Content uploaded by Heidi M Sosik
Author content
All content in this area was uploaded by Heidi M Sosik on Jun 17, 2017
Content may be subject to copyright.
195
Introduction
Several research planning groups have noted that time series
sensors able to monitor plankton community structure, in
addition to those measuring bulk properties such as chlorophyll,
are needed to adequately investigate questions about coastal
ocean ecosystems (SCOTS Steering Committee 2002; Daly et al.
2004; Jahnke et al. 2004). With the goal of better understand-
ing how coastal plankton communities are regulated, we have
begun a high-resolution, long-term plankton monitoring pro-
gram at the Martha’s Vineyard Coastal Observatory (MVCO)
(Austin et al. 2000; Austin et al. 2002). FlowCytobot, a
submersible flow cytometer that uses fluorescence and light
scattering signals from a laser beam to characterize the smallest
phytoplankton cells (~1–10 μm) (Olson et al. 2003; Sosik et al.
2003), has been deployed at MVCO for most of the past 3
years. Other instruments, such as the Autonomous Vertically
Profiling Plankton Observatory (Thwaites et al. 1998) are capa-
ble of monitoring plankton at the other end of the size spec-
trum (mainly zooplankton >100 μm). However, plankton in the
size range 10 to 100 μm are not well sampled by either of these
instruments. This is a critical gap because phytoplankton in
this size range, which includes many diatoms and dinoflagel-
lates, can be especially important in coastal blooms, and micro-
zooplankton, such as protozoa, are critical to the diets of many
grazers including copepods and larval fish.
Nano- and microplanktonic organisms can be studied in
the laboratory or on board ships with a commercially available
imaging flow cytometer, the FlowCAM (Sieracki et al. 1998).
Other submersible flow cytometers have been developed, such
as the CytoSub (e.g., Cunningham et al. 2003), but to our
knowledge none has the necessary resolution and field
endurance for the ecological studies we wish to carry out.
Therefore we developed our own submersible imaging flow
cytometer, based on FlowCytobot.
A submersible imaging-in-flow instrument to analyze nano-
and microplankton: Imaging FlowCytobot
Robert J. Olson and Heidi M. Sosik
Biology Department, MS 32, Woods Hole Oceanographic Institution, Woods Hole, Massachusetts, USA
Abstract
A fundamental understanding of the interaction between physical and biological factors that regulate plank-
ton species composition requires, first of all, detailed and sustained observations. Only now is it becoming pos-
sible to acquire these types of observations, as we develop and deploy instruments that can continuously mon-
itor individual organisms in the ocean. Our research group can measure and count the smallest phytoplankton
cells using a submersible flow cytometer (FlowCytobot), in which optical properties of individual suspended
cells are recorded as they pass through a focused laser beam. However, FlowCytobot cannot efficiently sample
or identify the much larger cells (10 to >100 μm) that often dominate the plankton in coastal waters. Because
these larger cells often have recognizable morphologies, we have developed a second submersible flow cytome-
ter, with imaging capability and increased water sampling rate (typically, 5 mL seawater analyzed every 20 min),
to characterize these nano- and microplankton. Like the original, Imaging FlowCytobot can operate unattend-
ed for months at a time; it obtains power from and communicates with a shore laboratory, so we can monitor
results and modify sampling procedures when needed. Imaging FlowCytobot was successfully tested for 2
months in Woods Hole Harbor and is presently deployed alongside FlowCytobot at the Martha’s Vineyard
Coastal Observatory. These combined approaches will allow continuous long-term observations of plankton
community structure over a wide range of cell sizes and types, and help to elucidate the processes and interac-
tions that control the life cycles of individual species.
Acknowledgments
This research was supported by grants from NSF (Biocomplexity
IDEA program and Ocean Technology and Interdisciplinary Coordination
program; OCE-0119915 and OCE-0525700) and by funds from the
Woods Hole Oceanographic Institution (Bigelow Chair, Ocean Life
Institute, Coastal Ocean Institute, and Access to the Sea Fund). We are
indebted to Alexi Shalapyonok for expert technical assistance, Tatiana
Orlova (Institute of Marine Biology, Russian Academy of Sciences) for
manual microscopic plankton analyses, Tom Hurst and Glenn McDonald
for electrical and mechanical engineering support, and the Martha’s
Vineyard Coastal Observatory operations team, especially Janet
Fredericks, for logistical support.
Limnol. Oceanogr.: Methods
5, 2007, 195–203
© 2007, by the American Society of Limnology and Oceanography, Inc
.
LIMNOLOGY
and
OCEANOGRAPHY: METHODS
Imaging FlowCytobot uses a combination of video and
flow cytometric technology to both capture images of organ-
isms for identification and measure chlorophyll fluorescence
associated with each image. Images can be automatically clas-
sified with software based on a support vector machine (Sosik
and Olson 2007), and the measurements of chlorophyll fluo-
rescence allow us to more efficiently analyze phytoplankton
cells by triggering on chlorophyll-containing particles. Quan-
titation of chlorophyll fluorescence in large phytoplankton
cells will also enable us to better interpret patterns in bulk
chlorophyll data and to discriminate heterotrophic from pho-
totrophic cells.
Imaging FlowCytobot’s design was intended to follow, as
far as possible, that of the original FlowCytobot, because of
that instrument’s history of reliability in field deployments.
A seawater sample is injected into the center of a sheath
flow of particle-free water; all the particles are thus confined
to the center of the flow cell, which ensures that each parti-
cle is in focus as it passes through the optical system. An
important feature of the design is that the sheath fluid is
recycled through a filter cartridge, which removes sample
particles after they have been analyzed. This feature allows
for the efficient use of antifouling agents so the system can
operate for months at a time without the need for mainte-
nance or cleaning. The instrument is contained in a watertight
housing, and it operates continuously and autonomously
under the direction of a computer whose programming can
be modified by a remote operator. Programmable operations
include data acquisition and transfer to shore, adjustment
of sampling frequency and rate of injection, injection of
internal standard beads, flushing the flow cell and/or sam-
ple tubing with detergent, backflushing the sample tubing
to remove potential clogs, adding sodium azide to the
sheath reservoir to prevent biofouling of the internal sur-
faces, and focusing the imaging objective lens.
Methods and Procedures
Mechanical/electrical—Imaging FlowCytobot is constructed
around an optical breadboard (20.32 by 60.96 cm) with
mostly off-the-shelf components; the fluid-handling and elec-
tronic components are mounted on opposite sides of the
breadboard (Fig. 1). The breadboard hangs from the instrument
end cap, which seals to the watertight housing (30.48 cm inner
diameter by 76.2 cm) via 2 nitrile o-rings and has external
connections to an MVCO guest port for power and Ethernet
communication with shore. Power is supplied as 36V DC (100 W).
Communication (10 megabits s–1) between the instrument and
the MVCO guest port are via category-5 cable, and between
the guest port and the shore laboratory via optic fiber (Austin
et al. 2000; Austin et al. 2002).
Olson and Sosik In situ imaging of nano- and microplankton
196
Fig. 1. Imaging FlowCytobot, removed from underwater housing. Three plastic standoffs prevent contact of the components and housing during instal-
lation. Left: front view, showing the “fluidics and optics” side of the optical breadboard. The flow cell (hidden by standoff) is located between the con-
denser and objective lenses. Right: side view; the optical breadboard is edge-on in the center, with fluidics/optics components mounted to the left and
electronics to the right.
Fluidics system—Imaging FlowCytobot’s fluidics system
(Fig. 2) is based on that of a conventional flow cytometer: hydro-
dynamic focusing of a seawater sample stream in a particle-free
sheath flow carries cells in single file through a laser beam
(and then through the camera’s field of view). The fluidics and
sampling system is similar to that of the original FlowCytobot
except that, to minimize problems due to settling of large par-
ticles, the syringe is mounted vertically rather than horizon-
tally and the flow through the flow cell is downward rather
than upward.
The sheath fluid, seawater forced through a pair of 0.2-μm
filter cartridges (Supor; Pall Corp.) by a gear pump (Micropump,
Inc. Model 188 with PEEK gears), flows through a conical cham-
ber to a quartz flow cell. The flow cell housing and sample injec-
tion tube is from a Becton Dickinson FACScan flow cytometer,
but the flow cell is replaced by a custom cell with a wider chan-
nel (channel dimensions 860 by 180 μm; Hellma Cells, Inc.).
Because the FACScan objective lens housing, which normally
supports the plastic flow cell assembly, is not used here, an alu-
minum plate (3.175 mm thick) is bolted to the assembly.
A programmable syringe pump (Versapump 6 with 48,000-
step resolution, using a 5-mL syringe with Special-K plunger;
Kloehn, Inc.) is used to sample seawater through a 130-μm
Nitex screen (to prevent flow cell clogging), which is protected
against biofouling by 1 mm copper mesh. The sample water is
then injected through a stainless steel tube (1.651 mm OD,
0.8382 mm internal diameter; Small Parts, Inc.) into the cen-
ter of the sheath flow in the cone above the flow cell. The tub-
ing is of PEEK material (3.175 by 1.575 mm external and inter-
nal diameter for sheath tubes, 1.588 by 0.762 mm for others;
Upchurch Scientific).
An 8-port ceramic distribution valve (Kloehn, Inc.) allows
the syringe pump to carry out several functions in addition
to seawater sampling. These include regular (~daily) addition
of sodium azide to the sheath fluid (final concentration
~0.01%) to prevent biofouling, and regular (~daily) analyses
of beads (20 or 9 μm red-fluorescing beads; Duke Scientific,
Inc.) as internal standards to monitor instrument perform-
ance. In addition, during bead analyses (~20 min d–1), the
sample tubing (which is not protected from biofouling by
contact with azide-containing sheath fluid) is treated with
detergent (5% Contrad/1% Tergazyme mixture) to remove
fouling. Finally, the syringe pump is used to prevent accu-
mulation of air bubbles (from degassing of seawater) in the
flow cell, which could disrupt the laminar flow pattern;
before each sample is injected, sheath fluid is withdrawn
(along with any air bubbles) through both the sample injec-
tion needle and the conical region above the flow cell, and
discarded to waste. Azide solution, suspended beads, and
detergent mixture are stored in 100-mL plastic bags with
Luer fittings (Stedim Biosystems).
Olson and Sosik In situ imaging of nano- and microplankton
197
Fig. 2. Schema of fluidics system of Imaging FlowCytobot.
Fig. 3. Schema of optical layout of Imaging FlowCytobot.
Optical system—Flow cytometric measurements are derived
from a red diode laser (SPMT, 635 nm, 12 mW, Power Tech-
nologies, Inc.) focused to a horizontally elongated elliptical
beam spot by cylindrical lenses (horizontal = 80 mm focal
length, located 100 mm from the flow cell; vertical = 40 mm
focal length, at 40 mm). Each particle passing through the
laser beam scatters laser light, and chlorophyll-containing
cells emit red (680 nm) fluorescence (Fig. 3). One of these sig-
nals (usually chlorophyll fluorescence) is chosen to trigger a
xenon flash lamp (Hamamatsu L4633) when the signal
exceeds a preset threshold; the resulting 1-μs flashes of light
are used to provide Kohler illumination of the flow cell. The
green component of the light (isolated by a bandpass filter) is
focused into a randomized fiber optic bundle (50 μm fibers,
6.35 mm diameter; Stocker-Yale, Inc.). At the bundle exit, the
light is collected by a lens, passes through a field iris, and is
focused onto a condenser iris located approximately at the
back focal plane of a 10×objective lens (Zeiss CP-Achromat,
numerical aperture [N.A.] 0.25), which is in turn focused on
the flow cell. A second 10×objective (Zeiss Epiplan, N.A. 0.2)
collects the light from both flash lamp illumination (green)
and laser (red, 635 nm scattered light and 680 nm chlorophyll
fluorescence). Green and red wavelengths are separated by a
dichroic mirror (590 nm short pass); green light continues to
a monochrome CCD camera (UniqVision UP-1800DS-CL,
1380 by 1034 pixels), and red light is reflected to a second
dichroic (635 LP), which directs scattered laser light and fluo-
rescence to separate photomultiplier (PMT) modules (Hama-
matsu HC120-05 modified for current-to-voltage conversion
with time constant = 800 kHz; the PMT for laser scattering also
incorporates DC restoration circuitry).
The optical path is folded by broadband dielectric mirrors
(Thorlabs BB1-E02) on either side of the flow cell to conserve
space. The flow cell assembly is fixed to the optical table, and the
light source/condenser and objective/PMT/camera assemblies
are each mounted on lockable translators (Newport Corp.) pro-
viding 3 degrees of freedom for adjustment. The objective focus-
ing translator is remotely controllable (see Instrument Control
below). Optical mounting hardware is from Thorlabs, Inc.
Data acquisition and instrument control—Imaging FlowCytobot
is controlled by a PC-104plus computer (Kontron MOPS-LCD7,
700 MHz) running Windows XP (Microsoft Corp.). Remote
operation is carried out via Virtual Networking Computing soft-
ware (www.realvnc.com). The camera is configured and the
syringe pump is programmed by software provided by the man-
ufacturers; all other functions (control, image visualization, and
data acquisition) are carried out by custom software written in
Visual Basic 6 (Microsoft Corp.).
A custom electronics board amplifies and integrates light
scattering and fluorescence signals, and also generates con-
trol pulses for timing purposes (Fig. 4). The signal from the
triggering PMT (typically chlorophyll fluorescence) is split,
with one part sent to a comparator circuit that produces a
trigger pulse if the signal is larger than a preset threshold
level. The other part of the signal, and the signal from the
other PMT, are delayed by 7 μs (by delay modules from a
Coulter Electronics EPICS 750 flow cytometer) and then split
and sent to paired linear amplifiers with 25-fold different
gains (to increase dynamic range) before integration (Burr-
Brown AFC2101). The delay modules allow the pretrigger
portions of the signals to be included in the integration. The
end of the integration window is also determined by the
comparator, with the provision that the signal remains
below the comparator threshold for 20 μs; this allows signals
from loosely connected cells such as chain diatoms to be
more accurately measured. Comparator output pulses are
also integrated to provide an estimate of the duration of each
signal. The PMT amplifier inputs are grounded by transistors
during flash lamp operation to avoid baseline distortion by
the very large signals from the flashes (Fig. 4A, D).
The trigger pulse is also sent to a frame grabber board
(Matrox Meteor II CL) to begin image acquisition, and, after a
delay of 270 μs, to the flash lamp, which illuminates the flow
cell for a 1-μs exposure. Integration of light scattering and flu-
orescence signals is limited to 270 μs to avoid contamination
by light from the flash lamp, so integrated signals from cells
or chains longer than ~600 μm are minimum estimates.
Olson and Sosik In situ imaging of nano- and microplankton
198
Fig. 4. Imaging FlowCytobot signals and controls. When a trigger signal
(from the chlorophyll fluorescence PMT) exceeds a preset level (A), a
comparator produces a logic signal (B); the decay of this signal is artifi-
cially slowed by a capacitor so that the integration window (C) does not
miss signals from closely following cells, as for chain diatoms. The chloro-
phyll fluorescence and side scattering PMT signals, delayed by 7 μs (D),
are integrated and held (E), as is the logic signal itself (to provide an esti-
mate of signal duration). At the end of the integration window, the held
signals are digitized (F). The original signal also triggers the CCD camera
(G) and a 270 μs flash lamp timer (H) (note the flash lamp signals in A
and D). The original signal also triggers a “reset” pulse (I) lasting 80 ms,
during which time no other triggers can be detected. (In the case of large
cells whose image processing requires >80 ms, this “dead” time is
extended through software-mediated grounding of the trigger signal
amplifier input and measured via a Visual Basic timer.)
A multifunction analog-digital board (104-AIO16-16E, Acces
I/O Products, Inc.) digitizes the integrated laser-derived signals
and the duration of the triggering signals, produces analog sig-
nals to control the PMT high voltages, and carries out digital I/O
tasks (e.g., motor control for focusing the objective and com-
munication between software and hardware, i.e., inhibiting
new trigger signals while the current image is being processed).
Data analysis—To minimize the resources needed for image
data storage, Imaging FlowCytobot utilizes a “blob analysis”
routine (Matrox Imaging Library 7.5) based on edge detection
(changes in intensity across the frame) to identify regions of
interest in each image. The subsampled images are transferred
to a remote computer for storage and further analysis. For tax-
onomic classification, we developed an approach based on a
support vector machine framework and several different fea-
ture extraction techniques; this approach is described else-
where (Sosik and Olson 2007), along with results of automated
classification of 1.5 ×106images obtained during Imaging
FlowCytobot’s test deployment in Woods Hole Harbor.
For each particle, 5 channels of flow cytometric signal data
are stored (integrated signals from fluorescence and light scatter-
ing detectors at 2 gain settings each, plus signal duration), along
with a time stamp (10-ms resolution). Accumulated images and
fluorescence/light scattering data are automatically transferred to
the laboratory in Woods Hole every 30 min. The data are ana-
lyzed using software written in MATLAB (The Mathworks, Inc.).
Deployment—Imaging FlowCytobot is currently deployed
by divers, who bolt the neutrally buoyant 70-kg instrument to
a mounting frame located at 4-m depth on the MVCO Air Sea
Interaction Tower (http://www.whoi.edu/mvco), and connect
the power and communications cable, which is equipped with
an underwater pluggable connector (Impulse Enterprise, Inc.).
Imaging FlowCytobot has been deployed at MVCO since 27
September 2006.
Assessment
Cell quantitation—Hydrodynamic focusing causes all the cells
in a sample to pass through Imaging FlowCytobot’s analysis
region, so cell concentrations can be calculated, to a first
approximation, by dividing the number of triggers by the vol-
ume of water analyzed (as determined by the analysis time
and the known rate of flow from the syringe pump). However,
this concentration is an underestimate, because during the
time required to acquire and process each image, sample con-
tinues to flow through the flow cell but no new triggers are
allowed. The minimum time required by the camera for image
acquisition is 34 ms (i.e., 30 frames s–1), but we determined
empirically that with image processing to locate and store the
region of interest, at least 86 ms was required by the system;
very large cells required even more time. We therefore mea-
sure the image processing period for each cell using a software
timer. By subtracting the sum of these periods from the total
elapsed time, we determine how much time is actually spent
“looking” for cells, and use this value to calculate cell concen-
tration in each syringe sample.
To evaluate cell quantitation by Imaging FlowCytobot, we
analyzed replicate samples with both Imaging FlowCytobot and
a Coulter EPICS flow cytometer, a nonimaging instrument capa-
ble of measuring cells at rates >103s–1. We used a laboratory cul-
ture of
Dunaliella tertiolecta
, a small (6 μm) phytoplankter, because
cells in this size range can be reliably analyzed by both instru-
ments. Using the measured analysis time as described above,
Imaging FlowCytobot–derived cell concentrations were indistin-
guishable from those of the EPICS flow cytometer (Fig. 5A).
Olson and Sosik In situ imaging of nano- and microplankton
199
Fig. 5. Quantitation of cell counting. (A) Concentrations of 6-μm
Dunaliella tertiolecta
cells measured with Imaging FlowCytobot were compared with those
measured with a Coulter EPICS flow cytometer modified for high flow rate. “Electronic dead time” during image capture and analysis caused progressive
undersampling as cell concentration increased (filled symbols), but by normalizing counts to the time actually spent examining sample water (open
symbols), counts indistinguishable from those of the EPICS were obtained (see 1:1 line). (B) Results of dilution series for
Dunaliella
and for the much larger
(~20 by 100 μm) diatom
Ditylum brightwellii
are linear, indicating that cell concentrations measured by Imaging FlowCytobot are accurate to >104cells mL–1.
Analyses of dilution series of
Dunaliella
and of a much larger
diatom (
Ditylum brightwellii
, ~20 by 100 μm), which often
required additional time (>86 ms) for image processing, indi-
cated that cell concentrations from Imaging FlowCytobot were
reliable for both sizes of cells, up to at least 1.5 ×104cell mL–1
(Fig. 5B), very high concentrations for marine nanoplankton.
Standard beads to assess flow cytometric measurements—
Measurements of uniform beads indicate that light scattering and
fluorescence data are quantitative (Fig. 6); signals are uniform
across the 150 μm–wide sample core, and the coefficient of vari-
ation of bead fluorescence signals is typically <10% even after
extended periods of deployment. Although the flow cytometric
measurements are probably of less interest than the images of
cells, it is important to note that the acquisition of each image is
initiated by the detection of a signal exceeding a threshold, so it
is important to monitor detection efficiency during operation.
Phytoplankton populations—Analysis of seawater samples by
Imaging FlowCytobot illustrates some advantages of the
approach over conventional flow cytometry and manual
microscopic analyses. First, flow cytometric sorting of particles
in seawater has shown that light scattering/fluorescence sig-
natures are rarely sufficient to identify nano- or microplank-
ton at the genus or species level. Discrete populations are
rarely discernible in a plot of light scattering vs. fluorescence
(e.g., Fig. 7), and even if they are, it is difficult to be sure of
their identity without cell sorting and examination. The
images associated with the flow cytometric data reinforce this
idea—different species do have characteristic light scattering/
fluorescence signatures, but these generally form a continuum
(and often overlap) and so are not very useful in determining
species composition. (The homogeneous populations of cells
indicated by the image groupings in Fig. 7 are not random
selections, but were obtained by trial-and-error searches of
small regions of the plot; other regions show mixtures of
species.) Thus, imaging allows us to greatly improve the accu-
racy of identification of different cells.
Imaging can also be used to study nonphytoplankton par-
ticles (Fig. 8), whose composition and abundance patterns are
almost unknown. Triggering from light scattering rather than
fluorescence signals reveals that the large majority of the par-
ticles in this seawater sample were not phytoplankton, but
included various forms of detritus, empty diatom frustules,
and heterotrophic organisms.
Preliminary comparisons of Imaging FlowCytobot’s per-
formance to traditional manual microscopy are encouraging
(Fig. 9). For the dominant and most easily recognized cells in
the water sample (the diatom
Guinardia
spp.), the counts were
Olson and Sosik In situ imaging of nano- and microplankton
200
Fig. 6. Analysis of uniform fluorescent beads (20 μm, Duke Scientific)
illustrates Imaging FlowCytobot’s measurements of fluorescence and scat-
tering; single beads are easily distinguished from doublets and clumps of
beads (A). This precision is the result of hydrodynamic focusing of the sam-
ple stream, which confines sample particles to the central core of the flow
cell; the core can be visualized (B) by plotting the position of each bead’s
image in the camera’s field of view. The flow (gray arrow) is downward at
2.2 m s–1 and imaging takes place 270 μs after a particle passes through
the laser beam (red arrow). An image from each population is shown.
Fig. 7. Flow cytometric measurements of side scattering and chlorophyll
fluorescence, and selected images of phytoplankton cells in a seawater
sample from Woods Hole Harbor (Dec. 2004), analyzed by Imaging Flow-
Cytobot (triggered by chlorophyll fluorescence). All images are shown at
the same scale; the smallest cells are ~5 μm. Different regions in the
scattering/fluorescence signature contain different species, but popula-
tion boundaries are indistinct. Cell images (clockwise from lower left):
mixed “small cells,”
Euglena
spp.,
Chaetoceros
spp.,
Ditylum
spp.,
Dactyliosolen
spp.,
Rhizosolenia
spp.
very close. For some categories, such as dinoflagellates, Imag-
ing FlowCytobot counts were lower, probably because the dis-
tinguishing features of dinoflagellate cells (e.g., cingulum)
were not always visible in the images (due to orientation) or
were insufficiently resolved. Dinoflagellates are often highly
pigmented relative to other cells of interest, so the illumina-
tion conditions used in Imaging FlowCytobot caused the cells
to appear very dark (even though green illumination, which is
not strongly absorbed by photosynthetic pigments, was used
to minimize this effect). It is likely that many dinoflagellates
were classified as “round 20 μm cells,” of which Imaging Flow-
Cytobot saw many more than the microscopist. For almost
all of the more rare categories, Imaging FlowCytobot found
more cells than microscopy, sometimes many more. Some-
times this was because the plankton groups were not counted
by the kind of microscopy/sample preservation method used
(ciliates, small dinoflagellates, flagellates), but others remain
unexplained (
Cylindrotheca
,
Licmophora
, pennate diatoms).
A second strength of Imaging FlowCytobot is the greatly
increased scope made possible by the automated nature of the
approach. A test deployment of Imaging FlowCytobot at 5 m
depth off the WHOI pier (Fig. 10) showed that the instrument
is capable of operating without external maintenance for at
least 2 months. The results presented here are simply cell
counts, showing long-term trends in cell abundance, with
superimposed higher-frequency periodicity (probably tidal).
Analyzing seawater at a nominal rate of 0.25 mL min–1, more
than 1.5 million images were collected during this deployment;
the analysis of these images with an automated approach is pre-
sented in a companion paper (Sosik and Olson 2007).
Image quality—The ultimate resolution of the optical sys-
tem is determined by the 10×microscope objective, which has
a theoretical resolution of ~1 μm. As presently configured, a
20-μm bead spans 68 pixels (3.4 pixels/μm), so the camera res-
olution is more than adequate for this objective. However,
image quality will be affected by several additional factors in
Imaging FlowCytobot, including cell motion, flash lamp pulse
duration, and location of cells in the flow cell.
Olson and Sosik In situ imaging of nano- and microplankton
201
Fig. 8. As for Fig. 7, but triggered by light scattering. Note the large number of detritus particles (including empty diatom frustules) relative to chloro-
phyll-containing cells.
Fig. 9. Microplankton community composition in a surface seawater
sample from Woods Hole Harbor (9 February 2006), as analyzed by Imag-
ing FlowCytobot and by manual microscopy. The sample was split and
100 mL was fixed with Lugol’s solution and settled in an Utermohl cham-
ber for examination by a trained taxonomist. The same sample volume
was analyzed by Imaging FlowCytobot, with subsequent manual image
classification. Only categories with 10 or more observations are shown.
Numbers of chain-forming cells were estimated as described in “Com-
ments and Recommendations.”
Movement of the subject due to sheath flow during the
camera exposure will tend to blur the image in the direction
of flow. Sample particle velocity was determined (by measur-
ing the image displacement caused by a known change in
strobe delay) to be 2.2 m s–1, so the subject moves 7.5 pixels
during the 1-μs exposure. The effect of this movement is visi-
ble in an image of a plastic bead as a thickening of the leading
and trailing edges, relative to the upper and lower edges (not
shown). In addition, although most of the light energy from
the xenon flash is emitted within 1 μs, the flash decays over
several microseconds, which produces a “shadow” down-
stream of high-contrast subjects. These factors limit the veloc-
ity of flow that can be employed, and thus the sampling rate
of the instrument (although a shorter flash, as from an LED or
pulsed laser, could be used to address this limitation).
The sample core in Imaging FlowCytobot is about 150 μm
wide (see Fig. 6B), so if we assume that the core has the same
shape as the channel, the thickness of the core would be ~33
μm. This is somewhat greater than the theoretical depth of
focus of a 10×objective with N.A. 0.2 (~10 μm). As the thick-
ness of the sample core increases, more particles will be out of
focus, which will limit both the sampling rate and the optical
resolution that can be employed. Finally, the illumination
conditions (e.g., condenser aperture, which is dictated by the
amount of light available during the flash) affect the resolu-
tion and contrast of the image.
Discussion
Plankton in the size range 10 to 100 μm, which includes
many diatoms and dinoflagellates, are critical components of
coastal ecosystems, but their regulation is relatively poorly
understood because it is difficult to sample them adequately in
the dynamic coastal environment. An important part of this
sampling problem is now addressed by Imaging FlowCytobot’s
unprecedented capabilities for autonomously obtaining quan-
titative data on nano- and microphytoplankton, with images
of sufficient quality to allow taxonomic resolution to genus or
even species level in some cases, high sampling resolution
(typically, 5 mL seawater analyzed every 20 min), and long
endurance (months). These capabilities, in combination with
the automated image classification approach described in the
companion paper (Sosik and Olson 2007), will allow oceanog-
raphers to carry out a wide variety of studies of species succes-
sion, responses of communities to environmental changes,
and bloom dynamics with vastly improved resolution and
scope. These improvements promise to lead to improved
understanding of many aspects of plankton ecology.
Comments and Recommendations
Limitations on deployment duration—The limiting factor for
Imaging FlowCytobot’s endurance in the field appears to be
wear of the syringe plunger seal, which eventually leaks water
into the housing. This is also the case with FlowCytobot, and
could conceivably be prevented by advances in materials used
for the seal. The problem is exacerbated by low temperatures,
as the syringe is designed for use at room temperature (Kloehn,
Inc., personal communication) and the seal shrinks signifi-
cantly as temperatures approach freezing. Even so, FlowCytobot
has had a successful 6-month deployment, and Imaging Flow-
Cytobot’s successful 2-month test deployment included peri-
ods with water temperatures approaching 0 °C. We have taken
the precaution of installing temperature and humidity sensors
inside the housing, which provides warning of slow leaks due
to syringe wear. Before closing, the housing is flushed with dry
nitrogen and a packet of silica gel desiccant is placed inside, to
prevent condensation. As a precaution against internal leaks in
the tubing or flow cell, the sample inlet and overflow ports are
fitted with solenoid valves (Reet Corp.) that close if humidity
rises above a preset level (or if power is interrupted). To help
analyze potential problems, we store temperature and humid-
ity data along with each image, so we can monitor the history
of conditions inside the instrument.
Beads as internal standards—The acquisition of each image is
initiated by the detection of a fluorescence (or scattering) sig-
nal exceeding a threshold, so it is important to monitor detec-
tion efficiency during operation. In the original FlowCytobot,
this is accomplished using periodic automated sampling of
1-μm beads. For Imaging FlowCytobot, we need to use beads
that are much larger and whose fluorescence is excited by red
light, and these have proved more difficult to use as internal
standards. During the 2-month test deployment of Fig. 10,
suitable standard beads were not available, but since then we
have obtained red-excited beads of 9- and 20-μm diameter,
which are now periodically analyzed as part of the sampling
program. Initially, the number of beads sampled decreased
dramatically after only a few days’ deployment, probably
because the relatively large size of these beads causes them to
sink rapidly, and because the beads can stick to the walls of the
Olson and Sosik In situ imaging of nano- and microplankton
202
Fig. 10. Phytoplankton cell concentrations measured by Imaging Flow-
Cytobot during test deployment in Woods Hole Harbor in 2005.
reservoir and/or tubing. Using a magnetic stirrer to mix the
suspension of beads before each sampling, and adding bovine
serum albumin and detergent to reduce stickiness (Velikov et
al. 1998), has reduced the loss of beads over time, but we are
still working on this problem.
Quantifying chain-forming cells—Chain-forming diatoms,
which often dominate coastal blooms, present a special count-
ing problem. Ideally, we want to know the number of cells
present, not simply the number of chains (which can vary
widely in length). For the comparison between microscopic
counts and Imaging FlowCytobot results (Fig. 9), therefore, we
manually counted each cell in images of the chain-forming
diatoms (
Guinardia
,
Thalassiosira
,
Nitzschia
,
Skeletonema
, and
Thalassionema
). We also used the measured duration of the
chlorophyll fluorescence signal for each chain, calibrated by
manual counts of the cells in each chain, to estimate cell num-
bers. This approach allows us to estimate the number of cells
in chains that extend out of the camera’s field of view (those
more than ~300 μm), by extrapolating the relationship
between cells per chain and signal duration. For the seawater
sample in Fig. 9, for example,
Guinardia
chains of up to 32
cells were observed by microscopy, whereas the maximum
number of cells visible in Imaging FlowCytobot images was
13; correction for these “unimaged” cells increased the Imag-
ing FlowCytobot estimate of
Guinardia
cells from 607 to 920.
This approach requires manual calibration for each chain-
forming species (although we plan to investigate automated
image analysis for this task) and will be affected by changes in
the size of the individual cells. In addition, the timing of the
flash lamp limits signal durations to ~270 μs, so cell numbers
for chains longer than 600 μm will be minimum estimates.
References
Austin, T., and others. 2002. A network-based telemetry archi-
tecture developed for the Martha’s Vineyard coastal obser-
vatory. IEEE J. Oceanic Eng. 27: 228-234.
———. 2000. The Martha’s Vineyard Coastal Observatory: A
long-term facility for monitoring air-sea processes. Proc.
Oceans 2000 3: 1937-1941.
Cunningham, A., D. McKee, S. Craig, G. Tarran, and C.
Widdicombe. 2003. Fine-scale variability in phytoplankton
community structure and inherent optical properties mea-
sured from an autonomous underwater vehicle. J. Mar. Sys.
43: 51-59.
Daly, K. L., R. H. Byrne, A. G. Dickson, S. M. Gallager, M. J.
Perry, and M. K. Tivey. 2004. Chemical and biological sen-
sors for time-series research: Current status and new direc-
tions. Marine Technology Society.
Jahnke, R. and others. 2004. Coastal Observatory Research
Arrays: A Framework for Implementation Planning. Report
on the CoOP CORA Workshop. Skidaway Institute of
Oceanography Technical Report Draft TR-03-01, 69.
Olson, R. J., A. A. Shalapyonok, and H. M. Sosik. 2003. An
automated submersible flow cytometer for pico- and
nanophytoplankton: FlowCytobot. Deep-Sea Res. I 50:
301-315.
SCOTS Steering Committee. 2002. SCOTS: Scientific cabled
observatories for time series, Workshop report, http://www.
coreocean.org/SCOTS/. 88.
Sieracki, C. K., M. E. Sieracki, and C. S. Yentsch. 1998. An
imaging-in-flow system for automated analysis of marine
microplankton. Mar. Ecol. Prog. Ser. 168: 285-296.
Sosik, H. M., and R. J. Olson. 2007. Automated taxonomic
classification of phytoplankton sampled with image-
in-flow cytometry. Limnol. Oceanogr. Methods in press.
———, R. J. Olson, M. G. Neubert, A. A. Shalapyonok, and A.
R. Solow. 2003. Growth rates of coastal phytoplankton
from time-series measurements with a submersible flow
cytometer. Limnol. Oceanogr. 48: 1756-1765.
Thwaites, F. T., S. M. Gallager, C. S. Davis, A. M. Bradley,
A. Girard, and W. Paul. 1998. A winch and cable for the
autonomous vertically profiling plankton observatory.
Oceans 98 - Engineering for sustainable use of the oceans:
conference proceedings 28 September - 1 October 1998 1:
32-36.
Velikov, K. P., F. Durst, and O. D. Velev. 1998. Direct observa-
tion of the dynamics of latex particles confined inside thin-
ning water-air films. Langmuir 14: 1148-1155.
Submitted 5 October 2006
Accepted 27 February 2007
Olson and Sosik In situ imaging of nano- and microplankton
203