Lensless high-resolution on-chip optofluidic
microscopes for
Caenorhabditis elegans
and cell imaging
Xiquan Cui*
†
, Lap Man Lee
†‡
, Xin Heng*, Weiwei Zhong
§
, Paul W. Sternberg
§
, Demetri Psaltis*
¶
, and Changhuei Yang*
‡
Departments of *Electrical Engineering and
‡
Bioengineering, and
§
Division of Biology, California Institute of Technology, Pasadena, CA 91125;
and
¶
School of Engineering, Ecole Polytechnique Federale de Lausanne, CH-1015 Lausanne, Switzerland
Communicated by Amnon Yariv, California Institute of Technology, Pasadena, CA, May 13, 2008 (received for review March 31, 2008)
Low-cost and high-resolution on-chip microscopes are vital for reduc-
ing cost and improving efficiency for modern biomedicine and bio-
science. Despite the needs, the conventional microscope design has
proven difficult to miniaturize. Here, we report the implementation
and application of two high-resolution (
0.9
m for the first and
0.8
m for the second), lensless, and fully on-chip microscopes based on
the optofluidic microscopy (OFM) method. These systems abandon
the conventional microscope design, which requires expensive lenses
and large space to magnify images, and instead utilizes microfluidic
flow to deliver specimens across array(s) of micrometer-size apertures
defined on a metal-coated CMOS sensor to generate direct projection
images. The first system utilizes a gravity-driven microfluidic flow for
sample scanning and is suited for imaging elongate objects, such as
Caenorhabditis elegans
; and the second system employs an electro-
kinetic drive for flow control and is suited for imaging cells and other
spherical/ellipsoidal objects. As a demonstration of the OFM for
bioscience research, we show that the prototypes can be used to
perform automated phenotype characterization of different
Caeno-
rhabditis elegans
mutant strains, and to image spores and single
cellular entities. The optofluidic microscope design, readily fabricable
with existing semiconductor and microfluidic technologies, offers
low-cost and highly compact imaging solutions. More functionalities,
such as on-chip phase and fluorescence imaging, can also be readily
adapted into OFM systems. We anticipate that the OFM can signifi-
cantly address a range of biomedical and bioscience needs, and
engender new microscope applications.
optofluidic microscopy
phenotype characterization
microfluidic
O
ptical microscopy pervades almost all aspects of modern
biomedicine and bioscience; to name a few key areas, optical
microscopes are vital instruments in microorganism studies, cell
biology, and clinical pathology. However, despite the long history
of microscopy and the remarkable range of optical tools that have
been developed since the invention of the first microscope in the
early 1600s, the fundamental design of microscopes has undergone
little change. A typical microscope still consists of an objective,
space for relaying the image, and an eyepiece or an imaging lens to
project a magnified image onto a person’s retina or a camera. In
addition to its relatively high implementation cost (precise and
expensive lenses are needed), the conventional microscope design
has also proven difficult to miniaturize (1, 2). A relatively modern
invention—digital inline holographic microscopy (DIHM) (3)—
showed that it is possible to render microscope-resolution images of
objects without the use of lenses; however, as a method, DIHM
requires significant postmeasurement computation and the use of
a coherent light source. In 2005, Lange
et al.
(4) reported a direct
projection method to implement compact and low-cost imaging
systems. In Lange’s method, the specimen is placed directly on a
CMOS image sensor, and the projection image is then recorded by
the sensor (Fig. 1
A
). The resolution in such a system is given by the
sensor pixel size. Because the typical pixel size of a commercial
CCD or CMOS sensor is
3
m, this approach is incapable of
yielding images that have resolution comparable with conventional
microscope images (resolution of 1
m or better). Despite their low
image qualities, recent works (5) showed that these pixelated images
are useful for certain high-throughput cell-identification
applications.
Our present objective is to determine whether it is possible to
modify the direct projection imaging accordingly so that micro-
scope-resolution images can be collected. We believe that if such
an approach can be found, it can be a viable low-cost and
compact replacement for the conventional microscope system
for a range of applications.
It is difficult to conceive or develop a direct projection imaging
strategy by which single-time-point images at resolution better than
the sensor pixel size can be acquired. However, if we permit
ourselves to exploit the time dimension during the image acquisition
process, it is possible to develop viable high-resolution direct
projection imaging strategies in which resolution and sensor pixel
size are independent. As an example, one can imagine covering a
sensor grid with a thin metal layer and etching a small aperture onto
the layer at the center of each sensor pixel. The sensor pixel will then
be sensitive only to light transmitted through the aperture. By
placing a target specimen on top of the grid, we can then obtain a
sparsely sampled image of the object (Fig. 1
B
). A ‘‘filled-in’’ image
can be generated by raster-scanning the specimen over the grid (or
equivalently, raster-scanning the grid under the specimen) and
compositing the time varying transmissions through the apertures
appropriately (Fig. 1
C
). We can see that in this case, the resolution
is fundamentally determined by the aperture size and not the pixel
size. Therefore, by choosing the appropriate aperture size, we can
achieve high resolution. The imaging strategy can be simplified by
tilting the aperture grid at a small angle (
) with respect to
x
axis
and replacing the raster-scan pattern with a single linear translation
of the specimen across the grid (Fig. 1
D
). As long as a sufficient
number of apertures span across the specimen completely in
y
axis,
and neighboring apertures overlap sufficiently along
y
axis, a
filled-in high-resolution image of the specimen will be achieved.
The design can be further simplified by replacing the tilted 2D
aperture grid with a long tilted 1D aperture array (Fig. 1
E
). This
imaging strategy (6) forms the basis of the optofluidic microscopy
(OFM) method. The OFM method shares a lot of similarities with
near-field scanning optical microscopy methods (7). In fact, the
OFM aperture array can be interpreted as a series of NSOM
apertures. Whereas NSOM sensors are generally raster-scanned
over the target objects, the OFM approach uses object translation
Author contributions: X.C., P.W.S., D.P., and C.Y. designed research; X.C., L.M.L., X.H., and
W.Z. performed research; X.C. and L.M.L. analyzed data; and X.C., L.M.L., X.H., W.Z., P.W.S.,
D.P., and C.Y. wrote the paper.
The authors declare no conflict of interest.
Freely available online through the PNAS open access option.
†
X.C. and L.M.L. contributed equally to this work.
To whom correspondence should be addressed at: M/C 136-93, 1200 East California
Boulevard, Pasadena, CA 91125. E-mail: chyang@caltech.edu.
© 2008 by The National Academy of Sciences of the USA
10670 –10675
PNAS
August 5, 2008
vol. 105
no. 31
www.pnas.org
cgi
doi
10.1073
pnas.0804612105
to accomplish scanning—in the microfluidic system, this is a far
simpler and more efficient strategy.
Here, we report the implementation and application of two
high-resolution, lensless, and fully on-chip microscope systems
based on the OFM method. The first OFM system is customized for
Caenorhabditis elegans
imaging. It utilizes a gravity-driven-flow
mechanism to translate
C. elegans
across the OFM sensing area.
The second system is optimized for cell and other spherical/
ellipsoidal object imaging. It utilizes an electrokinetic drive to
translate samples in the OFM system. This approach minimizes
potential rotations or tumbles of the cells during the scanning
process.
In the next section, we will describe the implementation of the
gravity-driven-flow-based OFM system, provide details about its
characteristics, and report on its use for automated
C. elegans
imaging and phenotype characterization. Next, we will describe
the electrokinetically driven OFM system, and report on its use
for imaging
Chlamydomonas
(single cellular algae), mulberry
pollen spores, and polystyrene spheres. We will then discuss the
OFM contrast mechanism and the resolution characteristics of
OFM systems. Finally, we conclude by briefly discussing the
potential applications of the OFM method.
Results
Gravity-Driven-Flow-Based Optofluidic Microscope System.
Design.
This on-chip OFM system was fabricated on a commercially
available 2D CMOS image sensor (Micron MT9V403C12STM)
with a 9.9-
m pixel size. We planarized the surface of the sensor
witha2-
m-thick SU8 photoresist and coated it with a 300-nm-
thick Al layer. We then milled two lines of apertures (1
m
diameter) separated by a single line of sensor pixels onto the Al
layer with a focused ion beam (FIB) machine (FEI Nova 200).
The apertures were spaced 9.9
m apart so that each aperture
mapped uniquely onto a single sensor pixel (Fig. 2
A
). Each line
consisted of 100 apertures. A 0.2-
m-thick poly(methyl methac-
rylate) (PMMA) layer was spin-coated on top of the Al film to
protect the OFM apertures. Finally, we bonded an optically
transparent poly(dimethylsiloxane) (PDMS) microfluidic chip
containing a channel (width of 50
m, height of 15
m) on top
of the sensor with a Karl Suss mask aligner (MJB3). The channel
was oriented at
0.05 radians with respect to the aperture
arrays. The top of the system was uniformly illuminated with
white light (
20 mW/cm
2
, approximately the intensity of sun-
light) from a halogen lamp.
Each line of apertures represents an OFM scanning array. The
metal layer blocks light from the underlying pixels; light can only be
transmitted through the apertures. The imaging process involves
uniformly flowing the specimen through the channel and recording
the time varying light transmission through each aperture as the
specimen passes. Each time scan represents a line profile across the
specimen. Because the specimen passes the apertures sequentially,
if the speed of the specimen is uniform, there will be a constant time
delay between adjacent line scans. By shifting the line scans with this
delay, we can obtain an accurate projection image of the specimen.
Specifically, unlike the physical sensing grid in a CCD or CMOS
image sensor, the OFM sampling scheme effectively establishes a
virtual sensing grid. The grid density is adjustable by changing the
number of apertures spanning the channel, the flow speed of the
specimen, and the OFM readout rate. Higher pixel density allows
the OFM to oversample the specimen, thereby preventing unde-
sirable aliasing artifacts in the images. Finally, as mentioned pre-
viously, the resolution of such system is fundamentally limited by
the aperture size.
Fig. 1.
Comparison of on-chip imaging schemes. (
A
) Direct projection im-
aging scheme. By placing the specimen directly on top of the sensor grid, we
can obtain a projection image with resolution equal to the sensor pixel size.
(
B
) By placing the specimen on a grid of apertures, we can obtain a sparse
image. However, for the same grid density, the obtained image will not be
much improved over that of
A
.(
C
) By raster-scanning the specimen over the
aperture grid, we can obtain a ‘‘filled-in’’ image. In this case, the image
resolution is limited by the aperture size. Grid density is no longer a factor in
resolution consideration. (
D
) The scanning scheme can be simplified into a
single-pass flow of the specimen across the grid by orientating the grid at a
small angle (
) with respect to the flow direction (
x
axis). (
E
) The aperture grid
can be simplified by substitution with a long linear aperture array. This scheme
is the basis for the optofluidic microscopy method.
Fig. 2.
OFM prototype. (
A
) schematic of the device (top view). The OFM
apertures (white circles) are defined on the Al (gray) coated 2D CMOS image
sensor (light gray dashed grid) and span across the whole microfluidic channel
(blue lines). (
B
) The actual device compared with a U.S. quarter. (
C
) Upright
operation mode. (
D
) Flow diagram of the OFM operation. Two OFM images of
the same
C.elegans
are acquired by the two OFM arrays, respectively (red arrows).
If the image correlation is
50%, the image pair is rejected. Otherwise, the area
and the length of the worms are measured automatically by evaluating the
contour (green dashed line) and the midline (yellow dashed line). (
E
) Cross-
section of the fabrication of an electrokinetically driven OFM device.
Cui
et al.
PNAS
August 5, 2008
vol. 105
no. 31
10671
CELL BIOLOGY
ENGINEERING
This on-chip OFM prototype (Fig. 2
B
and
C
) utilizes two parallel
OFM arrays for two reasons. First, by measuring the time difference
between when the specimen first crosses each array and knowing
the separation between the two arrays along the channel axis
x
,we
can determine the flow speed of the specimen
v
, which is important
for correct OFM image construction. (Note that the flow speed is
determined for each specimen independently. As such, speed
variations between specimens have no impact on our ability to
perform correct OFM image reconstruction.) Second, significant
differences between the two OFM images acquired by the two
OFM arrays for the same specimen will indicate shape changes,
flow speed variations, and/or rotations of the specimen during the
data acquisition. Because accurate OFM imaging requires the
absence of these variations, discrepancy between the images is a
possible criterion for rejecting that image pair (Fig. 2
D
). We set our
rejection criterion at the image-pair correlation threshold of
50%.
During our experiments,
50% of the specimens were rejected
based on this criterion. We note that this processing approach was
highly conservative in that it also rejected a large proportion of
acceptable images in which image-pair correlation was low because
of small variations in flow speed and slight sample shifts. We believe
that better flow controls (such as smoother channels and better
speed tracking) and better image-processing algorithms can signif-
icantly lower the rejection rate. This is an area that is worth further
study.
C. elegans
imaging demonstration.
We demonstrated the proper
functioning of this on-chip microscope system by employing it to
image
C. elegans
larvae. To facilitate efficient flow of the
specimens through the system, we took the following steps in
preparing the microfluidic channel.
The PDMS microfluidic channel was designed with two smooth
funnels at both ends, and oxygen plasma was used to render the
inner surface of the PDMS channel hydrophilic. Before use, we
conducted a surface treatment process (detailed in
Materials and
Methods
) to reduce sample adhesion to the channel walls. We
operated the prototype in the upright mode (Fig. 2
C
), so that
gravity could drive the flow; this eliminated the need for bulky
external pumps. When the specimen solution (newly hatched
C.
elegans
L1 larvae in S-basal buffer,
20 worms per microliter) was
injected into the top funnel, the solution wetted the channel and the
specimens were continuously pulled into the channel by the gravity-
driven microfluidic flow. To prevent the worms from wiggling, we
immobilized them by subjecting them to a 70°C heat bath for 3 min.
Because of the sedimentation of the worms in the solution, the
throughput of OFM imaging was not constant. The maximum
observed throughput was approximately five worms per minute.
However, the flow speed of worms
v
in the channel was fairly
uniform (
500
m/sec). The data readout rate
f
of each OFM array
was 1,000 frames/sec, and imaging of each worm required
2.5 sec.
The spacing of the OFM virtual grid along the
x
and
y
axes are both
0.5
m (less than the 1-
m aperture size).
Fig. 3
A
shows a pair of OFM images acquired by the two OFM
arrays from the same wild-type
C. elegans
. The image correlation
between them was 56%. Consistent internal structures were
found in both OFM images. Fig. 3
B
shows an image collected
from a similar worm that was placed directly onto an unproc-
essed CMOS sensor (note that the pixel size is 9.9
m); the worm
was barely distinguishable in this low-resolution direct projection
image. Fig. 3
C
shows a conventional microscope image of a
similar worm acquired through a
20 Olympus objective lens
(650-nm resolution for 555-nm wavelength under Sparrow’s
criterion). Similar internal structures of
C. elegans
appeared in
both the microscope and the OFM images. This confirmed that
the OFM can render images comparable in quality to those of a
conventional microscope with similar resolution.
Quantitative phenotype characterization of
C. elegans.
The function of
a gene must manifest itself in a certain phenotype to be observed.
However, the formidable number of genes and their combina-
tions imposes a difficult challenge to systematic phenotype
characterization (8, 9). Inexpensive, automated, and quantitative
phenotype characterization devices are critical to comprehensive
biology studies. Motivated by the extensive use of phenotype
characterization, especially morphology, in the genetic studies of
microorganisms and cells, we used the OFM prototype to image
and analyze phenotypes of
C. elegans
.
To compare body sizes of the wild-type (N2),
sma
-
3
(e491),
and
dpy
-
7
(e88)
C. Elegans
, we imaged 25 specimens of each
strain. The
sma
-
3
(e491) gene is part of a family of transforming
growth factor
pathway components (10). The
dpy-7
gene
encodes a cuticular collagen required for proper body form (11).
The typical OFM images of the three strains (Fig. 4
A
–
C
) show
that the
sma
-
3
worm is smaller and thinner than the wild-type
Fig. 3.
Images of wild-type
C. elegans
L1 larvae. (
A
) Duplicate OFM images
acquired by the two OFM arrays for the same
C. elegans
.(
B
) Direct projection
image on a CMOS sensor with 9.9-
m pixel size. (
C
) Conventional microscope
image acquired with a
20 objective.
Sma-3
50μm
Wild-type
50μm
0
50
100
150
200
250
300
Length (μm)
7
8
9
10
11
12
13
Width (μm)
Dpy-7
50μm
DE
C
B
A
Fig. 4.
Phenotype characterization of
C. elegans
L1 larvae. (
A
–
C
) Typical
OFM images of wild-type,
sma
-
3
, and
dpy
-
7
worms, respectively. (
D
and
E
)
The length (
D
) and effective width (
E
)of
wild
-
type
,
sma
-
3
, and
dpy
-
7
worms, respectively (color-coded). The columns represent the mean values
in the population; the hatched areas correspond to the confidence intervals
of the mean values; and the error bars are the standard deviations indi-
cating the variation between individuals in the population. Twenty-five
worms were evaluated for each phenotype.
10672
www.pnas.org
cgi
doi
10.1073
pnas.0804612105
Cui
et al.
worm, and that the
dpy
-
7
worm is fatter and shorter than the
wild-type worm. These observations are consistent with those
made under a conventional microscope.
Because OFM images are naturally digitalized, we can perform
large volume and automatic quantitative information extraction by
computer assisted postprocessing. We developed a MATLAB
program (the algorithm is described in
Materials and Methods
)to
determine the area and length of the worms in batches (Fig. 2
D
).
From those two quantities, we then computed an effective width for
each worm by dividing the area by the length. In Fig. 4
D
and
E
the
columns represent the mean length and width of the three
C. elegans
strains; the hatched areas correspond to the confidence intervals of
our mean length and width estimates. The standard deviations (blue
error bars) of the measurement indicate the variation between
individuals within the strain. The measured mean length and width
were 252.9
3.1
m and 11.7
0.1
m for wild-type, 214.3
2.9
m and 11.5
0.1
mfor
sma
-
3
, and 199.1
4.3
m and 12.1
0.1
mfor
dpy
-
7
. They were consistent with reported data (12). The
three strains have distinct length (
P
0.01 for each pair; Student’s
t
test).
dpy
-7 mutants are significantly wider than N2 and
sma
-
3
(
P
0.05 and
P
0.01, respectively), but we observed no statistically
significant width difference between
sma
-
3
and N2 for the sample
size used.
Electrokinetic-Drive-Based OFM System.
Flow dynamic difference be-
tween pressure and electrokinetic drive.
This next OFM system was
customized for imaging cells and other spherical/ellipsoidal
objects. Pressure-driven liquid flow in microfluidic channel
typically develops a parabolic velocity profile (Poiseuille flow)
due to the nonslip boundary condition on the channel side walls.
An object flowing in the channel will receive a torque from this
nonuniform velocity profile and start to tumble if it is slightly
off-center or if it is nonsymmetric. This nonuniform translational
movement can prevent the OFM system from acquiring an
accurate image of the object.
We prepared the following experiment to observe this effect. A
PDMS microfluidic channel, of dimension 3 mm in length, 40
m
in width, and 13
m in height was bonded with a PMMA-coated
glass slide. The channel was then passivated by using the process
described in
Materials and Methods
. We then injected heat-killed
(70°C water bath for 30 min) Chlamydomonas cells into the channel
by a syringe. A difference in fluid column height between the
channel inlet and outlet induced a pressure differential in the
channel and actuated the flow (13). The parabolic velocity profile
exerted an unsymmetrical distribution of drag force on the objects
flowing along the microfluidic channel and caused the cells to
rotate.
We found that the use of dc electrokinetics provides a simple and
direct way to control the motion of biological cells in the on-chip
OFM system to suppress object rotation. By imposing a uniform
electric field in a PDMS channel (3 mm in length, 40
m in width,
and 13
m in height, bonded with a PMMA-coated glass slide), a
dipole can be induced on target ellipsoidal
Chlamydomonas
cell
(heat-killed by 70°C water bath for 30 min). The dispersed dipole
moment can only be stabilized when its major axis is aligned along
the electric field lines. In other words, the cell will experience an
electroorientation force (14). At the same time, because the cell
Chlamydomonas
cell is likely to carry a net electrical charge, the
external electric field will also exert an electrophoretic force on
the cell.(15). This induces the cell to move along the channel. The
velocity-dependent viscous Stokes drag will eventually match this
force and result in a constant rotation-free translational motion of
the cell. The application of external electric field also causes the
translation of the electric double layer (EDL) at the channel walls;
this phenomenon is known as electroosmosis (16). Under the thin
EDL assumption, the electroosmotic plug-like velocity profile will
exert a symmetrical shear stress distribution on the cells. In steady-
state situations, this movement is also nonrotational. At a constant
voltage of 25 V to a pair of platinum electrodes at the channel inlet
and outlet, we found that an average translational speed of 270
m/sec was achieved for the cells.
In addition, we observed that the cells typically settled into
their steady-state orientation via electroorientation within a
flow distance of 200
m. This information was useful in inform-
ing the OFM system design because it indicated that we need to
allow for an extra flow channel length before the OFM aperture
array for the specimens to reach steady state before the image
acquisition. Finally, we note that we did not encounter any cell
lysis issues over the voltage range (10–65 V for a 3-mm-long
microfluidic channel) we tested.
To compare the performance of pressure drive and electroki-
netic drive for creating rotation-free sample translation, we per-
formed the following experiment. We flowed 100
Chlamydomonas
cells, with size ranged from 8 to 16
m in width, through the channel
by pressure drive and electrokinetic drive. We chose a region of
interest that is 500
m in length and 1.25 mm from the inlet for
observation. This region of interest matches with the location and
length of the OFM array in our second OFM system. A statistical
distribution of cell rotation events in that region observed through
a conventional microscope is recorded. We define a cell as expe-
riencing rotation if it has rotated by
3° during its passage through
the region of interest. In total, 83 cells experienced rotation under
pressure drive whereas only 6 rotating cells were observed under
electrokinetic drive. In fact, these errant movements were mainly
observed after a prolonged experiment period, typically
30 min,
and were likely caused by the presence of debris on the channel
floor after extended channel usage.
Design.
This system was similar to the gravity-driven OFM system
in its basic layout with a couple key differences and a few minor
noncritical design variations (Fig. 2
E
). For brevity, we shall
presently enumerate only the significant differences. This system
was fabricated on a 2D CMOS imaging sensor (Micron
MT9M001C12STM). The CMOS chip comprised of a grid lattice
of 1,280
1,024 square pixels with the pixel size at 5.2
m. We
planarized the sensor surface with PMMA and coated it with a
15-nm-thick chromium seed layer and a 300-nm-thick gold layer.
The aperture array consisted of 120 holes of diameter 0.5
m and
separation 10.4
m. We next deposited a second layer of PMMA
(0.4
m thick) on the gold layer to insulate the metal layer from
the electric field that would be subsequently applied (17). Then,
we emplaced a PDMS chip containing a microfluidic channel of
dimension 2.4 mm in length, 40
m in width, and 13
m in height
onto the chip. Finally, we inserted a pair of platinum electrodes
into the inlet/outlet of the channel.
Chlamydomonas
, mulberry pollen spores, and polystyrene sphere imaging
demonstration.
We imaged three different samples, namely,
Chlamydomonas
cells (8–16
m; Carolina Scientific), mulberry
pollen spores (11
mto16
m; Duke Scientific), and polystyrene
microspheres (10
m; PolyScience) with this system.
Before the experiment, we treated the channel surface using
the procedure detailed in
Materials and Methods
. The
Chlamy-
domonas
cells were heat treated (70°C water bath for 30 min)
before the experiment. The mulberry pollen spores and poly-
styrene microspheres were dissolved in deionized water and
sonicated for 5 min before use.
During imaging acquisitions, the potential difference between
the electrodes was set at
20 V. The typical translation speed of the
objects in the OFM system was
1.5 mm/sec. The data readout rate
was 2,000 frames/sec. For an object of dimension 15
m, this
implied a typical OFM image acquisition time of 0.3 sec. The
higher-than-expected translation speed was attributable to residual
pressure differential in the channel. Nevertheless, the electroori-
entation force was sufficiently strong to suppress sample rotations.
Several OFM images of the three different samples are shown
respectively in Fig. 5
A
–
E
in comparison with images acquired
Cui
et al.
PNAS
August 5, 2008
vol. 105
no. 31
10673
CELL BIOLOGY
ENGINEERING
with an inverted microscope (Olympus IX-71) under a
20
objective in Fig. 5
F
–
J
.
Contrast and Resolution.
OFM contrast mechanism.
The OFM image
contrast shares similar origins with the contrast in conventional
microscopy images. The OFM achieves its highest resolution in the
plane that is just above the aperture array. In effect, the OFM is
similar to a conventional microscope in which the focal plane is
locked at the plane that is just below the target object. The light field
at that plane consists of the combination of the unscattered
component of the illumination and the light fields that are scattered
by scattering sites in the object. The presence of a scattering site
immediately above a specific point in that plane will typically result
in a dark patch in the image as the illumination light is scattered
away by the scatterer. At other locations, the constructive interfer-
ence of scattered light and the illumination field can result in a
higher-than-average light field brightness. The dark boundary in
Fig. 5
D
is attributable to a diminished light field from the presence
of the pollen boundary scattering light away. The bright boundary
is attributable to the constructive interference of that scatter light
component with the illumination at those locations.
However, we note that the above-mentioned similarity between
the OFM and a conventional microscope with a fixed focal plane
only holds if near-field components are insignificant. Because the
OFM samples the wavefront without resorting to propagative
projection, it is also sensitive to near-field light components.
Therefore, it is possible for an OFM system to achieve better
resolution by using smaller apertures.
Object thickness limit.
It is worth noting that, similar to a conventional
transmission microscope, there is, in principle, no upper limit to the
sample thickness that the OFM can process. In practice, the OFM
will fail to acquire an image if the sample is too optically scattering
or absorptive to permit sufficient light to be transmitted through the
OFM apertures. This practical limit exists for the conventional
transmission microscope as well.
OFM resolution.
The resolution characteristics of the OFM are
useful for informing on system design and image interpretation.
We began by characterizing the point spread functions (PSF)
associated with individual OFM apertures. We measured the
PSF of our prototypes by laterally scanning a near-field scanning
optical microscope (NSOM) (Alpha-SNOM; WITec) tip across
the apertures (1 and 0.5
m in diameter) at various heights
H
and
measuring the signal detected by the underlying pixel (Fig. 6
A
).
We approximated the NSOM tip, which was
100 nm in
diameter, as a point source. Fig. 6
B Inset
shows representative
OFM PSF plots at
H
0.1, 1.5, and 2.5
m for the 1-
m-
diameter aperture. The PSF broadened as a function of
H
.We
quantified the height-dependent resolution of our prototype by
the PSF’s width. Fig. 6
B
shows the resolution [Sparrow’s crite-
rion (18)] as a function of
H
. From the plot, we can see that the
ultimate resolution of the gravity-driven-flow-based OFM sys-
tem was 0.9
m (with 0.2-
m-thick PMMA above the metal layer
accounted for) and the resolution degraded to 3
mat
H
2.5
m. The ultimate resolution of the electrokinetic drive based
OFM system was 0.8
m (with the 0.4-
m-thick PMMA above
the metal layer accounted for) and the resolution degraded to 2
mat
H
2.5
m. The result was consistent with our more
detailed study on the light collection characteristics of small
apertures (19).
We note that, given the OFM’s contrast mechanism, a better
approach for resolution characterization will be to translate a
10
μ
μ
m
OFM
conventional
microscope
Chlamydomonas
mulberry
pollen spores
polystyrene
microsphere
AB
C
D
E
F
G
H
I
J
Fig. 5.
Cell and microsphere images. (
A
–
E
) Images taken from the on-chip OFM driven by dc electrokinetics of
Chlamydomonas
(
A
and
B
), mulberry pollen (
C
and
D
), and a 10-
m polystyrene microsphere. (
F
–
J
) Images taken from a conventional light transmission microscope with a
20 objective of
Chlamydomonas
(
F
and
G
), mulberry pollen (
H
and
I
), and a 10-
m polystyrene microsphere (
J
). (Scale bars: 10
m.)
4
μ
m)
0.5
1
smission (a.u.)
r
2
3
n of OFM (
μ
-5
0
5
0
Tran s
r (
μ
m)
H
Point source
Z
1μm
0.5μm
1
Resolution
Al
X
Y
SU8
1
2
3
4
0
H (
μ
m)
R
50μm
CMOS pixel
50
0
B)
0
B)
μm
-10
0
plitude (d
-3dB
-10
0
plitude (d
-3dB
0
0.25
0.5
0.75
1
-20
f
r
(1/
μ
m)
Amp
0
0.25
0.5
0.75
1
-20
f
r
(1/
μ
m)
Amp
B
A
E
F
CD
Fig. 6.
Resolution of the OFM prototype. (
A
) Schematic of the PSF measure-
ment. (
B
) Resolution of the prototype at various heights
H
above a 0.5- and a
1-
m-diameter aperture under Sparrow’s criterion. (
Inset
) Representative
OFM PSF plots at
H
0.1, 1.5, and 2.5
m for the 1-
m-diameter aperture. (
C
and
D
) OFM images of wild-type
C. elegans
L1 larvae in a 15- and 25-
m-tall
microfluidic channels, respectively. (
E
and
F
) Radial frequency spectra of OFM
images in
C
and
D
, respectively. The
3-dB bandwidths (dashed red lines) are
at 0.62 and 0.38
m
1
, respectively.
10674
www.pnas.org
cgi
doi
10.1073
pnas.0804612105
Cui
et al.
point scatterer across the aperture under a uniform illumination
field and measure the light collected by the aperture. However,
we further note that the point source and point scatterer
configurations are optically similar in the context of resolution
considerations. Under Sparrow’s criterion (18) and in the small
scatterer limit, the point source resolution computation is di-
rectly translatable for point scatter consideration.
We also verified the degradation of the OFM resolution with
height through a
C. elegans
imaging experiment in which we varied
the channel height. Fig. 6
C
and
D
shows OFM images of wild-type
C. elegans
in 15- and 25-
m-tall channels, respectively. A shallow
channel was able to better confine the specimen close to the
aperture array and thus was able to provide better resolved images
(Fig. 6
C
). Fig. 6
E
and
F
shows the radial frequency spectrums of
the OFM images, which revealed that the
3-dB bandwidths were
at 0.62 and 0.38
m
1
, respectively, for the 15- and 25-
m channels.
Discussion
The application of OFM for cell imaging is a particularly
promising area. As an automatic on-chip cell microscopy
method, OFM can potentially be used in applications such as
blood fraction analysis (20), urine screening for infection (21,
22), stem cell screening and sorting (23, 24), tumor cell counting
(25, 26), and drug screening (27).
The compact, simple, and lensless OFM can significantly
benefit a broad spectrum of biomedicine applications and bio-
science researches, and also change the ways we conduct certain
experiments. For example, the availability of tens or even
hundreds of microscopes on a chip can allow automated and
massively parallel imaging of large populations of cells or
microorganisms. An on-chip microscope system can also poten-
tially provide low cross-contamination risk (by being cost-
effective enough to be disposable) point-of-care analysis in the
clinical settings. In a Third World environment, a complete,
low-cost, and compact microscope system suitable for malaria
diagnosis can be a boon for a health worker who needs to travel
from village to village.
Materials and Methods
Culture of
C. elegans
for Imaging.
The alleles used were
dpy
-
7(e88)
,
sma
-
3(e491)
, and wild-type (N2).
C. elegans
were maintained and handled as
described in ref. 28. Briefly, all strains were cultured at 20°C. Bleaching was
used to synchronize the development of
C. elegans
L1 larvae.
PEG Grafting Process to Promote the Flow of Samples in OFM Imaging.
The
microfluidic channel was filled up and flushed with a 10% poly(ethylene
glycol) (PEG) solution, 0.5 mM NAIO
4
, and 0.5% (by weight) benzyl alcohol.
Under the activation of UV light for 1 h, the channel surface was conjugated
with the PEG molecules. The process is similar to the one described in ref. 29.
The PEG grafted surface prevented nonspecific adsorption with biological
entities and lubricated the object flow. The chip can be rinsed with deionized
water, dried, and stored under ambient condition because the PEG grafted
surface has a long-term stability.
The Algorithm for Automatic Determination of
C. elegans
Length and Width.
(
i
) Delineate the boundary around each
C. elegans
from the OFM images. (
ii
)
Calculate the area occupied by each
C. elegans
based on the boundary. (
iii
)
Segment the
C.elegans
image along its length and calculate the centroid for each
segment. (
iv
) Connect the centroids by a continuous line. The length of the
C.
elegans
is given by the length of this line. (
v
) The width of the
C. elegans
is
calculated by dividing the area occupied by the nematode with its length.
ACKNOWLEDGMENTS.
We are grateful for the constructive discussions with
and the generous help from Professor Axel Scherer, Jigang Wu, Dr. Zahid
Yaqoob, Dr. Claudiu Giurumescu, Oren Schaedel and Dr. Xiaoyan Robert Bao.
We appreciate the assistance of the Caltech Watson clean-room. This work is
funded by DARPA’s Center of Optofluidic Integration, the Wallace Coulter
Foundation, National Science Foundation Career Award BES-0547657, and
National Institutes of Health Grant R21 PA03-107. L.M.L. thanks the Croucher
Scholarship for financial support. P.W.S. is an Investigator of the Howard
Hughes Medical Institute.
1. Whitesides GM (2006) The origins and the future of microfluidics.
Nature
442:368 –373.
2. El-Ali J, Sorger PK, Jensen KF (2006) Cells on chips.
Nature
442:403– 411.
3. Garcia-Sucerquia J,
et al.
(2006) Digital in-line holographic microscopy.
Appl Opt
45:836 – 850.
4. Lange D, Storment CW, Conley CA, Kovacs GTA (2005) A microfluidic shadow
imaging system for the study of the nematode Caenorhabditis elegans in space.
Sensors Actuators B
107:904 –914.
5. Ozcan A, Demirci U (2008) Ultra wide-field lens-free monitoring of cells on-chip.
Lab Chip
8:98 –106.
6. Heng X,
et al.
(2006) Optofluidic microscopy—A method for implementing a
high resolution optical microscope on a chip.
Lab Chip
6:1274 –1276.
7. Courjon D (2003)
Near-Field Microscopy and Near-Field Optics
(Imperial College
Press, London).
8. Bochner BR (2003) New technologies to assess genotype-phenotype relation-
ships.
Nat Rev Genet
4:309 –314.
9. Zhong WW, Sternberg PW (2006) Genome-wide prediction of C-elegans genetic
interactions.
Science
311:1481–1484.
10. Savage C,
et al.
(1996) Caenorhabditis elegans genes sma2, sma-3, and sma-4
define a conserved family of transforming growth factor beta pathway com-
ponents.
Proc Natl Acad Sci USA
93:790 –794.
11. Gilleard JS, Barry JD, Johnstone IL (1997) cis regulatory requirements for hypo-
dermal cell-specific expression of the Caenorhabditis elegans cuticle collagen
gene dpy-7.
Mol Cell Biol
17:2301–2311.
12. Savage-Dunn C,
et al.
(2000) SMA-3 Smad has specific and critical functions in
DBL-1/SMA-6 TGF
-related signaling.
Dev Biol
223:70 –76.
13. Beebe DJ, Mensing GA, Walker GM (2002) Physics and applications of microflu-
idics in biology.
Annu Rev Biomed Eng
4:261–286.
14. Hughes MP (2003)
Nanoelectromechanics in Engineering and Biology
(CRC
Press, Boca Raton, FL), 1st Ed.
15. Grossman PD, Colburn JC (1992)
Capilliary Electrophoresis
(Academic, London),
1st Ed.
16. Probstein RF (2003)
Physicochemical Hydrodynamics
(Wiley InterScience, New
York), 2nd Ed.
17. Park JH, Hwang DK, Lee J, Im S, Kim E (2007) Studies on poly(methyl methacry-
late) dielectric layer for field effect transistor: Influence of polymer tacticity.
Thin Solid Films
515:4041– 4044.
18. Jones AW, Bland-Hawthorn J (1995) Towards a general definition for spectro-
scopic resolution.
Astronomical Data Analysis Software and Systems IV
, Astro-
nomical Society of the Pacific Conference Series (Astronomical Society of the
Pacific, Provo, UT), Vol 77, pp 503–507.
19. Heng X,
et al.
(2006) Characterization of light collection through a subwave-
length aperture from a point source.
Opt Express
14:10410 –10425.
20. Cui W,
et al.
(2003) Expression of lymphocytes and lymphocyte subsets in
patients with severe acute respiratory syndrome.
Clin Infect Dis
37:857– 859.
21. Marrazzo JM,
et al.
(1997) Community-based urine screening for
Chlamydia trachomatis with a ligase chain reaction assay.
Ann Intern Med
127:796 – 803.
22. Carroll KC,
et al.
(1994) Laboratory evaluation of urinary-tract infections in an
ambulatory clinic.
Am J Clinical Pathol
101:100 –103.
23. McCowage GB,
et al.
(1998) Multiparameter-fluorescence activated cell sorting
analysis of retroviral vector gene transfer into primitive umbilical cord blood
cells.
Exp Hematol
26:288 –298.
24. Buchstaller J,
et al.
(2004) Efficient isolation and gene expression profiling of
small numbers of neural crest stem cells and developing Schwann cells.
J Neu-
rosci
24:2357–2365.
25. Fujiwara T,
et al.
(1994) Therapeutic effect of a retroviral wild-type P53 expres-
sion vector in an orthotopic lung-cancer model.
J Natl Cancer Inst
86:1458 –1462.
26. Fox SB,
et al.
(1993) Relationship of endothelial-cell proliferation to tumor
vascularity in human breast-cancer.
Cancer Res
53:4161– 4163.
27. Perlman ZE,
et al.
(2004) Multidimensional drug profiling by automated micros-
copy.
Science
306:1194 –1198.
28. Sulston JE, Hodgkin JA (1988)
The Nematode Caenorhabditis elegans
(Cold
Spring Harbor Lab Press, Cold spring Harbor, NY).
29. Hu S,
et al.
(2003) Cross-linked coatings for electrophoretic separations in
poly(dimethylsiloxane) microchannels.
Electrophoresis
24:3679 –3688.
Cui
et al.
PNAS
August 5, 2008
vol. 105
no. 31
10675
CELL BIOLOGY
ENGINEERING