy and fluorescent probe methodologies used in this
work allowed for multiparameter analysis of the overall
biofilm structure. In addition, the application of image
analysis procedures to generate realistic 3-D representations
of biofilm structure provided effective simulation of physical
transport properties within biofilm segments. Sampling large
(mm 2 ) 2-D areas at high optical resolution provided
information on the changing patterns of colonization of
specific organisms with respect to the biomass over time.
Results clearly indicated dominance of bacterial adhesion
mechanisms for the early development of oral biofilms on
HA in vitro. The application of confocal microscopy and
image analysis allowed for reconstruction of biofilm volume
segments, and, in conjunction with species-specific probes,
direct visualization of single-species micro-colonies was
demonstrated. Biofilm segments were visualized under
different nutrient conditions, and first steps toward
parameterization of the 3-D morphological and structural
distribution characteristics of representative biofilm segments
indicated that different biofilm structures could be
parameterized to give useful comparative data. Extraction of
realistic representations of biofilm structure allowed the
influence of this structure on transport properties to be
calculated. These simulations were capable of readily
identifying specific areas of the biofilm with different
diffusion characteristics. Modeling of flow in extracted
microstructures showed that the physical forces experienced
by individual biofilm components in the flow field could be
determined. Future work will address the additional
complexity of reactive modalities such as metabolic
processes through further development of imaging and
simulation techniques in parallel with biochemical
measurements.
ACKNOWLEDGMENT
We thank Dr. R. Gmiir for advice on antibody staining
procedures.
REFERENCES
Amann RI, Stromley J, Devereux R, Key R, Stahl DA (1992).
Molecular and microscopic identification of sulfate-
reducing bacteria in multispecies biofilms. Appl Environ
Microbiol 58:614-623.
Anwar H, Strap JL, Costerton JW (1992). Establishment of
ageing biofilms: possible mechanism of bacterial
resistance to antibiotic therapy. Antimicrob Agents
Chemother 36:1347-1351.
Barber P, Boyd A, Newman HN, Challacombe SJ,
Vrahopoulos TP, Gill S (1990). Immunogold labelling of
Porphyromnas gingivalis in pure culture and in apical
border plaque. Microb Ecol Health Dis 3:217-222.
Baudhuin P, Leroy-Houyet M-A, Quintart J, Bethert P
(1979). Application of cluster analysis for characterisation
of spatial distributions of particles by stereological
Downloaded from adr.sagepub.com at MCMASTER UNIV LIBRARY on February 21, 2016 For personal use only. No other uses without permission.
147
methods. / Microsc 115:1-17.
Behrend O, Harris R, Warren PB (1994). Hydrodynamic
behaviour of lattice Boltzmann and lattice Bhatnager-
Gross-Krook models. Phys Rev E 50:4586-4595.
Birmingham JJ, Hughes NP, Treloar R (1995). Diffusion and
binding measurements within oral biofilms using
fluorescence photobleaching recovery methods. Phil
Trans R Soc Lond B 350:325-343.
Bloem J, Veninga M, Shepherd J (1995). Fully automatic
determination of soil bacterium numbers, cell volumes,
and frequencies of dividing cells by confocal laser
scanning microscopy and image analysis. Appl Environ
Microbiol 61:926-936.
Bos R, Mei HC, Meinders JM, Busscher HJ (1994). A
quantitative method to study coadhesion of micro-
organisms in a parallel plate flow chamber: basic
principles of the analysis. / Microbiol Meth 20:289-305.
Caldwell DE, Korber DR, Lawrence JR (1992a). Imaging of
bacterial cells by fluorescence exclusion using scanning
confocal laser microscopy. J Microbiol Meth 15:249-261.
Caldwell DE, Korber DR, Lawrence JR (1992b). Confocal
laser microscopy and computer image analysis in micro-
bial ecology. In: Advances in microbial ecology. Vol. 12.
Marshall KC, editor. New York: Plenum Press, pp. 1-67.
Caldwell DE, Korber DR, Lawrence JR (1993). Analysis of
biofilm formation using 2D vs 3D digital imaging. / Appl
Bact 74(Symp Suppl):52s-66s.
Cobb CM, Killroy WJ (1990). Microbial colonization in
human periodontal disease: An illustrated tutorial on
selected ultrastructural and ecologic considerations.
Scanning Microsc 4:375-691.
Costerton JW, Cheng K-J, Geesey GG, Ladd TI, Nichel JC,
Dasgupta MK, et al. (1987). Bacterial biofilms in nature
and disease. Ann Rev Microbiol 41:435-464.
Costerton JW, Lewandowski Z, DeBeer D, Caldwell D,
Korber D, James G (1994). Biofilms, the customized
microniche. J Bacteriol 176:2137-2142.
Cowan M, Busscher HJ (1993). Flow chamber study of the
adhesion of Prevotella intermedia to glass after
preconditioning with mutans streptococcal species:
kinetics and spatial arrangement. Microbios 73:135-144.
Cronenberg CCH, Van den Heuvel JC (1991). Determination
of glucose diffusion coefficients in biofilms with micro-
electrodes. Biosens Bioelectro 6:255-262.
Deutch JM, Oppenheim I (1987). The concept of Brownian
motion in modern statisical mechanics. Faraday Disc
Chem Soc 83:1-20.
Dibdin GH (1981). Diffusion of sugars and carboxylic acids
through human dental plaque in vitro. Arch Oral Biol
26:515-523.
Dibdin GH (1990). Plaque fluid and diffusion: study of the
cariogenic challenge by computer modeling. / Dent Res
69:1324-1331.
d'Humieres D, Lallemand P, Frisch U (1986). Lattice gas
models for 3D hydrodynamics. Europhys Lett 2:291-297.
Eils R, Bertin E, Saracoglu K, Rinke B, Schrock E, Parazza
F, et al. (1995). Application of confocal laser microscopy
and three-dimensional Voronoi diagrams for volume and
148
surface estimates of interphase chromosomes. / Microsc
177:150-161.
Frisch U, Hasslacher B, Pomeau Y (1986). Lattice-gas
automata for the Navier-Stokes equation. Phys Rev Lett
56:1505-1508.
Gibbons RJ, van Houte J (1980). Bacterial adherence and the
formation of dental plaque. In: Bacterial adherence.
Beachey EH, editor. London: Chapman and Hall.
Gmiir R (1988). Applicability of monoclonal antibodies to
quantitatively monitor subgingival plaque for specific
bacteria. Oral Microbiol Immunol 3:187-191.
Gundersen HJG, Bagger P, Bendtsen TF, Evans SM, Korbo
L, Marcussen N, et al. (1988a). The new stereological
tools: Dissector, fractionator, nucleator and point sampled
intercepts and their use in pathological research and
diagnosis. Ada Pathol Microbiol Immunol Scand 96:857-
881.
Gundersen HJG, Bendtsen TF, Korbo L, Marcussen N,
M0ller A, Nielsen K, et al. (1988b). Some new, simple
and efficient stereological methods and their use in
pathological research and diagnosis. Acta Pathol
Microbiol Immunol Scand 96:379-394.
Hahn U, Lorz U (1994). Stereological analysis of the spatial
Poisson-Voronoi tessellation. / Microsc 175:176-185.
Hiemenz PC (1977). Principles of colloid and surface
chemistry. New York: Marcel Dekker, Inc.
Jernot JP, Prasad PB, Demaleprade P (1992). Three
dimensional simulation of flow through a porous medium.
J Microsc 167:9-21.
Kepner RL, Pratt JR (1994). Use of fluorochromes for direct
enumeration of total bacteria in environmental samples:
past and present. Microbiol Rev 58:603-615.
Kolenbrander PE, London J (1993). Adhere today, here
tomorrow: Oral bacterial adherence. / Bacteriol 175:3247-
3252.
Korber DR, Lawrence JR, Hendry MJ, Caldwell DE (1993).
Analysis of spatial variability within mot + and mot"
Pseudomonas fluorescens biofilms using representative
elements. Biofouling 7:339-358.
Ladd AJC (1994). Numerical simulations of particulate
suspensions via a discretized Boltzmann equation. 1.
Theoretical foundation. / Fluid Mech 271:285-309.
Lawrence JR, Wolfaardt GM, Korber DR (1994).
Determination of diffusion coefficients in biofilms by
confocal laser microscopy. Appl Environ Micobiol
60:1166-1173.
Lewandowski Z, Walser G, Characklis WG (1991). Reaction
kinetics in biofilms. Biotechnol Bioeng 38:877-882.
Lie T (1978). Ultrastructural study of early dental plaque
formation. JPeriodont Res 13:391-409.
MacDonald IF, Kaufmann PM, Dullien FAL (1986).
Quantitative image analysis of finite porous media. I.
Development of genus and pore map software. / Microsc
144:277-296.
Marsh PD (1991). Sugar, fluoride, pH and microbial
homeostasis in dental plaque. Proc Finn Dent Soc 87:515-
525.
Marsh PD (1994). Microbial ecology of dental plaque and its
Downloaded from adr.sagepub.com at MCMASTER UNIV LIBRARY on February 21, 2016 For personal use only. No other uses without permission.
SINGLETON ET AL. ADV DENT RES APRIL 1997
significance in health and disease. Adv Dent Res 8:263-
271.
Marsh PD, Bradshaw DJ, Watson GK, Allison C (1995).
Community development in a mixed species oral biofilm.
In: The life and death of biofilm. Wimpenny J, Handley P,
Gilbert P, Lappin-Scott H, editors. Cardiff, UK: Bioline,
pp. 65-69.
Mattfeldt T, Frey H, Rose C (1993). Second-order stereology
of benign and malignant alterations of the human
mammary gland. / Microsc 171:143-151.
McKee AS, McDermid AS, Ellwood DC, Marsh PD (1985).
The establishment of reproducible complex communities
of oral bacteria in the chemostat using defined inocula. /
Appl Bacteriol 59:263-275.
Mergenhagen SE, Sandberg AL, Chassy BM, Brennan MJ,
Yeung MK, Donkersloot JA, et al (1987). Molecular
basis of bacterial adhesion in the oral cavity. Rev Infect
Dis 9:5467-5476.
Moncla BJ, Motley ST, Braham P, Ewing L, Adams TH,
Vermeulen MJ (1991). Use of synthetic oligonucleotide
DNA probes for identification and direct detection of
Bacteroides forsythus in plaque samples. / Clin Microbiol
29:2158-2162.
Newman HN (1980). Plaque formation: structure and
biology. In: Dental plaque: The ecology of the flora of
human teeth. Newman HN, editor. Springfield, IL: Charles
C. Thomas, Pub., pp. 22-53.
Newman HN, Barber PM (1995). Dental plaque structure in
vivo. In: The life and death of biofilm. Wimpenny J,
Handley P, Gilbert P, Lappin-Scott H, editors. Cardiff,
UK: Bioline, pp. 27-32.
Pancorbo M, Anguiano E, Aguilar M (1994). Fractal
characterization by frequency analysis: III. Effect of noise.
J Microsc 176:54-62.
Qian YH, d'Humieres D, Lallemand P (1992). Lattice BGK
models for Navier-Stokes equation. Europhys Lett 17:479-
484.
Ripley BD (1977). Modeling spatial patterns. / R Stat Soc B
39:172-192.
Ripley BD (1979). Test of randomness for spatial point
patterns. JR Stat Soc B 41:368-374.
Rogers AH, de Jong MN, Zilm PS, van der Hoeven JS
(1986). Estimation of growth parameters for some oral
bacteria grown in continuous culture under glucose-
limited conditions. Infect Immun 52:897-901.
Russ JC (1993). Effects of noise and anisotropy on the
determination of fractal dimensions. / Microsc 172:239-
248.
Sahimi M (1994). Applications of percolation theory.
London: Taylor & Francis Ltd.
Savitt ED, Strzempko MN, Vaccaro KK, Peros WJ, French
CK (1988). Comparison of cultural methods and DNA
probe analyses for the detection of Actinobacillus
actinomycetemcomitans, Bacteroides gingivalis, and
Bacteroides intermedius in subgingival plaque samples. /
Periodontol 59:431-438.
Schwarz H, Exner HE (1981). The characterisation of the
arrangement of feature centroids in planes and volumes. /
VoLll(l)
Microsc 129:155-169.
Singleton S, Albiston L, Treloar R, Mahers E, Hodgson R,
Watson K, et al. (1995). Optical imaging and
characterisation of oral biofilm structures using vital stains
and specific antibody probes. In: The life and death of
biofilm. Wimpenny J, Handley P, Gilbert P, Lappin-Scott
H, editors. Cardiff, UK: Bioline, pp. 33-36.
Socransky SS, Manganiello AD, Propas D, Orava J, Van
Houte J (1977). Bacteriological studies of developing
supragingival plaque. / Periodont Res 12:90-106.
Tatevossian A (1985). The effect of heat inactivation,
tortuosity, extracellular polyglucan and ion exchange sites
on the diffusion of 14C-sucrose in human dental plaque
residue in vitro. Arch Oral Biol 30:365-371.
Torquato S, Kim IC (1989). Efficient simulation technique to
compute effective properties of heterogeneous media.
Appl Phys Lett 55:1$47-1S49.
Van Beelen P, van der Hoeven JS, de Jong MA,
Hoogendoorn H (1986). The effect of oxygen on the
growth and acid production of Streptococcus mutans and
Downloaded from adr.sagepub.com at MCMASTER UNIV LIBRARY on February 21, 2016 For personal use only. No other uses without permission.
MICROSCOPIC CHARACTERIZATION OF BIOFILMS 149
Streptococcus sanguis. FEMS Microbiol Ecol 38:25-30.
Van Loosdrecht MCM, Lyklema J, Norde W, Zehnder AJB
(1989). Bacterial adhesion: a physico-chemical approach.
Microbial Ecol 11:1-15.
Van Loosdrecht MCM, Lyklema J, Norde W, Zehnder AJB
(1990). Influence of interfaces on microbial activity.
Microbial Rev 54:75-87.
Wolfaardt GM, Lawrence JR, Roberts RD, Caldwell SJ,
Caldwell DE (1994). Multicellular organisation in a
degradative biofilm community. Appl Environ Microbiol
60:434-446.
Zambon JJ, Reynolds HS, Chen P, Genco RJ (1985). Rapid
identification of periodontal pathogens in subgingival
plaque: Comparison of indirect immunofluorescence
microscopy with bacterial culture for detection of
Actinobacillus actinomycetemcomitans. J Dent Res
64:793-798.
Zhao HQ, MacDonald IF (1993). An unbiased and efficient
procedure for 3-D connectivity measurement as applied to
porous media. J Microsc 172:157-162.