We find that several key endogenous protein structures
give rise to intense second-harmonic generation (SHG)
nonabsorptive frequency doubling of an excitation laser line. Second-harmonic imaging
microscopy (SHIM) on a laser-scanning system proves, therefore, to be a
powerful and unique tool for high-resolution, high-contrast, three-dimensional studies of live cell and tissue architecture. Unlike
fluorescence, SHG suffers no inherent photobleaching or toxicity and
does not require exogenous labels. Unlike polarization microscopy, SHIM
provides intrinsic confocality and deep sectioning in complex tissues.
In this study, we demonstrate the clarity of SHIM optical sectioning
within unfixed, unstained thick specimens. SHIM and two-photon excited
fluorescence (TPEF) were combined in a dual-mode nonlinear microscopy
to elucidate the molecular sources of SHG in live cells and tissues.
SHG arose not only from coiled-coil complexes within connective tissues
and muscle thick filaments, but also from microtubule arrays within
interphase and mitotic cells. Both polarization dependence and a local
symmetry cancellation effect of SHG allowed the signal from species
generating the second harmonic to be decoded, by ratiometric
correlation with TPEF, to yield information on local structure below
optical resolution. The physical origin of SHG within these tissues is addressed and is attributed to the laser interaction with dipolar protein structures that is enhanced by the intrinsic chirality of the
protein helices.
 |
INTRODUCTION |
Second-harmonic generation (SHG) is beginning to
emerge as a powerful contrast mechanism in nonlinear optical
microscopy. SHG was first demonstrated by Kleinman in crystalline
quartz in 1962 (Kleinman, 1962
), and, since that time, has been
commonly used to frequency double pulsed lasers to obtain shorter
wavelengths. Shortly thereafter, SHG from interfaces was discovered by
Bloembergen in 1968 (Bloembergen et al., 1968), and, since, has become
a standard spectroscopic tool for characterizing surfaces and probing
dynamics at interfaces (for reviews, see Shen, 1989; Eisenthal, 1996). In 1974, Hellwarth first integrated SHG into an optical microscope to
visualize the microscopic crystal structure in polycrystalline ZnSe
(Hellwarth and Christensen, 1974). This concept was also demonstrated
by Sheppard in 1977 (Sheppard et al., 1977) and again more recently
with modern imaging equipment and laser sources (Gauderon et al.,
1998). In this paper, we describe the use of SHG to obtain
high-resolution three-dimensional (3D) images of endogenous arrays of
collagen, acto-myosin, and tubulin in a wide variety of species and
cell and tissue types.
SHG is a second-order nonlinear optical process that has symmetry
constraints confining signal to regions lacking a center of symmetry.
Membranes lack such inversion symmetry, and these interfaces can
therefore be imaged with great specificity by SHG. In prior work, we
showed the utility of second-harmonic generation imaging microscopy
(SHIM) in imaging the plasma membranes of several tissue-culture cell
lines (Ben-Oren et al., 1996; Peleg et al., 1999; Campagnola et al.,
1999). We pointed out that, because SHG is a nonlinear optical process,
this form of excitation retains the benefits of two-photon excitation
fluorescence (TPEF) microscopy: due to the peak power requirements for
these processes, sufficient power density in a microscope only occurs
at the focal point. This results in intrinsic 3D sectioning without the
use of a confocal aperture (Denk et al., 1990), and, as a consequence,
out-of-plane photobleaching and photo-toxicity are greatly reduced. In
this prior work, the cells were stained with a lipophilic dye to
generate sufficient contrast. Although out-of-plane photobleaching and photo-toxic effects are reduced in nonlinear optical excitation schemes, in-plane damage still results, largely from formation of
singlet oxygen free radicals upon photo-bleaching of fluorescent dyes.
Thus, it would be desirable to obtain 3D, high-resolution (near
diffraction limited) images at high contrast without relying upon fluorescence.
In a series of rigorous experiments, Freund et al. (1986) used SHG
microscopy in 1986 to study the endogenous collagen structure in a rat
tail tendon at ~50-µm resolution. More recently, in a reflection
mode setup, Alfano and coworkers used stage-scanning laser excitation
to image SHG within muscle and connective tissue, where frame rates of
several hours were required (Guo et al., 1997, 1999). Here, we present
work that uses a laser-scanning transmission-mode microscope and
extends this imaging concept to higher resolution (~1 µm) and much
higher rates of image acquisition (1 frame per second). We show that
bright, high-resolution 3D SHG images of structural proteins within
connective tissue, muscle, and mitotic spindles can be obtained at
confocal-like frame rates.
Several aspects make this form of microscopy very powerful. Because the
excitation uses near-infrared wavelengths, this method is well suited
for studying intact tissue samples because excellent depths of
penetration can be obtained. For example, we have acquired optical
sections throughout 550 µm of mouse muscle tissue. Information about
the organization of protein matrices at the molecular level can be
extracted from SHG imaging data in several ways. Because the SHG
signals have well-defined polarizations, SHG polarization anisotropy
can be used to determine the absolute orientation and degree of
organization of proteins in tissues. In addition, two-photon-excited fluorescence images can be collected in a separate data channel simultaneously with SHG. Correlation between the SHG and TPEF images
provides the basis not only for molecular identification of the SHG
source but also for probing the radial and lateral symmetry within
structures of interest. Based on the physical nature of SHG contrast,
we suggest that the SHG signals arise from dipolar interactions that
are enhanced by the intrinsic chirality in higher-order protein
helices, making SHIM quite distinct from other optical-imaging techniques.
| |
MATERIALS AND METHODS |
Microscopy equipment and physical measurements
A detailed description of the SHG/TPEF microscope has been given
previously (Campagnola et al., 1999) and only a brief outline will be
provided here. The SHG imaging experiments were performed on modified
Biorad MRC600 confocal scan head mounted on a Zeiss upright microscope.
The laser system is a Coherent argon ion (Innova 310, Coherent, Santa
Clara, CA) pumped femtosecond titanium sapphire oscillator
(900-F, Coherent), characterized by pulse width of approximately100 fs at 76 MHz repetition rate at 850 nm. Average powers
at the sample were between 1 and 50 mW. Because SHG is a coherent
process, the signal wave copropagates with the laser and is collected
in a transmitted light configuration. A long working distance 40 × 0.8 N.A. water immersion lens (Zeiss, Zeiss, Jena, Germany)
and a 0.9 N.A. condenser (Zeiss) are used for excitation and signal
collection, respectively. The 1 mrad divergence of the ti:sapphire
laser was compensated before the scan head. There is essentially no
dispersion at 850 nm and thus no external precompensation was used to
compensate for the minimal group delay in the scan head or objective.
When applicable, the TPEF signal is descanned and collected with the
pinhole aperture fully opened and detected simultaneously with the SHG.
The SHG signal is first reflected with a 425-nm hard reflector
(bandwidth ± 25 nm), then isolated from the laser fundamental and
any fluorescence by 1 or 2 mm color glass (BG-39, CVI, Albuquerque,
NM) and a short-wave pass filters (450 nm), respectively, and
detected by a photon counting photomultiplier module (Hamamatsu 7421, Hammamatsu, Bridgewater, NJ). The TTL pulses from this module
are integrated by the Biorad acquisition electronics. Data acquisition
times were between 1 and 4 s per 768 × 512 frame, and images
were acquired either as single frame, or as a result of 3 Kalman
averages. The microscope is equipped with a fiber-based spectrometer to
also collect spectra (Ocean Optics, USB2000, Ocean Optics, Dunelin,
FL). Spectra are collected in the following way: A low-zoom SHG
image is acquired, and then zoomed in 100-fold, corresponding to an
area of 6 µm2. The signal is then directed onto
the fiber and the spectrum is acquired, with integration times of 25 ms
to 2.5 s, depending on the sample.
SHG polarization anisotropy measurements were made with a Glan
Laser Polarizer (GLP, CVI), where the data were obtained
by maintaining the same input laser polarization and obtaining images with the GLP oriented both parallel and perpendicular to the laser fundamental. The intensities of the resulting images were integrated using Imagequant (Molecular Dynamics, Sunnyvale, CA).
Specimens and sample preparation
Scales plucked from live black tetra fish
(Gymnocorymbus ternetzi) were provided by Vladimir Rodionov
(University of Connecticut Health Center).
Mouse tissues were dissected from freshly sacrificed animals (provided
by Ivo Kalajzic, University of Connecticut Health Center), mounted in 3% agarose on a glass slide under a 1.5-thickness
coverslip, and imaged within 1 h of sample preparation.
Caenorhabditis elegans worms were mounted live on a glass
slide upon a pad of 3% agarose, and were sealed under a 1.5-thickness
coverslip using a fillet of melting-point temperature bath oil (Sigma,
St. Louis, MO). N2 (Bristol) was used as a wild type strain.
C. elegans strain RW1596 expressing GFP::MHC A
was constructed as follows. A green fluorescent protein (GFP) tag was
inserted at the initiator methionine in a cloned copy of the
myo-3 gene, resulting in deletion of two amino acids from
myosin heavy chain A (MHC A). The coding sequence for GFP was
amplified from PD95.69 (A. Fire, S. Xu, J. Ahnn, G. Seydoux, personal
communication) using primers (written 5' to 3') L2617
TAGATCCATCTAGAAATGAGTAAAGGAGAAG and M3381
TTCGAATGCGTCTGGATTTTTGTATAGTTCATCCAT, where italicized bases
represent myo-3 sequence. The PCR product was digested with XbaI and BsmI, and cloned into digested pPH23aP1. The
full-length MHC A construct was generated using sequences from pJK26a
as described (Hoppe and Waterston, 1996). Transgenic lines were
generated by injecting wild-type N2 worms with a 200 ng/µl
DNA solution containing 1% myosin construct DNA from two independent
clones, and 99% the rol-6 co-injection marker pRF4 (Mello
et al., 1991). The extra-chromosomal array stEx30 was
crossed into the null mutant myo-3(st386) to make strain
RW1596. The array has sufficient rescuing activity to maintain the null
mutant line and displays normal GFP::MHC A localization in
the adult, although expression of the transgene causes some frequency
of embryonic lethality (P. E. Hoppe and R. H. Waterston,
unpublished observations). Strain FC34 was created by crossing the
array stEx30 from RW1596 into the MHC B null mutant unc-54(e190). The preparation of C. elegans
strain WH204 expressing 
-tubulin::GFP was previously described
(Strome et al., 2001).
Image processing
Images and image stacks acquired in BioRad .PIC format
were imported into NIH Image v1.62 (developed by Wayne Rasband and the
National Institutes of Health and available at
http://rsb.info.nih.gov/NIH-image/). Contrast enhancement,
maximum-point projections, and reslicing were performed using
existing functions of NIH Image. Dual-channel overlays and final
figures were created and edited within Adobe Photoshop v5.0.
| |
RESULTS |
Biological observations of SHG within tissues and cells
Connective tissue
Prior imaging and spectroscopic work by Freund et al. (1986) and
Kim et al. (1999, 2000), respectively, showed that collagen possesses a
large second-order nonlinear susceptibility. This observation suggested
the possibility of using SHG to image connective tissue at high
resolution. Figure 1 shows a single-frame
SHG image of a fish scale where the size scale is 330 × 220 µm
and acquisition time was 1 s. The fish scale is composed in large
part of dense collagen fibrils, and is expected to yield a bright
signal. Indeed the laser was attenuated to ~1 mW average power to
avoid saturation of the integration electronics. By contrast,
essentially no autofluorescence was observed at the incident intensity
(data not shown). These SHG signals from connective tissues are the
brightest that we have observed from any sample, far exceeding those
obtained from stained lipid membranes, and, in fact, are sufficiently
bright to see by eye if projected onto a white card.
View larger version (166K):
[in this window]
[in a new window]
|
FIGURE 1
SHG optical section from an isolated scale of black
tetra fish, G. ternetzi. This specimen was used for most
quantitative and spectroscopic studies of optical response. Laser power
was attenuated substantially to avoid saturation of the detector. Scale
bar = 50 µm.
|
|
The SHG and simultaneously acquired two-photon excited autofluorescence
from an intact region of a mouse ear containing hair follicles are
shown in Fig. 2,
A and B, respectively. A two-color overlay of the
two channels is shown in Fig. 2 C, where the SHG and TPEF
channels are violet and green, respectively. The SHG (A)
arises from the dermal collagen matrix, whereas the TPE
autofluorescence (B) is detected both in the hair shaft and
in cells of the epidermis (regular polygons). These images provide
totally separate but complimentary information. Although keratin is
well known to be strongly autofluorescent, collagen produces
essentially no TPEF signal. By contrast, collagen is an efficient SHG
source, whereas the keratin in the hair shafts and epidermis are
invisible by SHG. Figure 2 D is an xz section
through a hair follicle, further demonstrating the spatial separation
of the SHG and TPEF from these two protein species.
View larger version (34K):
[in this window]
[in a new window]
|
FIGURE 2
Dual-mode 3D imaging of endogenous SHG and fluorescence
in tissue layers of a mouse ear. A full-thickness, freshly dissected
sample of mouse ear was imaged simultaneously by SHG and TPEF.Stacks of optical sections were recorded for each channel.
(A) SHG signal from deep, collagen-rich layer within
dermis. Note the complete absence of signal from the central hair
follicle, in contrast with the dense arrangement of fibrils in the bulk
of the deep tissue. Extremely high absorbance by the overreaching
hairshafts (see B) creates a shadowing of SHG signal in
the upper right and left corners of the frame. (B) TPEF
autofluorescence signal from keratin-rich hair shafts and epidermis.
Note the continuous layer of epidermal cells closely surrounding the
central hair shaft. (C) Color overlay of both SHG
(violet) and TPEF (green) channels,
allowing spatial correlation of the complementary signals.
(D) xz section through hair follicle in
center of panel C. In both C and
D, it is apparent that the keratin-rich epidermal layer
completely encloses the region above the collagen-negative hair
follicle. Scale bar = 50 µm.
|
|
Muscle tissue
The low-resolution observations of Guo et al. (1997) indicated
that a significant source of SHG lies within striated muscle. We
examined mouse skeletal muscle to ask more precisely which structures
were responsible for SHG emission. A remarkable attribute of SHG
imaging in muscle tissue was the intrinsic deep optical sectioning.
Because SHG is a nonlinear form of excitation, signal arises only from
the point of focus. Furthermore, excitation in the near infrared
suffers less scattering than visible light and results in improved
resolution deep within dense tissue. We found that detailed,
high-contrast features could be resolved in SHIM optical sections
throughout the full ~550-µm thickness of a freshly dissected,
unfixed sample of mouse lower leg muscle (Fig.
3 A). Figure 3 B shows an optical section at a depth 14 µm into
the sample, revealing intense SHG from a matrix of collagen fibrils (black arrow) in the epimysium layer surrounding muscle
fibers. In the same image, contractile muscle cells grazed by this
optical section produced bright SHG in a pattern clearly related to the period of sarcomeric repeats in the myofilament lattice (white arrowhead). A second optical section at a depth of 24 µm (Fig. 3 C) shows the sarcomere band pattern of SHG extending
continuously across the full width of muscle fibers. Other fine details
of the internal tissue structure are also visible. A breach in the band
pattern (white arrow) indicates a tear in the myofilament lattice. A thin layer of extracellular matrix (perimysium) separating fiber groups within the muscle appears in cross-section (white arrowhead). A third optical section at a depth of 250 µm (Fig. 3 D) showed the orientation and packing of a more proximal
(right side of image) and a more distal (left
side) muscle body, separated by a thick layer of matrix
(white arrow). Although SHG from individual collagen bundles
was substantially brighter than myofilament SHG, signals were similar
enough for resolution of both types of structures in the same image
scan. Furthermore, because collagen-based features were much rarer than
the ubiquitous sarcomeric structure, the majority of the SHG produced
by muscle tissue arises from the myofilament lattice.
View larger version (52K):
[in this window]
[in a new window]
|
FIGURE 3
SHG imaging of extracellular and intracellular
structures within native muscle tissue. A dissected sample of mouse
lower leg muscle was optically sectioned through its full thickness of
~550 µm. (A). Graphic representation of the imaged
volume of tissue. xy (top),
yz (left), and xz
(right) sections from the boundaries of the volume are
shown. Dimensions are indicated, and are not portrayed proportionally;
note that the z axis is particularly compressed in this
model. (B) Extracellular matrix (epimysium) surrounding
muscle fibers. An optical section at a depth 14 µm into the sample
shows a web of presumptive collagen fibrils (black
arrow) in the layer above muscle fibers. Note the muscle
sarcomere pattern(white arrowhead) obtained by grazing of this
optical section just beneath the surface of some fibers. Relative
signal intensities from collagen and actomyosin can be compared in
(B, C, D).
(C) Myofilament lattice structure and interfiber matrix
arrangement within muscle tissue. An optical section at a depth of 24 µm into the sample shows continuous repetitive sarcomeric pattern of
SHG across the full width of muscle fibers. A breach in continuity of
the pattern (white arrow) most likely resulted from
tensile stress imposed during sample preparation. A thin layer of
extracellular matrix (perimysium) separating fiber groups within the
muscle is seen in cross-section (white arrowhead).
(D). Deep section within muscle tissue. An optical
section at a depth of 250 µm into the sample again shows the
sarcomeric pattern and extracellular matrix. A transition between two
distinct muscles is apparent in the different orientations of more
proximal (right side of image) and more distal
(left side) obliquely sectioned fibers separated by a
thick layer of matrix (white arrow). Scale bar = 50 µm in (B, C, D).
|
|
To characterize the physical nature and molecular source of
muscle-derived SHG more specifically, we then imaged muscle within the
optically compliant and genetically tractable model organism C. elegans. Polarization microscopy has often been used to observe the birefringence of ordered structural proteins, including those of
contractile muscle (Inoue, 1986; Waterston, 1988). To examine the
relationship of birefringence to SHG, we compared images of C. elegans muscle produced by a polarization microscope to optical sections produced by SHIM (Fig. 4,
A and B, respectively). Although somewhat
comparable, the polarization and SHG microscopes do not produce exactly
the same patterns of contrast. Both modes reveal bright bands with dim
central stripes. Yet the SHG bright bands appear relatively broader
than the anisotropic A bands seen by polarization microscopy. The dark
region between bright SHG bands (arrow) is devoid of bright
nodular signals corresponding to dense bodies in the isotropic band of
the polarization image. Thus, there is no simple one-to-one
correspondence between SHG and polarization images, although the same
structures (in this case, thick filaments) often generate both effects.
This may be because the contrast in a polarization scope image arises
from linear birefringence and is only seen between orthogonal
polarizers. By contrast, SHG signals can be seen with various
combinations of input and output polarizations, and indeed, these data
can be used to extract the molecular orientation (Shen, 1989).
Furthermore, SHG is a second-order phenomenon and thus depends on the
square of the molecular concentration, rather than the linear process
observed in a polarization microscope.
View larger version (142K):
[in this window]
[in a new window]
|
FIGURE 4
Comparison of polarization and SHG microscopy in
nematode muscle. Body wall muscle of live C. elegans was
imaged by both methods and is shown at identical magnification in
(A) and (B). Double arrows indicate
spacing between centers of adjacent sarcomere A bands in both panels.
(A) Polarization image. Bright A bands are characterized
by a dim central stripe (arrowhead). The dark I-band
region is punctuated by birefringent dense bodies
(arrow). (B). SHG image. Bright bands
show a pronounced dark central stripe (arrowhead). The
bright bands appear relatively broader than A bands seen by
polarization microscopy. The dark region between bright bands
(arrow) is devoid of nodular signal corresponding to
dense bodies. Note that the SHG image was acquired in a region
immediately adjacent to the pharygeal muscle, visible in grazing
section as a dim radial array in the bottom of (B).
Acquisition of similar detail in this region by polarization microscopy
would be impossible because of intense out-of-focus glare from the
underlying pharynx. Scale bar = 10 µm.
|
|
The benefits of SHIM optical sectioning are apparent even in relatively
thin (~100 µm) C. elegans. The SHG image in Fig.
4 B was acquired in a region overlying and immediately
adjacent to the pharyngeal muscle, visible in grazing section as a dim radial array at the bottom. Acquisition of similar detail in this region by conventional polarization microscopy would be difficult or
impossible because of intense out-of-focus glare from the underlying pharynx.
To explore the molecular source of SHG within the myofilament lattice,
we combined SHG/TPEF to observe muscle in a C. elegans strain expressing GFP::MHC A, the minor MHC isoform known to
be restricted to the longitudinal mid-zone of thick filaments
(Waterston, 1988; Miller et al., 1983). Figure
5, A and B, show
simultaneously acquired SHIM and TPEF images, respectively. High
magnification reveals that the signals are spatially distinct and
complementary, as seen in the two color overlay in Fig.
5 C. Fluorescence from GFP::MHC A is observed in
the dim middle portion of the SHG bright band, whereas SHG is strongest
in regions not containing GFP::MHC A.
View larger version (58K):
[in this window]
[in a new window]
|
FIGURE 5
Correlative structural localization of the SHG
source in nematode myofilament lattice. (A) Endogenous
SHG and (B) TPEF fluorescence from GFP::MHC A
in C. elegans body wall muscle are shown separately and
(C) in color overlay (SHG = violet,
TPEF = green). A central green/white stripe in the
center of each violet band in (C) indicates that
GFP::MHC A is localized at the center of the SHG-bright
band. Scale bar = 10 µm.
|
|
The dim stripe down the center of the bright bands seen in SHG may
arise from several causes. First, there is a nonuniform distribution of
MHC isoforms along thick filaments in C. elegans. The other
major isoform in C. elegans body wall muscle, MHC B, is
known to be restricted to distal portions of thick filaments (Miller et
al. 1983; Waterston, 1988). To test whether this isoform was a major
contributor to SHG from thick filaments, we imaged a C. elegans strain homozygous for a null mutant allele of the gene
encoding MHC B, unc-54(e190), that disrupts the sarcomeres. The data are shown in Fig. 6,
A, B, and C for the SHG, TPEF, and two-color overlay, respectively. Loss of MHC B resulted in substantial diminution of SHG compared to the wild type, indeed; the laser power
needed to be increased to obtain these data. Although irregularly patterned, the source of SHG is still largely spatially distinct from
GFP::MHC A. Thus, MHC B contributes significantly to the normal signal, but may do so in combination with both MHC A and paramyosin, the other major coiled-coil component of thick filament outer arms (Epstein et al., 1985; Waterston, 1988).
View larger version (55K):
[in this window]
[in a new window]
|
FIGURE 6
Characterization of abnormal myofilament lattice in
myosin mutant C. elegans. (A) Endogenous
SHG and (B) TPEF fluorescence from GFP::MHC A
in body wall muscle of an unc-54 homozygous mutant
C. elegans are shown separately and (C)
in color overlay (SHG = violet, TPEF = green). Although both signals are structurally
associated, the lattice has lost the characteristic order seen in Fig.
5. Note that the SHG signal in this figure required
substantially higher laser power than that from the normal muscle shown
in Fig. 5 A. Scale bar = 10 µm.
|
|
Other factors may also explain the dim central stripe. Only the distal
portions of the thick filament that comingle and interact directly with
actin-based thin filaments may induce SHG. Alternatively, a
symmetry-induced cancellation effect (as will also be seen for tubulin
structures, below) may arise at the center of the thick filament, where
proximal portions of oppositely oriented filament arms are close enough
to break down the conditions of local asymmetry required for SHG. This
is because SHG is a coherent process, and signals will only arise from
objects separated at distances on the order of the optical coherence
length, Lc, or larger,
|
(1)
|
where 
k is the difference in wave vectors between
the fundamental and second harmonic waves. This coherence length is
material independent and is used to relate the size scale or distance
separation over which objects (e.g., beads) that, on a macroscopic
scale, are centro-symmetric (or possess inversion symmetry) can produce
SHG due to local asymmetries (e.g., each side of a bead). Due to the
use of visible or near-infrared laser wavelengths, this effect occurs
on approximately the micrometer and submicrometer size scale. For SHG
arising from the electric dipole interaction, the lower limit for
producing SHG empirically appears to be about 
/10 (note that
electric quadrupole interactions can produce SHG at smaller particle
sizes but will not be important for biological samples). At smaller
distances, the SHG signals undergo complete destructive interference.
This size constraint has recently been well documented in the SHG
literature for several samples, including oil droplets, beads,
liposomes, cell membranes, and metallic spheres. (Hua and
Gersten, 1986; Ostling et al., 1993; Yan et al., 1998; Vance et
al.; 1998; Dadap et al., 1999; Campagnola et al., 1999; Moreuax et al.,
2000). It should be noted that there are actually two very different
operative coherence lengths governing SHG. The other is the material
coherence length, which characterizes bulk anisotropies, and, for
tissues, is on the order of 20 µm (Kim et al., 1999). For
bulk, anisotropic materials, SHG arises from constructive interference
within this length and drops off for larger lengths. Note that the
maximum material coherence length will not be reached on a
laser-scanning nonlinear microscope.
Tubulin
Birefringence is characteristic of microtubule-based structures
and myofilaments (Cassimeris et al., 1988). To test whether the second
harmonic could also be generated within complexes that are not composed
of coiled-coil structures, we imaged tubulin structures in C. elegans embryos by SHIM. To examine the specific contexts giving
rise to SHG from microtubules, we imaged by simultaneous SHG/TPEF a
strain of C. elegans expressing a GFP-tagged tubulin (-tubulin::GFP). In Fig. 7,
A, B, and C, the SHG, TPEF, and
overlay, respectively, show a gravid adult hermaphrodite bearing eight embryos within its uterus. SHG signal was detected from tubulin arrays
at all cell-cycle stages, including pronuclear rotation in the zygote
(far left), mitosis in the first cell division (far right), interphase in the two-cell stage (second from
left), and various later phases of early embryonic cleavage.
However, at high magnification, it was clear that SHG arises from only
a subset of the microtubules seen by TPEF. In Fig.
8, A, B, and
C, the SHG, TPEF, and overlay, respectively, show an embryo
during the first mitosis. Fluorescent microtubules compose the dense
spindle array and sparse astral arrays surrounding both spindle poles (Fig. 8, B and C). Yet SHG within the spindle
(Fig. 8, A and C) is interrupted by a discrete
dark space at the spindle midzone, and the spindle poles appear
distinctly more hollow by SHG than by TPEF. In Fig.
9, A-C, two interphase
centrosomes in early embryonic cells are shown. Although, centrosomes
are uniformly labeled with fluorescent tubulin (Fig. 9 B),
they yield hollow double-crescent SHG profiles (Fig. 9, A
and C). In contrast, SHG-bright autofluorescent granules in
the intestine of the mother are uniform (Fig. 9, A and
C, arrows). It should be noted that wild-type
worms produced comparable SHG images of asters and spindles (not
shown), proving that SHG only arises from the endogenous microtubules
and does not arise from the GFP. Two important dependencies of SHG upon orientation are manifested in the signal from tubulin. First, when
using a linearly polarized laser source, second harmonic waves will
only be produced by the subset of molecules whose dipoles are aligned
with the laser polarization. Because microtubules are radially arrayed
around the organizing center, the intensity of the SHG signal varies
around the centrosome with the coincidence of the laser polarization
and the orientation of the microtubules, giving rise to the observed
crescents (Fig. 9, A and C). Second, because SHG
is a coherent process, oppositely oriented second harmonic waves
produced by oppositely oriented protein structures destructively
interfere within a distance less than the optical coherence length
(~
= 860 nm). In the spindle midzone, the microtubules interdigitate counter-parallel to each other (Haimo, 1985; Euteneuer and McIntosh, 1980). Therefore, these regions may appear dark by SHG
(Fig. 8, A and C) because of symmetry
cancellation even though they contain abundant microtubules, as
evidenced in the TPEF images (Fig. 8, B and C).
View larger version (75K):
[in this window]
[in a new window]
|
FIGURE 7
SHG from tubulin structures in C.
elegans embryos. (A) SHG and (B)
TPEF, and (C) color-overlay optical section taken
through early embryos within the uterus of live C.
elegans allow comparison between signal profiles for the two
imaging modes. Eight embryos are shown within the uterus of an adult
worm at low magnification. Stages seen include pronuclear rotation in
the zygote (far left), mitosis in the first cell
division (far right), interphase in the two-cell stage
(second from left), and various stages of early
embryonic cleavage. Scale bar = 50 µm.
|
|
View larger version (29K):
[in this window]
[in a new window]
|
FIGURE 8
(A) SHG, (B) TPEF, and
(C) color overlay of an embryo shown at higher
magnification during the first mitosis. In (B),
fluorescent microtubules compose the dense spindle array and sparse
astral arrays surrounding both spindle poles. In (A),
however, SHG arises from only a subset of the microtubules seen in
(B); bright SHG in the spindle is interrupted by a
discrete dark space at the spindle midzone; only proximal portions of a
few astral microtubule bundles are seen; and the spindle poles appear
distinctly more hollow by SHG than by TPEF. Scale bar = 10 µm.
|
|
View larger version (24K):
[in this window]
[in a new window]
|
FIGURE 9
(A) SHG, (B) TPEF, and
(C) color overlay of two interphase centrosomes in early
embryonic cells are shown at high magnification
(arrowheads in A and B).
In (B), centrosomes are uniformly labeled with
fluorescent tubulin. In (A), centrosomes yield hollow
SHG profiles, with an angular dependence on the coronal brightness. In
contrast, SHG-bright granules in the intestine of the mother are
uniform (arrows). Scale bar = 10 µm.
|
|
Thus, important information about the structural context of
chromophoric molecules is encoded in their SHG signal, and is best
comprehended in comparison to a structure-independent imaging mode,
such as TPEF from a flexibly tethered GFP domain. Because both
nonlinear effects occur at precisely the same point of laser focus, the
separate data channels collected are absolutely contemporaneous and
colocalized. Figure 10 shows intensity
plots along the axis of the spindle shown in Fig. 8. Traces are drawn
for the separate SHG and TPEF channels and for the TPEF/SHG ratio. In
both representations, the ratio signal is clearly maximized (i.e.,
strong fluorescence with weak SHG) in zones of high microtubule overlap
or symmetry. Therefore, the combination of these modalities is well
suited to a ratio-imaging scheme revealing structural detail that is invisible by either mode alone and otherwise apparent only via electron
microscopy. This ability to infer molecular context by correlation of
multimode imaging profiles has also been recently demonstrated for
simultaneous SHG/TPEF imaging of contrast-generating dyes within giant
unilamellar vesicles (Moreaux, et al., 2000, 2001).
View larger version (44K):
[in this window]
[in a new window]
|
FIGURE 10
Correlation between SHG and TPEF tubulin signals by
ratio imaging. (A) Self-normalized plots of SHG
(violet), TPEF (green), and TPEF/SHG
(black) along the length of the spindle shown in Fig. 8.
(B) A TPEF/SHG ratio image of the data from Fig. 8.
Pseudocolor intensity scale for (B) is shown to the
right. Scale bar = 10 µm.
|
|
Optical experiments confirming and characterizing
SHG emission
Physical description
In general, the total polarization for a material can be expressed
as
|
(2)
|
where P is the induced polarization,
(n) is the nth-order nonlinear
susceptibility tensor, and E is the electric field vector. The first term describes normal absorption and reflection of light; the
second term, SHG, sum and difference frequency generation; and the
third term, two-photon absorption, third-harmonic generation, and
stimulated Raman processes. SHG is a second-order nonlinear optical
process that can only arise from media lacking a center of symmetry.
This criterion can be satisfied at interfaces and by anisotropic
crystals. The second-harmonic intensities per laser pulse are expected
to scale as (Shen, 1989)
|
(3)
|
where p is the pulse energy, a is the area
of the focused spot, 2 is the second-order
nonlinear susceptibility of the protein, and 
is the laser pulse
width. The inverse dependence on the pulse width arises because,
although the signal scales as the square of the peak power, second
harmonic will only be produced within the duration of the laser
fundamental pulse. Note that SHG and two-photon excitation
probabilities have the same inverse pulse width dependence. Similarly,
because the signal is proportional to the square of the intensity
(photons/area), second harmonic will only be generated within the focal
area. The squared dependence of 2 on the
signal magnitude is manifested in a squared dependence of the molecular concentration.
Several control measurements were performed to verify that the SHG
images only consisted of the second harmonic and no residual laser
light or autofluoresence. First, the laser was taken out of mode
locking and the signals vanished, indicating that the signals arose
from a nonlinear process, which was also verified by a quadratic
dependence on the laser power (data not shown). To show that the images
were free of autofluorescence, SHG spectra of all the specimens were
collected using a fiber-based spectrometer. The data for 850-nm
excitation for five specimens are shown in Fig.
11 and clearly shows that the signal
arose exclusively at the expected 425-nm wavelength, without any
autofluorescence. Further, all the spectra have the same bandwidth
(~10 nm), and this width is consistent with the expectation for
100-fs pulses: a Fourier transform-limited 100-fs pulse will have a
full-width half-maximum (FWHM) bandwidth of ~10 nm, and, for a
Gaussian beam, the second harmonic will have a FWHM of 1/
2 of the
fundamental or ~7 nm. Within experimental limitations (spectrometer
resolution and nontransform limited pulses), the observed value is in
good agreement with that predicted. Next, the laser was scanned between 820 and 880 nm to ensure that the SHG spectrum tracked with the excitation wavelength, and the resulting spectra are shown in Fig.
12. Note that the corresponding SHG
wavelengths are the only wavelengths efficiently passed by the 425-nm
hard reflector in the beam path after the condenser. The spectra do
indeed follow the excitation with the proper bandwidth, and are again
free of any autofluorescent contamination. Finally, the transmitted
light was dispersed through a 0.25-m monochromator and then imaged. Although this scheme is very inefficient as the laser is scanning, images were obtained from both a fishscale and adult C. elegans. If uncontaminated by autofluorescence, the SHG signal
should only be present at half the laser fundamental wavelength, and
then tune when the laser fundamental is scanned. Measurements were taken at fundamental wavelengths of 800, 845, and 885 nm, and, indeed,
the SHG did tune accordingly. Additionally, a bound was placed on the
extent of autofluorescence occurring within this spectral window. With
the laser fundamental at 840 nm, the SHG image was obtained at 420 nm.
Then the laser was tuned to 860 nm, and the corresponding image was
obtained at 420 nm. Following background subtraction, the ratio of the
intensities of these two images places an upper bound on
autofluorescence in the fish scale, where a value of 1:500 was
obtained. Finally, the SHG image appeared bleach resistant, and data
will be presented in a latter section to provide quantitative evidence
of this observation.
View larger version (22K):
[in this window]
[in a new window]
|
FIGURE 11
SHG spectra for the species used in this work. The
laser excitation wavelength was 850 nm. The spectra are free of any
contaminating autofluorescence.
|
|
View larger version (24K):
[in this window]
[in a new window]
|
FIGURE 12
SHG spectra of pharyngeal muscle of C.
elegans tracks the excitation wavelength over the range of
820-880 nm. The spectra are all normalized to the same maximum value.
No contaminating autofluorescence is observed in this spectral range.
|
|
Polarization anisotropy
An additional aspect of SHIM is the capability of probing the
orientation of structural proteins in tissues. The vector nature can be
exploited to provide data on the spatial organization, i.e., long-range
order, of the helices from the size range of 500 nm and larger by using
SHG polarization anisotropy measurements. This is analogous to the more
familiar fluorescence anisotropy to probe organized structures, however
the SHG approach is somewhat more specific and does not require
exogenous labels. For processes arising from an electric dipole process
(e.g., absorption, and SHG), the angular distribution of excitation is
given by the second Legendre polynomial (a cos2
function). The anisotropy parameter, , describing the molecular orientation, is derived from this function, and, when using linearly polarized light, is given by
|
(4)
|
where Ipar and
Iperp are the intensities of the
signals whose polarizations are parallel and perpendicular to the
polarization of the incident laser. This parameter can vary between
0.5, and 1. The special case of 0 represents the isotropic situation
where Ipar and
Iperp are equal and would physically
correspond to having complete randomization or disruption of the
helices. The value of 1 would correspond to complete ordering relative
to the incident laser, i.e., having well-aligned, well structured
helices. Images acquired from collagen in a fish scale with the output
polarization oriented parallel and perpendicular to the excitation
laser are shown in Fig. 13,
A and B, respectively. Integrating these data yields a value of 0.7. The actual value is probably somewhat closer
to unity because the use of medium or high numerical objective lenses
results in a slight loss of polarization (Axelrod, 1979). This result
indicates the highly ordered nature of the collagen fibrils. Note that
the use of the appropriate combinations of input laser polarization and
output polarization analysis will also yield the absolute molecular
orientations by determining the value of all the contributing matrix
elements of the second-order nonlinear susceptibility tensor (Shen,
1989).
View larger version (87K):
[in this window]
[in a new window]
|
FIGURE 13
SHG polarization anisotropy measurements of an
isolated scale of black tetra fish, G. ternetzi, from
the Glan Laser polarizer oriented (A) parallel and
(B) perpendicular to the laser fundamental.
|
|
Molecular symmetry can also be observed by rotating the input laser
polarization and collecting the resulting SHG signal. SHG images of
interphase centrosomes from early C. elegans embryos are
shown in Fig. 14,
where the data in A, B, and C were
collected from the same sets of centrosomes with varying excitation
polarization obtained by rotation of a wave plate. The rotation of the
crescents follows all the way through 90°, showing that the
interaction of the laser with the radial symmetry of the microtubule
array strongly influences the image contrast. Note that it would be predicted that the use of circularly polarized light would equally excite all orientations and result in continuous circles of constant intensity. In our current geometry, we were unable to achieve this
condition because substantial ellipticity is introduced in the
galvo scanning system due to the non-normal angles of incidence and the
metal mirror coatings. The use of a full wave plate following the
scanner would eliminate this residual ellipticity.
View larger version (32K):
[in this window]
[in a new window]
|
FIGURE 14
SHG images of C. elegans early embryo
centrosomes (similar to Fig. 10), where the polarization was
effectively rotated with a quarter-wave plate, where (A)
and (C) were the result of 45° of plate rotation
(i.e., 90° optical rotation), resulting in orthogonal projections of
the array; and (B) was no rotation.
|
|
| |
DISCUSSION |
Physical origin of SHG in tissues
SHG relies on the lack of inversion symmetry, and this condition
can be satisfied in several ways, including the molecular alignment in
a crystal or the intrinsic asymmetry at an interface. In prior work, we
exploited the fact that the plasma membrane in a cell provides a
biological liquid-liquid interface, and that high resolution images of
living cells could be obtained when this membrane was stained with a
contrast- increasing dye (Ben-Oren et al., 1996; Peleg et al., 1999;
Campagnola et al., 1999). The work presented here is fundamentally
different in that the contrast does not arise from cell membranes but
rather from endogenous structural proteins present in tissue. From
polarization microscopy, muscle and connective tissues are known to be
highly birefringent. However, an analogy cannot be directly drawn to
frequency doubling in a birefringent uni-axial crystal, such as KDP or
BBO. In such crystals, the laser fundamental and resulting SHG waves
have orthogonal polarizations, because this is the only nonvanishing
matrix element of (2) (Yariv, 1989). However,
we measured the polarization anisotropy of the SHG signal both parallel
and perpendicular to the laser fundamental in both fish scales and
C. elegans muscle and found the SHG signals in both cases
were strongly polarized parallel to the fundamental (see Fig. 13).
Furthermore, no Type 1 phase-matching condition exists for these
proteins at these wavelengths (Bolin et al., 1989). However,
phase-matching requirements are not rigorously applicable for the case
of strongly focused Gaussian beams. In their 1986 work, Freund et al.
showed that the collagen from rat tail tendon had dipolar structures
that gave rise to the SHG signals. Our data from the tubulin structures
(Fig. 7-10, 14) are consistent with the SHG arising from an electric
dipole interaction. However, these structural proteins are all
intrinsically chiral, and chirality is known to enhance SHG. This has
been shown recently by Hicks and coworkers (Beyers et al., 1994) for
chiral molecules at an interface and in our prior work comparing the
SHG efficiencies of chiral and nonchiral dyes (Campagnola et al.,
1999). This concept was also demonstrated by Verbiest et al. (1998) in
studying SHG in synthetic helicenes. They showed that, although racemic
mixtures of the chiral helicenes did produce second harmonic signals,
the pure enantiomers were upwards of 50-fold more efficient. Note that,
although chirality alone can give rise to SHG, this process is
inefficient and would lead to isotropic angular distributions. This is
in strong contrast to the strongly anisotropic data on fish scale
collagen and C. elegans centrosome asters shown in Figs. 13
and 14, respectively. Thus, the SHG can be ascribed to a dipolar
process that is chirality enhanced, where the signal arises from a
volume rather than surface effect. A direct analogy can be drawn from
the results in helicenes (Verbiest et al., 1998) to the work here on
endogenous proteins. The brightest signals were from the fish scale and
mouse ECM, where, in both cases, the signal arose from a collagen
matrix and were considerably brighter than either the mouse or C. elegans sarcomeres. The coiled-coils in collagen, for example,
will display supra-molecular chirality, i.e., although each of
individual helices in the triple-helical collagen structure is expected
to produce second-harmonic signals, the three helices will
cooperatively produce even larger signals. By contrast, myosin has two
coiled-coils and would be expected to be less SHG efficient. This is
consistent with work by Kim et al. (1999, 2000), where, in bulk
measurements (nonimaging), they compared relative efficiencies of
several forms of connective tissue and muscle tissue and found ratios
in the range of 50/1.
The signal strengths observed from the structural proteins in this work
are much greater, ~10-50 fold, than were obtained in our prior work
on imaging cellular membranes stained with a voltage-sensitive dye.
This is an interesting result in that it is expected that the molecular
hyperpolarizability for a highly polarizable dye will be much greater
than that of a protein. This is because highly conjugated pi systems,
in general, lead to greater second-order properties (Tykwinski et al.,
1998), and proteins lack such extended pi networks. For example,
collagen typically has a sparse density of aromatic residues and the
polarizability of single bonds is significantly less than that of
double bonds. Even more so, our prior SHG imaging work (Campagnola et
al., 1999) was done on resonance to provide sufficient contrast where
the resonant component was at least an order of magnitude larger than the pure surface term. By contrast, the apparent lack of spectral dependence on the signal levels from these structural proteins indicates that there is little resonance enhancement in any of these
images. However, myosin and collagen probably occur in millimolar abundance, whereas cellular membranes can only be stained in the 1-10
µM level to avoid aggregation and toxicity. Thus the quadratic dependence of the second-harmonic signals on concentration leads to
unexpectantly intense images from structural proteins.
Symmetry effects on SHG
We showed the effects of both radial and lateral symmetry on the
SHG images of tubulin structures in C. elegans. It is
interesting to compare this to the fluorescent image arising from
-tubulin::GFP. In Fig. 8, SHG is excluded from the center
of the spindle midzone, where GFP-labeled microtubules are abundant but
an anti-parallel arrangement of microtubules exists, in contrast to
more harmonogenic distal portions of the spindle. The centrosomal
region of both spindle poles and asters is also devoid of SHG signal,
even when microtubule density appears highest there by fluorescence
(Fig. 9). The centrosome is characterized by a 3D radially symmetric array of microtubules in close proximity. Cylindrical symmetry cancellation is not observed in the TPEF channel because the GFP domain
is flexibly tethered to the tubulin domain. Thus, because of the lack
of rigidity, it is expected to be randomly oriented and generate
fluorescence at all polarization angles. Similarly, the interdigitated
counter-parallel GFP-labeled microtubules produce strong fluorescence
in the midzone. Because fluorescence is an incoherent process, it is
not subject to the distance constraints that SHG experiences within the
optical coherence length.
Comparison with other forms of microscopy
Second-harmonic imaging of tissues has distinct advantages over
imaging specimens stained with fluorescent dyes. The most obvious
advantage lies in the lack of any staining preparation. Extensive
processingfixation, staining, sectioningis typically required to
image the fine 3D structure of complex biological specimens. Similarly,
there are no cytotoxic or phototoxic effects from the addition of
exogenous labels. Bleaching of fluorescent dyes results in the creation
of toxic free radicals, and, although multiphoton excitation of
fluorescence greatly reduces this problem away from the focal plane,
in-plane excitation will still result in phototoxicity, and limit
studying dynamics over long time courses or limit repeat acquisition to
increase signal-to-noise and image quality. It should be noted that, at
the wavelengths used (~850 nm), there is no apparent resonant
component of the SHG arising from these structural proteins by either
two- or three-photon absorption, as evidenced by the lack of clear
spectral dependence. To further verify this lack of photobleaching, we
irradiated C. elegans adults expressing
GFP::MHC A construct at high zoom with high laser power over
several scans. The data are shown in Fig. 15, where low zoom images were obtained
before and after irradiation at high zoom. The darkened rectangle in
the TPEF channel (B) arises from photo-bleaching the GFP,
whereas the SHG image (A) is unaffected, confirming the
bleach-free nature of SHG signal.
View larger version (74K):
[in this window]
[in a new window]
|
FIGURE 15
SHG signal is not subject to photobleaching,
demonstrated using the same C. elegans strain shown in
Fig. 5. Top and bottom panels are SHG and TPE fluorescence of GFP,
respectively, before (left) and after
(right) high-intensity, high-zoom scanning of a central
portion of the field. The SHG is unaffected whereas the GFP
fluorescence is strongly bleached after exposure at high zoom. Scale
bar = 50 µm.
|
|
The use of both GFP and highly selective chemical fluorescent probes
has added new dimensions to the capabilities of imaging live cells.
However, genetically engineered expression or chemical administration
of fluorescent labels can be a major limiting step in studies of
nonmodel organisms. Because SHIM takes advantage of signals
generated by endogenous structures, it should be universally applicable
to all animal species, and is perhaps ideally suited for development of
medical imaging strategies. We have shown here that a wide range of
proteins in several tissue types is readily visualized by SHIM without
the effort of exogenous labeling. Furthermore, because SHG
arises within the endogenous proteins of interest, it encodes
information on molecular orientation and assembly that cannot be
elicited from the fluorescence of GFP-labeled variants (see Figs. 9 and
10). GFP is typically tethered to the end of a domain, with a
pronounced lack of rigidity, and hence cannot provide information on
molecular-level orientation and structure. This is because the GFP will
be freely rotating and the fluorescence polarization anisotropy will
become isotropic. Similarly, information regarding both radial and
lateral symmetry will be lost.
Because it has been known for some time from polarization microscopy
that structural proteins organize to form highly birefringent structures (Inoue, 1986), it is useful to compare the data obtained by
SHIM with that from polarization microscopy. As shown in the Results
section (Fig. 4), while the images from these two modalities may appear
similar, there are significant differences in the method of contrast
generation. Polarization images probe linear birefringence and can only
achieve contrast by having orthogonal polarizers for excitation and
collection. Thus, it is difficult to obtain absolute orientation of
molecules. By contrast, SHG signals can be acquired for arbitrary
output polarizations relative to the excitation laser, and both
relative and absolute molecular orientations can be obtained through
the proper choice of input and output polarizations (Shen, 1989).
Additionally, it is possible to estimate the width of a distribution of
orientations through this analysis. Finally, because of the nonlinear
nature, SHIM provides intrinsic optical sectioning, whereas this is
difficult by polarization microscopy. This feature is especially
important in imaging whole tissue samples and extending this scheme to
in vivo specimens.
Limits on data acquisition
The data-acquisition rate here was typically 1 frame per second,
each frame consisting of 768 × 512 pixels. This was generally limited by the scanner speed rather than image intensity. With a
different experimental arrangement, line scans could be acquired to
obtain images on physiological time scales. Alternatively, much faster
frame rates could be realized with a resonant galvo scanner system. The
depth of imaging through thick tissue is limited by the Rayleigh
scattering of both the fundamental and SHG waves. We have readily
achieved depths of >500 µm through muscle tissue.
| |
CONCLUSIONS |
Our results demonstrate that SHIM, alone and in combination with
TPEF, offers novel opportunities for analyses of endogenous protein
polymer structure within living, 3D specimens. The simultaneous imaging
of SHG and two-photon excited GFP enhances our understanding of the
protein species and structural contexts giving rise to the
second-harmonic signal. Polarization analyses can yield data regarding
the molecular organization and symmetry of these matrices. We expect
that these methods will have significant impact on in vivo studies in
various fields of biology, including tissue organization, wound-healing, myofilament assembly, muscle development and disease, and the division cycle of normal and cancerous cells in situ. Already,
we have begun to extend these methods to the analysis of fibrillar
species in connective tissue and studies of skin and muscle pathology.
We gratefully acknowledge financial support under the National
Science Foundation Academic Research Infrastructure DBI-9601609, the
State of Connecticut Critical Technology program, and National Institutes of Health R01- GM35063 (to A.C.M.). This work was also supported by a New Scholar in Aging grant from the Ellison Medical Foundation and a Research grant from the Muscular Dystrophy
Association to W.A.M., National Institute of General Medical Sciences
grant R01-GM60389 to M.T.
We thank Prof. Vladimir Rodionov for providing the fish scales, Prof.
David Rowe and Ivo Kalajzic for mouse tissues, and Prof. Andrew Z. Fire
for gifts of DNA. We also thank Prof. Leslie Loew, Prof. John White,
and Prof. Ann Cowan for helpful technical discussions.
Address reprint requests to Paul J. Campagnola, University of
Connecticut Health Center, Center for Biomedical Imaging Technology
MC-1507, Farmington, CT 06030. Tel.: 860-679-4354; Fax: 860-679-1039;
E-mail: campagno{at}neuron.uchc.edu.
TOP
ABSTRACT
INTRODUCTION
-tubulin in early C. elegans embryos.
Mol. Biol. Cell.
12:1751-1764
